d4 - is-wireless · alcatel lucent deutschland ag alud germany commissariat a l energie atomique et...

The 5GNOW Project Consortium groups the following organizations:

Partner Name Short name Country FRAUNHOFER-GESELLSCHAFT ZUR FOERDERUNG DER ANGEWANDTEN FORSCHUNG E.V. HHI Germany ALCATEL LUCENT DEUTSCHLAND AG ALUD Germany COMMISSARIAT A L ENERGIE ATOMIQUE ET AUX ENERGIES ALTERNATIVES CEA France IS-WIRELESS ISW Poland NATIONAL INSTRUMENTS NI Hungary TECHNISCHE UNIVERSITÄT DRESDEN TUD Germany

Abstract: The screening process of T3.1 will generate a candidate list of waveforms and signal formats, described in this IR3.1.

Final MAC/Networking Concepts

D4.2

‘5GNOW_D4.2_final’ Version: 1.0

Last Update: 9/3/2015

Distribution Level: CO

• Distribution level PU = Public, RE = Restricted to a group of the specified Consortium, PP = Restricted to other program participants (including Commission Services), CO= Confidential, only for members of the 5GNOW Consortium (including the Commission Services)

D4.2

5GNOW FP7 – ICT– GA 318555

Page: 2 of 72

The 5GNOW Project Consortium groups the following organizations:

Partner Name Short name Country FRAUNHOFER-GESELLSCHAFT ZUR FOERDERUNG DER ANGEWANDTEN FORSCHUNG E.V. HHI Germany ALCATEL LUCENT DEUTSCHLAND AG ALUD Germany COMMISSARIAT À L’ENERGIE ATOMIQUE ET AUX ENERGIES ALTERNATIVES CEA France IS-WIRELESS ISW Poland NATIONAL INSTRUMENTS NI Hungary TECHNISCHE UNIVERSITÄT DRESDEN TUD Germany

Abstract: This report contains the final MAC and networking interface for non-orthogonal PHY layer.

Partner Author name

HHI

Gerhard Wunder Martin Kasparick Peter Jung

ALUD Thorsten Wild Yejian Chen

CEA Nicolas Cassiau Dimitri Kténas

ISW

Mateusz Buczkowski Marcin Dryjański Slawomir Pietrzyk

TUD

D4.2


Page: 3 of 72

“The research leading to these results has received funding from the European Community's Seventh Framework Programme (FP7/2007-2013) under grant agreement n° 318555”

D4.2


Page: 4 of 72

Document Identity

Title: Final MAC Concept WP: WP4 – MAC layer and networking interface WP Leader ISW Main Editor Mateusz Buczkowski (ISW), Gerhard Wunder (HHI), Dimitri Ktenas (CEA) Number: D4.2 File name: 5GNOW_D4 2_final Last Update: Monday, March 09, 2015

Revision History

No. Version Edition Author(s) Date 1 0.1 Marcin Dryjanski (ISW) 10.07.14

Comments: Providing initial template and skeleton 2 0.2 Marcin Dryjanski (ISW) and Mateusz Buczkowski (ISW) 29.09.14

Comments: Providing outline and input 3 0.3 Mateusz Buczkowski (ISW) 25.11.14

Comments: General revision and updates 4 0.4 Mateusz Buczkowski (ISW), Gerhard Wunder (HHI) 15.12.14

Comments: Added System Level Simulations chapter 5 0.5 Mateusz Buczkowski (ISW), Gerhard Wunder (HHI) 12.01.15

Comments: Added appendix 6 1 Gerhard Wunder 09.03.15

Comments: Final review 7

Comments: 8

Comments: 9

Comments: 10

Comments: 11

Comments: 12

Comments: 13

Comments: 14

Comments: 15

Comments:

D4.2


Page: 5 of 72

Table of Contents

1 INTRODUCTION ................................................................................................................................................. 7 1.1 HETNET LTE-ADVANCED REL. 11 AND REL. 12 FEATURES ............................................................................................ 7

HetNet dual connectivity ............................................................................................................................. 7 1.1.1 CoMP in 3GPP Rel. 11 and Rel. 12 ............................................................................................................... 9 1.1.2

1.2 5GNOW CONCEPTS IN THE CONTEXT OF MULTI-CELL / HETNET SCENARIOS ................................................................... 11 5GNOW unified frame structure ............................................................................................................... 11 1.2.1 5GNOW robustness framework ................................................................................................................ 12 1.2.2 5GNOW “one shot” transmission .............................................................................................................. 14 1.2.3

2 UNIFIED MAC CONCEPT ....................................................................................................................................15 2.1 DESIGN SPACE .................................................................................................................................................... 15 2.2 5GNOW UNIFIED MAC DESIGN ........................................................................................................................... 16

Unified MAC building blocks and procedures............................................................................................ 17 2.2.1 Unified MAC modes (heterogeneous services and waveforms) ................................................................ 26 2.2.2

3 PHY / MAC LAYER INTERFACE ISSUES ...............................................................................................................27 3.1 UNIFIED MAC INTERFACE TO PHY LAYER ................................................................................................................ 27

Dynamic frame structure adaptation........................................................................................................ 28 3.1.1 Dynamic resource allocation ..................................................................................................................... 28 3.1.2 Contention parameters steering/semi-static resource allocation............................................................. 28 3.1.3 Dynamic feedback control......................................................................................................................... 28 3.1.4

3.2 ALGORITHMS AND TRANSMISSION PROCEDURES ........................................................................................................ 29 D-PRACH random access algorithm .......................................................................................................... 29 3.2.1 CoMP delay detection algorithm .............................................................................................................. 33 3.2.2 MU-MIMO/CoMP robustness framework ................................................................................................. 34 3.2.3

4 SYSTEM LEVEL SIMULATION .............................................................................................................................44 4.1 5GNOW WAVEFORM PHY ABSTRACTION MODELLING EXAMPLE ................................................................................. 44

5GNOW PHY abstraction modelling .......................................................................................................... 44 4.1.1 Performance evaluation of 5GNOW waveforms in system level environment (OFDM and FBMC 4.1.2

comparison)............................................................................................................................................................. 45 4.2 PERFORMANCE EVALUATION OF D-PRACH AND ATA IN SYSTEM LEVEL SIMULATION ....................................................... 50

Simulation description .............................................................................................................................. 50 4.2.1 Analytical calculations and simulation results .......................................................................................... 56 4.2.2

5 APPENDIX A – DETAILED DESCRIPTION OF DUAL CONNECTIVITY AND COORDINATED MULTIPOINT ................60 5.1 HETNET LTE-ADVANCED REL. 11 AND REL. 12 FEATURES .......................................................................................... 60

HetNet dual connectivity ........................................................................................................................... 60 5.1.1 CoMP in 3GPP Rel. 11 and Rel. 12 ............................................................................................................. 63 5.1.2

6 ABBREVIATIONS AND REFERENCES ...................................................................................................................68

D4.2


Page: 6 of 72

Executive Summary

The final report in WP4 describes Medium Access (MAC) and Networking/Heterogeneous Networking (HetNet) concepts for the 5GNOW project. It is divided into four parts: HetNet concepts for LTE and 5GNOW respectively, unified MAC architecture, PHY/MAC layer interfacing, and system level simulation results. The first part discusses 3GPP Release 12 networking features, including dual connectivity with lean carrier and coordinated multipoint (CoMP) with non-ideal backhaul. After the basic description of the functionalities is provided, we put ideas on how 5GNOW approaches complement and fit within 3GPP framework including the use of unified frame structure (UFS) and robustness framework. The second part provides a detailed description of the unified MAC for 5GNOW system. It begins from the MAC design space definition and then present the proposed overall unified MAC architecture, main building blocks, tasks and MAC modes. It includes the following MAC elements: general scheduler/coordinator; individual scheduler approaches; timing advance, random access and feedback management. The third part defines an interface between the unified MAC and the 5GNOW PHY layer where the unified frame structure serves as the merging point. The interface is presented in the form of MAC/PHY transmission procedures and algorithms, feedback, and waveform parameters. The chapter is finalized by a description of system level simulations, evaluating 5GNOW waveforms, and a comparison against OFDM/LTE. The last part of the document describes results of the system level simulations, where improvements for machine type communications (MTC) are shown.

D4.2


Page: 7 of 72

1 Introduction

This chapter introduces briefly two LTE-Advanced (LTE-A) features of 3GPP Releases 11-12, namely dual connectivity and CoMP and shows how 5GNOW concepts can be applied within their context. Extended description of dual connectivity and CoMP can be found in Appendix A.

1.1 HetNet LTE-Advanced Rel. 11 and Rel. 12 features

HetNet dual connectivity 1.1.1

According to [3GP13], dual connectivity refers to the operation “where a given UE is allocated radio resources provided by at least two different network points connected with non-ideal backhaul”. Each enhanced NodeB (eNB) involved in dual connectivity for user equipment (UE) may have different role. These roles are not necessarily depending on the given eNB transmit power and can vary among various UEs in the cell. This idea utilizes close communication between macro and pico nodes, where the UE is connected for basic signalling and data transmission to macro cell and sets high-speed data link to pico eNB (and/or for DL reception to macro cell and transmits in UL to pico cell as described further on in this chapter). Figure 1 below shows the concept of dual connectivity; it illustrates the terminal that is connected to the both [ADF+13]:

• Wide area layer through an anchor connection (to the macro-node) used for system information, basic radio resource control (RRC) signalling, and possibly low-rate/high-reliability user data.

• Local area layer through an assisting connection (to the low power node) used for large amounts of high-rate user data.

Figure 1: Soft cell / Asymmetric node assignment concept from 3GPP.

Dual connectivity to the anchor node and assisting node.

The Rel. 10-Rel. 11 enhanced inter-cell interference coordination (eICIC) mechanisms are replaced by dynamic creation of such “virtual” or “soft” cells, i.e. physical eNB for DL and physical eNB for UL create a logical eNB that is not seen by the UE. The small cells provide a fast data pipe with improved link budget due to the close proximity of the UE to the network node. The macro cell on the other hand is providing cell wide system information as well as radio resource control including traffic steering, carrier selection and supporting the UE to detect a small cells that are close by. The challenge of this solution is, that the architecture requires a tight integration of the small cells into the network by ultra-fast interfaces or preferable by a central baseband processing unit. Another perspective of this solution is presented in [RMC12] and raises the issue of Channel Quality Indicator (CQI) reporting, i.e. CQI is reported to the closest node and thus lower transmit power is required, thus energy is saved. However in such scenario changes are required to power control since DL and UL path loss will be different.

D4.2


Page: 8 of 72

The possible scenarios for dual connectivity are as follows [3GP13]:

• Scenario A: Both macro and pico eNBs are equipped with the same two carriers. The macro and pico eNBs apply Rel-10 CA to aggregate both carriers.

• Scenario B: Both macro and pico eNBs are equipped with one carrier which differs from one another. The macro and pico eNBs apply inter-node radio resource aggregation.

According to [ADF+13] dual connectivity concept may imply to the system the following:

• Control and data separation where, e.g., the control signalling for mobility is provided via the macro layer and at the same time low-power layer provides high-speed data connectivity.

• A separation between DL and UL, where DL and UL connectivity is provided via different layers. • Diversity for control signalling, where RRC signalling may be provided over multiple links, which

may further enhance mobility performance. Dual connectivity concept may provide the following benefits (over the macro assistance) [ADF+13]:

• Enhanced support for mobility – by maintaining the mobility anchor point in the macro layer, it is possible to maintain seamless mobility between macro and low-power layers, as well as between low-power nodes.

• Low overhead transmissions from the low-power layer – by transmitting only information required for individual user experience, it is possible to e.g. avoid overhead coming from supporting idle-mode mobility within the local-area layer.

• Energy-efficient load balancing – by turning off the low-power nodes when there is no ongoing data transmission, it is possible to reduce the energy consumption of the low-power layer.

• Per-link optimization – by being able to select the termination point for UL and DL separately, the node selection can be optimized for each link.

3GPP proposes for Release 12 a so called “lean carrier”, which is a DL frame structure not compatible backwards with Rel. 8 [NNB+13] [HLK+13]. Lean carrier does not include CRS, cell specific synchronization signals and broadcast channel – they are controlled by macro-cell signaling (i.e. visibility of the legacy type of info is provided by macro cell). [NNB+13] [HLK+13] present concept of lean carrier within dual connectivity architecture as shown in the Figure 2 below.

Figure 2: Lean Carrier concept within dual connectivity architecture

Use case scenario is the following: UE is synchronized to both, macro eNB and to low-power node at lean carrier, i.e. UE uses legacy carrier at macro cell for signaling and lean carrier at pico/home eNB for high speed data. This concept may utilize carrier aggregation (i.e. legacy component carrier at lower frequency band and supporting lean component carrier at higher frequency band strictly used for IP data).

D4.2


Page: 9 of 72

CoMP in 3GPP Rel. 11 and Rel. 12 1.1.2

The latest 3GPP documents of Rel. 11 and Rel. 12 discussing CoMP issues include [3GP14] [3GP13a] [3GP13b]. As for now, both UL and DL CoMP have been considered by 3GPP. Due to their complexity and different requirements, these two directions are separated in the standardization process under two independent work sets. The following DL CoMP processing schemes are considered currently in 3GPP for PDSCH [3GP13b] (also shown in a simplified way in the Figure 3):

• Joint Processing (JP). In that case data for a UE is available at more than one point in the CoMP cooperating set (Set of geographically separated points directly and/or indirectly participating in data transmission to a UE in a time-frequency resource) for a time-frequency (TF) resource.

• Coordinated scheduling / beamforming (CS/CB). In that case, data for an UE is only available at and transmitted only from one point in the CoMP cooperating set for a TF resource, but UE scheduling/beamforming decisions are made with coordination among CoMP cooperating set points. The transmitting points are chosen in a semi static manner.

• Muting. Muting may be applied in dynamic and semi-static manner with the transmission schemes described above.

• Hybrid category of JP and CS/CB (possible). In that case, data for a UE may be available only in a subset of points in the CoMP cooperating set for a TF resource, but UE scheduling/beamforming decisions are made with coordination among CoMP cooperating set points. E.g. some points in the cooperating set may transmit data to the target UE according to JP scheme, while other points in the same time m may perform CS/CB.

Figure 3: DL CoMP schemes

The following UL CoMP processing schemes are considered currently in 3GPP for PUSCH [3GP13b] (also shown on the Figure 4):

• Joint reception (JR). In that case, PUSCH transmitted by UE is received jointly at multiple points (part or entire CoMP cooperating set) at a time (e.g. to improve the Rx signal quality).

• Coordinated scheduling/coordinated beamforming (CS/CB). In that case UE scheduling and precoding selection decisions are made with coordination among CoMP cooperating set points while data is intended for one point only at a time.

D4.2


Page: 10 of 72

Figure 4: UL CoMP schemes

The main differences between DL and UL CoMP in terms of standardization are the following [ADF+13]:

• UL CoMP is mostly implementation-specific with little need for additional work from specification perspective. However some smaller enhancements improving the UL orthogonality are included in Rel. 11.

• DL CoMP requires large specification effort, mainly due to CSI feedback issues. Therefore multipoint CSI feedback framework was introduced in 3GPP Rel. 11. Additionally due to many open issues seen around CoMP Rel. 12 will continue with multipoint CSI feedback and enhancements to DL control channels.

According to [3GP14], 3GPP triggers the following study in Release 12 to consider CoMP with the relaxed requirements for backhaul (non-ideal backhaul) to broaden the practical applicability of CoMP. CoMP specifications in Rel. 11 did not considered the specified support of a network interface for CoMP involving multiple separated eNBs with non-ideal backhaul. Due to this limitation, the operators having non-ideal backhaul in their networks may not be able to take advantage of performance benefit from inter-eNB CoMP operation [3GP14]. The study on CoMP for LTE with Non-Ideal Backhaul identified the specific cases in which CoMP can provide performance enhancements. Due to that, network interface enhancement and signalling messages shall be specified to allow implementation of both centralized and distributed CoMP focusing on HetNets (including macro/pico scenario). This issue is discussed in [3GP14], where the following objectives were raised for this study item with respect to evaluation:

• Evaluate CS/CB including semi-static point selection/muting as candidate CoMP scheme for multiple eNB with non-ideal but typical backhaul. If there is performance benefit from such evaluation – it is requested to recommend for which CoMP technique(s) the signaling should be specified for inter-eNB operation;

• Consider level of backhaul delay achievable with non-ideal backhaul; • Evaluation should be on the CoMP operation between macro eNBs, between macro eNB and small

cell eNB and between small cell eNBs with non-ideal backhaul. • The evaluations shall take into account estimation errors, DL overhead, complexity, feedback

overhead, backward compatibility and practical UE implementations. • Rel. 11 MMSE-IRC shall be used as a baseline receiver for evaluating performance gain.

According to discussions and performance evaluation present in [3GP13b], the following conclusions were made:

• CoMP offers performance benefits in both homogeneous networks (in scenarios 1 and 2) and heterogeneous networks (in scenarios 3 and 4);

• Performance of CoMP schemes that rely on spatial information exchange is sensitive to delay between two transmission points (while level of sensitivity depends on particular CoMP scheme).

D4.2


Page: 11 of 72

Due to the above observations 3GPP recommends that further work on CoMP shall focus on: Joint Transmission, DPS including dynamic point blanking and coordinated scheduling/beamforming including dynamic point blanking.

1.2 5GNOW concepts in the context of multi-cell / HetNet scenarios

The two features of 3GPP Release 12 including Dual Connectivity with Lean Carrier and CoMP, presented in Section 1.1 of this work, serve as framework for Multi-Cell / HetNet scenario in which 5GNOW concepts may be introduced. These two features could be enhanced / complemented by such features as unified framework structure and robustness framework presented in [5GNOWD4.1]. Merging of mentioned 3GPP and 5GNOW ideas is presented below under the assumption to set the HetNet concepts for 5GNOW project.

5GNOW unified frame structure 1.2.1

5GNOW focuses on the approach, where various types of non-orthogonal waveforms or waveform parameters’ sets may be evaluated within a single system. This enables to extend the legacy LTE system requirements with concept of bypassing synchronicity requirements under unified frame structure (UFS). 5GNOW vision of the UFS concept, aims at collecting various requirements / characteristics of traffic types within a single framework of frame. The frame then results in differentiation of waveform parameters among traffic classes. This will enable to handle the large set of approaches / requirements within a single 5G system. A (filtered) multicarrier approach will enable the mix of synchronous / asynchronous and orthogonal / non-orthogonal traffic types, where the reduced side-lobe levels of the waveform minimize inter-carrier interference (ICI) (see details in [5GNOWD4.1], [5GNOWD3.1]. UFS is a general idea of 5GNOW interface between PHY and MAC layer, while fitting into this HetNet scenario with DL/UL split. Typically it is assumed that the closed-loop synchronous traffic will cover a contiguous portion of the band in order to avoid or minimize the need for guard subcarriers. Adjacent portions of the spectrum can be filled by the non-perfectly synchronized traffic types. Certain low-end devices may even operate fully asynchronous. The UFS, according to Figure 5, can use multiple superimposed signal layers, e.g. separated by IDMA or CDMA and processed with multi-user detection receivers.

Figure 5: Unified frame structure

As 3GPP opens up for new carrier concepts starting from introduction of non-backward compatible Rel. 12 lean carrier for DL (see Section 1.1.1 above), 5GNOW ideas fit in the next releases of 3GPP standards e.g.

