csp-net breakout sessionmmwrcn.ece.wisc.edu/wp-uploads/2017/05/csp-net-breakout1.pdf · csp-net...
TRANSCRIPT
CSP-NET Breakout Session
NSF-RCN mmWave WorkshopJuly 2017
Goals
• Identify the most critical aspects of mmW PHY-abstraction
• WISHLIST: simulation/emulation capabilities for mmW networking research
• Opportunity for a white paper?
Agenda
• Introductions (25’)
• Open discussion (60’)
• Wrap-up (5’)
Introductions
• Haitham Hassanieh (U Illinois)
• Ismail Guvenc (U North Carolina)
• Tom Henderson (U Washington)
• Marco Mezzavilla (NYU)
• Jing Zhu (Intel)
• Matthew Andrews (Nokia)
• Rui Yang (Interdigital)
• Yuichi Kakishima (NTT DOCOMO)
Pasternack 60 GHz Evaluation Kit:
Cost ≈ 14,000 USD+Full PHY Control+Can Synchronize −Only Radio Front End!−Horn Antenna −Unidirectional
Haitham Hassanieh (UIUC)
TP-Link Talon AD7200Cost ≈ 380 USD
Dell Latitude Series
Acer Travelmate Series
Intel 18260 WiGig Cards
Qualcomm QCA WiGig Cards
Cost ≈ 1000 USD Cost = ?+Phased Array (4 ant.) 64 antennas+802.11ad stack 802.11ad+Bidirectional+Beam Selection−Rang ≈ 2m ≈ 100m
Tensorcom 60 GHz: Products
Cost ≈ 142K–314K USD +PHY Control, 2GHz+MIMO+Uni/Bidirectional−Cost−Horn Antennas
NI Transceiver (28,70 GHz):
PicoCell PlatformEvaluation Kit
Cost ≈ 10s of USD
−Designed mainly for laptop docking stations
Cost > 700 USD
Analog Devices 60 GHzEvaluation Kit
Cost ≈ 3700 USD x2
+/−Same as Pasternack
Millimeter Wave Platforms
Build 24 GHz mmWave Front-End from components: Cost ≈ 10,000 USD
10
15
20
25
30
35
40
10 30 50 70 90
SN
R (
dB
)
Distance (m)
256 QAM
16 QAM
I
Q
I
Q
256 QAM
16 QAM
Millimeter Wave PlatformsHaitham Hassanieh (UIUC)
+Full PHY Control+Can Synchronize +Phased Array (8 element)+Range−Unidirectional−RF Front End Only
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 100 120 140 160 180No
rma
lize
d P
ow
er
Angle (deg)
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 100 120 140 160 180No
rma
lize
d P
ow
er
Angle (deg)
Wit
ho
ut
Cal
ibra
tio
nW
ith
C
alib
rati
on
Phased Array Calibration
Range (>90m)
Challenge: All components available and can be found at other frequencies
except: Phase Shifters
Phase Shifters we used:+Analog Phase Shifters+3𝑜Accuracy−Discontinued!−Need Calibration−High Loss 6dB
Ray Tracing (RT) Simulations for mmWave PropagationIsmail Guvenc, North Carolina State University
Altair WinProp RT Software Remcom Wireless Insite RT Software
• Urban/indoor multipath simulations• Prediction of maximum 5G data rate• mmWave propagation simulation up to 75 GHz • Considers oxygen absorption at 60 GHz• Built-in graphical editors for buildings, tunnels
stadiums, antenna pattern, etc.• Website:
http://www.altairhyperworks.com/product/FEKO/WinProp-Propagation-Modeling
• Urban/indoor multipath simulations• Massive MIMO support• mmWave propagation simulation up to 100 GHz • Considers oxygen absorption at 60 GHz, diffuse
scatterring• Built-in graphical editors • Website: https://www.remcom.com/wireless-insite-
em-propagation-software/
Benefits of Ray Tracing• Can capture diffuse scattering and spatial consistency
(important for mobility studies)• Convenient and quick way to generate data for specific
environments when compared to experiments
Drawbacks of Ray Tracing• May not fully characterize subtle details of real world
environments• Precise modeling of an environment requires modeling
of scatterer locations/material (time consuming)
Use of Ray Tracing Tools in the Recent Literature• # of papers in Google Scholar with the query (Winprop
mmWave): ~25• # of papers in Google Scholar with the query (Remcom
mmWave): ~88• Example: ICC 2016 paper by Qualcomm [1]
• Winprop is used for studying RMS delay spread with omni and beamformed links, results compared with real world experimental data
• RMS-DS measurements with WinProp: ~60-100ns (omni) and ~20ns (beamformed)
• Beamformed measurements: 50ns (NYU), 30ns (Samsung) – higher in practice
• Diffuse scattering critical at higher frequency
[1] Zhang, Zhenliang, Jung Ryu, Sundar Subramanian, and Ashwin Sampath, “Coverage and channel characteristics of millimeter wave band using ray tracing,” in Proc. IEEE Int. Conf. Commun. (ICC), pp. 1380-1385, 2015.
