future mobile communications: - lte radio scheduler ... · domain scheduler 6/25 comnets ... enodeb...
TRANSCRIPT
Future Mobile Communications:LTE Radio Scheduler Analytical Modeling
Dr. Yasir Zaki
New York University Abu Dhabi (NYUAD)FFV Workshop 2013
15th of March 2013
1/251/25
ComNetsCommunication Networks
Overview
1 Introduction
2 LTE Radio Scheduler
3 LTE Scheduler Analytical Models
4 Conclusion
2/252/25
ComNetsCommunication Networks
Long Term Evolution (LTE)LTE is the newest 3GPP1 standard (Release 8)
UE
eNodeB
MME
S-GW
HSS
PDN
GW
PCRF
Uu
S1-U
S1-MME
S6a
S11 Gx
Rx
S5/S8
eNodeB
UE E-UTRAN EPC Services
IP connectivity Layers, The EPS
Service connectivity Layer
User plane
Control plane
SGi
Operator’s
IP services
(e.g. IMS, PSS)
The main motivation of the work presented here is the analyticalmodeling of the LTE QoS aware radio scheduler.And to validate our simulation results by comparing it to the analyticalresults.
13rd Generation Partnership Project
3/253/25
ComNetsCommunication Networks
LTE OPNET Simulation Model
Application
configProfile
configMobility &
Channel
Global UE
Database
Remote_Server
IPcloud
PDN_GW aGW
eNB1
eNB2
UE1 UE2 UE3 UE4 UE5
UE6 UE7 UE8 UE9 UE10
UE11 UE12 UE13 UE14 UE15
UE16 UE17 UE18 UE19 UE20
UE21 UE22 UE23 UE24 UE25
UE26 UE27 UE28 UE29 UE30
Remote_Server
IPcloudPDN_GW
aGW
R2
1 2 3
4 5 6
7 8 9
10 11 12
13 14 15
16 17 18
19 20 21
22 23 24
25 26 27
28 29 30
31 32 33
34 35 36
Cell 1
Cell 2
Cell 3
R2
R3
eNB1
R1
eNB2
eNB3
eNB4
Application
TCP/UDP
IP
PDCP
RLC
MAC
PHY
RLC
MAC
PHY
UDP
IP
L2
L1
UDP
IP
L2
L1
UDP
IP
L2
L1
UDP
IP
L2
L1
GTP
IP
L2
L1
Application
TCP/UDP
IP
L2
L1
Relay
PDCPGTP
Relay
GTPGTP
4/254/25
ComNetsCommunication Networks
LTE Physical Resource StructureResources in LTE consist of both: time and frequency dimensions
The smallest resource the scheduler can allocate is called aPhysical Resource Block (PRB).A PRB consists of the following:
72 OFDM symbols in the time domain12 sub-carriers in the frequency domain
0
1
2
3
4
5
6
1 S
lot
7 S
ymb
ols
1 2 3 4 5 6 7 8 9 10 11
Resource Block 0
Resource Block 1
Resource Block 2
Resource Block N1
Slo
t7
Sym
bo
ls
Resource Block 3
180 kHz 15 kHz x 12 sub-
carriers15 kHzSymbols
Resource Element
Sub-Carriers
N represents the maximum number of resource blocks
which depend on the defined spectrum/bandwidth
26 or 7 symbols depending on the length of the cyclic prefix
5/255/25
ComNetsCommunication Networks
LTE Dynamic Packet Scheduling
The packet scheduling is divided into two different stages:Time Domain Scheduler (TDS): deals with QoS requirements anduser/bearer prioritizationFrequency Domain Scheduler (FDS): deals with spectrumallocation and multi-user diversity exploitation
Time Domain Scheduling
Frequency Domain
Scheduling
Buffer informationQoS informationLink Adaptation
HARQ information...
bearers for FDS
Time Domain Updates
Schedule bearers
In LTE literature such a split is called decoupled time and frequencydomain scheduler
6/256/25
ComNetsCommunication Networks
Time Domain Scheduler TDSTwo of the classical TDS schedulers are:
Blind Equal Throughput (BET): give a fair chance of resources so thatusers can achieve similar throughput
Maximum Throughput (MaxT): maximize the cell/system throughput byscheduling users with the best channel conditions
The TDS prioritization is done by calculating the priority factor.