D4.2


Page: 12 of 72

for providing vision for enhanced UL. This can also be extended to the dual connectivity concept as presented in 1.1.1 above. A potential improvement includes the use of 5GNOW’s waveforms provided within a form of UFS, to be utilized in the HetNet scenario to serve as supporting cell/resources, while still using legacy LTE carrier for signalling (see Figure 6 below). The work takes into account the necessity to decrease backhaul signalling and outdated CSI that is in the scope of 5GNOW framework. UFS is designed especially for the UL direction, therefore may be treated as a complementary new carrier type for 3GPP standardization lean carrier (while lean carrier supports DL transmission direction) as presented in Figure 7 below.

Figure 6: Dual Connectivity concept from 3GPP adopted for 5GNOW

Figure 7: DL / UL split concept from 3GPP adopted for 5GNOW

5GNOW robustness framework 1.2.2

The original approach of 3GPP to CoMP mechanisms required huge additional overhead, e.g. backhaul message sharing, eNB synchronization, feedback of CSI and exchange of control information among CoMP cooperating nodes. Practical realizations of initial CoMP algorithms show very low gains far away from theoretical limits [5GNOWD4.1]. 5GNOW has developed Robustness Framework (see details in [5GNOWD4.1] and Section 3.2.3 of this document) that is designed to overcome current CoMP limitations (because it is about limited feedback). Thus it can be utilized within the current 3GPP considerations for non-ideally backhauled CoMP scenarios as indicated in Section 1.1.2 of this work. Robustness framework is dedicated for CoMP/MU-MIMO processing including delay sensitivity (or low sensitivity to backhaul delay) and limited feedback approach (see Figure 8).

Figure 8: Robustness Framework in CoMP scenario

D4.2


Page: 13 of 72

3GPP Rel. 12 considers both [3GP13a], macro homogeneous CoMP scenario and HetNet scenarios with limited backhaul and 5GNOW targets them with Robustness Framework to solve current (3GPP Rel. 12 and Rel. 13) standardization needs. 5GNOW may go beyond PHY processing algorithms and feedback approach and could propose the required parameters set as objectives for “CoMP with non-ideal backhaul” 3GPP Work Item as pointed and referenced in Section 1.1.2 of this work (e.g. CSI reports, power parameters, PHY parameters etc.).

1.2.2.1 Limited feedback

The major limiting factor in 4G CoMP (OFDM) systems is properly sharing channel state information (CSI) and other overhead among cells. This so-called limited feedback problem has been greatly analysed for multiuser MIMO [CJK+10], as well as for joint transmission [KG12] in terms of the rate distance Δ𝑟𝑚 of node 𝑚 to capacity subject to some offset independent of SNR. Hence, these results essentially provide a systems' degrees-of-freedom (DoF) analysis assuming infinite SNR regime. It is now highly relevant how the overall throughput scales with the feedback overhead for both practical systems in more relevant regimes. Classical analyses in literature reveals a scaling of the throughput degradation in the number of feedback bits in the order of 2−𝑏(𝑝)/(𝑁𝑇−1) where 𝑁𝑇 denotes the number of transmit antennas while 𝑝 and 𝑏 = 𝑏(𝑝) denote the SNR and feedback budget per user and resource block (in bits/channel use) as a function of SNR, respectively. Very recent 5GNOW studies [SWJ15][SWJ13] indicate that these results are fragile and that, in fact, the tradeoffs actually behave very differently in more practical regimes. Classical analysis of the scaling of per-node capacity in the number of feedback bits falls short due to several reasons: (i) it assumes an infinite SNR regime where achieving DoF is optimal. In this operational regime, interference mitigation instead of signal enhancement is the primary goal; (ii) it asserts that the transmitter can optimally allocate rates while, in practice, the transmitter allocates rates according to the available CSI and corresponding scheduler decisions (real versus ideal link adaption); (iii) the optimal scheduling decision is known a priori, which is unrealistic since limited feedback not only affects the choice of spatial precoding but also user selection and resource allocation. In [SWJ15][SWJ13] it is shown that for any finite SNR point 𝑝 and for any scaling in 𝑏 the per-node capacity degradation is actually in the order of 2−𝑏/�2(𝑁𝑇−1)� , which actually doubles the required number of bits. Moreover, it is well known that 4G-OFDM systems are sensitive to impairments such as Carrier Frequency Offset (CFO) and Sampling Frequency Offset (SFO). For joint transmission CoMP, synchronization conditions are substantially more challenging than for transmissions from a single base station, as the distributed base stations are driven by their local oscillators. This leads to individual CFOs and SFOs between the base stations as well as between each base station and each terminal. In contrast to OFDM systems, the FBMC waveform should be less sensitive to timing errors between the different cells, due to a better frequency localization of the prototype filter [MTR+11]. As part of a possible robustness framework, this feature was used to investigate the robustness of DL CoMP with respect to asynchronicities in a multi cell scenario including the FBMC waveform at the physical layer [CKD+13]. It was demonstrated that the UE can robustly estimate very large delays and arbitrary high carrier frequency offsets. The correction of the delays required only a few bits of feedback when the compensation of the offset can be realized at the UE side.

1.2.2.2 Channel aging

To counter the aforementioned effects a 5G robustness framework should provide a reliable feedback link and be able to cope with MAC related asynchronicities such as outdated CSI. In practical systems there is always a delay between the channel estimation at the mobile terminal and the time instant when this

D4.2


Page: 14 of 72

estimate is used for composing the following DL transmission. CSI aging is an important point to consider in the robustness framework. The CSI aging is the time between the measurement and the actual transmission based on this measurement. It includes the processing time at each node, the transmission delay on the backhaul, the scheduling delay at the cooperating base station, and is also a function of the CSI feedback periodicity. As the channel may change during this delay time, the channel observation may be outdated for precoding the next transmission. Thus, the accuracy of CSI feedback and the impact of the refresh rate of CQI feedback on the performance of precoding techniques will be taken into account in our attempt to provide an analysis of robust transmission in 5G under limited feedback. This will be further elaborated in Chapter 3. If perfect CSI is assumed, FBMC presents higher energy efficiency when compared to OFDM since it does not require a cyclic prefix. Nevertheless in practical systems there is always a delay between the channel estimation at the mobile terminal and the time instant when this estimate is used for composing the following DL transmission. Channel State Information aging is an important point to consider in the robustness framework. The CSI aging is the time between the measurement and the actual transmission based on this measurement. It includes the processing time at each node, the transmission delay on the backhaul, the scheduling delay at the cooperating base station, and is also a function of the CSI feedback periodicity. As the channel may change during this delay time, the channel observation may be outdated for precoding the next transmission. Thus, the accuracy of CSI (Channel State Information) feedback and the impact on performance of refresh rate of CQI feedback on the precoding techniques needs to be taken into account. Therefore a new multi-cell operation scheme needs to carefully designed to support both asynchronicity and a reduced level of control information. The multi-cell FBMC-based CoMP operation that is proposed in the project brings more robustness thanks to a careful design of the feedback link (e.g. limitation of CSI feedbacks, synchronization) and also incorporates other MAC related asynchronicities such as outdated CSI. This will be further described in chapter 3 of this report.

5GNOW “one shot” transmission 1.2.3

Sporadic traffic due to MTC will dramatically increase in the 5G market and, obviously, cannot be handled with the bulky 4G random access procedures. Two major challenges must be addressed: 1) unprecedented number of devices asynchronously access the network over a limited resource and 2) the same resource carries control signalling and payload. Dimensioning the channel according to classical theory results in a severe waste of resources which even worse does not scale towards the requirements of the IoT. On the other hand, since typically user activity, channel profiles and message sizes are compressible within a very large receive space the sparse signal processing methodology is a natural framework to tackle the sporadic traffic. One of the main ideas is a common control channel to enable “one shot” transmission as proposed in our “core” paper [WJK+14]. One shot transmission is described in more details in Section 3.2.1 of this document.

D4.2


Page: 15 of 72

2 Unified MAC concept

This chapter includes a detailed description of unified MAC concept proposed by 5GNOW, which follows the UFS concept set as a baseline for PHY/MAC interfacing. We start with the general overview relating to MAC design, and MAC design space. This is followed by presentation of the overall MAC architecture and building blocks. Then each important MAC building block is described, including scheduling, feedback steering, timing advance and random access algorithms. The chapter is finalized with the description of MAC layer procedures and proposed MAC modes for the targeted 5GNOW traffic scenarios.

2.1 Design space

While designing the MAC protocol for a single cell the following three key issues need to be addressed [SS10]:

• Efficiency - How can we schedule the transmission of different devices, in order to operate the network with the aims of maximization its efficiency?

• Stability - How can we stabilize the entire network to sustain the desired operating points to combat the randomness of the traffic load, channel conditions, etc.?

• Fairness - How should we design the protocols such that the devices can access the medium in a fair manner?

From higher perspective (network view) it’s desirable to design such protocol, so that by being distributed (each MAC entity works at different network node), it still can improve system performance (in terms of these three issues – efficiency, stability and fairness), while minimizing the required feedback and information exchange among different nodes [SS10]. Going to more specific issues of MAC protocol, one of the most important parts is the design of the scheduler. Typically, the gain from opportunistic scheduling comes at a cost, i.e. the receiving UEs must be told in what resources (time, frequency, code, etc. slots) the base station allocates their useful data. This information takes out the resources and is known as control signalling (e.g. in LTE system up to around 30% of the radio frame resources are consumed by control signalling, where the major portion of it is the signalling related to scheduling information). General observations from scheduling info design are as follows:

• the finer the granularity of the scheduling block is, the better the scheduling algorithm is tuned to each channel realization.

• the finer the granularity of the scheduling block is, the more control signalling and consumed resources that otherwise can be used by payload data.

According to [EML+09], there are three main parameters that affect the performance of the scheduling:

• Scheduling granularity – that determines how small part of the time / frequency grid (resource grid) can be allocated to a specific UE. (One extreme is where a single UE gets entire resource grid – i.e. minimum granularity. The other extreme would be, that each UE gets a set of individual resources elements – i.e. full granularity).

• Scheduling method – that determines how the base station decides, to which UE it should give each scheduling block (e.g. PF, RR, MaxCQI, M-LWDF).

D4.2


Page: 16 of 72

• Control signalling strategy – that determines the way in which the base station conveys the control information (scheduled resources) to different UEs. The control signaling is typically placed in the first part of the scheduled frame (or TTI).

Within 5GNOW a various sets of these 3 parameters may be selected, to define scheduling and control signalling strategies, depending on the particular waveform design constraints, as well as on the application scenarios. One example could be the selection of the two contrasting ways of conveying the scheduling info to the scheduled UEs [EML+09]:

• Compress and Broadcast (CBC) – in this option all the control information is jointly broadcasted to all UEs / portion of the UEs fully utilizing the correlation in the scheduling info between UEs in the compression stage, obtaining small amount of data to be transmitted. In the context of 5GNOW This strategy can be utilized in traffic types III and/or IV (see UFS concept and traffic types), where a set of MTC devices is scheduled in UL by using a single ID where the scheduling info includes a frequency range and a TTI number where there are selected resources for contention based traffic.

• Compress and Transmit Separately (CTS) – in this strategy, the scheduling info of an individual UE is treated separately from the scheduling info of all other UEs in the compression stage and it is transmitted separately to each UE. CTS strategy doesn’t utilize the correlation in the scheduling info, but transmits separately over individual channels to the UEs (adapting to the channel quality). This scenario may be used in a typical high data rate synchronized scenario to e.g. maximize throughput / system efficiency.

Speaking of scheduling – a typical design rules for a scheduler are the following [SJT09]: • Fulfill QoS parameters – the scheduler shall assure QoS requirements for the given service/traffic

classes. The example metrics are: minimum reserved / guaranteed bitrate, maximum allowable delay and tolerated jitters.

• Throughput optimization – due to the character of the resources in wireless networks, which are limited, an important consideration is how to maximize the total system throughput. The example metrics are: maximum number of supported UEs or utilization of the link.

• Fairness – after the QoS requirements are fulfilled the scheduler shall allocate the left-over resources fairly. A typical parameter of fairness is the convergence time (fairness may be defined as a short term or long term).

• Energy consumption and Power Control (PC) – scheduler shall consider the maximum allowable power at the UE or BS. Given the BER vs SNR that the service requires, the scheduler can calculate the suitable power to use for each UE taking into account their location (based on feedback info). From the other side, the UE scheduler shall optimize transmission power of a device.

• Implementation complexity – BS has to handle multiple simultaneous connections and the decisions need to be made fast (e.g. in LTE - within 1ms TTI), the scheduling algorithms need to be simple, fast and using minimum resources (e.g. a memory size can be a factor here).

• Scalability – the scheduling algorithm should work efficiently for both small and large number of connections / services.

2.2 5GNOW unified MAC design

The overall unified MAC architecture is presented in the Figure 9 below. According to [5GNOWD4.1] the architecture / framework is based on the following two assumptions:

D4.2


Page: 17 of 72

• Hybridity – use well defined resource management mechanisms targeted to specific / individual traffic type / application scenario. This allows to using optimized well-known algorithms to fulfil traffic types’ requirements and specifics.

• Hierarchy – firstly select the traffic types that need to be served and then select proper resource management algorithms (e.g. individual schedulers) and let them run. This allows to adjusting the policies and operation according to the current traffic situation (e.g. large MTC traffic in a specific part of the day or a long huge data download sessions). This concept defines the architecture where there is a general MAC scheduler acting as a coordinator of individual schedulers, allowing to select proper one and deals with proper communication between modes of internal schedulers.

Figure 9: Overall unified MAC architecture

This chapter describes the details of the unified MAC design including: the protocol entity structure, the description of each building blocks, the definition of the unified MAC tasks and MAC modes (i.e. binding of the use case / traffic profile with resource management policy / profile).

Unified MAC building blocks and procedures 2.2.1

Figure 10 below shows the block diagram of the unified MAC, i.e. the architecture of MAC protocol incorporating asynchronous aspects of the 5GNOW (e.g. non-ideal / limited feedback, open-loop synchronization, PRACH procedure etc.).

D4.2


Page: 18 of 72

Figure 10: Unified MAC architecture

According to the figure, unified MAC consists of the following elements and their corresponding roles:

• Scheduling coordinator – coordinates selection of the transmission modes and policies, selection of the schedulers and data / traffic prioritization (e.g. decision of the user type – contention / scheduled user; scheduling mechanism selection / waveform parameter set selection, etc.);

• Individual scheduler – responsible for resource allocation and MCS selection (an individual scheduling algorithm);

• Timing Advance module – responsible for management of the timing of the individual traffic profile, e.g. reused closed loop mode for synchronous traffic; open loop timing advance as an example of asynchronous MAC concept;

• RACH management module – responsible for handling random access procedure, settings of parameters (e.g. split of RA preambles into normal / MTC traffic), PRACH dimensioning, etc.

• Feedback control module – responsible for feedback periodicity and content adaptation; • Cognitive MAC / Learning module – UE / MTC / QoS / traffic type learning / gathering block. Helps

to adjust parameters with self-optimization nature (e.g. observation of the static MTC may imply the following decisions: decrease feedback rate, decide on SPS mode usage, pilot pattern selection);

• Data handling module – responsible for preparation of the data for transmission, including data buffering, multiplexing of the packets, etc.

• Interface towards other MAC entities – in HetNet / multi-cell scenario exchange of the MAC messages / protocol primitives (e.g. feedback adaptation for CoMP);

• Data channels – service access point (SAP) for transmission / reception of the multiplexed / formed data packets to / from PHY layer;

• Control / steering channels – SAP for exchanging primitives and control messages between PHY/MAC (e.g. control messages for partitioning the UFS into traffic areas; resource allocation data; feedback data reception, etc.).

In the subchapters each of the main building blocks is described in more details.

D4.2


Page: 19 of 72

2.2.1.1 General scheduler / scheduling coordinator / priority handler

An adaptive and hybrid approach on MAC design is taken to dynamically select specific algorithm or frame parameters according to the traffic pattern. General scheduler will decide on scheduling strategies in the traffic modes. Additionally an approach is taken to dynamically select between different MAC protocols/protocol operation ”on the fly” based on the observation of runtime network topology, traffic changes, user mobility [HR11]. We can build separate schedulers, policies and strategies with the switch of e.g. the MTC vs normal mode. 5GNOW may propose to have several scheduling algorithms implemented and switch between them to adjust the algorithm depending on what is the traffic load to e.g. assure everyone gets resources and that energy is minimized. Algorithm provides switching, that depends on the traffic type we support and we might consider the solution of various algorithms for various traffic types, where switching is implemented to assign different resources for MTC and normal (i.e. data hungry services) and on top of that, we may put a new general algorithm for “scheduling of the schedulers”. The presented concepts are referenced in [SYM+12] [HA12] [HR11]. The following strategies may be incorporated in the scheduling coordination to help the UEs/Scheduling coordinator to decide whether to request MAC switching after PHY adaptation fails [HR11]:

• performance degradation threshold: UE computes its throughput performance (or scheduling coordinator do it alternatively) in an observation interval while data traffic flows continue. If performance is degraded below some threshold, the UE will request (Scheduling coordinator will perform) MAC switch. In this approach, a focus is on maintaining the UE’s performance in order to satisfy application requirements such as throughput or delay.

• predicted future traffic patterns: Each UE checks its (or scheduling coordinator check all UE’s) future traffic patterns by periodically examining the MAC transmit buffer. If the average packet size is large, the UE may request (or scheduling coordinator may initiate) a TDMA MAC switch in order to reduce contention. On the other hand, for smaller average packet size light load, CSMA is preferable to prevent unnecessary wastage of slot time. Besides, AMAC is able to provide better performance under conditions of different traffic types. Bursty traffic, for instance, can be transmitted efficiently with CSMA; however, streaming traffic is better suited for TDMA transmission because of improved reliability and lower contention.

The following assumptions (treated also as tasks) for general scheduler are given:

• Traffic dependent selection of the resource allocation mechanism and feedback; • Traffic / device dependent selection of the waveform parameters according to UFS structure • Dynamic frame parameters adjustment according to number / type of users / traffic.

Scheduling coordinator may operate (i.e. dynamically changing MAC behaviour on-the-fly and configuring PHY parameters) in a following algorithmic approach (in conjunction with Cognitive / learning MAC module) [HR11]:

• Baseline MAC selection: Each node starts with the default MAC protocol which handles the nominal network traffic condition.

• PHY adaption: UEs/Scheduling coordinator first try to discover if they can adapt PHY parameters (such as operating channel, power or modulation type) when the performance of data transmission drops significantly.

• MAC adaptation: When the PHY adaptation is not able to meet performance goals, scheduling coordinator initiate a switch of the MAC protocol operation.

Scheduling coordinator is interfacing with:

• Individual schedulers;

D4.2


Page: 20 of 72

• Other cell unified MAC entities; • Timing advance module; • RACH management module; • Feedback control module; • Learning / cognitive MAC module.

