CommI Queue-Aware Beam Scheduling for Half-Duplex mmWave Relay - - PowerPoint PPT Presentation

commi
SMART_READER_LITE
LIVE PREVIEW

CommI Queue-Aware Beam Scheduling for Half-Duplex mmWave Relay - - PowerPoint PPT Presentation

Communications and Information Theory Chair CommI Queue-Aware Beam Scheduling for Half-Duplex mmWave Relay Networks Xiaoshen Song , Giuseppe Caire Technische Universitt Berlin June 21-26, 2020 Xiaoshen Song (TU Berlin)


slide-1
SLIDE 1

Queue-Aware Beam Scheduling for Half-Duplex mmWave Relay Networks

Xiaoshen Song, Giuseppe Caire Technische Universität Berlin June 21-26, 2020

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 1 / 25

CommI

Communications and Information Theory Chair

slide-2
SLIDE 2

Outline

1

Background

2

System Model

3

Proposed Beam Schedulers HC-EC scheduler BP scheduler newBP scheduler

4

Numerical Results

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 2 / 25

CommI

Communications and Information Theory Chair

slide-3
SLIDE 3

Outline

1

Background

2

System Model

3

Proposed Beam Schedulers HC-EC scheduler BP scheduler newBP scheduler

4

Numerical Results

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 3 / 25

CommI

Communications and Information Theory Chair

slide-4
SLIDE 4
  • 1. Background

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 4 / 25

mmWave communication

Being considered: The key enabler for 5G and beyond. Challenges: Severe path loss & vulnerability to obstacles. Solution: Beamforming + relaying CommI

Communications and Information Theory Chair

slide-5
SLIDE 5
  • 1. Background

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 4 / 25

mmWave communication

Being considered: The key enabler for 5G and beyond. Challenges: Severe path loss & vulnerability to obstacles. Solution: Beamforming + relaying

  • X. Song, S. Haghighatshoar, and G. Caire, "A scalable and statistically robust beam alignment

technique for mm-wave systems," IEEE Transactions on Wireless Communications, 2018.

  • X. Song, S. Haghighatshoar, and G. Caire, "Efficient beam alignment for mmWave single-carrier

systems with hybrid MIMO transceivers," IEEE Transactions on Wireless Communications, 2019.

  • X. Song, T. Kühne, and G. Caire, "Fully-/Partially-Connected Hybrid Beamforming Architectures

for mmWave MU-MIMO," IEEE Transactions on Wireless Communications, 2019.

CommI

Communications and Information Theory Chair

slide-6
SLIDE 6
  • 1. Background

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 4 / 25

mmWave communication

Being considered: The key enabler for 5G and beyond. Challenges: Severe path loss & vulnerability to obstacles. Solution: Beamforming + relaying

  • X. Song, S. Haghighatshoar, and G. Caire, "A scalable and statistically robust beam alignment

technique for mm-wave systems," IEEE Transactions on Wireless Communications, 2018.

  • X. Song, S. Haghighatshoar, and G. Caire, "Efficient beam alignment for mmWave single-carrier

systems with hybrid MIMO transceivers," IEEE Transactions on Wireless Communications, 2019.

  • X. Song, T. Kühne, and G. Caire, "Fully-/Partially-Connected Hybrid Beamforming Architectures

for mmWave MU-MIMO," IEEE Transactions on Wireless Communications, 2019.

CommI

Communications and Information Theory Chair

slide-7
SLIDE 7

Outline

1

Background

2

System Model

3

Proposed Beam Schedulers HC-EC scheduler BP scheduler newBP scheduler

4

Numerical Results

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 5 / 25

CommI

Communications and Information Theory Chair

slide-8
SLIDE 8
  • 2. System Model
  • 1Y. H. Ezzeldin, M. Cardone, C. Fragouli, and D. Tuninetti, "Efficiently finding simple schedules in Gaussian

half-duplex relay line networks," in 2017 IEEE International Symposium on Information Theory (ISIT), 2017, pp. 471-475.

  • 2Y. H. Ezzeldin, M. Cardone, C. Fragouli, and G. Caire, "Gaussian 1-2-1 networks: Capacity results for

mmwave communications," in 2018 IEEE International Symposium on Information Theory (ISIT), 2018, pp. 2569-2573. Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 6 / 25

Network model

S D (a) S D (b)

(a) The line network L.1 (b) The diamond network D.2

CommI

Communications and Information Theory Chair

slide-9
SLIDE 9
  • 2. System Model
  • 3Y. H. Ezzeldin, M. Cardone, C. Fragouli, and D. Tuninetti, "Efficiently finding simple schedules in Gaussian

half-duplex relay line networks," in 2017 IEEE International Symposium on Information Theory (ISIT), 2017, pp. 471-475. Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 7 / 25

Capacity results

The line network L

S D (a)

Relay nodes: i ∈ [N]. Source node: i = 0. Destination node: i = N + 1. Link capacity: li. The approximate capacity 3: CL. Within a gap: O(N).

