weighted cdf-based scheduling for an ofdma relay...
TRANSCRIPT
IntroductionSystem model and performance
Extension - with RelaysConclusions
Weighted CDF-based Scheduling for anOFDMA Relay Downlink with Partial Feedback
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao.University of California, San Diego
November 14, 2012
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Outline
1 Introduction
2 System model and performanceSystem modelAnalysisWeights settingExperimental results
3 Extension - with RelaysFast fadingSlow fading
4 Conclusions
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Outline
1 Introduction
2 System model and performanceSystem modelAnalysisWeights settingExperimental results
3 Extension - with RelaysFast fadingSlow fading
4 Conclusions
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
OFDMA system - Multiuser diversity
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
10−4
10−3
10−2
10−1
100
101
102
Frequency (MHz)
|h|2
Frequency selective fading
Best of 30 usersUser 1User 2
Effects of fading on channels of usersMultiuser diversity and frequency selectivity
Multiuser diversity: channels to different users are differentFrequency selectivity: channels on different frequency are different
Goal: Exploit the diversity to improve system’s performance
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Performance measures vs. Challenges
Measures of System’s PerformanceThroughputFairness: meet users’ requirementsFeedback constraint
ChallengesA large number of usersA large number of resource blocksDiversity in users’ characteristics
Users’ location: different channel gainand channel statistic...Users’ requirements: types of service,power, data rate, delay tolerance,...
At first, we introduce the system model
Multimedia
News
Pictures
…..
OFDM: NRB resource blocks
Web surfing
Download
Youtube …..
Voice
SMS …..
resource
block 4
resource
block 2,3,5
resource
block 1,6
Relay node
Figure: A multiuser system
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
System modelAnalysisWeights settingExperimental results
Outline
1 Introduction
2 System model and performanceSystem modelAnalysisWeights settingExperimental results
3 Extension - with RelaysFast fadingSlow fading
4 Conclusions
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
System modelAnalysisWeights settingExperimental results
System, Feedback and Scheduling
System: Groups of users with differentpriority
Groups of macro usersGroups of cell edge users, each group isserved by a relay
OFDM with Partial feedbackUsers feed back the best M amongchannels on N resource blocksOn resource block r , there are subset ofusers to feed back
SchedulingUsers are selected based on weightedcdf of the SNR.
Base Station
Selected user on resource bock r
Group 1
Group i
Users provide feedback
weighted w1
weighted wi
W
f1 f2 f3 f4 f5
best M=2
N=5 resource blocks
Figure: partial feedback in anOFDMA signal
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
System modelAnalysisWeights settingExperimental results
Weighted CDF scheduling
How weighted CDF work [5]From SNR Yki of user ki
Obtain uki = FYki(x),
Uniformly distributed in [0, 1]Identical for all users
Compare and prioritize usersSelect the user with the largest weighted CDFk∗ = arg maxk{FYki
(x)}1
wi
Advantage: can control precisely selection probability of all the users
[5] D. Park, H. Seo, H. Kwon, and B. Lee, “Wireless packet scheduling based on the cumulativedistribution function of user transmission rates", IEEE Trans on Communications, Nov. 2005.
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
System modelAnalysisWeights settingExperimental results
System’s performance - average sum rate
The average system sum rate is
R =1N
N∑r=1
E log(1 + Xr ) = E log(1 + Xr ) (1)
where Xr is SNR to the selected user on resource block r
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
System modelAnalysisWeights settingExperimental results
Performance analysis
Steps in analyzing system performance [4]Framework CQI feedback
Random variable OutputStep 1 Zk ,r : CQI at a receiver FZk
Step 2 Yk ,r : SNR seen at a transmitter FYk
Step 3 Xr : SNR of a selected user FX |cond
Step 4a FX = EcondFX |cond
Step 4b EcondEXr [log(1 + Xr )|cond ]k: user index, r: block index
[4] Seong-Ho Hur, and Bhaskar Rao, “Sum rate analysis of a reduced feedback
OFDMA system employing joint scheduling and diversity", IEEE Transactions on Signal
Processing, 2011.
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
System modelAnalysisWeights settingExperimental results
Performance analysis
TheoremIn an OFDMA system where all equipments have a single antenna, with Lgroups of users, each group i has Ki users, if only CQI on M best among Nresource blocks is fed back, the CDF of system’s throughput is
FR(ζ(x)) =L∑
l=1
Kj∑nj =0
j=1,...,L;n̄ 6=0
L∏j=1
Pr{|Sr ,j | = nj}
×
nl wl∑Lj=1 nj wj
∑∞t=1 e3(αl , t)
∑t(M−1)m=0 e2(m)FZk
(xρ
)Nt−mαl 6= 1∑M−1
m=0 e1(m)∑N−m
k=0 FZk
(xρ
)N−mαl = 1
,
(2)
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
System modelAnalysisWeights settingExperimental results
Performance analysis
where
ζ(x) = log(1 + x)
FZk
(xρ
)is CDF of SNR from the BS to user k
Pr{|Sr ,l | = nl} =( Kl
nl
) (MN
)nl (1− MN
)Kl−nl
αl =∑L
j=1 wj nj
nl wl;
e1(m) =∑M−1
i=mM−i
M ( Ni )(
im )(−1)i−m
e2(0) = e1(0)t ; e2(m) = 1me1(0)
∑min(M−1,m)k=1 (kt −m + k)e1(k)e2(m − k);
e3(αl , t) =∑∞
i=t (αli )(
it
)(−1)i−t .