2.2.1.2 Cognitive / learning module

The a-priori knowledge of the type of MTC device (e.g. as defined in [5GNOWD4.1]) is useful to support direct decisions in the multiple access approach in order to minimize signalling and learning part (e.g. a type of MTCD provided to the network during NW attach). As an example, the knowledge that a MTCD is low mobile (or even static) can help to:

• Decrease feedback rate and granularity; • Decide on scheduling approach – e.g. SPS, dynamic scheduling, contention based transmission; • Define the granularity of pilots (e.g. decrease the amount of pilots) – to be incorporated in the UFS

design (e.g. define the different pilot patterns for different traffic classes). The above considerations may lead to definition of a learning module (or a Cognitive MAC module) to be incorporated in the overall unified MAC design.

2.2.1.3 Individual schedulers / resource allocation mechanisms

Table 1 below includes the traffic types defined within UFS together with resource allocation strategies proposed for each type.

Table 1: Resource allocation strategies

Traffic Type Synchronization Access Type Resource allocation approach

I

closed-loop scheduled Dynamic, channel adaptive resource scheduling

II

open-loop scheduled Semi-static/persistent scheduling (SPS)

III

open-loop sporadic, contention-based

One shot transmission (low amount of data + pilots) with contention based approach

IV

open-loop/none*

contention-based

The approaches can be summarized as follows:

• Traffic type 1 is a standard scheduled access type where we can utilize / reuse standard LTE / OFDM based mechanisms with dynamic, channel adaptive resource allocation or semi-static, pseudo-random non-channel adaptive resource allocation approaches, such as e.g. proportional fair, max CQI/SINR, MLWDF, round robin etc.

D4.2


Page: 21 of 72

• Traffic type 2 is still scheduled access but with the use of open loop synchronization, where the concepts of IDMA and ATA mechanisms may be utilized. In such a case scheduler is not channel aware / channel adaptive but decides on the resource location on the time-frequency grid and the amount of required resources. For example: IDMA may be used for MTC in UL with medium traffic volume. MAC - decides on amount of resources allocated. Scheduler not channel aware (may also be used for high speed UEs). IDMA fits here – doesn’t provide channel adaptive scheduling and doesn’t need spreading code to be allocated to an individual UE.

• Traffic type 3 and/or 4 are open loop or non-synchronized sporadic and contention based types of traffic. For this a new PRACH may be utilized such as one-shot transmission (low amount of data) with contention based transmission algorithm. MAC defines/steers the amount of resources dedicated for such data usage. Also differentiation of emergency/normal data may be considered with various approaches.

2.2.1.4 Adaptive feedback steering

5NOW may apply a concept of adaptive feedback steering, which bases on changing periodicity and amount of bits of the feedback depending on HARQ ACK/NACK, mobility profile and service type. To set a background for dynamic feedback steering, an example of adaptive feedback steering is presented in [AB11] known as Dynamic CQI Resource Allocation (DCRA) algorithm. The number of slots for CQI is limited, thus each UE is allowed to send its CQI once in every “n” frames (“n” is referred to as feedback window). However, a UE with good channel doesn’t need to send its feedback as frequently as a UE with bad channel. In such context, assigning a fixed w for each UE is not optimal. The main objective is to select a subset of UEs to send CQI among all the UEs to be scheduled in next frame while keeping the system throughput maximum. [AB11] describes cross layer approach for DCRA including: HARQ, QoS, and mobility profile for multiuser OFDMA system. The DCRA algorithm is implemented at BS side and choses adaptively the UEs which are to send feedback to maintain their QoS constraints. Based on the CQI, feedback rate for each UE is determined exploiting the signalling of MAC layer ARQ protocol, UE’s application (MAC layer service classifier) and mobility profile (Doppler based mobility tracking). DCRA derives the optimal average feedback window values based on the UE velocity and service class for a target PER. Using this information, the algorithm then allocates the CQI resource dynamically on frame-by-frame basis. DCRA uses cross layer strategy to combine information from two levels (PHY and MAC) which jointly give a complementary insight on the channel quality of the UE and its effect on transmission at MAC level in terms of PER, delay, etc. As each application reacts differently at MAC layer for the errors encountered at the physical layer, DCRA takes into account information from both MAC and PHY levels to derive best feedback window periodicity to meet QoS application constraints. The two opposite settings of the feedback window are:

• Smaller feedback window provides more knowledge about the channel state and helps to achieve higher throughout.

• Larger feedback window saves UL resources with the tradeoff of DL throughput degradation. The algorithm adjusts the parameters among the two boundary conditions described above to achieve good balance between the robustness and feedback overhead (e.g. measuring the number of consecutive ACKs from UE, the algorithm may decide to increase the feedback periodicity). The practical approach takes into account the following considerations: in the real-time network operations, the cell and UE dynamics (number of UEs, session duration, mobility) change very frequently, so that that it might be too late for the BS station to determine optimal feedback window size for all UEs. The DCRA scheme proposes to deal with these dynamics, so that the BS can act on frame-by-frame basis. For

D4.2


Page: 22 of 72

each cell, the network operator should learn about the appropriate average feedback window size to be used for a UE experiencing a specific application and mobility profile. This knowledge can be acquired during the network planning phase and then can be used in real-time operations phase. The analysis shows that the feedback window can be averaged according to UEs service class and their mobility profile for a given cell environment. Another concept of controlling the feedback update period and feedback rate in MIMO systems is presented in [KLC10]. Update period is steered by the temporality of the spatial channels correlation (by measuring a lower bound of averaged effective SNR). Feedback rate and update period shall be minimized with respect to overhead, but shall be able to follow the channel correlation changes. Feedback control strategy adapts feedback update period “n” (measured in subframes) and feedback load “B” (measured in bits) as a function of the amount of time correlation. From the other hand in a standard LTE system even when the UE is in a stable condition, the updated CSI does not improve the system performance due to exactly the same content. [HZL09] proposes an example approach for improving the standard LTE CQI feedback. Using the same amount of bits within a single CQI message (e.g. 11 bits), [HZL09] proposes to improve the accuracy of feedback by changing the content. A quantized channel is fed back from receiver to transmitter with incremental updating of the feedback over time (e.g. the accurate CSI is present at eNB after 4TTIs). Taking into account the LTE option, the scheme increases RTT (because of the waiting time of larger period), but may be good for low UE speeds. This approach does not aim at reducing the amount of signalling, but rather improving the feedback quality. The scheme may be also adjusted e.g. by adapting the number of increments depending on UE speed and channel conditions. To realize that, it requires additional mechanism to dynamically decide on how many concurrent TTIs will be used for feedback messages. This is possible due to the nature of incremental approach, i.e. already the first CSI feedback gives possibility to estimate the channel profile (of course with very rough approximation). Figure 11 below presents the general concepts of adaptive feedback steering that may be applied in 5GNOW. BS can act on frame-by-frame basis. An example mechanism to be applied may work as follows: BS measures the number of consecutive ACKs from the UE; if multiple NACKs are received the algorithm may decide to increase the feedback periodicity.

Figure 11: Adaptive feedback steering

2.2.1.5 Autonomous timing advance

As outlined in general by 5GNOW [5GNOWD3.2], today’s cellular networks, like LTE, operate strictly synchronously, at least within each cell. A device wanting to transmit a few bits of data has to enter the network via the random access procedure. This includes an ongoing closed-loop timing advance control. MAC is responsible for handling timing alignment of an UL transmission that includes setting and updating timing advance parameters. The idea and basic closed-loop operation of timing alignment is shown in the Figure 12 below.

D4.2


Page: 23 of 72

Figure 12: Timing alignment mechanism (left side - an UL transmission without timing advance; right side - an UL transmission

with timing advance)

This typical closed loop mechanism fits in traffic type I (regarding UFS). However for traffic types III and IV as well as II an open loop timing advance is proposed in [5GNOWD3.2] as one of the asynchronous MAC concepts. The 5th generation of wireless systems, in order to support the upcoming internet of things, will include a large fraction of machine-initiated communication. We will have a large number of devices, where many of them only have sporadic traffic. In this case, the extensive random access procedures, like in LTE, generate a painful signalling overhead. Furthermore, existing closed-loop time synchronization consumes a notable amount of energy. This shortens the battery life time of e.g. cheap sensors. Our vision here is to use open-loop synchronization for devices which just have sporadic MTC traffic (type III and IV in the UFS). Furthermore open-loop synchronization may also be used for scheduled access (type II traffic) for energy and overhead saving reasons, multi-cell scenarios applying UL CoMP mechanisms and in the light of user-centric instead of cell-centric networking [5GNOWD3.2]. The device listens to the DL and synchronizes itself coarsely. Here, the usage of a multi-carrier waveform helps: The division of the available bandwidth into smaller chunks (sub-carriers) generates comparatively long symbols, improving robustness against delay spread but also against timing misalignments. The symbol duration can be designed such that the misalignments are just fractions of the multi-carrier symbols. Furthermore, the devices may apply some autonomously derived timing advance called call autonomous timing advance (ATA). Two variants of ATA are possible [5GNOWD3.2]:

• ATA Type I: On the one hand the single devices may estimate the round trip time their signals are suffering (e.g. based upon GPS data or making use of pathloss estimations) and apply a respective timing advance. When doing so, performance of the respective transmission is improved. Additionally the base station may broadcast its own GPS position within system information).

• ATA Type II: Another simpler variant is for all type II/III devices to apply the same timing advance NOLmax/2 with NOLmax being the highest round trip time appearing within the cell (or alternatively e.g. the 95%-tile of all existing round trip times). When using this technique, inter-traffic interference at the frequency borders is reduced. Various means are available for the devices

D4.2


Page: 24 of 72

in order to get knowledge on NOLmax. The simplest one is for the base station to broadcast the value based on measurements (e.g. during closed-loop synchronization of type I devices).

Figure 13 below illustrates the timing offset ranges at the receiver side caused by different timing alignment strategies from closed-loop timing advance to ATA to completely asynchronous transmission.

Figure 13: Illustration of different degrees of synchronicity w.r.t. transmission in an anticipated frame structure: (1) Perfectly synchronized, (2) Fully asynchronous, (3) Open-loop (OL) synchronization, (4) ATA type 1 (OL sync. + shift), (5) ATA type 2 (OL

sync. + timing estimate + shift)

In that context unified MAC is deciding on selecting the following scenarios, depending on the traffic type (i.e. part of the UFS) and base station / UE capabilities:

• Closed-loop TA; • Open-loop ATA Type I; • Open-loop ATA Type II.

2.2.1.6 “One-shot” transmission for sporadic data

Another building block of unified MAC is random access mechanism handler. While again depending on the traffic profile (closely related to UFS) a selection of a typical random access procedure for closed loop synchronized UEs with typical large data pipe and a new fast and scalable architecture allowing sporadic data (Type III and IV) to be transmitted in “one shot” in PRACH [5GNOWD3.2][5GNOWD4.1] [WJK+14][KWJ+14]. Clearly, by doing so, sporadic traffic is removed from standard UL data pipes resulting in drastically reduced signalling overhead. Another issue that is closely related to the signalling overhead is the complexity and power consumption of the devices. Notably, waveform design in such a setting is necessary since the OFDM waveform used in LTE cannot handle the highly asynchronous access of different devices with possible negative delays or delays beyond the CPs [5GNOWD3.2] [WJK+14][KWJ+14]. Clearly, guards could be introduced between the individual (small) data sections, however, which make the approach again very inefficient.

t

Anticipated frame structure Multi-Carrier symbols

Timing offset range of arriving signals (1) Perfect sync. (e.g. closed-loop)

(2) Fully async.

(3) Open-loop sync. (delay is max. prop. Round-trip)

(4) ATA type 1 (Open-loop sync + shift by e.g. half of max. prop. Round-trip)

(5) ATA type 2 (Open-loop sync + timing estimate + shift)

D4.2


Page: 25 of 72

The design problem is as follows: terminals seek access to the system over the PRACH. We assume that in the presence of a common broadcast channel some rough synchronization is available. Each user has a data and control signal part which overlap in time with each other. Obviously, loosely speaking, to allow for channel equalization, the control must be ”close” to the data to probe the channel conditions ”nearby”. Conversely, detectability becomes erroneous the more control is interfering with the data, i.e. when separation between pilots and data cannot be achieved in a stable sense. This seems a contradicting, irresolvable task at first sight. However, with sparse signal processing methodology we can cope with that task thanks to the sparse structure of 1) user activity, 2) channel impulse response, 3) spatial correlations 4) sparse topology etc. A possible configuration is shown in Figure 14. Here we spread the signalling over the whole signal space, so-called underlay signalling, and collect it back within the observation window 𝔅. Let us denote the control part as C-PRACH and the data part as D-PRACH of the PRACH. In order to estimate the channel for the users in D-PRACH then, we will spread C-PRACH over the whole signal space (underlay signalling) and collect it back within the observation window 𝔅. The clue to proper detecting the channel is that specific pilot signal plus the sparse channel serve as a spreading signature which can be detected in a much lower space so that the noise floor per estimated component becomes smaller with increasing dimension. The approach has some intriguing advantages: 1) Instead of placing distinguishable pilot patterns in the data section for a large number of users based

on the classical sampling theorem the control signals can overlap in the large space but are then also distinguishable in a small observation window tailored to the actual number of degrees of freedom. This allows easy overloading of the channel as well.

2) The control is necessarily interfering with the data (and vice versa) but due to the spreading of the control signal the noise from the data is scaled down by a factor defined by the size of the observation window.

In the following sections, the details of this “sparsity” aware random access scheme are outlined. We have published the concept in [WJW14][WKW+14][WBS+15].

Figure 14: Underlay signaling in “one shot transmission” concept. Red area indicates the window 𝕭 from which the whole space

is “illuminated “ due to underlay signalling (shaded area)

D4.2


Page: 26 of 72

The D-PRACH advantages in the unified MAC over standard PRACH are: • one-shot transmission of the preamble and data enabling to measure the signal and demodulate

data; • shortens the procedure for transmission of small amount of data; • may be used to transmit feedback messages (e.g. 40bits of CQI feedback); • due to sparsity and possibility of illumination of the larger BW part with sparse PRACH we may get

rid of sounding reference signal (SRS) known from LTE, to measure channel frequency response of the channel.

Overall using of sparse processing and D-PRACH concept we decrease amount of signaling (data with PRACH), decrease number of PHY channels (getting rid of SRS and PUCCH) and save battery life of MTC device (shortening the connection time for small data transmission and using less processing). The tasks of unified MAC in random access context include:

• handling both (i.e. typical random access and new D-PRACH) mechanisms of random access; • selection of one dedicated random access mechanism dedicated to a specific scenario; • distribution of random access preambles among different traffic classes; • resource management of random access, i.e. preamble parameters settings, definition of time-

frequency slots for random access opportunities.

Unified MAC will handle the following main tasks and procedures: • Dynamic Frame structure adjustment, including:

o Resource distribution among traffic classes / UE distance / speed o Waveform parameters selection/adjustments o Frame parameters setting (e.g. pilot patterns)

• Scheduling mechanisms: o Resource distribution within a traffic class o Scheduling policy / strategy selection o MCS control

• Feedback control: o Feedback periodicity and content adaptation (e.g. in a CoMP scenario)

• Contention mechanism adjustments: o Contention measurements and PRACH/D-PRACH dimensioning o D-PRACH procedure

• Timing advance o Closed loop and open loop synchronization mechanisms (e.g. ATA)

Unified MAC modes (heterogeneous services and waveforms) 2.2.2

The approach of Unified MAC and UFS is that various algorithms for resource management are taken and various parameters sets are used to suit different traffic types or application scenarios. According to that, the subchapters below classify them the modes, where the application / traffic is linked with the selected algorithm approach. The following classes are defined: MTC modes (including single one shot transmission of small data in the UL, as well as medium sized asynchronous traffic), CoMP mode and normal / synchronized data transmission mode. The following MTC modes are seen by 5GNOW system:

• Very low data traffic periodic or aperiodic (i.e. event triggered) MTC mode – corresponding to D-PRACH;

• Medium sized asynchronous traffic (corresponding to traffic type III or IV according to definitions in UFS) – corresponding to IDMA/ATA.

D4.2


Page: 27 of 72

3 PHY / MAC layer interface issues

Chapter 3 describes the PHY / MAC interfacing (and also gathers WP3 / WP4 merging point) including UFS and algorithms that are running on the PHY layer and are treated as building blocks to be controlled by MAC layer. This section starts with the review of the UFS with respect to system scenarios. Then we describe algorithms and transmission procedures, such as novel D-PRACH approach, ATA and CoMP related mechanisms. This is followed by two PHY / MAC interfacing issues, namely: feedback and waveform parameters. These two can be developed as primitives / messages / parameters exchanged by the two layers to configure the system with respect to certain traffic scenario. The chapter is finalized with the example of link-to-system modelling of 5GNOW concepts in multiuser scenario, where the provided PHY abstraction model of FBMC is compared with standard LTE OFDMA in a system level environment.

3.1 Unified MAC interface to PHY layer

Unified MAC is responsible for controlling PHY resources. PHY layer is responsible for processing the data-bits and transmission using proper time-frequency placement. A bridge between MAC and PHY layer is done with the use of UFS. MAC layer is steering the PHY layer with the use of an interface by primitives / parameters. Figure 15 below shows a structure of this interface. The most important steering elements include (from top to down):

• Data handling – transport block creation according to required QoS / available resources; • Contention parameters steering / semi-static allocations – management of the traffic type II, III and

IV resources placement and e.g. RACH settings; • Dynamic frame structure adjustments – steering of the resource distribution among traffic profiles; • Dynamic resource allocation – steering of the resource placement for individual allocation in a TTI

for traffic type I and II; • Dynamic feedback control – setting parameters of the feedback for individual frame part.

Figure 15: Unified MAC interfacing

D4.2


Page: 28 of 72

Dynamic frame structure adaptation 3.1.1

According to the description of UFS (see Section 3.1 in this document) which brings a flexible frame structure dynamic adaptation of the parameters is feasible. These adaptations are steered by unified MAC over MAC-PHY interface primitives. [SA12] presents OFDMA based frame structure that is adapted dynamically with respect to OFDM parameters that are adjusted to UE mobility and distance from the base station. 5GNOW may extend this approach with respect to parameters of the selected waveforms and traffic specifics, in the following example areas:

• Frame structure design - adaptive UFS depending on UEs mobility and distance from the base station (subcarrier separation, ΔF and max channel spread Tmax respectively). In that context the number of resources is adapted within frame structure depending on the service distribution. The distance from the base station is calculated by eNB with the use of RACH signal – while calculating TA and sending Tadvance with the frame parameters and allocation. Distribution of the UEs based on power measurements. Some statistics of the cell may be taken into account to have the ΔF and Tguard (CP size) calculated or they can be adapted based on the actual power and TA measurements. Additionally while designing the UFS, it is needed to differentiate between control channels and data channels (e.g. CS/CB/JP COMP solutions for data and some ICIC for control also for the design of the antenna solutions). On top of these detailed parameters a TTI can be dynamically configured.

• MAC layer considerations - Additionally having multiple CP durations and subcarrier spacing within the frame impacts the MAC layer. While the UEs are assigned anywhere within the frame in the conventional scheduling, the resource blocks which are suitable for the UEs are limited in the proposed frame structure. This might impact the flexibility of the MAC protocols. Additionally, the time and frequency spread statistics are required for the proposed structure. The static adaptation with the proposed frame structure might be problematic considering several issues which requires dynamic adaptations. The work / investigations on extension of the concepts presented in [SA12] including e.g. required feedback information, dynamic adaptation of the frame parameters and scheduling that considers the CQI.