Case Study: Ray Tracing for UAV mmWave Propagation ChannelsUsing RemCom Wireless Insite [2]
RSS at 28 GHz
RMS-DS at 28 GHz
[2] W. Khwaja, O. Ozdemir, and I. Guvenc, “UAV Air-to-Ground Channel Characterization for mmWaveSystems”, to appear in IEEE VTC Workshops, Sep. 2017.
Sub-urban scenario
Urban scenario
Over sea Rural
Suburban Urban
Over sea Rural
Suburban Urban
Received Signal Strength (RSS):• Over sea: Closely follows a two ray model• Rural/suburban: Fluctuation from scatterers• Urban areas: Fluctuations become higher• Fluctuation rate of the two ray propagation
model increases with UAV height• Fluctuation rate higher for 60 GHz (not shown)
versus 28 GHz• Rate of RSS decay with respect to distance at 60
GHz is higher than at 28 GHz indicating higher path loss
Root mean Square Delay Spread (RMS-DS):• RMS-DS largest for urban environment for most
of UAV heights (high scatterer density)• RMS-DS increases as a function of the UAV
height in the urban environment (can observe larger scattering at higher UAV heights)
• For suburban, rural environments, RMS-DS reduces with UAV altitude (buildings are not as tall, and hence less scattering)
• RMS-DS is lower at 60 GHz compared to 28 GHz
ns-3 wireless challenges
Thomas R. Henderson and Sumit Roy
CSP-NET Interface Breakout Session
NSF-RCN Millimeter-Wave Workshop, July 2017
ns-3 overview
• ns-3 is a leading open source, packet-level network simulator oriented towards network research, featuring a high-performance core enabling parallelization across a cluster (for large scenarios), ability to run real code, and interaction with testbeds
Runsonasinglemachine
orpartitionedacrossacluster
Challenges for 5G wireless simulations
• Largely categorized as scope and scale• Scope issues: pace of standards development, level of detail, implementation
issues, scenario support
• Scope approaches: federated development, common scenario support, vendor assistance
• Scale issues: managing abstractions, parallel wireless simulations
• Scale approaches: link-to-system mapping, simple wireless models
NSF-RCN Millimeter-Wave Workshop July 2017
Recent work on Phy abstractions for ns-3
• ns-3 implementation of TR 38.900 channel model above 6 GHz [1]• including MIMO beamforming architecture, spatial consistency and blockage models, integration with
ns-3 buildings module
• Wi-Fi OFDM error models for TGn fading channels • leveraging MATLAB WLAN System Toolbox
• maps link system simulation results into AWGN tables in the ns-3 simulator
NSF-RCN Workshop July 2017
[1] Zhang et al., NYU Wireless, arXiv: 1702.04822v1, Feb 2017[2] Patidar et al., Univ. Washington, WNS3 2017, June 2017
End-to-end mmWaveDr. Marco mezzavilla, research scientist, NYU
University OF WISCONSIN-MADISON, JULY 19 2017
End-to-end Network Simulator
14
Open source
Indoor/outdoor, urban macro/micro, rural scenarios
Mobility (including vehicular / hi-speed transportation)
Traffic and channel models (including real measurements)
Customizable frame structures, frequency bands, schedulers
Core network (S1/X2) latency
TCP/IP protocols and other technologies (e.g., IEEE 802.11)
© 2017 NYU WIRELESS
mmWave Channel
15© 2017 NYU WIRELESS
• Pathloss (3GPP channel models, NYU, QUADRIGA, ..)
