slide11 icc2015
TRANSCRIPT
User Scheduling for Massive MIMO OFDMASystems with Hybrid Analog-Digital Beamforming
Tadilo Endeshaw Bogale
Institute National de la Recherche (INRS), Canada
June 9, 2015, (ICC 2015)
Presentation outline
Presentation outline
1 IntroductionScenario and ObjectiveSystem Model and Problem Formulation
2 Proposed Solution
3 Performance Analysis
4 Simulation Results
5 Conclusions
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 2 / 11
Introduction Scenario and Objective
System Scenario and Objective
BS
a1 · · · aM
MS1
MS2
MSKt
h 1
h2
hK
t
...
System Scenario
MS1, MS2, MSKt are decentralized in spaceDownlink Communications⇒ Downlink Multiuser systemMS1, MS2, MSKt have single antennas⇒ Downlink Multiuser MISO systemBS has N antennas but Na < N RF chains⇒ BS use hybrid analog-digital architecture
(To get better performance (Clear later!))Channel between Tx and Rx is Freq. Selective
BS
a1 · · · aM
MS1
MS2
MSKt
h 1
h2
hK
t
...
System Scenario
MS1, MS2, MSKt are decentralized in spaceDownlink Communications⇒ Downlink Multiuser systemMS1, MS2, MSKt have single antennas⇒ Downlink Multiuser MISO systemBS has N antennas but Na < N RF chains⇒ BS use hybrid analog-digital architecture
(To get better performance (Clear later!))Channel between Tx and Rx is Freq. Selective
Objective
Schedule Kt MSs (Ki ,∀i)To maximize the Overall Data RateS.t. Per sub-carrier Power constraint
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 3 / 11
Introduction Scenario and Objective
System Scenario and Objective
BS
a1 · · · aM
MS1
MS2
MSKt
h 1
h2
hK
t
...
System Scenario
MS1, MS2, MSKt are decentralized in spaceDownlink Communications⇒ Downlink Multiuser systemMS1, MS2, MSKt have single antennas⇒ Downlink Multiuser MISO systemBS has N antennas but Na < N RF chains⇒ BS use hybrid analog-digital architecture
(To get better performance (Clear later!))Channel between Tx and Rx is Freq. Selective
BS
a1 · · · aM
MS1
MS2
MSKt
h 1
h2
hK
t
...
System Scenario
MS1, MS2, MSKt are decentralized in spaceDownlink Communications⇒ Downlink Multiuser systemMS1, MS2, MSKt have single antennas⇒ Downlink Multiuser MISO systemBS has N antennas but Na < N RF chains⇒ BS use hybrid analog-digital architecture
(To get better performance (Clear later!))Channel between Tx and Rx is Freq. Selective
Objective
Schedule Kt MSs (Ki ,∀i)To maximize the Overall Data RateS.t. Per sub-carrier Power constraint
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 3 / 11
Introduction System Model and Problem Formulation
System Model and Problem FormulationSource Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 4 / 11
Introduction System Model and Problem Formulation
System Model and Problem Formulation
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 4 / 11
Introduction System Model and Problem Formulation
System Model and Problem Formulation
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 4 / 11
Introduction System Model and Problem Formulation
System Model and Problem Formulation
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 4 / 11
Introduction System Model and Problem Formulation
System Model and Problem Formulation
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 4 / 11
Introduction System Model and Problem Formulation
System Model and Problem Formulation
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
REREFENCES1 X. Zhang, A. F. Molisch, and S-Y. Kung, ”Variable-phase-shift-based RF-Baseband codesign for MIMO antenna
selection” IEEE Trans. Signal Process., vol. 53, no. 11, pp. 4091 - 4103, Nov. 2005.
2 T. E. Bogale, L. Le, A. Haghighat, and L. Vandendorpe ”On the Number of RF Chains and Phase Shifters, andScheduling Design with Hybrid Analog-Digital Beamforming,” IEEE Trans. (Submitted),http://arxiv.org/abs/1410.2609.