P BETk (t ) = argmaxk
[1
θk [t ]
]= BET TD priority factor
PMaxTk (t ) = argmaxk [SINRk [t ]] = MaxT TD priority factor
θk [t ] is the normalized average throughput of bearer k ranging between 0 and 1
SINRk [t ] is the instantaneous SINR value of bearer k
7/257/25
ComNetsCommunication Networks
Frequency Domain Scheduler FDS
Distributes the radio resources (PRBs) among the highest prioritybearers obtained from the TDS
Exploits the multi-user diversity and tries to enhance the overallspectral efficiencyThe FDS used in this work is an optimized round robin scheduler
Serves first strictly the GBR3 bearersServes the highest ψ4 nonGBR priority bearers with the remainingresourcesSchedules the PRBs using an iterative approach
3Guaranteed Bit Rate4ψ is chosen to be 5 in this work
8/258/25
ComNetsCommunication Networks
Optimized Service Aware Scheduler
The proposed OSA scheduler provides:
QoS guarantees
Fairness among users
System performance maximization
eNodeB per UE Buffers
MAC-QoS-Class-1
HARQ
MAC-QoS-Class-2
MAC-QoS-Class-3
MAC-QoS-Class-4
MAC-QoS-Class-5
TDS FDS
QCI Classification Time-Domain Scheduling
Frequency-Domain Scheduling
Cla
ssif
ier PRB
assignmentGBR
Candidate List
nonGBR Candidate List
9/259/25
ComNetsCommunication Networks
OSA Scheduler: TDS and FDS
The OSA TDS priority factor is:
P nonGBR−OSAk (t ) = argmaxk
[WQoS j ×
γk [t ]
θk [t ]
]
WQoS j is the QoS weight of the j th QoS class
θk [t ] is the normalized EMA5 throughput of bearer kγk [t ] is the normalized EMA channel condition of bearer k
The OSA serves the GBR bearers (i.e., VoIP) with strict prioritybefore the non-GBR bearers
The OSA FDS is an optimized round robin scheduler
5Exponential Moving Average
10/2510/25
ComNetsCommunication Networks
LTE Scheduler Analytical Models
The presented model is an extension of the model found in [1]
The scheduler has a fixed amount of PRBs to distribute amongactive UEs every TTI 6
The UEs have different SINRSupport different Modulation and Coding Schemes (MCS)
[1] S. Doirieux, ... “An efficient analytical model for the dimensioning of WiMAX networks supporting multi-profile best efforttraffic”, Computer Communications, Volume 33, Issue 10, 15 June 2010
6Transmission Time Interval = 1ms
11/2511/25
ComNetsCommunication Networks
Model Assumptions
UEs are statistically identical with respectto the channel
The channel is divided into 8 states(MCSk ) (0 < k < 7)
Each MCSk has a static probability Pk
(Mobility model dependent)
UEs use the same ON/OFF elastic traffic:
XON is the average file size in bitstoff is the average OFF duration in s
0 1 2 3 4 5 6 70
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
Modulation and Coding Scheme (MCS)
Pro
bab
ility
of
MC
S (
Pk)
Random WayPointRandom Direction
12/2512/25
ComNetsCommunication Networks
General Analytical Model
The model is based on the Continuous Time Markov Chain (CTMC)n: Markov chain state representing the number of active UEs in a TTIN: total number of UEs in the system
The resulting CTMC consists of N+1 statesArrival transition is performed with rate (N − n )λ with λ = 1/toffDeparture transition is performed with a generic rate µ(n )
10 2
Nλ (N-1)λ λ
µ(1) µ(2) µ(N)
nn-1 n+1
(N-n+1)λ (N-n)λ
µ(n) µ(n+1)
N
(n0, n1, …, nk)Σi ni = n
13/2513/25
ComNetsCommunication Networks
LTE Frequency Domain Scheduler
The FDS is a round robin scheduler
It serves up to ψ=5 users per TTI
This means, that the UEs served per TTI are equal to η:
η = min [n − n0, ψ]
with n being the number of active users in a TTI, and n0 is thenumber of users in outage7
7User is in bad channel condition that does not allow any transmission
14/2514/25
ComNetsCommunication Networks
Steady State Probability
The steady state probability π(n ) can be calculated from:
π · Q = 0 (1)
π=[π0, π1, ...πN ] and∑
i πi = 1Q is the infinitesimal generator matrix
According to Little’s law, the average ON period duration tON8 is:
ton =
∑Nn=1 nπ(n )∑N
n=1(N − n + 1)λ · π(n )
8tON is the duration of an active transfer, or file download time
15/2515/25
ComNetsCommunication Networks
Departure Rate µ(n )
µ(n ) =
∑(n ,...,n )(n0,...,nK )=(0,...,0)|
n0+...+nK=nn0 6=n
TBS (n0, ..., nK )(
nn0, ..., nK
)[∏Kk=0 P
nkk
]Xon · TTI
Xon is the average file size of the ON period(n
n0, ..., nK
)is the multinomial coefficient
Pk is the stationary probability of MCSk
nk is the number of active users in MCSk
TBS (n0, ..., nK ) is the total number of bits transmitted for all servedusers under a certain combination (n0, ..., nK )
16/2516/25
ComNetsCommunication Networks
TBS (n0, ..., nK )
TBS (n0, ..., nK ) =∑
i=[i1,...,iη]
TBSi (η) (2)
η is the number of users served under (n0, ..., nK )
[i1, ..., iη] represents the index of the chosen η users out of the nactive users considered for scheduling within a TTI
TBSi (η) denotes the number of bits of user i using his MCS
the choice of the η users is determined by the TD scheduler.