CL = min

i∈[N]

li · li+1 li + li+1

  • CommI

Communications and Information Theory Chair

slide-10
SLIDE 10
  • 2. System Model
  • 4Y. H. Ezzeldin, M. Cardone, C. Fragouli, and G. Caire, "Gaussian 1-2-1 networks: Capacity results for

mmwave communications," in 2018 IEEE International Symposium on Information Theory (ISIT), 2018, pp. 2569-2573. Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 8 / 25

Capacity results

The diamond network D

S D (b)

Relay nodes: p ∈ [N]. Link capacity: lp,1, lp,2. Path capacity: Cp =

lp,1·lp,2 lp,1+lp,2 .

Path activation time fraction: xp. The approximate capacity 4: CD. Within a gap: O(N log N). CD = max

  • p∈[N]

xpCp s.t. 0 ≤ xp ≤ 1, ∀p ∈ [N]

  • p∈[N]

xp Cp lp,j ≤ 1, ∀j ∈ [2]

CommI

Communications and Information Theory Chair

slide-11
SLIDE 11

Outline

1

Background

2

System Model

3

Proposed Beam Schedulers HC-EC scheduler BP scheduler newBP scheduler

4

Numerical Results

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 9 / 25

CommI

Communications and Information Theory Chair

slide-12
SLIDE 12

Outline

3

Proposed Beam Schedulers HC-EC scheduler BP scheduler newBP scheduler

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 10 / 25

CommI

Communications and Information Theory Chair

slide-13
SLIDE 13

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 11 / 25

HC-EC (edge coloring) scheduler

S D (a)

Link Capapcity: N = 3, l1 = 8, l2 = 8, l3 = 12, l4 = 4.

CommI

Communications and Information Theory Chair

slide-14
SLIDE 14

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 11 / 25

HC-EC (edge coloring) scheduler

S D (a)

Link Capapcity: N = 3, l1 = 8, l2 = 8, l3 = 12, l4 = 4. The common multiple of li: M.

CommI

Communications and Information Theory Chair

slide-15
SLIDE 15

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 11 / 25

HC-EC (edge coloring) scheduler

S D (a)

Link Capapcity: N = 3, l1 = 8, l2 = 8, l3 = 12, l4 = 4. The common multiple of li: M. Assign parallel edges to each link: ni = M

li .

CommI

Communications and Information Theory Chair

slide-16
SLIDE 16

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 11 / 25

HC-EC (edge coloring) scheduler

S D (a)

Link Capapcity: N = 3, l1 = 8, l2 = 8, l3 = 12, l4 = 4. The common multiple of li: M. Assign parallel edges to each link: ni = M

li .

The running example: M = 24, n1 = 3, n2 = 3, n3 = 2, n4 = 6. Maximum node degree: ∆L = 8.

CommI

Communications and Information Theory Chair

slide-17
SLIDE 17

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 12 / 25

HC-EC (edge coloring) scheduler

S D (a)

The number of parallel edges: n1 = 3, n2 = 3, n3 = 2, n4 = 6.

Links i i = 1 i = 2 i = 3 i = 4 Assigned colors Ci

L

1 2 1 2

CommI

Communications and Information Theory Chair

slide-18
SLIDE 18

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 12 / 25

HC-EC (edge coloring) scheduler

S D (a)

The number of parallel edges: n1 = 3, n2 = 3, n3 = 2, n4 = 6.

Links i i = 1 i = 2 i = 3 i = 4 Assigned colors Ci

L

1 2 1 2 Links i i = 1 i = 2 i = 3 i = 4 Assigned colors Ci

L

1 2 1 2 3 4 3 4

CommI

Communications and Information Theory Chair

slide-19
SLIDE 19

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 12 / 25

HC-EC (edge coloring) scheduler

S D (a)

The number of parallel edges: n1 = 3, n2 = 3, n3 = 2, n4 = 6.

Links i i = 1 i = 2 i = 3 i = 4 Assigned colors Ci

L

1 2 1 2 Links i i = 1 i = 2 i = 3 i = 4 Assigned colors Ci

L

1 2 1 2 3 4 3 4 5 6 5

CommI

Communications and Information Theory Chair

slide-20
SLIDE 20

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 12 / 25

HC-EC (edge coloring) scheduler

S D (a)

The number of parallel edges: n1 = 3, n2 = 3, n3 = 2, n4 = 6.