• Real measurements + raytracing (U Bristol)
mmWave Beamforming
16© 2017 NYU WIRELESS
• For each TX-RX pair we store the following variables
MultiModelSpectrumChannel
MmWaveSpectrumPhy MmWaveSpectrumPhy
MmWaveEnbPhy MmWaveUePhy
MmWaveUeNetDeviceMmWaveEnbNetDevice
LteRlcAm/UmLtePdcpLteEnbRrc
MmWave
Beamforming
MmWavePropagationLossModel
AntennaArrayModel
MiErrorModel
IPTCP/UDPApplication
EpcEnbApplication
UDPIP
MmWaveEnbMacScheduler
MmWaveUeMac
LteRlcAm/UmLtePdcpLteEnbRrc
mmWave eNodeB mmWave UE
UDPIP
IP
EpcPgwSgwApp
PGW/SGW
IPTCP/UDPApplication
RemoteHost
ns3/LENA
built-incustommmW
PHY/MACcustommmW
channelmodel
MultiModelSpectrumChannel
MmWaveSpectrumPhy MmWaveSpectrumPhy
MmWaveEnbPhy MmWaveUePhy
MmWaveUeNetDeviceMmWaveEnbNetDevice
LteRlcAm/UmLtePdcpLteEnbRrc
MmWave
Beamforming
MmWavePropagationLossModel
AntennaArrayModel
MiErrorModel
IPTCP/UDPApplication
EpcEnbApplication
UDPIP
MmWaveEnbMacScheduler
MmWaveUeMac
LteRlcAm/UmLtePdcpLteEnbRrc
mmWave eNodeB mmWave UE
UDPIP
IP
EpcPgwSgwApp
PGW/SGW
IPTCP/UDPApplication
RemoteHost
ns3/LENA
built-incustommmW
PHY/MACcustommmW
channelmodel
Research Platform
17© 2017 NYU WIRELESS
Dynamic channel sounder Channel emulator End-to-end network simulator
TCP Optimization over mmWave* Jing Zhu
Senior Staff Research Scientist
Wireless Communication Research / Intel Lab In collaboration with NYU Wireless (Menglei Zhang; Marco Mezzavilla)
Research Interest / Areas• Multi Radio Network / Convergence
• WLAN/WPAN/WWAN coexistence (e.g. 802.11, 802.16, LTE, etc.)
• Multi-RAT convergence & integration (e.g. LTE WiFi Aggregation, MP-TCP, etc.)
• Interference Mitigation & Spatial Reuse in Ultra-Dense Wi-Fi Networks
• Adaptive Carrier Sensing & TX Power Control
• Cross-Layer Optimization & QoS for Real-Time Internet Apps (e.g. Skype)
• Intra-Flow Prioritization & Scheduling
• 5G/5G+ mmWave Network current focus
• MAC protocol /System Design: mobility management/handover & multi-connectivity/multi-hop relay & beam management
• TCP/cross-layer optimization (bufferbloat, linkoutage)
• Edge Computing & Mobile Compute Offload
TCP over 5G mmWave: Problems• Packet Loss over wireless link is NO longer the
main problem, thanks to advanced link-layer retransmission scheme, e.g. LTE RLC AM.
• Bufferbloat/Overfow [1]: when the link speed drops suddenly (e.g. LOSNLOS), RLC buffer occupancy may increase significantly leading to long latency, packet loss
• Long Link Outage [1] [2]: if link outage is too long (say 1 second), TCP sender may trigger RTO.
• ACK Congestion [3]: the UL congestion (“TCP ACKs”) may limit the maximum DL throughput of a TCP flow
>200ms RTT due to bufferbloat
[1] L. Zhang, et al., Transport Layer Performance in 5G mmWave Cellular, http://arxiv.org/abs/1603.02701[2] R2-1701686, Impact of blockage on TCP performance in high frequency scenarios, MTK, RAN2#97[3] R2-168036, Potential hurdle in maximizing DL TCP throughput NTT DOCOMO, INC., Fujitsu, RAN2#96
Solutions & Preliminary Results
• Active Queue Management [1]: AQM detects congestion by monitoring the instantaneous or average queue size. When the average queue size exceeds a certain threshold but is still less than the capacity of the queue, AQM algorithms infer congestion on the link and notify the end systems to back off by proactively dropping some of the packets arriving at a router. How effective will AQM solve the bufferbloat/overflow problem in 5G mmWave?