17/2517/25
ComNetsCommunication Networks
Choosing η UEs Procedure- MaxT chooses the η UEs with the highest MCS
MCS0 MCS1 MCS2 MCS3
n0 n1 n2 n3
Find η UEs
MCS4 MCS5 MCS6 MCS7
n4 n5 n6 n7
- BET chooses the η UEs from the MCS that achieves the avg. throughput:
MCS0 MCS1 MCS2 MCS3
n0 n1 n2 n3
Find (η-1) UEs
MCS4 MCS5 MCS6 MCS7
n4 n5 n6 n7
Find 1 UE
- OSA chooses the η UEs from the MCS as follows:
MCS0 MCS1 MCS2 MCS3
n0 n1 n2 n3
Find (η-1) UEs
MCS4 MCS5 MCS6 MCS7
n4 n5 n6 n7
Find 1 UE
AverageThroughput =K∑
k=1TBSk · Pk
0 1 2 3 4 5 6 70
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
Modulation and Coding Scheme (MCS)P
rob
abili
ty o
f M
CS
(P
k)
Random WayPointRandom Direction
18/2518/25
ComNetsCommunication Networks
Two Classes Model
The single class model is extendedinto a two dimensional Markov Chain,with state (i,j)
i represents the number of UEswithin class 1j represents the number of UEswithin class 2
The combinations within each statewill increase to:(i0, i1, i2, i3, i4, i5, i6, i7, j0, j1, j2, j3, j4, j5, j6, j7)where
∑r ir = N1 and =
∑r jr = N2
0,1
1,1
0,0
1,0
0,2
1,2
0,N2
1,N2
N1,1N1,0 N1,2 N1,N2
N2λ2 (N2-1)λ2 (N2-2)λ2 λ2
µ2(N1,1) µ2(N1,2) µ2(N1,3) µ2(N1,N2)
N1λ 1
λ 1(N
1-1)λ
1
µ1(1,0)
µ1(2,0)
µ1(N
1,0)
19/2519/25
ComNetsCommunication Networks
Three Classes Model
Similary the single class model canbe extended into a three dimensionalMarkov Chain, with state (i,j,z)
i represents the number of UEswithin class 1j represents the number of UEswithin class 2z represents the number ofUEs within class 3
The combinations within each statewill increase to:(i0, i1, i2, i3, i4, i5, i6, i7, j0, j1, j2, j3, j4, j5, j6, j7, z0, z1, z2, z3, z4, z5, z6, z7)where
∑r ir = N1 and
∑r jr = N2
and∑
r zr = N3
20/2520/25
ComNetsCommunication Networks
Scenarios Configuration
General parameter ValueBandwidth (# PRBs) 5 MHz (i.e., 25 PRBs)Radio QoS weights WQoS 1=5, WQoS 2=2 and WQoS 3=1Traffic model FTP with different IATs:
uniform (0, 30) s, uniform (15, 45) s,uniform (30, 60) s and uniform (45, 75) s
Simulation run time 8000 s, 10 seeds with 95% confidence interval
Analysis Class1 Class2 Class3
MaxT 10 UEs with5, 10 and 15 MByte
3 UEs with 6 UEs withw-MaxT 5 MByte 10 MByte
3 UEs with 3 UEs with 3 UEs withw-MaxT 5 MByte 10 MByte 15 MByte
21/2521/25
ComNetsCommunication Networks
Average FTP Download Time: MaxT
15 30 45 600
20
40
60
80
100
120
140
160
Inter−arrival−time IAT (sec)
Dow
nloa
d tim
e (s
ec)
MaxT Average FTP download time (10UEs scenario)
5MB simulation5MB analytical10MB simulation10MB analytical15MB simulation15MB analytical
22/2522/25
ComNetsCommunication Networks
Average FTP Download Time: w-MaxT
15 30 45 600
10
20
30
40
50
60
70
Inter−arrival−time IAT (sec)
Dow
nloa
d tim
e (s
ec)
w−MaxT Average FTP download time
class1 simulationclass1 analyticalclass2 simulationclass2 analytical
23/2523/25