Source Tx (Digital part) RF Chain Tx (Analog part)
Freq. Dom.
data source
D(1,:)
D(2,:)
D(K,:)
••
•
Freq. Dom.
BF
1
2
Na
••
•
IFFT (row)
& add CP
1
2
Na
••
•
RF1
analog
RF2
analog
RFNa
analog
••
•
AnalogBF
AnalogBF
AnalogBF
••
•
N
N
N
••
•
2
2
2
1
1
1
1
2
N
••
•
••
•
••
•
••
•
••
•
••
•
H
1 Discard CP& take FFT
di1 Decoded11, · · · , dNf1
2 Discard CP& take FFT
di2 Decoded12, · · · , dNf2
K Discard CP& take FFT
diK Decoded1K , · · · , dNfK
••
•
d1, · · · ,dNfB1, · · · ,BNf A
dik = hHikABidi + nik
FH F
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
For one Sub-carrier (Flat fading)
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax
Kt : Total number of usersK : Scheduled usersAssume that we have Na = K RF chains
First Possibility
maxA,B
K∑k=1
log(1 + γk )
s.t tr{BHAHAB} ≤ Pmax , |Aij |2 = 1
Disadvantage:How far from digital scheduler which isan optimal approach?(Not clearly known for general channel)
Second Possibility
maxBd
K∑k=1
log(1 + γk )
s.t tr{(Bd )HBd} ≤ Pmax
Exploit the solution of digital scheduler:Fact: rank(Bd ) ≤ K for any schedulerClever Method [1]: Bd = UD, U ∈ CN×2K ,D ∈ R2K×K , D = blkdiag matrix∴ 2K RF chain: No performance lossVIP: No need to constrain |Aij |2 = 1
Observations
U and D are not uniqueQuestion: Can we make U and D uniqueand reduce RF chain < 2KAnswer: Yes, by employing [2]x = ej cos−1( x
2 ) + e−j cos−1( x2 ), for −2 ≤ x ≤ 2
⇒ Bd = UD, with |Uij |2 = 1 andD = blkdiag two consecutive diag elementsare the same⇒ K RF chain is enough [2]∴ Scheduling without |Aij |2 = 1 (Simpler)
Problem Formulation
maxA,Bi
Nf∑i=1
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (ABi)
s.t tr{BHi AHABi} ≤ Pi
Ki : Users served by sub-carrier iB = [B1,B2, · · · ,BNf ], Bi ∈ CNa×Si
γik : SINR of i th sub-carrier k th userA ∈ CNa×N : Realized using PSs only
(Expected to be more constrained!)
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 4 / 11
Proposed Solution
Proposed Solution
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Optimize Bi for fixed A
For fixed A: Each sub-carrier canperform scheduling independently
maxBi
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (Bi)
s.t tr{BHi AHABi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamforming
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Optimize Bi for fixed A
For fixed A: Each sub-carrier canperform scheduling independently
maxBi
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (Bi)
s.t tr{BHi AHABi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamforming
Conclusions
A user may be scheduledin one (more) sub-carriersProposed solution maynot be global optimal
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 5 / 11
Proposed Solution
Proposed Solution
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Optimize Bi for fixed A
For fixed A: Each sub-carrier canperform scheduling independently
maxBi
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (Bi)
s.t tr{BHi AHABi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamforming
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Optimize Bi for fixed A
For fixed A: Each sub-carrier canperform scheduling independently
maxBi
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (Bi)
s.t tr{BHi AHABi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamforming
Conclusions
A user may be scheduledin one (more) sub-carriersProposed solution maynot be global optimal
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 5 / 11
Proposed Solution
Proposed Solution
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Optimize Bi for fixed A
For fixed A: Each sub-carrier canperform scheduling independently
maxBi
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (Bi)
s.t tr{BHi AHABi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamforming
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Optimize Bi for fixed A
For fixed A: Each sub-carrier canperform scheduling independently
maxBi
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (Bi)
s.t tr{BHi AHABi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamforming
Conclusions
A user may be scheduledin one (more) sub-carriersProposed solution maynot be global optimal
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 5 / 11
Proposed Solution
Proposed Solution
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Optimize Bi for fixed A
For fixed A: Each sub-carrier canperform scheduling independently
maxBi
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (Bi)
s.t tr{BHi AHABi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamforming
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Optimize Bi for fixed A
For fixed A: Each sub-carrier canperform scheduling independently
maxBi
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (Bi)
s.t tr{BHi AHABi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamforming
Conclusions
A user may be scheduledin one (more) sub-carriersProposed solution maynot be global optimal
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 5 / 11
Proposed Solution
Proposed Solution
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Optimize Bi for fixed A
For fixed A: Each sub-carrier canperform scheduling independently
maxBi
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (Bi)
s.t tr{BHi AHABi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamforming
Input: Hki ,Ki ,Pi ,Na
Solve Relaxed Problem
rank(B) ≤ Na?