• AQM (e.g. CoDel) will trigger TCP congestion avoidance, causing underutilization (low TPT) when the link recovers
• TCP retransmissions may be delayed or lost during the bufferbloat period
• TCP Receiver Window Adjustment seems to be more effective than AQM, but today’s UE-based solution is hard managed, and may not have all the info, e.g. RTT, available bandwidth
• Splitting TCP/TCP Proxy[3] [4]: split the TCP connection at BS/AP into two parts: wireless and wired; Base Station (BS) to mitigate impact of link outage & bufferbloat. How effective will “split TCP” (e.g. TCP proxy) solve the link outage problem and bufferbloat in 5G mmWave?
[1]: https://en.wikipedia.org/wiki/Active_queue_management[2]: Menglei Zhang, et al., The Bufferbloat Problem over Intermittent Multi-Gbps mmWave Links, https://arxiv.org/pdf/1611.02117.pdf[3]: K. Brown and S. Singh, “M-TCP: TCP for Mobile Cellular Networks,” ACM Computer Communications Review (CCR), vol. 27, no. 5, 1997[4]: Jing Zhu, et al. Performance Modelling of TCP Enhancements in Terrestrial-Satellite Hybrid Networks, IEEE/ACM Transactions on Networking, Aug. 2006.
E2E Simulator (like NS3) is essential for cross-layer research (PHY/MAC/NET) in mmWavenetwork
Source: [2]
22
New Real-time Services are envisioned by 2020~2025!
Autonomous Services enabled by ultra-responsive and reliable 5G network
Remote Controlled Drones Self-Driving Cars
Augmented/Virtual Reality Autonomous Robots
Compute and Communication Frameworks
Intelligent, learning-enabled wireless platforms
and networks that can reliably sense, interpret
and act in real-time (<1 ms).
Source: ITU Watch Report on Tactile Internet
Experimental Trials for5G and Beyond
DOCOMO Innovations, Inc.
Yuichi Kakishima
Biography
• Yuichi Kakishima
• Received the B.S. and M.S. degrees from Tokyo Institute of Technology in 2005 and 2007, respectively.
• Joined NTT DOCOMO, INC. in 2007• Engaged in R&D of wireless access technologies
including MIMO technologies for LTE and LTE-Advanced.
• Currently a Manager in DOCOMO Innovations, Inc., California, USA.• Involved in research of 5G cellular network.
• Involved in 3GPP standardization as a delegate to 3GPP RAN1.
Collaborations with Global Vendors• Experimental trials are being started since 2014/Q4
• Including wide frequency band up to 70GHz
• Expanding collaborations to vertical industries
• E.g., Advanced driving support, virtual reality (VR), security systems, etc.
Outdoor Trial • LoS: 2Gbps at 480m away from BS
• N-LoS (in front of building): around 3Gbps
Frequency band: 14.9 GHz
System bandwidth: 730 .5MHz
Radio access: OFDM
BS: 4Tx/Rx (Max 4 streams)
MS: 8Tx/Rx
Indoor Trial
High-Mobility Test in Poster Session• Achieved more than 3Gbps with MS speed over 150km/h
Please visit our booth after this break out session for details!!
Interesting MmWave Networking Issues
Matthew Andrews
19 July 2017
vs
PPP Structured
Traffic Multihop
Half duplex constraints and dynamic TDD
interference
channel
GoB, Interference Modeling, Beam Coordination
vs
Antenna Abstraction
Mm Wave Standardization
© 2017 InterDigital, Inc. All Rights Reserved.
Important Standards for mm Wave Communications
• 3GPP• The global standard body that has defined 3G UMTS and 4G LTE cellular
technologies
• It’s where all of the wireless companies (terminal and infrastructure vendors, operators, …) meet to define how to connect devices to networks
• 3GPP is currently developing 5G cellular networks – the New Radio (NR)
• IEEE 802.11• 802.11 is the de facto Wireless LAN Standard
• Wi-Fi is based on 802.11, implemented in nearly all mobile devices
• Wi-Fi will be a key component of 5G • More than half of all traffic from mobile-connected devices will be offloaded to the fixed
network by means of Wi-Fi
© 2017 InterDigital, Inc. All Rights Reserved.