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
System modelAnalysisWeights settingExperimental results
Weights for groups of users
The probability of selection of user kl is
Pr{k∗r = kl} =
∑π(n̄)
wl∑Lj=1 njwj
nl
Kl
( K1n1
)( KL
nL
)(MN
)∑Lj=1 nj
(1− M
N
)∑Lj=1(Kj−nj )
. (3)
Initialize w(0) = [Palloc,1
Kl, . . . ,
Palloc,LKL
] which isthe weights for the full feedback case.
Solving δw(t)φ(w(t)) = −∇φ(w(t)).
Update w(t + 1) = w(t) + δw(t). Normalizew so that ‖w‖2 = 1 which does not changeφ(w).
End ‖φ(w)‖2 < ε.
Set ε = 10−10
Target Palloc = [0.2, 0.8]
Found weightw = [0.318, 0.682] after 4iterations
Iteration Norm ‖φ(w)‖2
0 2.4691512e-0011 6.2304622e-0022 3.0325582e-0033 7.4881615e-0064 4.5758897e-011
Table: Convergence behaviorAnh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
System modelAnalysisWeights settingExperimental results
Experimental results
We consider an OFDMA system withN = 10 resource blocks and groups ofusers with different priority
Group 1K1 = 10 users, weight w1 = 0.4,located at d1 = 414m
Group 2 - cell edgeK2 = 5 users, weight w2 = 0.6,located at d2 = 834m
Base Station
Selected user on resource bock r
Group 1
Group 2
weighted w1
weighted w2
Figure: A partial feedback OFDMAsystem
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
System modelAnalysisWeights settingExperimental results
System’s performance with partial feedback
27 28 29 30 31 32 33 34 35 36 37 383
3.5
4
4.5
5
5.5
6
6.5
7
Power (dBm)
Thr
ough
put b
ps/H
z
Achievable throughput with weighted priotization with K=[10,5], N=10, w=[0.4,0.6]
M=1 simulatedM=2 simulatedM=3 simulatedM=4 simulatedM=5 simulatedM=1 analyzedM=2 analyzedM=3 analyzedM=4 analyzedM=5 analyzed
Figure: Analyzed and simulated performance of a partial feedback OFDMAsystem with N = 10, K = [10,5], w = [0.4,0.6]
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
System modelAnalysisWeights settingExperimental results
System’s performance with different number of users
5 10 15 20 25 303
4
5
6
7
8
9
10
11
Number of users K (K=K1+K2; K1=2K2)
Thr
ough
put b
ps/H
z
Achievable throughput with weighted priotization with N=10, M=5 w=[0.4,0.6]
P=32 dBm simulatedP=37 dBm simulatedP=42 dBm simulatedP=47 dBm simulatedP=32 dBm analyzedP=37 dBm analyzedP=42 dBm analyzedP=47 dBm analyzed
Figure: Analyzed system throughput of a partial feedback OFDMA systemwith N = 10, M = 5, w = [0.4,0.6], K1 = 2K2
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Fast fadingSlow fading
Outline
1 Introduction
2 System model and performanceSystem modelAnalysisWeights settingExperimental results
3 Extension - with RelaysFast fadingSlow fading
4 Conclusions
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Fast fadingSlow fading
Fast fading BS-RS links
We consider an OFDMA system withN = 10 resource blocks and groups ofusers with different priority
Group 1K1 = 10 users, weight w1 = 0.4,located at d1 = 414m
Group 2 - cell edgeK2 = 5 users, weight w2 = 0.6,located at d2 = 834m
A relay located at dr = 815m
Base Station
Selected user on resource bock r
Group 1
Group 2
weighted w1
weighted w2
relay
Figure: A partial feedback OFDMAsystem
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Fast fadingSlow fading
System’s performance with relays
2 3 4 5 6 7 8 9 100
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1Achievable throughput with weighted priotization
Throughput bps/Hz
cdf
w=[0.5 0.5]T
w=[0.4 0.6]T
w=[0.2 0.8]T
without relaywith relay
Figure: Performance tradeoff due to the biased treatment with users; OFDMA systemwith N = 10, K = [10, 5], P = 37dBm, the distance BS-macro MS is d1 = 414m andthe distance BS-MS group 2 is d2 = 834m which are aided by a relay with power30dBm located 815m from the BS, full feedback is provided
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Fast fadingSlow fading
Slow fading BS-RS links
Group 1Macro users, located at d1 = 414m
5 groups (2,...,6) - at the cell edgeUsers, each group served by arelaysLocated at d2 = 834m
A relay located at dr = 815mParameters
Number of users K = [2 2 2 3 3 3]Weightsw = [0.05 0.1 0.15 0.2 0.2 0.3]Log-normal shadowing fading 8dB...