ComNetsCommunication Networks
Average FTP Download Time: w-MaxT
15 30 45 600
10
20
30
40
50
60
70
80
90
100
Iner−arrival−time IAT (sec)
Dow
nloa
d tim
e (s
ec)
w−MaxT Average FTP download time
class1 simulationclass1 analytical
class2 simulation
class2 analytical
class3 simulationclass3 analytical
24/2524/25
ComNetsCommunication Networks
Conclusion and Outlook
The proposed analytical model shows very accurate resultscompared to the simulation results
The model can support up to three different QoS classes
The analytical model only requires a couple of minutes to run (muchfaster than simulations)The model can be extended to:
support more QoS classesa GBR class can also be modeledother types of TDS can also be modeled
25/2525/25
ComNetsCommunication Networks
Backup issues
LTE QoS bearers
3GPP QoS bearer classification
Channel dependent scheduling
Channel model
Analytical model scaling
Analytical model TBS(n)
Analytical model Q-matrix
3GPP QoS Bearer Classification
UE eNodeB S-GW P-GW Peer Entity
E-UTRAN EPC Internet
End-to-End Service
External BearerEPS Bearer
S5/S8 BearerS1 BearerRadio Bearer
LTE-Uu S1 S5/S8 SGi
back
LTE QoS BearersQCI Bearer type Priority Packet delay Packet error Example services
budget (ms) loss rate
1 GBR 2 100 10−2 Conversational voice2 GBR 4 150 10−3 Conversational video
(live streaming)3 GBR 5 300 10−6 Non-conversational
video (buffered streaming)4 GBR 3 50 10−3 Real time gaming5 non-GBR 1 100 10−6 IMS signaling6 non-GBR 7 100 10−3 Voice,
video (live streaming),interactive gaming
7 non-GBR 6 300 10−6 Video (buffered streaming)8 non-GBR 8 300 10−6 TCP based (e.g., www,
e-mail, chat, FTP, p2p)9 non-GBR 9 300 10−6
Nine predefined QoS classesfour Guaranteed Bit Rate (GBR)five non Guaranteed Bit Rate (nonGBR)
Operators are free to choose their own classificationback
Channel Dependent Scheduling
User #2 scheduled
Frequency
User #1 channel
User #1 scheduled
Time
User #2 channel
back
Channel Model
Path loss:
128.1+ 37.6 · log10(distance )
Correlated slow fading:Log-normal distributed
Fast fading:Jakes-like methodTwo extended ITU channel models
Extended pedestrian AExtended vehicular A
20002004
20082012
20162020
0
200
400
600
800
1000−40
−30
−20
−10
0
10
20
30
Frequency (MHz)time (msec)
Fast fading (dB)
back
Steady State Probability
Another solution wiil be to use the closed loop solution from thebirth-and-death process of the Markov chain:
π(n ) =
[n∏
i+1
(N − i + 1)λµ(i )
]π(0)
π(0) is obtained by normalization
back
TBS (2)if we have two active UEs i.e. n=2those two users can be at any MCS (from MCS0 to MCS7)if we represent the number of users in each MCSk by nk we get:(n0, n1, n2, n3, n4, n5, n6, n7) with n0 + ...+ n7 = nthen we have the following possible combinations
back
TBS (2)
TBS (2) =(2,...,2)∑
(n0,...,nK )=(0,...,0)|n0+...+nK=0
n0 6=2
TBS (n0, ..., nK )(
2n0, ..., nK
)[K∏
k=0
P nkk