Links i i = 1 i = 2 i = 3 i = 4 Assigned colors Ci

L

1 2 1 2 Links i i = 1 i = 2 i = 3 i = 4 Assigned colors Ci

L

1 2 1 2 3 4 3 4 5 6 5 6 7 8

CommI

Communications and Information Theory Chair

slide-21
SLIDE 21

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 12 / 25

HC-EC (edge coloring) scheduler

S D (a)

The number of parallel edges: n1 = 3, n2 = 3, n3 = 2, n4 = 6.

Links i i = 1 i = 2 i = 3 i = 4 Assigned colors Ci

L

1 2 1 2 Links i i = 1 i = 2 i = 3 i = 4 Assigned colors Ci

L

1 2 1 2 3 4 3 4 5 6 5 6 7 8

Scheduling decision: ΛL(t)i = 1{ˆ

t∈Ci

L}, ˆ

t = (t − 1) mod ∆L.

CommI

Communications and Information Theory Chair

slide-22
SLIDE 22

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 13 / 25

HC-EC (edge coloring) scheduler

S D (b)

Link capacity: Path activation time fraction: Parallel edges: l1,1 = 3, l1,2 = 3 l2,1 = 2, l2,2 = 3 l3,1 = 3, l3,2 = 2 l4,1 = 2, l4,2 = 2

x1 = 1 x2 = 0.5 x3 = 0.5 x4 = 0

n1,1 = 5, n1,2 = 5 n2,1 = 3, n2,2 = 2 n3,1 = 2, n3,2 = 3 n4,1 = 0, n4,2 = 0

CommI

Communications and Information Theory Chair

slide-23
SLIDE 23

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 13 / 25

HC-EC (edge coloring) scheduler

S D (b)

Parallel edges: n1,1 = 5, n1,2 = 5 n2,1 = 3, n2,2 = 2 n3,1 = 2, n3,2 = 3 n4,1 = 0, n4,2 = 0

Hops j j = 1 j = 2 Assigned colors Cp,j

D

C1,j

D

1 2 3 4 5 6 7 8 9 10 C2,j

D

2 1 4 3 6 C3,j

D

8 5 10 7 9

CommI

Communications and Information Theory Chair

slide-24
SLIDE 24

Outline

3

Proposed Beam Schedulers HC-EC scheduler BP scheduler newBP scheduler

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 14 / 25

CommI

Communications and Information Theory Chair

slide-25
SLIDE 25

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 15 / 25

BP (Backpressure) scheduler

S D (a)

Queuing backlog at each node: Ui(t). Link rate allocation: ri(t). Differential backlog weights: W (t)[i] = max{Ui−1(t) − Ui(t), 0}.

CommI

Communications and Information Theory Chair

slide-26
SLIDE 26

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 15 / 25

BP (Backpressure) scheduler

S D (a)

BIP: ΛL(t) = arg max

N+1

  • i=1

W (t)[i] · ri(t) s.t. ri(t) = ¯ ri(t) · ΛL(t)[i] ¯ ri(t) = min{Ui−1(t), li} ΛL(t)[i] ∈ {0, 1} ΛL(t)[j] + ΛL(t)[j+1] ≤ 1, j ∈ [N],

CommI

Communications and Information Theory Chair

slide-27
SLIDE 27

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 16 / 25

BP (Backpressure) scheduler

S D (b)

Queuing backlog at each node: Ui(t). Link rate allocation: ri,j(t). Differential backlog weights: W (t)[i,1] =max{U0(t)−Ui(t), 0}, W (t)[i,2] =Ui(t).

CommI

Communications and Information Theory Chair

slide-28
SLIDE 28

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 16 / 25

BP (Backpressure) scheduler

S D (b)

BIP: ΛD(t) = arg max

N

  • i=1

2

  • j=1

W (t)[i,j] · ri,j(t) s.t. ri,j(t) = ¯ ri,j(t) · ΛD(t)[i,j] ¯ ri,1(t) = min{U0(t), li,1}, ¯ ri,2(t) = min{Ui(t), li,2} ΛD(t)[i,j] ∈ {0, 1} ΛD(t)[i,:]1 ≤ 1, i ∈ [N], ΛD(t)[:,j]1 ≤ 1, j ∈ [2]

CommI

Communications and Information Theory Chair

slide-29
SLIDE 29

Outline

3

Proposed Beam Schedulers HC-EC scheduler BP scheduler newBP scheduler

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 17 / 25

CommI

Communications and Information Theory Chair

slide-30
SLIDE 30

3.Proposed Beam Schedulers

  • 5S. Wang and N. Shroff, "Towards fast-convergence, low-delay and low-complexity network optimization,"

Proceedings of the ACM on Measurement and Analysis of Computing Systems, vol. 1, p. 34, 2017. Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 18 / 25

newBP scheduler5

S D (a)

Choose parameters: ρ > 0, τ > 0, βi > 0 Physical queue backlog: Ui(t). Virtual queue backlog: Vi(t), Vi(t) = Vi(t − 1) − ρτDi(t) + ρτAi(t). New weights: zi(t) = (1 + 1

τ )Vi(t − 1) − 1 τ Vi(t − 2).