Base Station
Selected user on resource bock r
Group 1
Group 2
weighted w1
weighted w2
relay
Group Ng
weighted wNg
Figure: A partial feedback OFDMAsystem
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Fast fadingSlow fading
Slow fading BS-RS links
Allocation of users does not meetrequirementsGroups in resource starvation
Out of service if the coherencetime is large
Short term adjustment can notmaintain fairness among users
2 4 5 6 3 10
0.1
0.2
0.3
0.4
User groups
Sel
ectio
n P
roba
bilit
y
Long Term Fairness
Conditional CDF − simulatedConditional CDF − analyzedCDF of whole link BS−RS−MS
1 2 3 4 5 60
0.1
0.2
0.3
0.4
0.5
User groups
Sel
ectio
n P
roba
bilit
y
Short Term Fairness
Conditional CDF − simulatedConditional CDF − analyzedCDF of whole link BS−RS−MS
Figure: Selection probability of usersin an OFDMA system
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Fast fadingSlow fading
Slow fading BS-RS links
The proposed solutions
Interpolate CDF to create anartificial CDF
ContinuousUniformly distributed [0,1]
Can, again precisely controlallocation probability for users
2 4 5 6 3 10
0.1
0.2
0.3
0.4
User groups
Sel
ectio
n P
roba
bilit
y
Long Term Fairness
Conditional CDF − simulatedConditional CDF − analyzedCDF of whole link BS−RS−MS
1 2 3 4 5 60
0.1
0.2
0.3
0.4
0.5
User groups
Sel
ectio
n P
roba
bilit
y
Short Term Fairness
Conditional CDF − simulatedConditional CDF − analyzedCDF of whole link BS−RS−MS
Figure: Selection probability of usersin an OFDMA system
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Fast fadingSlow fading
Slow fading BS-RS links
Tradeoff in system’s performanceGoal: propose a modified technique
Similar allocation managementBetter tradeoff in performance
27 28 29 30 31 32 33 34 35 36 37 383
3.5
4
4.5
5
5.5
6
6.5
7
Power (dBm)
Sys
tem
thro
ughp
ut
Performance tradeoff − CDF based scheduling in Relay OFDMA networks
long term fairnessshort term fairness
Figure: Tradeoff in system’sthroughput
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Outline
1 Introduction
2 System model and performanceSystem modelAnalysisWeights settingExperimental results
3 Extension - with RelaysFast fadingSlow fading
4 Conclusions
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
Conclusion
Consider a Weighted CDF-based scheduling OFDMA system withpartial feedback
Ensures fairnessSupports user priorityExploits multiuser diversity users
Developed an analytical expression for system throughputThe simulations verify the performance of the scheduling methodon OFDMA systems
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego
IntroductionSystem model and performance
Extension - with RelaysConclusions
ReferencesAnh H. Nguyen, and Bhaskar D. Rao, “User selection schemes for maximizingthroughput of multiuser MIMO systems using Zeroforcing Beamforming",ICASSP, 2011.
Anh Nguyen, Yichao Huang, and Bhaskar Rao, “Weighted CDF-basedScheduling for an OFDMA Relay Downlink with Partial Feedback", AsilomarConference on Signals, Systems, and Computers, Nov. 2012.
T. Yoo, A. Goldsmith, "On the optimality of multiantenna broadcast schedulingusing zero-forcing beamforming", IEEE Journal on Selected Areas in Comms,Mar. 2006.
Seong-Ho Hur, and Bhaskar Rao, “Sum rate analysis of a reduced feedbackOFDMA system employing joint scheduling and diversity", IEEE Transactions onSignal Processing, 2011.
D. Park, H. Seo, H. Kwon, and B. Lee, “Wireless Packet Scheduling Based on theCumulative Distribution Function of User Transmission Rates", IEEE Trans onCommunications, Nov. 2005.
I. Gradshteyn, and I. Ryzhik, “Table of Integrals Series and Product", AcademicPress 2007.
G. Senarath et al. “Multi-hop relay system evaluation methodology (channelmodel and performance metric)",http://ieee802.org/16/relay/docs/80216j-06_013r3.pdf, Feb.2007.
Thank you!
Anh H. Nguyen, Yichao Huang, and Bhaskar D. Rao. University of California, San Diego