Get A,Bi : From SVD(B)
FINISH
Yes
Determine A
Given A: Optimize Bi
FINISH
No
Determination of A
Sort f (B1) ≥ f (B2) ≥, · · · ,≥ f (BNf )
Get Bd = [B1, B2, · · · , BS]
S: Min No of SC with rank(B) ≥ Na
Compute SVD(B) = UΛVH
Λ arranged in decreasing order.Set A: First Na columns of U
Relaxed Problem
Assume the system as if it is DB(i.e., implicitly uses Na = N and A = I)⇒ Scheduling independently for each SC
maxBi
Ki∑k=1
log(1 + γik ) , f (Bi),
s.t tr{BHi Bi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamformingSet B = [B1, B2, · · · , BNf ]
Optimize Bi for fixed A
For fixed A: Each sub-carrier canperform scheduling independently
maxBi
Ki∑k=1
log(1 + γik ) ,Nf∑
i=1
f (Bi)
s.t tr{BHi AHABi} ≤ Pi
Gready based SchedulerZero forcing (ZF) beamforming
Conclusions
A user may be scheduledin one (more) sub-carriersProposed solution maynot be global optimal
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 5 / 11
Performance Analysis
Performance AnalysisApproaches
Three approaches examined:Antenna Selection Beamforming (ASB):Implicitly assumes A = IN×Na
Proposed Hybrid Beamforming (HB)Digital Beamforming (DB):Assumes N RF chains and A = IN
Intuitive expectationP(ASB) ≤ P(HB) ≤ P(DB)
To what extent?In what scenario?
Approaches
Three approaches examined:Antenna Selection Beamforming (ASB):Implicitly assumes A = IN×Na
Proposed Hybrid Beamforming (HB)Digital Beamforming (DB):Assumes N RF chains and A = IN
Intuitive expectationP(ASB) ≤ P(HB) ≤ P(DB)
To what extent?In what scenario?
Performance: Rayleigh Channel
Rayleigh Fading Channel:
Lemma 1: ZF beamforming, Rayleigh fading hHk and large Kt ,
we have
RHBi ≥ RASB
i when KHBi = KASB
i , and
RHBi = RDB
i when KHBi = KDB
i
KHBi = KDB
i may not hold, ⇒ Average performance makes senseTheorem 1: ZF beamforming with Pi = P,Ki = K and Rayleighfading channel hH
k , we have
E{RASB} ≤ KNf log2
(1 +
PK
E{χNa−K+1max (Kg)}
)E{RDB} ≤ KNf log2
(1 +
PK
E{χN−K+1max (Kg)}
)E{RHB} ≤ K S log2
(1 +
PK
E{χN−K+1max (Ks)}
)+ K (Nf − S) log2
(1 +
PK
E{χNa−K+1max (Kg)}
)where E{χM
max (L)}: Max. of L independent χ2 each with M DOF,S ≥ 1, Kg = dKt
K e and Ks = dKt NfKNae
Simulation: Bound is tight
Approaches
Three approaches examined:Antenna Selection Beamforming (ASB):Implicitly assumes A = IN×Na
Proposed Hybrid Beamforming (HB)Digital Beamforming (DB):Assumes N RF chains and A = IN
Intuitive expectationP(ASB) ≤ P(HB) ≤ P(DB)
To what extent?In what scenario?