Dynamic resource allocation 3.1.2

MAC resource management parameters within traffic I and/or II radio frame part will include the time-frequency or time-frequency-space resource block identifiers. The resource block definition will very much depend on the defined structure of a particular 5GNOW waveform. The parameters may include: time-frequency resource block index, modulation and coding scheme, MIMO precoding matrix index, transport block size, etc.

Contention parameters steering/semi-static resource allocation 3.1.3

The radio-frame part for traffic types III and IV is dedicated for semi-static resource allocation for e.g. periodic or event-triggered fully asynchronous small data transmission. The parameters that can be seen under this part may include e.g.: periodicity and frequency range for transmission, parameters of the waveform, size of the message, RACH parameters (e.g. min power, power ramp, preamble set, etc.).

Dynamic feedback control 3.1.4

Feedback mechanisms are also under considerations of adaptation to traffic / service / environmental needs. According to this assumption the following parameters would be required for feedback control: periodicity of the feedback (may depend on number of consecutive ACK/NACKs from UEs; may depend on the UE type, mobility profile / UE speed, service type); quantized feedback value; ACK/NACK management

D4.2


Page: 29 of 72

(e.g. number of possible retransmissions; decision on not using ACK/NACK); type of feedback (e.g. for CoMP).

3.2 Algorithms and transmission procedures

This subchapter includes detailed description of the selected algorithms and transmission procedures that operate either on PHY layer and are steered by MAC layer or on the MAC / PHY interface. The set of algorithm includes: D-PRACH access algorithm, ATA operation, CoMP delay detection algorithm and robustness mode operation. They are all developed within 5GNOW WP3 and WP4 and deal with asynchronous aspects of the system at both single cell and multi-cell aspects.

D-PRACH random access algorithm 3.2.1

3.2.1.1 Underlay signalling and sparsity aware detection

To explain the overall idea we start by considering first a generic single–user system model. Let 𝑝 ∈ ℂ𝑛 be a pilot (preamble) sequence which is unknown but from a given set 𝒫 ⊂ ℂ𝑛 and 𝑥 ∈ ℂ𝑛 be an unknown (coded) data vector. Both are transmitted simultaneously and use potentially the same resource. We set 𝔼 1𝑛‖𝑝‖𝑙2

2 = 𝛼, and 𝔼 1𝑛‖𝑥‖𝑙2

2 = 1 − 𝛼. Hence, the control signalling fraction of the power is 𝛼 and, due to

the random nature of x we have 𝔼 1𝑛‖𝑝 + 𝑥‖𝑙2

2 = 1, i.e. the total transmit power is unity. Note that in typical systems n is large, say n=24576 as in LTE/LTE-A with 20 Mhz bandwidth and 𝒫 represents the Frank-Zadoff-Chu sequence set and no data is transmitted. The primary goal at the receiver is to estimate the data vector x from the observations y whereby also the vector h of channel coefficients is unknown. A possible strategy is to estimate separately first the channel coefficients ℎ� = 𝑄ℎ(𝑦|𝑥 ∈ 𝑋), under certain assumptions on the data x. A simple approach here is for example to treat the data as noise. In a second phase then the data 𝑥� = 𝑄𝑥�𝑦�ℎ�� conditioned onto ℎ� has to be estimated. Obviously, this procedure can then be iterated with or without data decoding. While for a classical receiver this procedure results in a huge interference for the channel estimation, a non-standard receiver can make use of the reduced dimensionality of the problem beyond the classical Shannon setting: 1) The communication in random access is sporadic so that out of the total set only an unknown small subset of users are actually active. Alternatively, we can assert certain probabilities to each node. 2) We assume to have a–priori support knowledge on h, i.e. 𝑠𝑠𝑝𝑝(ℎ) ⊆ Τ with T denotes, e.g., a subset of [0, … ,𝑇𝑐𝑝] where 𝑇𝑐𝑝.represents the cyclic prefix of LTE-A (i.e. 𝑇𝑐𝑝=8192). Moreover, we assume sparsity of channel profile with only k=6 relevant paths and supp(h) < 300 (corresponding to a 1.5 km cell). A key is the illumination and partial Fourier sampling concepts from [WJW14][WBS+15] with the limitation to a relatively small observation frequency window, i.e. frequency indices in the set 𝔅 of cardinality 𝑚 = |𝔅| so that the data noise floor is effectively suppressed. Let us denote with 𝑃𝐵:ℂ𝑛 → ℂ𝑚 the corresponding projection matrix. It is known that fixed Fourier windowing has several drawbacks. To introduce certain randomization we consider point wise multipliers 𝜉 ∈ ℂ𝑛 in time domain and we denote the corresponding 𝑛 × 𝑛 diagonal matrix as 𝑀𝜉 ≔ 𝑑𝑑𝑑𝑑(𝜉). Summarizing, the 𝑚 × 𝑛 sampling matrix Φ = 𝑃𝐵𝑊𝑀𝜉 will be considered. The 𝑚 × 𝑛 matrix 𝑃𝐵 ⋅ 𝑐𝑑𝑟𝑐(𝜉), is usually called partial circulant matrix and it can be shown that it is applied to ĥ (instead of ℎ) so that ĥ is actually sparse in the inverse Fourier domain. Preferable for sparse channel vectors h (in whatever domain) is the 𝑙1–penalized least square method or quite efficient greedy methods like CoSAMP with precise guarantees in reconstruction performance [WJW14][WBS+15].

D4.2


Page: 30 of 72

It can be proved that altogether the rate per subcarrier channel is lower bounded by [WJW14][WBS+15]

𝑅(𝛼) ≥ log�1 + SNR ⋅ (1 − 𝛼)� − log�1 +𝑚 ⋅ �4�1 + 𝛿2𝑘

1− �1 + √2�𝛿2𝑘�2

⋅ �SNR ⋅1 − 𝛼𝛼

+1𝛼��

where 𝛿2𝑘 is small for m>klog5(n) measurements. The details can be found in [WJW14][WBS+15]. Performance results are depicted in Figure 16, where we show symbol error rates (SER) over the pilot-to-data power ratio 𝛼. Recall the extremely challenging scenario of only 839 subcarriers in the measurement window versus almost 24k data payload subcarriers. Moreover, in Figure 17, we show the false detection probability 𝑃𝐹𝐹 (some user is detected while not active) over the missed detection probability 𝑃𝑀𝐹 (user is active while not detected). We observe that, although the algorithms do not yet capture the full potential of the idea, reasonable data detection performance can be achieved by varying 𝛼, cf. [WJW14]. In the 4G LTE-A standard a minimum 𝑃𝐹𝐹 = 10−3 is required for any number of receive antennas, for all frame structures and for any channel bandwidth. For certain SNRs a minimum 𝑃𝑀𝐹 = 10−2 is required. It can be observed from the simulations that the requirements can be achieved [WJW14][WBS+15]. In the context of 5GNOW we have also extended this to the multi-antenna case which is shown in Figure 18 and summarized in [RW15].

Figure 16: Avg. BPSK SER in 5G “one-shot” random access (in 20MHz LTE-A setting) at SNR=20dB. Out of n=24576 dimensions, m=839 are used for CS. This limits the control overhead to below 5%. In the second case,pilots and data are fully separated so

that the performance is greatlyimproved at the expense of a slightly increased control overhead (<14%). Note that the curves do not cross for the same number of active users so the performace is always better with separation.

D4.2


Page: 31 of 72

Figure 17: P_FD over P_MD for the 5G “one-shot” random access with varying detection thresholds. The box illustrates the LTE

feasible region.

Figure 18: D-PRACH SER with multiple antennas

D4.2


Page: 32 of 72

3.2.1.2 D-PRACH transmission

The overall operation of for D-PRACH transmission (using RA preamble with data bits) in MTC application including feedback is the following (see Figure 19 below):

• The D-PRACH preamble is sent with data. • Base station sends ACK/NACK confirmation to acknowledge the UE if it need retransmission. • If NACK is received, it means, the receiver failed to decode the data, but received the preamble

sequence. In that case a notification is sent together with resource allocation and then retransmission of the data is sent over PUSCH.

• If ACK is received, it means, BS received data and preamble. In that case nothing else need to be send and MTC device is going to idle;

• If nothing (no ACK/NACK) is received, it means, the D-PRACH preamble need to be sent with higher power (possibly including backoff time).

Figure 19: D-PRACH operation procedure

Figure 20 below shows the signalling flows comparison for the synchronization standard LTE-like UL transmission with the proposed operation of D-PRACH.

Figure 20: Signaling flows comparison for non-synchronized UE transmission: a) LTE-like UL; b) MTC scenario for D-PRACH

D4.2


Page: 33 of 72

CoMP delay detection algorithm 3.2.2

Another component of MAC architecture is a delay detection algorithm in the CoMP scenario. Received signals from multiple cells may not be aligned in time at the receiver due to the difference of propagation delays and to the timing alignment errors between base stations. This time offset causes pilots rotations at the receiver that makes the estimation of the channel difficult. As shown in D4.1, the deterioration of the BER for real channel estimation with high delays (higher than 120 samples) is due to the phase rotations introduced by the delay 𝜏 that increases the frequency selectivity of the equivalent channel. Indeed when the coherence bandwidth of the channel is much smaller than the pilot spacing, channel interpolation cannot be reliably computed. Cooperative MIMO-OFDM deals with Inter Symbol Interference (ISI) thanks to the use of a CP: when delay difference between the cooperative BSs is shorter than the CP, all ISI that may deteriorate the quality of service at the receiver is suppressed. For cooperative MIMO-OFDM, due to high propagation distances, the CP must be longer than for non-cooperative OFDM. In LTE 10MHz the short CP is 72 samples long and the long CP has duration of 256 samples. Contrary to OFDM symbols, FBMC symbols structurally overlap in the time domain at the transmitter, the use of a CP thus becomes useless. In [5GNOWD4.1] it was shown that FBMC, with real channel estimation, can cope with delays of up to 120 times samples (7.8µs) without any need for correction. For the parameters of LTE with 10MHz channel bandwidth, this corresponds to a distance of 2340m. FBMC modulation is thus very resistant to time propagation differences between signals from the two BSs, due to its overlapping structure. Recalling that FBMC does not need any Cyclic Prefix to cope with time propagation differences; the spectral efficiency is then preserved without impact on time synchronization. As stated in [3GP13b], in the context of Rel-11 DL CoMP, UE performance requirements are defined by assuming a typical timing offset in the range of [-0.5, 2] µs. Delays higher than 7.8μs are unlikely to occur in most of cooperative systems; nevertheless it could be useful to decrease the virtual delay (i.e. the actual delay minus the estimated delay) seen at the UE in order to keep the channel flat over a Resource Block (12 adjacent carriers). Furthermore releasing the time synchronization between multiple points allows to consider non-coherent joint transmission schemes. A delay detection algorithm has therefore been designed. It is described below. Firstly, UE computes the estimate of the channel in the time domain (thanks to the IFFT of the estimate in the frequency domain) and seeks for the peak that corresponds to the delay. Figure 21 shows an example for two values of the actual delay 𝜏. Note that the range of the estimation is limited by the aliasing of the IFFT; i.e. the peaks at 𝜏 = 0 and 𝜏 = 256 both correspond to the arrival time of signal from the closer BS. The goal of the CoMP delay detection algorithm is then to return a value �̃� such that 0 ≤ 𝜏 − �̃� ≤ 30 with 𝜏 the actual delay. The choice of the value 30 for the maximum virtual delay �̂� = 𝜏 − �̃� has been motivated by the study of the coherence bandwidth of the channel. Such a value indeed ensures that the channel is flat over 12 consecutive carriers (one PRB). The ‘latter’ BS is then asked to rotate each carrier of the transmitted data with a factor exp (𝑗2𝜋𝜋�̃�/𝑁), with 𝑁 the number of carriers of the symbol. The set of values taken by �̃� must be as small as possible in order to limit the feedback from UE to BS. The algorithm, depicted in Figure 22 below, consists of filtering the estimated channel |ℎ(𝑛)|2 and finding the index of the maximal value. Based on this index the algorithm generates a value �̃� chosen in the set {0,20n+5}, n=1,2,…,11 . This set is composed of 12 values; it therefore requires 4 bits of feedback, plus the identifier of the BS that must rotate the data.

D4.2


Page: 34 of 72

Figure 21. Time estimation of the channel, example.

Figure 22: Delay detection algorithm

Probabilities of false alarms and non-detection for this algorithm are defined has follows:

• The probability of non-detection of a delay (𝑝𝑛𝑑) is defined as the probability that �̂� = 𝜏 − �̃� ≥ 30.

• The probability of false alarm of a delay (𝑝𝑓𝑓) is defined as the probability that �̂� = 𝜏 − �̃� < 0.

𝑝𝑛𝑑 and 𝑝𝑓𝑓 were computed for different values of 𝜏 (ranging from 0 to 220 samples). Those probabilities were shown to be null for all tested values. The algorithm is therefore very robust to non-detections and false alarms with very limited feedback information [CKD+13].

MU-MIMO/CoMP robustness framework 3.2.3

3.2.3.1 Robustness framework

In [WSJ12][WS10] (single cell)and in the context of 5GNOW in [WRS+13] (CoMP), a new direction has been developed which bears some great potential for the envisioned robustness framework. The main idea in this work is to use the structure of the transmit signals (in this case the spatial transmit codebook) and to incorporate as much information as possible in the design of the control channels. The collection of all

D4.2


Page: 35 of 72

information is the key to tailor the metrics used in the network to generate control messages as close as possible to the underlying performance indicators (rather than close to e.g. the mobile channel coefficients). In the following, we summarize 5GNOW theoretical results on the rate loss due to limited feedback [SWJ15][SWJ13] and provide a connection to the outdated CSI problem [CKW+14]. To capture the full effect of partial CSI on the user rates we need to look at an appropriate metric. The rate loss gap analysis needs to capture the following effects: (i) the rate loss due to scheduling and beamforming based on partial CSI; (ii) the rate loss due to link adaption based on partial CSI. Let 𝑁𝐵𝐵 be the number of base stations. Moreover we define 𝑆𝑏 to be the set of users selected at base station b, and let 𝜋 denote a mapping from the set of scheduled users to the corresponding precoding vectors. Then, we can define the following per user (𝑚) performance metric [SWJ15][SWJ13]:

Δ𝑟𝑚(𝝅𝐻 ,𝝅𝑉 ) ≜ max{𝑟𝑚(𝝅𝐻 ,𝑆𝐻;𝐻) − 𝑟𝑚(𝝅𝑉,𝑆𝑉;𝐻), 𝑟𝑚(𝝅𝐻,𝑆𝐻;𝐻) − 𝑟𝑚(𝝅𝑉, 𝑆𝑉;𝑉)}

= 𝑟𝑚(𝝅𝐻 ,𝑆𝐻;𝐻) − min{𝑟𝑚(𝝅𝑉,𝑆𝑉;𝐻), 𝑟𝑚(𝝅𝑉,𝑆𝑉;𝑉)},

with 𝑉 denoting partial CSI available at the base station (in contrast to the true CSI 𝐻), and consequently 𝑆𝐻 ,𝑆𝑉 and 𝝅𝐻 ,𝝅𝑉 being the corresponding optimal (with respect to sum-throughput) user selections and precoder mappings, respectively, based on true and partial CSI. The rate loss gap captures the following effects: (i) the first part 𝑟𝑚(𝝅𝐻,𝑆𝐻;𝐻) − 𝑟𝑚(𝝅𝑉 ,𝑆𝑉;𝐻), which is usually considered in the literature (see e.g. [CJK+10]), describes the rate loss due to beamforming based on partial CSI and assumes perfect link adaption, i.e., transmission with a rate equal to 𝑟𝑚(𝝅𝑉,𝑆𝑉;𝐻); (ii) the second part, 𝑟𝑚(𝝅𝐻 ,𝑆𝐻;𝐻)− 𝑟𝑚(𝝅𝑉,𝑆𝑉;𝑉) models the rate loss due to link adaption based on partial CSI, i.e., allocation of the rates 𝑟𝑚(𝝅𝑉,𝑆𝑉;𝑉). Even though Δ𝑟𝑚(𝜋𝐻 ,𝜋𝑉) is still optimistic since it does not account for the case 𝑟𝑚(𝝅𝐻,𝑆𝐻;𝐻) < 𝑟𝑚(𝝅𝑉,𝑆𝑉;𝑉) it is strong enough to address some of the drawbacks of the conventional analysis and leads to indeed very different results. Please note that, due to the per-user formulation, Δ𝑟𝑚(𝝅𝐻,𝝅𝑉 ) is not necessarily positive for all channels 𝐻.

3.2.3.2 Influence of quantization

In [CJK+10] a broadcast channel is considered, with |𝑆| = 𝑁𝑇 single antenna users and zero forcing (ZF) beamforming 𝝅𝑍𝐹,𝐻, 𝝅𝑍𝐹,𝑉 with respect to 𝐻,𝑉, respectively. Then, limited feedback with 𝑏 feedback bits per user incurs a throughput loss relative to ZF with perfect CSI according to

𝐸�𝑟𝑚�𝝅𝑍𝐹,𝐻 ,𝑆,𝐻� − 𝑟𝑚�𝝅𝑍𝐹,𝑉 ,𝑆,𝐻�� < log(1 + 2−𝑏

𝑁𝑇−1). In [SWJ15][SWJ13] it was shown that the scaling law is overly optimistic under the premises of the more realistic metric provided above. Furthermore the main result in [SWJ15][SWJ13] is an upper and a lower bound on the expected rate gap introducing an iterative algorithm quantizing only the effective channels. More precisely, in any iteration of [SWJ15, Algorithm 1][SWJ13], the average rate loss per user is upper bounded by

𝐸[Δ𝑟𝑚(𝝅𝐻 ,𝝅𝑉)] ≤ 3𝑝 �𝑁𝐵𝐵2 𝑁𝑅22−𝑏

𝑁𝑇−1 +𝑁𝐵𝐵𝑁𝑅2− 𝑏2(𝑁𝑇−1)�

Moreover, for some 𝑐2 > 0 and for sufficiently high SNR, the average rate loss is lower bounded by

𝐸[Δ𝑟𝑚(𝝅𝐻 ,𝝅𝑉)] ≥ 𝑐2 log�1 + 𝑝𝑝�𝜇𝑚,𝑏2 �

|𝐵𝑏|+�1+𝑝𝑝�𝜇𝑚,𝑏2 ��

2− 𝑏2�𝑁𝑇−1��+ o�2

− 𝑏2�𝑁𝑇−1��,

with 𝜇𝑚,𝑏 = �𝒉𝑚,𝑏 �2. Hence, it is shown that the scaling cannot be essentially improved. These theorems including the iterative algorithms are a main result of the robustness framework.

D4.2


Page: 36 of 72

Figure 23: Comparison of centralized and distributed algorithms.

Figure 23 illustrated the anticipated distributed algorithm in comparison with a centralized variant. For further details, please refer to [SWJ15][SWJ13].