3GPP 5G Standardization Activities and Timeline
© 2017 InterDigital, Inc. All Rights Reserved.
20212017 2018 2019 2020
LTE Release 16
New Radio WIs (R16) New Radio WIs (R17)
LTE Release 17LTE Release 15R14
Phase 2 of R15 Specifications
Release 16 Specifications
Phase 1 of R15 - Early Freeze
Today
New Radio SIs (R15) New Radio Study beyond R16 IMT-2020 Submission
New Radio WIs (R15)
5G LTE + NB-IoT(required for
mMTC use cases for IMT-2020)
5G New Radio (NR)…
WRC-19 mmW bands (>6GHz)
eMBB, Low Latency Ultra Reliable, Satellite, mMTC, NOMA, V2X, …
Ch
ann
el M
od
el a
bo
ve 6
GH
z
Non standalone (depend on LTE) NR standalone
Proof of Concept Design e.g., Beam centric operations (< 52.6GHz)
Road to WRC-19
© 2017 InterDigital, Inc. All Rights Reserved.
Studies carried out by ITU on above 6 GHz Bands Technical feasibility of IMT in bands above 6 GHz are being studied by ITU-R WP-5D
The ITU report describes a number of channel measurement campaigns by industry and academia
Investigation aimed at the propagation channel characteristics under different propagation condition in these bands: 10, 18, 28, 38, 60, and 72 GHz
ITU & 3GPP plans for future studies on above 6 GHz Bands 3GPP RAN has completed a channel modelling project for above 6 GHz (< 52.6GHz)
New candidate bands will be proposed in the First Conference Preparatory Meeting for the WRC-19
The earliest opportunity for allocation of any band above 6 GHz is in 2019 at WRC-2019
A joint task group within ITU-R of existing license holders from other industries has been formed to closely examine the reallocation of above-6GHz band to mobile communications.
Mm Wave Communications in IEEE 802 Standards• IEEE802.15.3c
• Published in Sept 2009• 57-66GHz frequencies and short range (10m)
• PHY and MAC for WPAN• Single carrier (up to 5.3 Gbps) and OFDM (up to 3.8 Gbps)
• Use Cases • High speed stream content download and wireless data bus
• IEEE 802.11ad (WiGig)• Published in 2012 and has become part of 802.11-2016 (SC part only)• Peak data rate 6.7Gbps• Enabling a set of comprehensive beam selection, refinement, and tracking
procedures• Support multiple antennas and Relay
© 2017 InterDigital, Inc. All Rights Reserved.
Mm Wave Communications in IEEE 802 Standards• IEEE 802.11ay
• Currently being developed by TGay• Working on specification draft D0.3 (D1.0 in Nov17, and D2.0 in Mar18)
• Supporting a maximum throughput of at least 20 gigabits per second• Backward compatible to 802.11ad• To include channel bonding (up to 8.64GHz), SU/MU MIMO (up to 8 SSs), analogue and
digital baseband hybrid beamforming, higher order and non-uniform modulation • Use cases: Indoor, Outdoor, Backhaul/front haul/Mesh network,…
• IEEE802.11aj• Chinese mm Wave frequency bands including the 59-64 GHz and the 45 GHz frequency
bands to enable multi-Gbps throughput and lower power• Peak data rate 14.175Gbps (4x4 SU MIMO)• Plan to be published at the end of 2017
© 2017 InterDigital, Inc. All Rights Reserved.
Open discussion1. Channel abstraction. The main challenges are (i) directional time-dynamics and (ii) blockage: traces or
statistical models?
2. Are there components of the channel models that we still need to understand to perform reasonable simulations? If so, what?
3. Antenna abstraction: how much should we simplify the antenna pattern? What's the impact on interference characterization, MAC, et cetera?
4. What are the appropriate tools for networking research? What simplifications are reasonable at the physical layer to scale up the number of nodes?
5. Some PHY-MAC layer procedures may be centralized. What is the appropriate interface to the coordinator? What are the bandwidth / delay requirements?
6. What traffic models should we use, in order to capture the key 5G requirements?
7. Network discovery / tracking. How do we model this at a network layer in order to model handover / cell selection?
8. Similarly, how are we going to design and test PHY/MAC for new use cases like vehicular and aerial communications?
9. More in general – what will be the next big PHY-abstraction challenge 5-10 years from now?