Performance: Rayleigh Channel
Rayleigh Fading Channel:
Lemma 1: ZF beamforming, Rayleigh fading hHk and large Kt ,
we have
RHBi ≥ RASB
i when KHBi = KASB
i , and
RHBi = RDB
i when KHBi = KDB
i
KHBi = KDB
i may not hold, ⇒ Average performance makes senseTheorem 1: ZF beamforming with Pi = P,Ki = K and Rayleighfading channel hH
k , we have
E{RASB} ≤ KNf log2
(1 +
PK
E{χNa−K+1max (Kg)}
)E{RDB} ≤ KNf log2
(1 +
PK
E{χN−K+1max (Kg)}
)E{RHB} ≤ K S log2
(1 +
PK
E{χN−K+1max (Ks)}
)+ K (Nf − S) log2
(1 +
PK
E{χNa−K+1max (Kg)}
)where E{χM
max (L)}: Max. of L independent χ2 each with M DOF,S ≥ 1, Kg = dKt
K e and Ks = dKt NfKNae
Simulation: Bound is tight
Performance: ULA Channel
Rayleigh:DB Scheduler: Likely chooses usersclose to orthogonal each otherProposed HB achieves lower sum rate
(Mainly due to rank(H) > Na)∴ if rank(H) ≤ Na then RHB
i ≈ RDBi
ULA: Condition rank(H) / Na existsLemma 2: When d = λ
2 and the AOD ofthe Kt users satisfy sin (θkm) ∈n sin (θ)[− 1
2N ,1
2N ], n = 1,2, · · · ,Na,where θ is an arbitrary angle,
KHBi = KDB
i and RHBi = RDB
i , ∀i
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 6 / 11
Performance Analysis
Performance Analysis
Approaches
Three approaches examined:Antenna Selection Beamforming (ASB):Implicitly assumes A = IN×Na
Proposed Hybrid Beamforming (HB)Digital Beamforming (DB):Assumes N RF chains and A = IN
Intuitive expectationP(ASB) ≤ P(HB) ≤ P(DB)
To what extent?In what scenario?
Approaches
Three approaches examined:Antenna Selection Beamforming (ASB):Implicitly assumes A = IN×Na
Proposed Hybrid Beamforming (HB)Digital Beamforming (DB):Assumes N RF chains and A = IN
Intuitive expectationP(ASB) ≤ P(HB) ≤ P(DB)
To what extent?In what scenario?
Performance: Rayleigh Channel
Rayleigh Fading Channel:
Lemma 1: ZF beamforming, Rayleigh fading hHk and large Kt ,
we have
RHBi ≥ RASB
i when KHBi = KASB
i , and
RHBi = RDB
i when KHBi = KDB
i
KHBi = KDB
i may not hold, ⇒ Average performance makes senseTheorem 1: ZF beamforming with Pi = P,Ki = K and Rayleighfading channel hH
k , we have
E{RASB} ≤ KNf log2
(1 +
PK
E{χNa−K+1max (Kg)}
)E{RDB} ≤ KNf log2
(1 +
PK
E{χN−K+1max (Kg)}
)E{RHB} ≤ K S log2
(1 +
PK
E{χN−K+1max (Ks)}
)+ K (Nf − S) log2
(1 +
PK
E{χNa−K+1max (Kg)}
)where E{χM
max (L)}: Max. of L independent χ2 each with M DOF,S ≥ 1, Kg = dKt
K e and Ks = dKt NfKNae
Simulation: Bound is tight
Approaches
Three approaches examined:Antenna Selection Beamforming (ASB):Implicitly assumes A = IN×Na
Proposed Hybrid Beamforming (HB)Digital Beamforming (DB):Assumes N RF chains and A = IN
Intuitive expectationP(ASB) ≤ P(HB) ≤ P(DB)
To what extent?In what scenario?