]
(3)
back
Q Infinitesimal Generator Matrix
0,1
1,1
0,0
1,0
0,2
1,2
0,N2
1,N2
N1,1N1,0 N1,2 N1,N2
N2λ2 (N2-1)λ2 (N2-2)λ2 λ2
µ2(N1,1) µ2(N1,2) µ2(N1,3) µ2(N1,N2)
N1λ 1
λ 1(N
1-1)λ
1
µ1(1,0)
µ1(2,0)
µ1(N
1,0)
N1,1N1,0 N1,2 N1,N20,1 1,10,0 1,00,2 1,20,N2 1,N2
back
Q Infinitesimal Generator Matrix
(0,0) (0,1) (0,2) (0,3) (0,4)
(0,0)
(1,0) (1,1) (1,2) (1,3) (1,4) (2,0) (2,1) (2,2) (2,3) (2,4)
(0,1)
(0,2)
(0,3)
(0,4)
(1,0)
(1,1)
(1,2)
(1,3)
(1,4)
(2,0)
(2,1)
(2,2)
(2,3)
(2,4)
4λ1 2λ2
μ1
(0,1)3λ1 2λ2
μ1
(0,2)2λ1 2λ2
μ1
(0,3)λ1 2λ2
μ1
(0,4)0 2λ2
μ2
(1,0)
μ1
(1,0)4λ1 λ2
μ2
(1,1)
μ1
(1,1)3λ1 λ2
μ2
(1,2)
μ1
(1,2)2λ1 λ2
μ2
(1,3)
μ1
(1,3)λ1 λ2
μ2
(1,4)
μ1
(1,4)0 λ2
μ2
(2,0)
μ1
(2,0)4λ1
μ2
(2,1)
μ1
(2,1)3λ1
μ2
(2,2)
μ1
(2,2)2λ1
μ2
(2,3)
μ1
(2,3)λ1
μ2
(2,4)
μ1
(2,4)
Markov chain state (i,j)
Ma
rko
v c
ha
in s
tate
(i,j)
Q (i , i ) = −∑j
Q (i , j )
back
Performance ParametersThe average number of active users Q is expressed as:
Q 1 =i=0∑N 1
j=0∑N 2
iπ(i , j )
The mean number of departures by unit of time is D and is obtainedas:
D 1 =i=0∑N 1
j=0∑N 2
µ1(i , j )π(i , j )
According to Little’s law, the average ON period duration tON (i.e.duration of an active transfer):
t1on =Q 1D 1
back
Departure Rate µ1(i , j )
µ1(i , j ) = TBS 1(i ,j )X 1on ·TTI
TBS 1(i , j ) is the average amount of bits sent by all served UEs in aTTI
X 1on is the average file size of class 1 UEs
TTI is the Transmission Time Interval (1ms)
back
Generic Average Bit Rate TBS 1(i , j )
TBS 1(i , j ) =(i ,...,i ,j ,...,j )∑
(i0,...,iK ,j0,...,jK )=(0,...,0)|i0+...+iK=ij0+...+jK=j
i0 6=i with j0 6=j
i !j !TBS 1(i0, ..., iK , j0, ..., jK )
[K∏
k=0
P ikk P jk
k
ik !jk !
]
Pk is the stationary probability of MCSk
ik is the number of class 1 active users in MCSk
jk is the number of class 2 active users in MCSk
TBS 1(i0, ..., iK , j0, ..., jK ) is the total bits transmitted for all served UEsunder a certain combination (i0, ..., iK , j0, ..., jK )
back
TBS 1(i0, ..., iK , j0, ..., jK )
TBS 1(i0, ..., iK , j0, ..., jK ) =∑
r=[r1,...,rδ1]
TBSr (η)
rδ1 + rδ2 = η
r is a vector representing the indices of the served users in class1
rδ1 is the number of users served in class1.
TBS 2(i0, ..., iK , j0, ..., jK ) can be calculated similarly
back
choosing eta Users
AverageThroughput =K∑
k=1
TBSk · Pk
K is the number of MCSs
TBSk is the transport block size using MCSk
Pk is the static probability of MCSk
Avg. throughput for RWP is MCS5
Avg. throughput for RD is MCS4
back
Analytical Model ScalingThese measurements were performed in MATLAB using a server with AMD Phenom(tm) 9850 Quad-Core Processor 2.50 GHz,8.00 GB of RAM, and 64-bit Windows 7 OS.
back
Model Number of UEs Combinations Run time
1-D 10 UEs 19448 2 seconds20 UEs 888029 79 seconds
2-D (3,6) UEs 205919 49 seconds3-D (2,3,5) UEs 3421440 13 minutes ??
1−D 2−D 3−D0
0.5
1
1.5
2
2.5
3
3.5
4x 10
6
# of Combinations