3.2.3.3 CoMP modes

A useful characterization of transmission schemes under perfect CSI in complex wireless networks where the exact capacity is unknown are the degrees of freedom (DoF). For Multiuser MIMO [CJK+10], as well as for Network MIMO [KG12], or CoMP, the asymptotical capacity behaves as

𝐶(𝑃) = 𝐵𝑛𝑡 log𝑃 + 𝑐2(𝐵,𝑛𝑡),

with B being the number of base stations. The slope of the curves in the asymptotic regime is determined by the pre-log factor which is identical to the DoFs of the system. CoMP promises to achieve huge gains. However, it requires message sharing among the cooperating base stations. This raises the problem that real-world constraints on the backhaul induce delay and consequently we have to deal with frequency offsets, time synchronization issues, etc. A possible solution to this problem can be Interference Alignment. The key is that transmitters use linear precoding in a way, such that the interference signals are in the same subspace at the receiver, i.e., multiple interferers appear as one single interferer. In [SW11] the DoFs of a �(𝐵,𝑁) × (𝐾,𝑀)� cellular system are characterized by

𝑑 = 𝐵𝑁𝐵𝑀𝐵𝑀+𝑁

.

These DoFs are achieved by averaging over multiple fading slots. A possible scheme to achieve these rates is Cellular Iterative Interference Alignment presented first in [SW12]. This approach has been proven to drastically increase the rates in a non-cooperative multi-cell network.

Recently in 5GNOW this was generalized to CoMP including joint transmission and coordinated beam-forming schemes such as interference alignment using non-standard (and robust) alignment conditions for imperfect CSI [SWJ15][SWJ13]. The algorithm is based on “alternating optimization of precoders and receive filters”. Thereby the intra-cell interference is aligned in the receive subspace spanned by the out-of-cell interference. The algorithm has three basic components. First the algorithm performs a receive filter optimization where only the out-of-cell interference is considered. The optimization depends only on precoders used by the other base stations and not on the actual user selection. Second the algorithm performs a Greedy user selection which is based on the estimated rates. Users are added as long as the

D4.2


Page: 37 of 72

estimated sum rate increases. Third the precoders are optimized, where intra-cell interference is aligned in the receive subspace spanned by the out-of-cell interference. Thereby only the active users are considered.

The situation is quite different with imperfect CSI. With limited feedback, the conclusion is that the algorithm performs close to joint transmission for a good fraction of feedback rates and hence the standard analysis using DoF does not reflect the true performance. Also the requirements on the backhaul network are significantly smaller than with block diagonalization.

In the following we present a new robust version. Let T, be the number of base stations. Moreover we define 𝑆𝑏 to be the set of users selected at basestation b. Furthermore 𝜋 denotes a mapping from the set of scheduled users to the corresponding precoding vectors. With this notation the overall algorithm is given as follows.

Algorithm 1: Iterative Alignment Scheme Begin of transmission frame:

• Transmit common pilots to all users and make an estimate of the channels from each user to each base station

• For each base station select a subset of users • Set 𝜋(𝜋) = 1

�𝑛𝑡(1,1, … ,1)𝑇 for all 𝜋 ∈ {𝑆1,𝑆2, … , 𝑆𝑇}

Repeat

• Transmit dedicated pilots For b=1,2,…,T do

• Compute receive filter matrix 𝑠𝑘 for all 𝜋 ∈ 𝑆𝑏 • Quantize and feedback the effective channels

End for For b=1,2,…,T do

• Compute precoding vectors 𝜋(𝜋) for all 𝜋 ∈ 𝑆𝑏 End for Until termination condition is satisfied

End of transmission frame

It can be observed in simulations, that the proposed IA algorithm converges rapidly, i.e., going from 2 to 6 iterations the performance is increased only slightly. This is a remarkable result since in each iteration, each user needs to feedback TB = 48 bits and, therefore, a small number of iterations keeps the feedback load small. It can be further observed that with 2 iterations and a codebook with 16 bit we already come very close to a centralized solution. Using a codebook with 16 bit and 6 iterations the algorithm clearly outperforms the centralized solution [SWJ15][SWJ13].

3.2.3.4 Influence of channel aging

In [MP11], where OFDM is considered, the authors demonstrate that joint transmission among multiple cells can provide significantly large throughput gains at the users’ side. Nevertheless, they point out the

D4.2


Page: 38 of 72

price to pay to achieve an efficient CoMP transmission; these identified drawbacks are below assessed against the assumptions and results of our study, which is based on FBMC:

• Precise synchronization of the oscillators (synchronous data exchange and clock synchronization between eNBs for proper timing of coordinated transmission). In our study cooperating BSs are considered synchronized in frequency but not a priori in phase. We have shown that both the CFO estimation and compensation can be realized at the UE side, in the frequency domain. These operations require no feedback from the UE to the BSs. We moreover demonstrate that FBMC allows an easier and more accurate estimation and compensation of the CFO than OFDM.

• Backhaul latency and capacity (need for fast exchange of scheduling information, CSI, BF weights, data). The backhaul link is the link on which cooperating BSs are exchanging, among other information, the data intended to the CoMP users. The rate on this link would be the same for OFDM and for FBMC. However one must have in mind that, as described in [MP11], the overall backhaul rate for two cooperating BSs can reach tens of Mbit/s/cell. BS sites thus need high-capacity backhaul (fiber or microwave).

• Need for potentially large CSI feedback (UL CSI feedback overhead for radio channels’ measurements to all cooperative points in the serving set). When dealing with MU-CoMP, most of the feedback is indeed devoted to providing to the BSs the necessary information for codeword and user selection. The robustness of FBMC to MAC related asynchronicities such as outdated CSI and feedback rate has been investigated in the next sections.

One last issue that is not mentioned in [MP11] is the necessary feedback for time synchronization at the UEs. Indeed in this book, as OFDM is considered, no time synchronization is required because all the ISI are suppressed by the use of a CP. To do so and when dealing with large cooperating areas, the size of the cyclic prefix can reach high values, leading to high spectral efficiency loss. FBMC does not require CP but as expected we have shown that time synchronization then become necessary. An algorithm has been provided that minimizes the feedback rate for this synchronization: only 4 bits are necessary on the UL to get to an acceptable synchronization. The refresh rate this information must be provided to the BSs depends on the speed of the UE. Of course, quantization is not the only source of imperfect or partial CSI. In fact, in a practical system, we are more concerned with the influence of CSI aging. In [SWJ13] it was also shown that, given the reported channel 𝝂 and the channel after feedback delay and processing time 𝒉,

𝐸[Δ𝑟𝑚(𝝅𝐻 ,𝝅𝑉)] ≤ 6log(1 + pNBS max𝛂||𝒉𝐻𝜶|2 − |𝝂𝐻𝜶|2|),

where 𝑁𝐵𝐵 is the number of base stations and 𝜶,𝒉,𝝂 ∈ 𝕊𝑁𝑇−1 with 𝕊𝑁𝑇−1 being the unit sphere in ℂ𝑁𝑇 . Further, it is shown that

max𝜶||𝒉𝐻𝜶|2 − |𝝂𝐻𝜶|2| = �1 − |𝒉𝐻𝝂|2 (1)

Considering (1), a straightforward option to incorporate the influence of channel aging is by replacing |𝒉𝐻𝝂| with |𝒉𝐻𝒉�|, where 𝒉� ≔ 𝒉−𝚫𝒉

‖𝒉−𝚫𝒉‖ where Δ𝒉 models the influence of outdated channel information.

Naturally, Δ𝒉 is difficult to capture analytically, even with very simple channel models, and needs to be evaluated by experiments. Alternatively, in literature (e.g., in [CJK+10]) there are approaches to use prediction methods to estimate Δ𝒉. However, this is out of the scope of this paper. In the following we investigate by simulations the influence of limited feedback, in particular the influence of outdated CSI, on the necessary feedback overhead and the throughput of practical systems using CoMP with different waveforms.

Algorithm to deal with channel aging

The MU-MIMO CoMP scheme considered here falls into the category of joint processing/joint transmission where simultaneous data transmission occurs from multiple points to multiple UEs using the same time-

D4.2


Page: 39 of 72

frequency resource. In this framework, on each Resource Block (RB), the two BSs schedule a MU-MIMO cooperative transmission with two UEs. To do so, for each RB and for each UE, the BSs apply a weight vector chosen from a finite set, named the codebook, of 2𝑁𝑇 vectors (the codewords). Codebooks for 𝑁𝑇 = 2 and 𝑁𝑇 = 4 can be found in [3GP13c], and relies on TM9 for PDSCH transmission for UEs beyond Rel-10. Codebooks are further divided into 𝑁𝑇 unitary codebooks (UCs) composed of 2𝑁𝑇/𝑁𝑇 codewords each. 𝐰𝑢,𝑖 ∈ ℂ𝑁𝑇×1,𝑠 ∈ 𝑆𝑈𝑈 , 𝑑 ∈ {1,2, … , 2𝑁𝑇/𝑁𝑇}, is the weight vector of index 𝑑 chosen in UC 𝑠. 𝑆𝑈𝑈 is the set of UCs that are allowed for feedback and transmission. Note that the same weights are applied on all the carriers of the RB. Let 𝐇𝑟,𝑘 ∈ ℂ𝑁𝑇×𝐿Hr,k ∈ CNT×L be the channel frequency response between both BSs and UE 𝜋 on RB 𝑟, with 𝐿 being the number of carriers in a RB. Consequently, 𝐡𝑟,𝑘(𝑓) ∈ ℂ𝑁𝑇×1 is the channel response between both BSs and UE 𝜋 on carrier 𝑓 of RB 𝑟. The proposed algorithm operates independently on each RB, in the following the subscript 𝑟 is therefore dropped for notational simplicity. UEs that are likely to be scheduled for MU-CoMP transmission from BS1 and BS2 are uniformly distributed at the edge of cells BS1 and BS2. It is assumed in the following that each UE 𝜋 has perfect knowledge of the channel 𝐇𝑟,𝑘 on all RBs r. We consider the following procedure for CSI feedback generation at the user side and corresponding scheduling at the base stations.

• At the UE: for each RB, each UE 𝜋 that is likely to be scheduled for MU transmission computes 𝑈 × 2𝑁𝑇(𝑁𝑇 − 1)/𝑁𝑇 SIR values 𝛾𝑖,𝑗𝑢 , with 𝑈 = |𝑆𝑈𝑈| being the number of unitary codebooks that are allowed for feedback and transmission. The SIR is given by

( )( ) { } { }

21

,,

0 ,

1 , , 1, 2,..., , ,HLu i ku

i j T UCHl u j k

li j N i j u S

L lγ

−

=

= ∈ ≠ ∈∑w hw h

In each set 𝑠 of 2𝑁𝑇(𝑁𝑇 − 1)/𝑁𝑇 SIR values, the UE selects the index (𝑠, 𝑑) of codewords 𝐰𝑢,𝑖 that maximizes 𝛾𝑖,𝑗𝑢 . The UE then computes the expected SINR at its input if 𝐰𝑢,𝑖 is used at the BSs, based on the SIR 𝛾𝑖,𝑗𝑢 and on the knowledge of the SNR. For each RB, each UE 𝜋 likely to be scheduled for MU transmission then reports the following quantities to the BSs:

o The 𝑈 indices (𝑠, 𝑑)(𝑘),𝑠 ∈ 𝑆𝑈𝑈 , 𝑑 ≤ 𝑁𝑇, of preferred weight vectors 𝐰𝑢,𝑖 o The 𝑈 MCS indices 𝜗𝑢,𝑖

(𝑘) that correspond to the U computed SINRs values.

• At the BSs: For each RB 𝑟, the BSs must find the best pair (𝜋, 𝜋′) of UEs for MU transmission. This pair is such that: o (𝑠, 𝑑)(𝑘) and (𝑠, 𝑗)�𝑘′� are the indices of two different codewords in the same UC, 𝑑 ≠ 𝑗. o min �𝜗𝑢,𝑖

(𝑘),𝜗𝑢,𝑗�𝑘′�� is maximized over 𝑠 and over the UEs.

On each RB, all UEs compute their preferred codewords based on the estimation of the channel at time 𝑡. The codewords and the corresponding MCS indices are then fed back to the BSs for scheduling. Let 𝜏 be the feedback delay (see Figure 24), which corresponds to the time needed for the feedback and the scheduling process plus the time between two computations of weights at the UE. At time 𝑡 + 𝜏, the channel conditions have changed, causing a change on the SINR value at the UEs and on the PER. Therefore, the feedback delay has an influence on the throughput of UEs, depending also on the velocities of these UEs.

D4.2


Page: 40 of 72

Figure 24: Delay between two computations and feedback of indices at the UE

The performance has been assessed in terms of achieved throughput at the UEs, [CKW+14]. We measured the influence of the feedback delay on the throughput for several velocities of the UEs. The BSs are equipped with two transmit antennas, 10 UEs are randomly drawn for each run in the cooperation area and four different speeds of UEs are considered: 1, 3, 10 and 20 km/h. Figure 25 shows the distribution of the throughputs of UEs for FBMC and OFDM (OFDM with dashed lines, FBMC with continuous lines), for 𝑆𝑈𝑈 = {1,2}. On each figure and for each modulation, 3 curves are plotted, corresponding to 𝜏 = {0,∆(𝑡), 2∆(𝑡)} ms. The correspondence between ∆(𝑡) and the speed of the UEs is given in Table 2. Two observations can be made from this figure. Firstly, FBMC outperforms OFDM whatever the feedback delay 𝜏. These gains from FBMC to OFDM are primarily caused by the removal of the cyclic prefix. Secondly, the degradation of the throughput when 𝜏 = ∆(𝑡) compared to 𝜏 = 0 is significant. At 20 km/h (∆(𝑡) = 13.5 ms), the weight vectors must therefore be updated frequently.

Figure 25. CDF of the throughput of UEs. (OFDM: dashed lines, FBMC: continuous lines)

D4.2


Page: 41 of 72

Table 2. Correspondance between the time interval and the velocity of the UEs.

velocity [km/h] ∆(𝒕) [ms]

1 270

3 90

10 27

20 13.5

The total throughput of the UEs vs. the feedback rate per UE was plotted in Figure 26 for two transmit antennas, for SUC = {1,2}. The feedback per RB per unitary codebook is NT bits for the prefered weight and 4 bits for the MCS. As |SUC| = 2 the feedback is 2(NT+4) bits per RB. It must be noted that FBMC once again outperforms OFDM. With a feedback rate of 20kbps, FBMC UEs moving at 1km/h reach a throughput of 24,7Mbps whereas the throughput reach by OFDM UEs is only 23.6Mbps. As intuitively expected, the speed of the UEs has a significant impact on the necessary amount of feedback: to ensure a throughput of 24.7Mbps, 60kbps are necessary at 3km/h and more than 100kbps at 10km/h. Furthermore, it can be observed that the FBMC curves have a slope similar to the slope of the corresponding OFDM curves. Thus, FBMC shows a comparable robustness against limited feedback.

Figure 26. Total throughput as a function of the feedback rate. (OFDM: dashed lines, FBMC: continuous lines)

When different users are scheduled on adjacent RBs with different precoders, as in the study above, the orthogonality between edge carriers of RBs, previously ensured by the O-QAM modulation, is broken by precoders. On Figure 27 (right hand side) the CDF of the SINR on three carriers has been plotted. The system configuration used for these measurements is shown on Figure 27 (left hand side): UES1.1 and UES1.2 are co-scheduled on the same GRBs p+i, i odd, of the set of GRBs S1. Orthogonal precoders wS1.1 and wS1.2 are used to separate these UEs. The UE of interest is UES2; it is scheduled on the GRBs p+i, i even, of the set of GRBs S2. S1 and S2 are interleaved as shown on Figure 27 (left hand side). SNR has been set to a high value, 33 dB. The Modulation and Coding Scheme (MCS) is 16QAM 1/2. SINR on carriers B and C is dominated by the noise whereas on carrier A the SINR is significantly lower than the SNR. SINR is thus dominated by interference on this carrier.

D4.2


Page: 42 of 72

Figure 27. (left) system configuration (right) cdf of the SINR on edge carriers. SNR = 33 dB.

An easy and immediate way to mitigate the interference depicted above is thus to set one Guard Carrier (GC) at the edge of each GRB: the last carrier on the right side of each GRB (carrier D on Figure 27 left) is left blank. Similarly to Figure 27 (right hand side), Figure 28 shows the CDF of the SINR on the three edge carriers A, B and C, but when GC are used. It must be noticed that all the three carriers now experience the same SINR, which is dominated by the noise level. One GC is therefore sufficient to remove all the interferences.

Figure 28. CDF of the SINR on edge carriers with GC. SNR = 33 dB.

The impact of the insertion of this GC on the capacity has been measured. The uncoded channel capacity as introduced in [Ung82] was used. The capacity for three configurations has been computed and was plotted on Figure 29:

• FBMC without interference, CNoInterf (i.e. only the user of interest is scheduled). Note that this capacity is also the capacity of the equivalent OFDM system.

• FBMC with GC, CNotBoostedGC. Here the carriers are not boosted, i.e. their power is the same than for CNoInterf.

• FBMC with GC and boosted carriers, CBoostedGC. The power of constellation symbols is boosted so that the total transmit power is the same than in the scheme without GC.

Here for each UE the scheduling unit is two consecutive RBs, i.e. 24 carriers. The required SNR to reach a BER of 10-3 for each modulation in the SISO uncoded AWGN channel has been computed and is specified on the figure (black circles). For each modulation, this SNR is assumed to be enough to obtain good

D4.2


Page: 43 of 72

performance in a frequency selective channel with channel coding. Firstly, it must be noticed that, as expected, whatever the constellation size and the SNR, CNotBoostedGC is always (23/24=) 95.8% of the reference capacity. Indeed in this scenario one carrier over the twenty four carriers of the two RBs is not carrying data. Because the boost of data is not taken into account in the SNR, one can secondly observe that for very low SNRs (less than 4 dB), CBoostedGC is higher than CNoInterf. Finally, the capacity CBoostedGC is shown to be close to the reference capacity for reference SNR: 99.4% for QPSK, 98.6% for 16QAM and 97.7% for 64QAM.

Figure 29. Capacity of UES2 in the uncoded SISO AWGN channel, compared to CNoInterf. Two RBs per user.

Red stars: CNotBoostedGCBlue squares:CBoostedGC

D4.2


Page: 44 of 72

4 System Level Simulation

4.1 5GNOW waveform PHY abstraction modelling example

In order to verify the performance of the new waveforms provided by 5GNOW WP3 it is necessary to combine system level simulations together with abstraction layer of a waveform for this purpose (e.g. FBMC/GFDM/UFMC abstraction layer). The abstraction layer requires analytical definition and Matlab implementation.

5GNOW PHY abstraction modelling 4.1.1

Issues such as multiplexing, channel coding and modulation, signal processing, equalization and detection, etc., are usually investigated using link level simulations. It would entail a tremendous overhead if this was to be carried out in system-level simulations which typically consider multiple users and cells and where the focus lies on scheduling, interference, mobility, etc. In order to avoid having to carry out actual signal processing, in system level simulations usually a link-to-system interface comprising a suitable PHY layer abstraction is used. The goal of this PHY layer abstraction is essentially to obtain a block error rate (BLER) for a transport block (given particular subcarrier channel realizations) without having to carry out real signal processing. This is usually implemented in form of a look-up table where packet or block error rates are given for a certain effective quality measure, often the SINR, taking into account modulation and coding, power control, resource allocation, HARQ, etc. [WKB+13]. Using the SINR as quality measure, a crucial element of the PHY abstraction is the Effective SINR Mapping (ESM). In order to obtain a BLER a so-called effective SINR for a whole transport block of a given user is needed, since in complex systems simply averaging the SINRs does not provide a satisfying accuracy of the interface. This comes with a number of challenges. First, the transport block of bits is spread across (possibly non-orthogonal) subcarriers and each individual subcarrier may experience different SINR. Second, interferences from different base stations may have different influence on different subcarriers. Also, each user may use different parts of the available bandwidth and the BLER for each of these parts may be calculated separately (since the user may use different modulation and coding schemes (MCS) on its allocated resources) [WKB+13]. The required elements from WP3 to define PHY abstraction for new 5GNOW waveform (Link-to-system interface) include:

• “SINR-per-subcarrier” calculation method (e.g. including additional interference from neighboring subcarriers due to “non-orthogonality”)

• SINR mapping model – to get SINR per resource block for CSI to be provided for link adaptation algorithm

• SINR mapping model – to get overall effective SINR for transmission evaluation • Resource block, frame parameters definition (Important to calculate throughput per resource

block. This may require also separate definitions for DL and UL or using just a single transmission direction.),

o e.g. defining how many “resource elements do we have per resource block” o what MCSs are used o what is the percentage of the signalling that need to be taken out from the resource

elements (however for the first iteration we may treat all the resources as available for data transmission and use the same approach for standard LTE system parameters).