Performance: Rayleigh Channel
Rayleigh Fading Channel:
Lemma 1: ZF beamforming, Rayleigh fading hHk and large Kt ,
we have
RHBi ≥ RASB
i when KHBi = KASB
i , and
RHBi = RDB
i when KHBi = KDB
i
KHBi = KDB
i may not hold, ⇒ Average performance makes senseTheorem 1: ZF beamforming with Pi = P,Ki = K and Rayleighfading channel hH
k , we have
E{RASB} ≤ KNf log2
(1 +
PK
E{χNa−K+1max (Kg)}
)E{RDB} ≤ KNf log2
(1 +
PK
E{χN−K+1max (Kg)}
)E{RHB} ≤ K S log2
(1 +
PK
E{χN−K+1max (Ks)}
)+ K (Nf − S) log2
(1 +
PK
E{χNa−K+1max (Kg)}
)where E{χM
max (L)}: Max. of L independent χ2 each with M DOF,S ≥ 1, Kg = dKt
K e and Ks = dKt NfKNae
Simulation: Bound is tight
Performance: ULA Channel
Rayleigh:DB Scheduler: Likely chooses usersclose to orthogonal each otherProposed HB achieves lower sum rate
(Mainly due to rank(H) > Na)∴ if rank(H) ≤ Na then RHB
i ≈ RDBi
ULA: Condition rank(H) / Na existsLemma 2: When d = λ
2 and the AOD ofthe Kt users satisfy sin (θkm) ∈n sin (θ)[− 1
2N ,1
2N ], n = 1,2, · · · ,Na,where θ is an arbitrary angle,
KHBi = KDB
i and RHBi = RDB
i , ∀i
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 6 / 11
Performance Analysis
Performance Analysis
Approaches
Three approaches examined:Antenna Selection Beamforming (ASB):Implicitly assumes A = IN×Na
Proposed Hybrid Beamforming (HB)Digital Beamforming (DB):Assumes N RF chains and A = IN
Intuitive expectationP(ASB) ≤ P(HB) ≤ P(DB)
To what extent?In what scenario?
Approaches
Three approaches examined:Antenna Selection Beamforming (ASB):Implicitly assumes A = IN×Na
Proposed Hybrid Beamforming (HB)Digital Beamforming (DB):Assumes N RF chains and A = IN
Intuitive expectationP(ASB) ≤ P(HB) ≤ P(DB)
To what extent?In what scenario?
Performance: Rayleigh Channel
Rayleigh Fading Channel:
Lemma 1: ZF beamforming, Rayleigh fading hHk and large Kt ,
we have
RHBi ≥ RASB
i when KHBi = KASB
i , and
RHBi = RDB
i when KHBi = KDB
i
KHBi = KDB
i may not hold, ⇒ Average performance makes senseTheorem 1: ZF beamforming with Pi = P,Ki = K and Rayleighfading channel hH
k , we have
E{RASB} ≤ KNf log2
(1 +
PK
E{χNa−K+1max (Kg)}
)E{RDB} ≤ KNf log2
(1 +
PK
E{χN−K+1max (Kg)}
)E{RHB} ≤ K S log2
(1 +
PK
E{χN−K+1max (Ks)}
)+ K (Nf − S) log2
(1 +
PK
E{χNa−K+1max (Kg)}
)where E{χM
max (L)}: Max. of L independent χ2 each with M DOF,S ≥ 1, Kg = dKt
K e and Ks = dKt NfKNae
Simulation: Bound is tight
Approaches
Three approaches examined:Antenna Selection Beamforming (ASB):Implicitly assumes A = IN×Na
Proposed Hybrid Beamforming (HB)Digital Beamforming (DB):Assumes N RF chains and A = IN
Intuitive expectationP(ASB) ≤ P(HB) ≤ P(DB)
To what extent?In what scenario?
Performance: Rayleigh Channel
Rayleigh Fading Channel:
Lemma 1: ZF beamforming, Rayleigh fading hHk and large Kt ,
we have
RHBi ≥ RASB
i when KHBi = KASB
i , and
RHBi = RDB
i when KHBi = KDB
i
KHBi = KDB
i may not hold, ⇒ Average performance makes senseTheorem 1: ZF beamforming with Pi = P,Ki = K and Rayleighfading channel hH
k , we have
E{RASB} ≤ KNf log2
(1 +
PK
E{χNa−K+1max (Kg)}
)E{RDB} ≤ KNf log2
(1 +
PK
E{χN−K+1max (Kg)}
)E{RHB} ≤ K S log2
(1 +
PK
E{χN−K+1max (Ks)}
)+ K (Nf − S) log2
(1 +
PK
E{χNa−K+1max (Kg)}
)where E{χM
max (L)}: Max. of L independent χ2 each with M DOF,S ≥ 1, Kg = dKt
K e and Ks = dKt NfKNae
Simulation: Bound is tight
Performance: ULA Channel
Rayleigh:DB Scheduler: Likely chooses usersclose to orthogonal each otherProposed HB achieves lower sum rate
(Mainly due to rank(H) > Na)∴ if rank(H) ≤ Na then RHB
i ≈ RDBi
ULA: Condition rank(H) / Na existsLemma 2: When d = λ
2 and the AOD ofthe Kt users satisfy sin (θkm) ∈n sin (θ)[− 1
2N ,1
2N ], n = 1,2, · · · ,Na,where θ is an arbitrary angle,
KHBi = KDB
i and RHBi = RDB
i , ∀i
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 6 / 11
Simulation Results
Simulation ResultsParameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
ULA: Selected phases, Ls = 8
0 5 10 15 200
10
20
30
40
50
60
70
80
SNR (dB)
Per
subc
arrie
r A
SR
(b/
s/hz
)
Existing ASBProposed HBDB
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
ULA: Selected phases, Ls = 8
0 5 10 15 200
10
20
30
40
50
60
70
80
SNR (dB)
Per
subc
arrie
r A
SR
(b/
s/hz
)
Existing ASBProposed HBDB
Other Results
Effect of power allocation policyObservation: Adaptive superior to Equal powerEffect of Kt
Observation: Rate increases as Kt increasesEffect of Na
Observation: Rate increases as Na increases
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 7 / 11
Simulation Results
Simulation Results
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
ULA: Selected phases, Ls = 8
0 5 10 15 200
10
20
30
40
50
60
70
80
SNR (dB)
Per
subc
arrie
r A
SR
(b/
s/hz
)
Existing ASBProposed HBDB
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
ULA: Selected phases, Ls = 8
0 5 10 15 200
10
20
30
40
50
60
70
80
SNR (dB)
Per
subc
arrie
r A
SR
(b/
s/hz
)
Existing ASBProposed HBDB
Other Results
Effect of power allocation policyObservation: Adaptive superior to Equal powerEffect of Kt
Observation: Rate increases as Kt increasesEffect of Na
Observation: Rate increases as Na increases
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 7 / 11
Simulation Results
Simulation Results
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
ULA: Selected phases, Ls = 8
0 5 10 15 200
10
20
30
40
50
60
70
80
SNR (dB)
Per
subc
arrie
r A
SR
(b/
s/hz
)
Existing ASBProposed HBDB
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
ULA: Selected phases, Ls = 8
0 5 10 15 200
10
20
30
40
50
60
70
80
SNR (dB)
Per
subc
arrie
r A
SR
(b/
s/hz
)
Existing ASBProposed HBDB
Other Results
Effect of power allocation policyObservation: Adaptive superior to Equal powerEffect of Kt
Observation: Rate increases as Kt increasesEffect of Na
Observation: Rate increases as Na increases
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 7 / 11
Simulation Results
Simulation Results
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
ULA: Selected phases, Ls = 8
0 5 10 15 200
10
20
30
40
50
60
70
80
SNR (dB)
Per
subc
arrie
r A
SR
(b/