D4.2


Page: 45 of 72

• SNR – BLER curves for selected FEC implemented within particular system • Interference modelling (if different from OFDMA systems).

Performance evaluation of 5GNOW waveforms in system level environment (OFDM and FBMC 4.1.2comparison)

The first scenario for system level evaluation in 5GNOW is comparison of original OFDM-based LTE system with FBMC [WKB+13]. The simulation scenario setup is presented in the Figure 30 below. The PHY layer of the LTE system for comparison is emulated and abstracted in ISW’s system level simulator, LTE MAC Lab. Additionally an interface is provided to incorporate other waveforms. LTE MAC Lab is firstly used for FBMC and OFDM comparison within WP4 of 5GNOW. All the parameters and models except PHY layer abstraction (e.g. schedulers, link adaptation, radio propagation, eNB/UE models, etc.) are reused for the simulation.

Figure 30: System Level simulation scenarios for OFDM and FBMC comparison

Answering for requirements imposed by Section 4.1.1 in this document, below we present a definition of the following elements of FBMC PHY abstraction definition:

• Frame structure; • “SINR per subcarrier”; • Effective SINR calculation; • SINR vs BLER evaluation; • Frequency offset impact.

D4.2


Page: 46 of 72

4.1.2.1 Frame structure

According to In order to carry out system-level simulations and for performance comparisons between different waveforms we need suitable frame structures in the simulation. The frame structure has to be matched in order to enable a fair comparison (here to have the same throughput) between the waveforms, that is, they have to support the same throughput. In the OFDM case we use 4 symbols for signalling/control etc. and 9 symbols for data with a total subframe duration of 0:93ms, while in FBMC we use 16 symbols for data with a total subframe duration of 1:63ms. A summary of the relevant modelling parameters is given in Table 3 below.

Table 3: Parameters of FBMC and OFDM/LTE frames

Parameter Value FBMC Value OFDM Unit Abreviation Useful bandwidth 107 107 Hz 𝐵 Number of carriers 1024 1024 𝑁𝑈 Number of active carriers 601 601 𝑁𝑓𝑐 Cyclic Prefix N/A 72 samples Carrier spacing 15x103 15x103 Hz ∆𝑓 Sampling rate 15.36x106 15.36x106 Hz 𝑓𝐵 Sampling period 65.1x10-9 65.1x10-9 s 𝑇𝐵 = 1/𝑓𝐵 Number of preamble symbols in the subframe

4 4

Number of data symbols in the subframe 16 9 𝐿 Subframe duration 1.63x10-3 0.93x10-3 s 𝐿𝑓𝑟𝑓𝑚𝑓 Overlapping factor 4 N/A 𝐾 Number of Resource Blocks 50 50

The comparison of the frame structures for FBMC and OFDM / LTE is shown in the Figure 31 below.

Figure 31: OFDM/LTE and FBMC frame structure

D4.2


Page: 47 of 72

4.1.2.2 “SNI(D)R per subcarrier” calculation

The SINR measure, which is usually applied in the evaluation of OFDMA systems, cannot capture the mutual dependencies between subcarriers of non-orthogonal multicarrier schemes. In our simulations we resolve to using the SNIDR model, as referenced in [WKB+13], where the “D” stands for the residual distortion. The SNIDR on subcarrier l is defined as:

SNIDR𝑘 =𝑃𝑠

𝑃𝑖 + 𝑃𝑛 + 𝑃𝑑

whereas

𝑃𝑑(𝜋) =2𝑃𝑠𝐾3 �

𝐻𝑘′ (𝜔)𝐻𝑘(𝜔)�

2

𝐶𝑓

In Rayleigh fading channel: 𝐻𝑘′(𝜔)

𝐻𝑘(𝜔) = 1𝑤𝑘− 𝑤𝑘

𝜎2, with 𝑤𝑘 = 2𝜋

𝑁(𝜋 − 1),𝜋 = 1, . . ,𝑁:

• 𝐻𝑘(𝜔), 𝐻𝑘′ (𝜔): channel frequency response and its derivative evaluated at the 𝜋′𝑡ℎ subcarrier

frequency.

• 𝑃𝑠,𝑃𝑖 ,𝑃𝑛: signal power, interference power, noise power.

• 𝐶𝑓: first order derivative of receive prototype pulse. The smoother the filter the lower 𝐶𝑓 and the lower is the distortion power.

4.1.2.3 Effective SINR calculation

The problem of finding an effective SINR arises with two granularities. First an average quality measure for a particular resource block needs to be found (mainly for providing CSI feedback). Here we can use either an ESM method or, if the coherence bandwidth is sufficiently large, a simple averaging over SNI(D)R values of the single resource elements. Second we need to get an overall SINR for the whole transmission (mainly to determine the block error probability). For this we definitely need to use the ESM mechanism. A widely used method is the so called exponential ESM (E-ESM). However this function is especially difficult to fit to the corresponding simulations since optimization over multiple parameters has to be carried out. By contrast it turns out that the mutual information ESM (MI-ESM) is particularly suited for our purpose since it provides a good match with AWGN curves without specific parameter tuning.

4.1.2.4 SNI(D)R vs BLER evaluation

The actual link-to-system interface is a mapping from an effective SINR (or SNIDR in case of FBMC) to a corresponding packet error rate. This mapping has to be evaluated and fitted against measured packet error rates coming from link-level simulations in an equivalent AWGN channel. Figure 32 below depicts the resulting curves generated with the particular elements discussed above. Thereby a convolutional encoder and Viterbi decoder was used. Already here, the gain in FBMC from abandoning the cyclic prefix can be clearly observed. Based on the generated results depicted in Figure 33, lookup tables can be generated for usage in system level simulations.

D4.2


Page: 48 of 72

Figure 32: SNR/BLER curves for OFDM and FBMC [WKB+13]

4.1.2.5 Frequency offset impact on system level evaluation

We also investigate the influence of (Doppler) frequency shifts (induced by the user velocity) on OFDM and FBMC. The details of incorporating the influence of frequency offsets in system level simulations for OFDM and FBMC are presented in [WKB+13]. We compare the same user velocities (3 and 120km/h) that we also consider in our system simulations. Figure 33, below shows the results for different original SNR values for OFDM and FBMC. On x axis we have the calculated average SNR, whereas y axis shows the SNR offset / SNR degradation by which the original SNR is decreased due to Doppler shift. It can be observed that the influence of frequency offsets in FBMC is much lower. In particular the influence of the Doppler offsets at lower SNR and lower speed is close to 0.3dB for OFDM and rather small for FBMC, while in a high-velocity, high-SNR regime the original SNR is diminished up to 1.2 dB for OFDM and 0.8 dB for FBMC.

Figure 33: SNR-offset vs. SNR for OFDM and FBMC with different user [WKB+13]

4.1.2.6 System level evaluation

The simulation is evaluating system performance for the two main scenarios: • Standard LTE scenario with LTE parameters’; • New waveform incorporated in the system level simulation using standard LTE mechanisms for

scheduling, link adaptation mechanism.

We use the following configuration: Two different user velocities, 3km/h and 120km/h and a total number of 50 users. Moreover we use a Max-SINR scheduler. The overall simulation time is 3:26s which amounts to 2000 FBMC TTIs and 3260 OFDM TTIs due to the frame structure and TTI definition as explained earlier. Additionally we average the results over six simulation runs. In this paper ideal feedback is used, thus CSI is

D4.2


Page: 49 of 72

instantaneously available at the base stations. We perform simulations with two different channel models, a 3GPP EVA channel with a coherence bandwidth of 398.41 kHz, and a 3GPP ETU channel which was modified to have a coherence bandwidth of only 108.06 kHz, in order to investigate the influence of high frequency selectivity. This is necessary since the additional distortion term in FBMC has only influence when the coherence bandwidth is smaller than the filter response in the frequency domain.

Figure 34, below shows CDFs of SNI(D)R values obtained with the two waveforms for the two channels in consideration. For a large coherence bandwidth we can observe small gains from FBMC, especially at higher velocity, which stems from the smaller susceptibility to frequency offsets, as described above. In case of high frequency selectivity these gains vanish due to the additional distortion term in FBMC which now becomes visible.

Figure 34: SINR-CDF comparison of OFDM and FBMC [WKB+13]

Figure 35 below, shows user throughput CDFs for the two channels. Gains from FBMC can be observed in both the “frequency flat” and in the “frequency selective” case. These gains are primarily caused by the removal of the cyclic prefix. In the case with higher frequency selectivity the relative gains are slightly smaller due to the additional distortion.

Figure 35: Throughput-CDF comparison of OFDM and FBMC [WKB+13]

This can be better observed in Figure 36 below where, for the ease of comparison, the average user throughputs for both waveforms and velocities are also depicted.

D4.2


Page: 50 of 72

Figure 36: Average throughput comparison of OFDM and FBMC [WKB+13]

4.1.2.7 Conclusions

Comparing the influence of frequency shifts induced by user velocities, it turns out that FBMC is less sensitive to such frequency offsets, although their overall impact is not strong enough to significantly change the performance. We expect a much more relevant impact of timing offsets, however, this is left for further investigations. From the first system simulation results using the described link-to-system interface, it can be concluded that FBMC is (as expected) more efficient with small frequency selectivity, since in this case the gains caused by not having cyclic prefix take the full effect. With larger frequency selectivity the gains expected from using FBMC become slightly smaller [WKB+13].

4.2 Performance evaluation of D-PRACH and ATA in system level simulation

Simulation description 4.2.1

The main concept of the system level simulation in the context of MTC D-PRACH is to combine the approaches of:

• one shot transmission utilizing concept of D-PRACH where there is PRACH preamble using BFDM and the data subcarriers carrying small amount of data

• assumption of autonomous time synchronization (open loop time synchronization) by utilizing concept of ATA

• above procedures to shorten the transmission of non-synchronized MTC device that has little data to transmit.

Simulations and analytical calculations were conducted to show that utilizing these three components (provided mainly by combination of Data and PRACH preamble) leads to the significant improvement in a MTC scenario versus LTE case:

• decrease of signalling overhead required to convey little messages in MTC scenario; • thus increasing available system capacity; • and additionally increasing the standby time of UE that does not need to wake up for a long time to

transmit the little amount of data thus saving battery power (and decrease of latency).

To evaluate the performance of D-PRACH concept according to the above we provide the system level simulation. The overall simulation includes certain MTC traffic model suggested by 3GPP with a single cell supporting large number of devices (e.g. 30k) in a specific scenario (e.g. stationary or low mobility meters). Simulations include the power ramp-up mechanism for PRACH preamble transmission. We measure the time of delivery by calculating the average difference between ACK time stamp and PRACH preamble transmission. For LTE we assume the need to use standard Timing advance procedure thus sending the PDSCH message whereas the 5GNOW D-PRACH does not require it due to utilization of the ATA procedure.

D4.2


Page: 51 of 72

4.2.1.1 Main simulation parameters and assumptions:

• UL simulation only; DL assumed to be ideal (for scheduling message and ACK); • Single omni-directional cell and 6 surrounding interferers (i.e. UEs belonging to different cells

transmitting in the UL); • Inter-Site-distance = 500m (according to 3GPP simulation recommendations) • eNB antenna gain = 15dB; • Carrier frequency = 2GHz; • Uniform UE distribution; • UE power max = 23dBm; • Starting from minimum PRACH preamble transmit power and power ramp-up step (according to

3GPP specifications); • 3GPP pathloss model; • 2 PRACH opportunity per frame • 5MHz (according to 3GPP recommendations for MTC evaluation); • 5000, 10000 and 30000 UEs in a cell • Stationary or very low mobile UEs (3km/h)

4.2.1.2 High level simulation description

Single cell environment is assumed. The simulations show the packet delay (time since the application layer sends the data to lower layers until the packet is received) when a different number of MTC devices access the network. The number of MTC devices is set to 5000, 10000 and 30000 devices (see Table 4). Users arrive randomly spread during one minute, transmit the packet and leave the system. The MTC devices are assumed to be in RRC Idle. Application layer packets with the size of 24 bytes are transmitted to the device using UDP/IP (see Table 4). RLC AM is configured [TR 37.868]. An error-free wireless channel without capture effect was considered. In the simulation, it is assumed that the network reserves L RAOs per second for MTC devices to transmit their access attempts. Time is divided into fixed-length slots. Some of the slots are reserved as RACH slots for MTC devices to transmit their access attempts. The RACH slots are uniformly distributed in time. In other words, the time intervals between any two successive RACH slots are equal. The MTC device which arrives in the time interval between two RACH slots shall wait and transmit its access attempt at the beginning of the next RACH slot. The access attempt is transmitted through a randomly chosen RAO in the RACH slot. The access attempts are collided if more than one access attempt choose the same RAO in the same RACH slot. The collided access attempts, which are referred as the backlogged access attempts, herein, shall perform random backoff for further retransmission [CWT12]. The UE performs a power ramping procedure, where it increases its power for the subsequent preamble transmission if the UE is not granted access due to a preamble detection miss or contention, as described in [TS 36.321].

4.2.1.3 MTC traffic model

Table 4 below shows selected In-Home M2M application traffic parameters, such as average transaction time, message size and equivalent data rate [EC10].

D4.2


Page: 52 of 72

Table 4 In-home MTC application traffic parameters[18]

MTC application / Use Case

Average Message

Transaction Rate/s

Average Message Size (Bytes)

Date Rate (b/s)

Overhead (Bytes) Number Per

Home

Traffic distribution

Smart electric Meter

1/300 (1 read message every 5 minute)

Meter periodic data read: 24bytes: *Real power import / export[kW] – 8B * Reactive power import / export[kW] – 8B * voltage – 4B * power factor – 4B

0.64 8bytes (TCP/IP profile with GPRS used in smart metering project in Netherlands max 60/min 20bytes. We assume 50bytes and RoHC going down to 8bytes) Total packet size = 24+8 = 32bytes = 256bits (QPSK 128REs)

1 Uniform

We assume MTC traffic model 1 from Table 5 [TR 37.868][LS+13][LSN+13].

Table 5 Traffic models for MTC [TR 37.868]

Traffic model 1 can be considered as a realistic scenario in which MTC devices access the network uniformly over a period of time, i.e. in a non-synchronized manner. Traffic model 2 can be considered as an extreme scenario in which a large amount of MTC devices access the network in a highly synchronized manner, e.g. after a power outage.

4.2.1.4 Impacts on/from H2H traffic

We assume an approach where there are separate resources for MTC and H2H – thus RACH preambles dedicated for MTC and H2H traffic presented in [TR 37.868] and [LKY11] as an option for RAN improvements for massive transmission opportunities. For such assumption and RACH capacity evaluation, all RACH attempts are assumed to be initiated by MTC devices with no background noise caused by H2H UEs. An assumption from [TR 37.868] is taken where in addition to the RACH intensity generated by smart meters, the cell also has to cope with the RACH intensity generated by normal UE traffic. An analysis of the busy hour traffic for the London area on the Vodafone UK 2G network shows that for the urban area under consideration, a busy hour traffic density of up to 100 Erlangs/SqKm can be expected. The mean call holding time was around 43 s. This means that the corresponding busy hour voice call arrival rate is 2.32 calls/s/sqKm. For a circular cell of typical radius of 2Km, the expected call arrival rate is 29 calls/s. This is equivalent to a RACH intensity of 29 RACH attempts/s. Hence around 15 RACH preambles are required to support such as voice traffic intensity. According to the above we assume that the split of the preambles among MTC and H2H is done in a such way, that H2H traffic has enough preambles and opportunities to handle its traffic. This split is signaled to the UEs using System Information in BCCH (i.e. to identify a group of preambles dedicated to smart meters).

Characteristics Traffic model 1 Traffic model 2

Number of MTC devices 5000, 10000, 30000 5000, 10000, 30000

Arrival distribution Uniform distribution over T Beta distribution over T,

Distribution period (T) 60 seconds 10 seconds

D4.2


Page: 53 of 72

4.2.1.5 LTE FDD RACH parameters / settings

Tables 6-7 gather the main simulation parameters for RACH capacity evaluation [TR 37.868].

Table 6 Basic simulation parameters for LTE RACH Parameter Setting

Cell bandwidth 5 MHz PRACH Configuration Index 6 (2 preambles per Frame) Total number of preambles 64

Number of preambles for MTC 49 = 64(all) -15(dedicated for H2H)

Maximum number of preamble transmission 10 Number of UL grants per RAR 3

Number of CCEs allocated for PDCCH 16 Number of CCEs per PDCCH 4

Ra-ResponseWindowSize 5 subframes mac-ContentionResolutionTimer 48 subframes

Backoff Indicator 20ms HARQ retransmission probability for PUSCH Data 10% Maximum number of HARQ TX for PUSCH Data

(non-adaptive HARQ) 5

PreambleInitialReceivedTargetPower Modelled by 4.7 equation 1 PowerRampingStep Modelled by 4.7 equation 1 PreambleTransMax 4

Table 7 PRACH configuration [3GP13c]

PRACH Configuration

Index

Preamble Format

System frame

number

Subframe number

6 0 Any 1, 6 In [TR 37.868] RACH intensity generated by smart meters is evaluated. The total number of RACH attempts per second depends on the PRACH configuration index as described in [3GP13c]. Table 8 below summarizes some possible values of RACH opportunities/s/preamble for different PRACH configuration index value for LTE frame structure type 1 (FDD).

Table 8 Number of RACH opportunities/s/Preamble for Frame Structure Type 1 (FDD)

PRACH Configuration Index

% resources consumed in a 5MHz bandwidth

Number of RACH opportunities/s/preamble

0 1.25 50 6– Our case according

to Table 6 above 5 200

9 7.5 300 12 12.5 500 14 25 1000

In our scenario LTE, the RACH could be configured to occur once every subframe up to once every other radio frame. We have assumed that the RACH occurs every 5 ms. 15 preambles are configured to be dedicated; therefore, the other 59 can be used for the random access for MTC. Considering these assumptions, we end up having 200 RACH opportunities per second and a total of 9800 preambles per second.

D4.2


Page: 54 of 72

4.2.1.6 RACH protocol level simulation methodology [TR 37.868][CWT12]

Firstly collision probability is calculated and in case of no collision – detection probability is calculated. For LTE FDD, if two (or more) MTC devices select the same preamble at the same time, it is assumed that the eNB will not be able to decode any of the preambles; hence, the eNodeB will not send the Random Access Response (RAR). MTC devices will only detect a collision if Msg2 (RAR) is not received in the RA-Response Window. Collision is verified independently of received power from individual UEs [TR 37.868][CWT12]. Handling of detection: In case of no collision, 1 − 1

𝑓𝑖 preamble detection probability is assumed, where i

indicates the i-th preamble transmission, to take into account the effects of radio channels, for example path-loss, fading, inter-cell interference, as well as power ramp-up procedure [TR 37.868][TS 36.321][TS 36.902].

4.2.1.7 PRACH transmission procedure

In our simulation, transmission is modelled in the following way: • Packet is randomly generated according to traffic model 1 (Table 4 above) • A UE is selected randomly (uniform distribution) to transmit this packet (each UE is assumed to

transmit packet within distribution period T. • Random preamble is assigned for a particular UE (one out of M2M_preamble_set) • The closest TTI is selected with PRACH opportunity • If at least two UEs are to transmit within the same PRACH opportunity, using the same preamble

index, collision occurs. In that case, the next PRACH opportunity (within backoff window) and a new preamble index are selected.

• If not we apply the detection probability formula (1).

Simplified algorithm flow is presented in the Figure 37.

Figure 37 Simulation algorithm flow

D4.2


Page: 55 of 72

4.2.1.8 D-PRACH procedure processing flows

Figure 38, Figure 39, Figure 40, and Figure 41 show the signalling flows for D-PRACH procedure.

Figure 38 D-PRACH procedure - successful reception of preamble and data in 1st transmission

Figure 39 D-PRACH procedure - successful reception of the preamble, not successful reception of the data (scheduled

retransmission needed)

Figure 40 D-PRACH procedure - non-successful reception of the preamble and data (D-PRACH retransmission)

Figure 41 D-PRACH procedure - non-successful reception of the preamble and data (D-PRACH retransmission and scheduled

retransmission)

D4.2


Page: 56 of 72

Analytical calculations and simulation results 4.2.2

4.2.2.1 Processing latency

[TR 36.912] and [ZNK12] presents describes the timing of UL channel access. The processing time and transmission delay of a particular message delivery is shown in Figure 42 [ZNK12] below for LTE-standard case.

Figure 42 UL channel access method timing [ZNK12]

The assumed LTE-standard timing parameters for procedure described in Figure 20 a) are gathered below. • Waiting time for preamble transmission – 0.5 up to 5ms • Preamble transmission time – 1ms • Preamble processing time at eNB – 2ms • RAR transmission time – 1ms • RAR processing time and PUSCH preparation at UE – 2ms • PUSCH transmission – 1ms • PUSCH processing at eNB – 3ms • PHICH (ACK/NACK) transmission – 1ms • PHICH processing time and PUSCH preparation for retransmission (if NACK) – 3ms • PUSCH transmission (if NACK) – 1ms • PHICH (ACK/NACK) transmission – 1ms

Total time for processing if first ACK = 11.5 up to 16ms Total time if second ACK = 16.5 up to 21ms Data packet size - 24bytes (see Table 4). Throughput for lowest transmission processing time – 24bytes * 8 bits/byte / 11.5ms = 17kb/s The assumed D-PRACH timing parameters for procedure described in Figure 20 b) are gathered below.

• Waiting time for preamble transmission – 0.5 up to 5ms • Preamble and data transmission time – 1ms • Preamble and data processing time at eNB – 3ms • ACK/NACK transmission (if NACK then NACK and scheduling grant is transmitted) – 1ms • Grant processing and retransmission preparation – 3ms • PUSCH transmission – 1ms

D4.2


Page: 57 of 72

• PUSCH processing at eNB – 3ms • ACK transmission – 1ms

Total time for processing if first ACK = 5.5 up to 10ms Total time if second ACK = 13.5 up to 18ms Data packet size - 24bytes (seeTable 4). Throughput for lowest transmission processing time – 24bytes * 8 bits/byte / 5.5ms = 35kb/s

4.2.2.2 Resource calculation / overheads and signalling

According to [TR 36.822] the following parameters shall be taken into account while evaluating MTC scenario:

• Signalling costs should be evaluated • System resource overheads (e.g. in terms of number or fraction of assigned/used/reserved control

channel resources and RBs) should be considered • Effects on the average time spent in connected mode (vs. idle) should be reported where

appropriate. Note that the average time spent in connected mode can also be used to derive the impacts on the number of simultaneous RRC connections required within a cell or eNB site.

The UL PRACH resource usage is given as follows: • PRB usage of single PRACH opportunity = 1.08MHz and 1TTI = 6PRB per slot * 2slots per subframe =

12PRBs • PRB usage of all PRACH opportunities in a frame = single PRACH opportunity * number of

opportunities = 12PRB * 2 = 24PRB • PRB usage for all PRACH opportunities in a second = all PRACH opportunities in a frame * number

of frames in a second = 24PRB * 100 = 2400PRBs • Total number of PRBs in a second = number of PRBs in BW in a slot * number of slots in a frame *

number of frames in a second = 25PRBs/slot * 20slots/frame * 100frames/s = 50000PRBs/s • PRACH ratio = PRACH PRB usage in a second / total number of PRBs in a second = 2400/50000 =

0.048 = 4.8% of resources consumed by PRACH in our scenario

The UL PUSCH data resource usage (assuming 2PRB (180kHz)) is given as: • Number of RE per single allocation = 7 symbols/PRB * 12 subcarriers * 2PRB = 168REs • Number of REs for pilots = 12 subcarriers * 1symbol*2PRB = 24REs • Number of REs for PUSCH data = 168 REs – 24REs = 144REs • Number of bits per allocation = Number of REs for PUSCH data * modulation Order = 144 REs *

2bits/RE (QPSK) = 288bits • Number of bits for data packet = 256bits (see Table 3) • Number of coded bits for data packet = number of bits / coding rate = 256 * 3 = 768bits • Number of REs for coded bits of data packet = number of coded bits / modulation order = 768/2 =

384 • Number of required SRBs (2xPRB) for data packet transmission = ceiling(number of coded bits for

data packet / number of bits per PRB) = ceiling(768/286) = 3SRB (6PRB)

The DL Random Access Response (RAR) resource usage is given as: • Number of bytes per RAR ([TR 36.822]) – MAC PDU Size = 8bytes

D4.2


Page: 58 of 72

• Number of bits per RAR (transport block size) = 8bytes * 8bit/byte = 64bits • Number of encoded bits = number of bits per RAR / coding rate * modulation order = 64bits / 1/3 *

2bits/symbol (QPSK) = 64 * 3 * 2 = 384symbols = 384REs • Required number of PRBs per subframe = 4PRBs

The additional signalling associated with a typical PRACH is given by • PDCCH – DL allocation for RAR • PHICH – ACK for PUSCH data

Regarding the overall LTE standard resource occupation, the following Calculation of the number of REs for the overall data delivery procedure can be made:

• UL transmission: PRACH – 12 PRB = 12*84RE = 1008RE • DL transmission: PDCCH allocation DL – 4CCE= 144RE • DL transmission: PDSCH RAR – 4PRB = 336RE • UL transmission: PUSCH Data – 6PRB = 504RE (including 384RE for actual data) • UL transmission: DRS for PUSCH – 12RE/PRB * 6PRB = 72RE • DL transmission: PHICH ACK/NACK – 12RE = 12RE

Total number of RE for signalling = 1008 +144+336+12+72+504-384 = 1692 Total number of RE for data = 384 Total number of used REs = 1692+384 = 2076 Ratio of data over all resources = 384/2076 = 0.185 (18.5% of resources is for data) Ratio of data over signalling = 384/1692 = 0.227 Regarding the overall D-PRACH standard resource occupation, we carry out the following calculation of the number of REs for the overall data delivery procedure (from figure Figure 20 b):

• UL transmission: D-PRACH preamble part – 12 PRB = 12*84RE = 1008RE • UL transmission: D-PRACH Data part – 384RE = 384RE • DL transmission: ACK/NACK – 12RE (taking LTE as example) = 12RE

Total number of RE for signalling = 1008 + 12 = 1020 Total number of RE for data = 384 Total number of used REs = 1020 + 384 = 1404 Ratio of data over all resources = 384/1404 = 0.274 (27.4% of resources is for data) Ratio of data over signalling = 384/1020 = 0.376

4.2.2.3 Retransmission and collision probability

During RACH procedure, where resources are not allocated by the base station, collisions may occur. Collision causes preamble retransmissions which lengthen the whole procedure thus its probability should be optimized. Reducing collision probability by increasing number of PRACH slots decreases overall throughput by occupying more resource elements by signaling. Increasing number of terminals connected to single base station affects the collision probability in manner shown in the Figure 43. If collision probability exceeds threshold assumed for the network it can be lowered by increasing either number of possible preambles dedicated to the MTC devices or RACH slots. During simulations occurrence of the collision is calculated in the same manner for both LTE case and new procedure. Also SINR values used for evaluating ACK/NACK response are the same.

D4.2


Page: 59 of 72

Figure 43: Collision probability vs number of PRACH preambles

System level simulations were also used to evaluate how interferences caused by lack of the closed-loop synchronization can be reduced using ATA. Simulation case includes Random Access procedure and packet transmission over PUSCH channel.

Figure 44: Average number of retransmissions

Results obtained during simulation shows that performance of the network can be kept on high level using ATA mechanism. Figure 44 shows how does the average number of retransmission is affected by the interferences caused by the open-loop synchronization in UFMC while using ATA mechanism. Even users that are located far from eNB does not suffer much from lack of closed-loop synchronization.

D4.2


Page: 60 of 72

5 Appendix A – Detailed description of Dual Connectivity and Coordinated Multipoint

Appendix contains extended description of the LTE-A features introduced in Section 1.1 of this document.

5.1 HetNet LTE-Advanced Rel. 11 and Rel. 12 features

HetNet dual connectivity 5.1.1

5.1.1.1 Downlink/Uplink imbalance

In HetNet scenario, the eNBs (enhanced network nodes) might have different DL output powers, e.g. macro-eNBs with high output power and pico-eNBs with low output power, and the cells may have also different UL power control settings. Due to this fact, a so called “DL/UL imbalance” situation may occur for some UEs. DL/UL imbalance is presented in the Figure 43 below. The location of the UE and macro/pico eNBs is depicted on the X axis, whereas the received signal strength is depicted on the Y axis of the two graphs. The DL scenario is presented in the top graph, whereas the UL scenario in the bottom graph. The imbalance comes from the fact that in the DL the UE receives higher power from the macro eNB due to higher macro transmit power, and in the UL the pico eNB receives higher Rx power, than macro eNB due to lower path loss to pico eNB. Such effect sets different cell borders to both directions. This also means that the node with the strongest received signal at the terminal/UE (that is typically used for DL transmission) may not be the node with the lowest path loss to the terminal/UE (and thus not the best node that can provide UL reception) [3GP13][ADF+13]. The solutions for this problem are presented below.

Figure 45: DL-UL imbalance in HetNet scenario

D4.2


Page: 61 of 72

5.1.1.2 Dual connectivity concept from 3GPP Release 12

The 3GPP originally used the concept of CRE (cell range extension) within Rel. 10 and Rel. 11 to overcome DL/UL imbalance and decrease interference. In CRE mechanism UE is connected to the node with lowest path loss. In such a case, if the UE was connected to pico eNB, the macro eNB decreased the DL transmit power in the sub-frames in which UE was allocated resources by pico eNB. While measuring path loss some offset value is added to the result, which causes virtual cell expansion. Such approach sets threshold somewhere between thresholds coming from DL and UL. Additionally 3GPP started Work Item of “dual connectivity” (also known as “soft cell” or “phantom cell”) as stated in [3GP14] and provided detailed description in [3GP13]. According to [3GP13], dual connectivity refers to the operation “where a given UE is allocated radio resources provided by at least two different network points connected with non-ideal backhaul”. Each eNB involved in dual connectivity for a UE may have different role. These roles are not necessarily depending on the given eNB transmit power and can vary among various UEs in the cell. This idea utilizes close communication between macro and pico nodes, where the UE is connected for basic signaling and data transmission to macro cell and sets high-speed data link to pico eNB (and/or for DL reception to macro cell and transmits in UL to pico cell as described further on in this chapter). Figure 44 below shows the concept of dual connectivity; it illustrates the terminal that is connected [ADF+13]:

• to the wide area layer through an anchor connection (to the macro-node) used for system information, basic radio resource control (RRC) signaling, and possibly low-rate/high-reliability user data.

• to the local area layer through an assisting connection (to the low power node) used for large amounts of high-rate user data.

Figure 46: Soft cell / Asymmetric node assignment concept from 3GPP.

Dual connectivity to the anchor node and assisting node.

The small cells extend the macro cell with the same PCI or with dynamically assigned VCI (virtual cell identities) which is re-used in the spatial domain. The Rel. 10-Rel. 11 eICIC mechanisms are replaced by dynamic creation of such “virtual” or “soft” cells, i.e. physical eNB for DL and physical PeNB for UL create a logical eNB that is not seen by the UE. The small cells provide a fast data pipe with improved link budget due to the close proximity of the UE to the network node. The macro cell on the other hand is providing cell wide system information as well as radio resource control including traffic steering, carrier selection and supporting the UE to detect a small cells that are close by. The challenge of this solution is, that the architecture requires a tight integration of the small cells into the network by ultra-fast interfaces or preferable by a central baseband processing unit. Another perspective of this solution is presented in [RMC12] and raises the issue of CQI reporting, i.e. CQI is reported to the closest node and thus lower transmit power is required, thus energy is saved. However in such scenario changes are required to power control since DL and UL path loss will be different. The possible scenarios for dual connectivity are as follows [3GP13]:

D4.2


Page: 62 of 72

• Scenario A: Both macro and pico eNBs are equipped with the same two carriers. The macro and pico eNBs apply Rel-10 CA to aggregate both carriers.

• Scenario B: Both macro and pico eNBs are equipped with one carrier which differs from one another. The macro and pico eNBs apply inter-node radio resource aggregation.

5.1.1.3 DL/UL split concept from dual connectivity

As already mentioned above a DL/UL imbalance may occur for certain UEs in HetNets due to large difference in transmit power of macro and pico nodes. This has a consequence of uneven load distribution between macro and pico cells (i.e. pico cells are less loaded than macro) and suboptimal UL performance (because UE is not necessarily connected to eNB with smallest path loss). Additionally there may be DL/UL traffic load imbalance: e.g. pico eNB may be highly loaded in DL while unloaded in UL. Therefore it may be beneficial to offload macro-cell UE’s data to the pico eNB while keeping UE’s DL traffic still in macro cell [3GP13]. An already mentioned the concept of DL/UL split comes together with Dual Connectivity where UE is connected to the nodes with separate connections for DL and UL. The nodes’ selection for the particular transmission direction is based on the lowest path loss. DL/UL split provides also the advantage to apply load balancing separately for DL and UL. This in turn enables to achieve optimal cell capacity in DL and UL. Due to that, network has the possibility to shift more UL traffic to the pico cell if the macro eNB is highly loaded in the UL, while keeping DL traffic still in the macro eNB [3GP13]. The architecture for DL/UL split is presented in Figure 45 below.

Figure 47: Example architecture for DL/UL split [3GP13]

5.1.1.4 Dual connectivity implications and benefits

To sum up, according to [ADF+13] dual connectivity concept may imply to the system the following: • Control and data separation where, e.g., the control signalling for mobility is provided via the

macro layer and at the same time low-power layer provides high-speed data connectivity. • A separation between DL and UL, where DL and UL connectivity is provided via different layers. • Diversity for control signalling, where RRC signalling may be provided over multiple links, which

may further enhance mobility performance. Dual Connectivity concept may provide the following benefits (over the macro assistance) [ADF+13]:

D4.2


Page: 63 of 72

• Enhanced support for mobility – by maintaining the mobility anchor point in the macro layer, it is possible to maintain seamless mobility between macro and low-power layers, as well as between low-power nodes.

• Low overhead transmissions from the low-power layer – by transmitting only information required for individual user experience, it is possible to e.g. avoid overhead coming from supporting idle-mode mobility within the local-area layer.

• Energy-efficient load balancing – by turning off the low-power nodes when there is no ongoing data transmission, it is possible to reduce the energy consumption of the low-power layer.

• Per-link optimization – by being able to select the termination point for UL and DL separately, the node selection can be optimized for each link.

5.1.1.5 Lean carrier

3GPP proposes for Release 12 a so called “lean carrier”, which is a DL frame structure not compatible backwards with Rel. 8 [NNB+13] [HLK+13]. Lean carrier does not include CRS, cell specific synchronization signals and broadcast channel – they are controlled by macro-cell signalling (i.e. visibility of the legacy type of info is provided by macro cell). [NNB+13] [HLK+13] present concept of lean carrier within dual connectivity architecture as shown in the Figure 46 below.

Figure 48: Lean Carrier concept within Dual Connectivity architecture

Use case scenario is the following: UE is synchronized to both, macro eNB and to low-power node at lean carrier, i.e. UE uses legacy carrier at macro cell for signalling and lean carrier at pico/home eNB for high speed data. This concept may utilize carrier aggregation (i.e. legacy component carrier at lower frequency band and supporting lean component carrier at higher frequency band strictly used for IP data).

CoMP in 3GPP Rel. 11 and Rel. 12 5.1.2

The latest 3GPP documents of Rel. 11 and Rel. 12 discussing CoMP issues include [3GP14] [3GP13a] [3GP13b].

5.1.2.1 CoMP scenarios in 3GPP

According to [3GP13b], the following four CoMP scenarios are considered for evaluation and further development for both DL and UL:

• Scenario 1: homogeneous network deployment with single eNB serving multiple sectors (i.e. intra-site CoMP) (Figure 47).

• Scenario 2: homogeneous network deployment with multiple high-transmit power RRH (Figure 48).

• Scenario 3: HetNet where macro-cell (high power RRH) and small-cell (with low power RRH) within macro-cell coverage are jointly deployed using different cell identities (PCI) (Figure 49).

D4.2


Page: 64 of 72

• Scenario 4: HetNet where macro-cell (high power RRH) and small-cell (low power RRHs) within macro-cell coverage are jointly deployed using the same cell identity (PCI) (Figure 49).

Figure 49: Scenario 1 - Homogeneous network with intra-site CoMP

Figure 50: Scenario 2 - Homogeneous network with high transmit power RRHs

Figure 51: Scenario 3/4 – HetNet with low power RRHs within the macro-cell coverage (with different – Scenario 3, or the same –

Scenario 4 PCIs)

5.1.2.2 CoMP processing options

As for now, both UL and DL CoMP have been considered by 3GPP. Due to their complexity and different requirements, these two directions are separated in the standardization process under two independent work sets. The following DL CoMP processing schemes are considered currently in 3GPP for PDSCH [3GP13b] (also shown in a simplified way in the Figure 50):

1) Joint Processing (JP). In that case data for a UE is available at more than one point in the CoMP cooperating set (Set of geographically separated points directly and/or indirectly participating in data transmission to a UE in a time-frequency resource) for a time-frequency (TF) resource. The following options of JP are possible:

a. Joint Transmission (JT). Simultaneous data transmission from multiple points (part of or entire CoMP cooperating set) to a single UE / multiple UEs in a TF resource. Data to a UE is simultaneously transmitted from multiple points, e.g. to improve the received signal quality

D4.2


Page: 65 of 72

and/or data throughput (JT generally requires low latency between transmission point, high bandwidth of the backhaul and low mobility of the UEs). This can be done in a coherent or non-coherent way.

i. Coherent JT: RRHs are coordinated by corresponding eNB and are transmitting the data in a time-synchronized manner.

ii. Non-coherent JT is using a non-synchronous transmission.

b. Dynamic point selection (DPS)/muting. Data transmission from one point (within CoMP cooperating set) in a TF resource. The transmitting / muting point may change from one subframe to another including varying over the PRB pairs within a TTI/subframe. Data is available simultaneously at multiple points. This includes also Dynamic cell selection (DCS).

c. DPS combined with JT. In this case multiple points can be selected for data transmission in the TF resource (different than DPS only scenario).