s/hz
)
Existing ASBProposed HBDB
Parameter Settings
Number of antennas: 64Channel: Lk = 8 tap, Nf = 64,Scheduled users: Ki = Kmax = 8Power: Pi = P,∀iSINR: SNR = Nf P
Kmaxσ2
Total number of users: Kt = 8Policy: Equal power allocation
Rayleigh: Theory Vs Simulation
−1.5 1.5 4.5 7.5 10.5 13.5 16.5 19.5 22.50
10
20
30
40
50
60
70
80
90
SNR (dB)
Per
Sub
carr
ier
AS
R (
b/s/
hz)
SimulationTheory (Upper bound rate)
DB
Existing ASB
Proposed HB
ULA: Selected phases, Ls = 8
0 5 10 15 200
10
20
30
40
50
60
70
80
SNR (dB)
Per
subc
arrie
r A
SR
(b/
s/hz
)
Existing ASBProposed HBDB
Other Results
Effect of power allocation policyObservation: Adaptive superior to Equal powerEffect of Kt
Observation: Rate increases as Kt increasesEffect of Na
Observation: Rate increases as Na increases
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 7 / 11
Conclusions
Conclusions
In this work, we accomplish the following main tasks.We propose greedy like user scheduling algorithm with hybridanalog-digital beamformingThe proposed hybrid scheduling achieves better than antennaselection approachWe analyze the performance of the proposed hybrid schedulingfor Rayleigh and ULA channelsTheoretical results are confirmed via extensive simulations
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 8 / 11
Conclusions
References I
O. E. Ayach, S. Rajagopal, S. Abu-Surra, Z. Pi, and R. W. Heath,Spatially sparse precoding in millimeter wave MIMO systems,IEEE Trans. Wireless Commun. 13 (2014), no. 3, 1499 – 1513.
T. E. Bogale and L. B. Le, Beamforming for multiuser massiveMIMO systems: Digital versus hybrid analog-digital, Proc. IEEEGlobal Telecommun. Conf. (GLOBECOM) (Austin, Tx, USA), 10 –12 Dec. 2014.
T. E. Bogale, L. B. Le, A. Haghighat, and L. Vandendorpe, On thenumber of RF chains and phase shifters, and scheduling designwith hybrid analog-digital beamforming, IEEE Trans. WirelessCommun. (Submitted) (2014).
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 9 / 11
Conclusions
References II
S. Hur, T. Kim, D. J. Love, J. V. Krogmeier, T. A. Thomas, andA. Ghosh, Millimeter wave beamforming for wireless backhaul andaccess in small cell networks, IEEE Trans. Commun. 61 (2013),no. 10.
T. L. Marzetta, Noncooperative cellular wireless with unlimitednumbers of base station antennas, IEEE Trans. Wireless Commun.9 (2010), no. 11, 3590 – 3600.
S. Thoen, L. Van der Perre, B. Gyselinckx, and M. Engels,Performance analysis of combined Transmit-SC/Receive-MRC,IEEE Trans. Commun. 49 (2001), no. 1, 5 – 8.
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 10 / 11
Conclusions
References III
V. Venkateswaran and A-J. V. Veen, Analog beamforming in MIMOcommunications with phase shift networks and online channelestimation, IEEE Trans. Signal Process. 58 (2010), no. 8, 4131 –4143.
T. Yoo and A. Goldsmith, On the optimality of multiantennabroadcast scheduling using zero-forcing beamforming, IEEE Trans.Sel. Area. Commun. 24 (2006), no. 3, 528 – 541.
E. Zhang and C. Huang, On achieving optimal rate of digitalprecoder by RF-Baseband codesign for MIMO systems, Proc.IEEE Veh. Technol. Conf. (VTC Fall), 2014, pp. 1 – 5.
X. Zhang, A. F. Molisch, and S-Y. Kung, Variable-phase-shift-basedRF-Baseband codesign for MIMO antenna selection, IEEE Trans.Signal Process. 53 (2005), no. 11, 4091 – 4103.
Tadilo (ICC 2015, London, UK) User Scheduling June 9, 2015, (ICC 2015) 11 / 11