2) Coordinated scheduling / beamforming (CS/CB). In that case, data for an UE is only available at and transmitted only from one point in the CoMP cooperating set for a TF resource, but UE scheduling/beamforming decisions are made with coordination among CoMP cooperating set points. The transmitting points are chosen in a semi static manner. The following option of CS/CB is possible:

a. Semi-static point selection (SSPS). Transmission to a specific UE is performed from one transmission point at a time. The transmitting point may only change in a semi-static manner.

3) Muting. Muting may be applied in dynamic and semi-static manner with the transmission schemes described above.

4) Hybrid category of JP and CS/CB (possible). In that case, data for a UE may be available only in a subset of points in the CoMP cooperating set for a TF resource, but UE scheduling/beamforming decisions are made with coordination among CoMP cooperating set points. E.g. some points in the cooperating set may transmit data to the target UE according to JP scheme, while other points in the same time m may perform CS/CB.

Figure 52: DL CoMP schemes

D4.2


Page: 66 of 72

The following UL CoMP processing schemes are considered currently in 3GPP for PUSCH [3GP13b] (also shown on the Figure 51):

1) Joint reception (JR). In that case, PUSCH transmitted by UE is received jointly at multiple points (part or entire CoMP cooperating set) at a time (e.g. to improve the Rx signal quality).

2) Coordinated scheduling/coordinated beamforming (CS/CB). In that case UE scheduling and precoding selection decisions are made with coordination among CoMP cooperating set points while data is intended for one point only at a time.

Figure 53: UL CoMP schemes

The main differences between DL and UL CoMP in terms of standardization are the following [ADF+13]:

• UL CoMP is mostly implementation-specific with little need for additional work from specification perspective. However some smaller enhancements improving the UL orthogonality are included in Rel. 11.

• DL CoMP requires large specification effort, mainly due to CSI feedback issues. Therefore multipoint CSI feedback framework was introduced in 3GPP Rel. 11. Additionally due to many open issues seen around CoMP Rel. 12 will continue with multipoint CSI feedback and enhancements to DL control channels.

5.1.2.3 Limited feedback and relaxed backhaul requirements in Rel. 12

According to [3GP14], 3GPP triggers the following study in Release 12 to consider CoMP with the relaxed requirements for backhaul (non-ideal backhaul) to broaden the practical applicability of CoMP. CoMP specifications in Rel. 11 did not considered the specified support of a network interface for CoMP involving multiple separated eNBs with non-ideal backhaul. Due to this limitation, the operators having non-ideal backhaul in their networks may not be able to take advantage of performance benefit from inter-eNB CoMP operation [3GP14]. The study on CoMP for LTE with Non-Ideal Backhaul identified the specific cases in which CoMP can provide performance enhancements. Due to that, network interface enhancement and signalling messages shall be specified to allow implementation of both centralized and distributed CoMP focusing on HetNets (including macro/pico scenario). This issue is discussed in [3GP14], where the following objectives were raised for this study item with respect to evaluation:

1) Evaluate CS/CB including semi-static point selection/muting as candidate CoMP scheme for multiple eNB with non-ideal but typical backhaul. If there is performance benefit from such evaluation – it is requested to recommend for which CoMP technique(s) the signalling should be specified for inter-eNB operation;

2) Consider level of backhaul delay achievable with non-ideal backhaul; 3) Evaluation should be on the CoMP operation between macro eNBs, between macro eNB and small

cell eNB and between small cell eNBs with non-ideal backhaul. 4) The evaluations shall take into account estimation errors, DL overhead, complexity, feedback

overhead, backward compatibility and practical UE implementations.

D4.2


Page: 67 of 72

5) Rel. 11 MMSE-IRC shall be used as a baseline receiver for evaluating performance gain. The considerations related to the above are provided in [3GP13a]. Additionally according to [3GP14], the following objectives were raised for this study item with respect to definitions in standard:

1) Specify signalling of information, e.g.: a. One or more sets of CSI reports (RI, PMI, CQI) of individual UEs b. One or more measurement reports (RSRP) of individual UEs c. SRS received power of individual UEs d. Resource utilization per cell e. Enhanced RNTP-type information in frequency/time/power/spatial domain f. Enhanced ABS information in power and spatial domain g. QCI h. Indication of resource coordination result or resource coordination request (resource

allocation in frequency/time/power/spatial domain) i. Used configurations of reference signals, CSI processes j. Indication of coordination result or coordination request for reference signal

configurations, CSI processes 2) Specify necessary procedures related to the above; 3) Determine whether the above signalling shall be introduced to the X2 interface or a new interface

shall be specified.

5.1.2.4 3GPP conclusions regarding CoMP

According to discussions and performance evaluation present in [3GP13b], the following conclusions were made:

• CoMP offers performance benefits in both homogeneous networks (in scenarios 1 and 2) and heterogeneous networks (in scenarios 3 and 4);

• Performance of CoMP schemes that rely on spatial information exchange is sensitive to delay between two transmission points (while level of sensitivity depends on particular CoMP scheme).

Due to the above observations 3GPP recommends that further work on CoMP shall focus on: Joint Transmission, DPS including dynamic point blanking and coordinated scheduling/beamforming including dynamic point blanking.

D4.2


Page: 68 of 72

6 Abbreviations and References

Abbreviations

3GPP 2G 3G

3rd Generation Partnership Project Second Generation Third Generation

4G Fourth Generation 5G Fifth Generation 5GNOW 5th Generation Non-orthogonal Waveforms for Asynchronous

Signaling BS BW

Base Station Bandwidth

CoMP CFO CP CQI CSI DCRA DL DRS eICIC eNB

Coordinated Multipoint Carrier Frequency Offset Cyclic Prefix Channel Quality Indication Channel State Information Dynamic CQI Resource Allocation Downlink Demodulation Reference Signal Enhanced Inter-Cell Interference Coordination Enhanced-NodeB

FBMC FEC

Filter Bank Multi-Carrier Forward Error Correction

FP7 7th Framework Programme GFDM GPS

Generalized Frequency Division Multiplexing Global Positioning System

H2H HetNet

Human-to-Human Heterogeneous Networking

ICI (I)FFT ISI

Inter-Carrier Interference (Inverse) Fast Fourier Transform Inter-Symbol Interference

LTE Long Term Evolution LTE-A M2M

Long Term Evolution Advanced Machine-to-Machine

MAC MIMO MTC OFDM OFDMA PDCCH PDSCH PHY

Medium Access (layer) Multiple Input Multiple Output Machine Type Communication Orthogonal Frequency-Division Multiplexing Orthogonal Frequency-Division Multiple Access Physical Downlink Control Channel Physical Downlink Shared Channel Physical (layer)

PRACH PUCCH PUSCH

Physical Random Access Channel Physical Uplink Control Channel Physical Uplink Shared Channel

D4.2


Page: 69 of 72

QAM QPSK RACH RE SFO SINR UE UL

Quadrature Amplitude Modulation Quadrature Phase-Shift Keying Random Access Channel Resource Element Sampling Frequency Offset Signal to Noise and Interference Ratio User Equipment Uplink

References:

[3GP13] 3GPP TR 36.842 V12.0.0 (2013-12) “Study on Small Cell enhancements for E-UTRA and E-UTRAN; Higher layer aspects” http://www.3gpp.org/DynaReport/36842.htm

[3GP13a] 3GPP TR 36.874 V12.0.0 (2013-12), “Coordinated multi-point operation for LTE with non-ideal backhaul (Release 12)”, http://www.3gpp.org/DynaReport/36874.htm

[3GP13b] 3GPP TR 36.819 V11.2.0 (2013-09), “Coordinated multi-point operation for LTE physical layer aspects (Release 11)”, http://www.3gpp.org/DynaReport/36819.htm

[3GP13c] 3GPP 36.211, V10.7.0 (2013-02) “Physical Channels and Modulation (release 10)”

[3GP14] “Overview of 3GPP Release 12 V0.1.2 (2014-03)”, 3GPP, http://www.3gpp.org/ftp/Information/WORK_PLAN/Description_Releases/Rel-12_description_20140316.zip

[5GNOWD3.1] 5GNOW, “5G Waveform Candidate Selection”, Deliverable 3.1, November 2013

[5GNOWD3.2] 5GNOW, “Intermediate 5GNOW Transceiver and frame structure concept”, Deliverable 3.2, May 2014

[5GNOWD4.1] 5GNOW, “Intermediate MAC Concept”, Deliverable 4.1, November 2013

[AB11] M. Awal, L. Boukhatem, “Dynamic CQI resource allocation for OFDMA systems”, WCNC

2011

[ADF+13] David Astely, Erik Dahlman, Gabor Fodor, Stefan Parkvall, and Joachim Sachs, “LTE Release 12 and Beyond”, Communications Magazine, July 2013

[CJK+10] G. Caire, N. Jindal, M. Kobayashi, N. Ravindran, Multiuser MIMO Achievable Rates With Downlink Training and Channel State Feedback, IEEE Transactions on Information Theory, Vol. 56, Issue 6, pp 2845-2866, 2010

[CKD+13] N. Cassiau, D. Kténas and J.B. Doré, “Time and Frequency Synchronization for Downlink CoMP with FBMC”, Tenth International Symposium on Wireless Communication Systems (ISWCS’13), Illmenau, Germany, August 2013

[CWT12] R. Cheng, C. Wei, S. Tsao, F. Ren , “RACH collision probability for MTC”, IEEE 2012

http://www.3gpp.org/DynaReport/36842.htm



http://www.3gpp.org/ftp/Information/WORK_PLAN/Description_Releases/Rel-12_description_20140316.zip

http://www.3gpp.org/ftp/Information/WORK_PLAN/Description_Releases/Rel-12_description_20140316.zip

D4.2


Page: 70 of 72

[EC10] Engage Consulting Limited, “High-level Smart Meter Data Traffic Analysis”, May 2010, Doc. REF: ENA-CR008-001-1.4

[EML+09] J. Eriksson, R. Moosavi, E. Larsson, N. Wiberg, P. Frenger, F. Gunnarsson, “On coding of scheduling information in OFDM”, VTC 2009

[HA12] C. Han, S. Armour, “Computational Complexity and Energy Consumption Analysis of Dynamic Resource Scheduling Algorithms for LTE”, VTC 2012

[HLK+13] C. Hoymann, D. Larsson, H. Koorapaty, J. Cheng, “A Lean Carrier for LTE”, IEEE Communications Magazine, Feb 2013

[HR11] K. Huang, D. Raychaudhuri, “MAC Protocol Adaptation in Cognitive Radio Networks”, WCNC 2011

[HZL09] A. Haghighat, G. Zhang, Z. Lin, “Full-Band CQI feedback by Haar Compression in OFDMA Systems”, VTC 2009

[KG12] P. de Kerret, D. Gesbert, Degrees of Freedom of the Network MIMO Channel With Distributed CSI, IEEE Transactions on Information Theory, Vol. 58, Issue 11, pp 6806-6824, 2012

[KLC10] T. Kim, D. Love, B. Clerckx, “A Feedback Update Control Scheme for Limited Feedback Multiple Antennas Systems”, GLOBECOM 2010

[KWJ+14] M. Kasparick, G. Wunder, P. Jung, D. Maryopi, „Bi-Orthogonal Waveforms for 5G Random Access with Short Message Support“, European Wireless Conference, Barcelona, Spain, May 14-16, 2014.

[LKY11] K. Lee, S. Kim, B. Yi, “Throughput comparison of random access methods for M2M service over LTE networks”, IEEE International Workshop on M2M communications, 2011

[LS+13] M. Laner, P. Svoboda, N. Nikaein, M. Rupp, K. Zhou, D. Drajic, M. Popovic, S. Krco, “Traffic Models for MTC”, ISWCS 2013

[LSN+13] M. Laner, P. Svoboda, N. Nikaein, M. Rupp, “Simple traffic modeling framework for MTC”, ISWCS 2013

[MP11] P. Marsch and G. P. Fettweis, “Coordinated Multi-Point in Mobile Communications, From Theory to Practice,” Cambridge University Press, 2011, p. 508

[MTR+11] Y. Medjahdi, M. Terre, D. Le Ruyet, D. Roviras, D., and A. Dziri, “Performance Analysis in the Downlink of Asynchronous OFDM/FBMC Based Multi-Cellular Networks”, IEEE Transactions on Wireless Communications, Vol. 10, August 2011

[NNB+13] T. Nakamura, S. Nagata, A. Benjebbour, Y. Kishiyama, T. Hai, S. Xiadong, Y. Ning, L. Nan, “Trends in Small Cell Enhancements in LTE Advanced”, IEEE Communications Magazine, Feb 2013

[RMC12] P. Rost, A. Maeder, X. Costa “Asymmetric Uplink-Downlink Assignment for Energy-Efficient Mobile Communication Systems”, VTC 2012

D4.2


Page: 71 of 72

[RW15] Mohammed Ramadan, “Compressive 5G Random Access”, Master thesis TU Berlin, March March 2015 supervised by G. Wunder and G. Caire

[SA12] A. Sahin, H. Arslan, “Multi-User Aware Frame Structure for OFDMA Based System”, VTC

2012

[SJT09] C. So-In, R. Jain, A. Tamimi, “Scheduling in IEEE 802.16e Mobile WiMAX Networks: Key Issues and a Survey”, Journal on selected areas in communications 2009

[SS10] Y. Su, M. Schaar, “Towards Efficient, Stable, and Fair Random Access Networks: A Conjectural Equilibrium Approach”, GLOBECOM 2010

[SW11] J. Schreck and G. Wunder, Distributed Interference Alignment in Cellular Systems: Analysis and Algorithms, Proc IEEE European Wireless, pp. 1-8, April 2011

[SW12] J. Schreck and G. Wunder, Iterative Interference Alignment for Cellular Systems with User Selection, IEEE 37th International Conference on Acoustics, Speech, and Signal Processing (ICASSP), pp. 1-4, March 2012

[SWJ13] J. Schreck, G. Wunder, and P. Jung, “Robust Iterative Interference Alignment for Cellular Networks with Limited Feedback", Accepted for International Workshop on Emerging Technologies for LTE-Advanced and Beyond-4G, IEEE Globecom'13, Atlanta, GA, USA, December 2013

[SWJ15] J. Schreck, G. Wunder, and P. Jung, “Robust Iterative Interference Alignment for Cellular Networks with Limited Feedback”, IEEE Transactions on Wireless Communications, February 2015, vol. 14, no. 2, pp. 882-894, http://arxiv.org/abs/1308.6750

[SYM+12] S. Sun, Q. Yu, W. Meng, C. Li, “A Configurable Dual-Mode Algorithm on Delay-Aware Low-Computation Scheduling and Resource Allocation in LTE Downlink”, WCNC 2012

[TR 36.822] 3GPP TR 36.822 V0.2.0, “LTE RAN Enhancements for Diverse Data Applications,” Nov. 2011

[TR 36.912] 3GPP TR 36.912 v9.0.0 “Feasibility Study for Further Enhancements for E-UTRA (LTE Advanced) (Release 9),” Sep. 2009

[TR 37.868] “Study on RAN improvements for MTC”, v.11.0.0

[TS 36.321] 3GPP TS 36.321, “Medium Access Control (MAC) Protocol Specification (Release 8)”, v8.5.0

[TS 36.902] “SON use cases and solutions”, v.9.3.1

[Ung82] Ungerboeck G., “Channel coding with multilevel/phase signals,” Information Theory, IEEE Transactions on , vol.28, no.1, pp.55,67, Jan 1982

[WBS+15] G. Wunder, H. Boche, T. Strohmer, P. Jung, “Sparse Signal Processing Concepts for Efficient 5G System Design”, IEEE Access, http://arxiv.org/abs/1411.0435, accepted, to appear 2015

[WJK+14] G. Wunder, P. Jung, M. Kasparick, T. Wild, F. Schaich, Y. Chen, S. ten Brink, I. Gaspar, N. Michailow, A. Festag, L. Mendes, N. Cassiau, D. Ktenas, M. Dryjanski, S. Pietrzyk, B. Eged, P. Vago, and F. Wiedmann, “5GNOW: Non-Orthogonal, Asynchronous Waveforms for Future

http://arxiv.org/abs/1308.6750

http://arxiv.org/abs/1411.0435

D4.2


Page: 72 of 72

Mobile Applications,” IEEE Communications Magazine, 5G Special Issue, February 2014, vol. 52, no. 2, pp. 97–105 (>70 citations including VTC conference version in 2014)

[WJW14] G. Wunder, P. Jung, C. Wang, „Compressive Random Access for Post-LTE Systems“, accepted at IEEE International Conference on Communications (ICC), Sydney, Australia, June 10-14, 2014

[WKB+13] G. Wunder, M. Kasparick, S. ten Brink, F. Schaich, T. Wild, I. Gaspar, N. Michailow, G. Fettweis, D. Ktenas, N. Cassiau, M. Dryjanski, S. Pietrzyk, K. Sorokosz, B. Eged, “System-level interfaces and performance evaluation methodology for 5G physical layer based on non-orthogonal waveforms”, Proc. 47th Annual Asilomar Conference on Signals, Systems, and Computers, Pacific Grove, CA, USA, Nov. 2013

[WKW+14] G. Wunder et al, “5GNOW: Intermediate Frame Structure and Transceiver Concepts”, IEEE Global Communications Conference (GLOBECOM’14) Workshop on Telecommunication Standards, Austin, USA December 2014, invited paper

[WRS+13] G. Wunder, S. Raghuprakash, J. Schreck and P. Jung, Multi-User MIMO Multi-Cell Systems: Finite SNR Rate Loss Gap Analysis, ITG/IEEE Workshop on Smart Antennas (WSA'13), Stuttgart, Germany, March 2013

[WS10] G. Wunder and J. Schreck, A Robust and Efficient Transmission Technique for the LTE Downlink, Proc. 44th Annual Asilomar Conference on Signals, Systems, and Computers, Monterey, USA, Nov. 2010

[WSJ12] G. Wunder, J. Schreck, and P. Jung, Nearly Doubling the Throughput of Multiuser MIMO Systems Using Codebook Tailored Limited Feedback Protocol, IEEE Trans. on Wireless Communications, December 2012, Vol. 11, No. 11, pp. 3921-3931, www.arxiv.org/pdf/1107.2101

[ZNK12] K. Zhou, N. Nikaein, R. Knopp, C. Bonnet, “Contention based access for MTC over LTE”, IEEE 2012

--- End of Document Main Body---

http://www.arxiv.org/pdf/1107.2101

d4 - is-wireless · alcatel lucent deutschland ag alud germany commissariat a l energie atomique et...

Documents