New differential backlog weights: ˜ W (t)[i] = zi−1(t) − zi(t).

CommI

Communications and Information Theory Chair

slide-31
SLIDE 31

3.Proposed Beam Schedulers

  • 5S. Wang and N. Shroff, "Towards fast-convergence, low-delay and low-complexity network optimization,"

Proceedings of the ACM on Measurement and Analysis of Computing Systems, vol. 1, p. 34, 2017. Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 18 / 25

newBP scheduler5

S D (a)

BIP: ΛL(t) =arg max

N+1

  • i=1

˜ W (t)[i]ri(t) − ρβi 2 [ri(t) − ri(t − 1)]2 s.t. as above,

CommI

Communications and Information Theory Chair

slide-32
SLIDE 32

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 19 / 25

newBP scheduler

S D (b)

Choose parameters: ρ > 0, τ > 0, βi,j > 0 Physical queue backlog: Ui(t). Virtual queue backlog: Vi(t). New weight: zi(t) = (1 + 1

τ )Vi(t − 1) − 1 τ Vi(t − 2).

New differential backlog weights: ˜ W (t)[i,1] = z0(t) − zi(t), ˜ W (t)[i,2] = zi(t).

CommI

Communications and Information Theory Chair

slide-33
SLIDE 33

3.Proposed Beam Schedulers

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 19 / 25

newBP scheduler

S D (b) BIP: ΛD(t) =arg max

  • i∈[N],j∈[2]

˜ W (t)[i,j] · ri,j(t) − ρβi,j 2 [ri,j(t) − ri,j(t − 1)]2 s.t. as above

CommI

Communications and Information Theory Chair

slide-34
SLIDE 34

Outline

1

Background

2

System Model

3

Proposed Beam Schedulers HC-EC scheduler BP scheduler newBP scheduler

4

Numerical Results

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 20 / 25

CommI

Communications and Information Theory Chair

slide-35
SLIDE 35

4.Numerical Results

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 21 / 25

Source input rate: Poisson distribution with mean ¯ A0. Number of arrivals: Ai(t). Number of departures: Di(t). Queuing evolution: Ui(t + 1) = max

  • Ui(t) − Di(t), 0
  • + Ai(t).

Performance metrics: Network stability & packet end-to-end delay.

CommI

Communications and Information Theory Chair

slide-36
SLIDE 36

4.Numerical Results

6[8] Y. H. Ezzeldin, M. Cardone, C. Fragouli, and D. Tuninetti, "Efficiently finding simple schedules in

Gaussian half-duplex relay line networks," in 2017 IEEE International Symposium on Information Theory (ISIT), 2017, pp. 471-475. Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 22 / 25

Network stability

1 1.5 2 2.5 3 50 100 150 200

Source input rate ¯ A0 (packet/s)

Average sum backlog ¯ U

(a)

EC in [8] HC-EC BP newBP

1 1.5 2 2.5 50 100 150

Source input rate ¯ A0 (packet/s) (b)

HC-EC BP newBP

The averaged sum backlog ¯ U w.r.t. different source input rates ¯ A0 for (a) the line network L, (b) the diamond network D.6

CommI

Communications and Information Theory Chair

slide-37
SLIDE 37

4.Numerical Results

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 23 / 25

End-to-end delay

10 15 20 25 30 0.8 0.85 0.9 0.95 1

End-to-end delay (slots)

CDF

(a)

HC-EC, ¯ A0 = 2.5 BP, ¯ A0 = 2.5 newBP, ¯ A0 = 2.5

10 20 30 40 50 60 0.8 0.85 0.9 0.95 1

End-to-end delay (slots)

CDF

(b)

HC-EC, ¯ A0 = 1.75 BP, ¯ A0 = 1.75 newBP, ¯ A0 = 1.75

The end-to-end delay comparison of the proposed three beam schedulers for (a) the line network L, (b) the diamond network D.

CommI

Communications and Information Theory Chair

slide-38
SLIDE 38

Work Extension

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 24 / 25

More general rely topology

S D (a) S D (b) S D (c)

  • X. Song, Yahya H. Ezzeldin, et al. "Joint Topology Simplification and Beam Scheduling for

Half-Duplex mmWave Relay Networks," (online available soon).

CommI

Communications and Information Theory Chair

slide-39
SLIDE 39

The End

Xiaoshen Song (TU Berlin) xiaoshen.song@campus.tu-berlin.de June 21-26, 2020 25 / 25

Stay Healthy!

[1] X. Song and G. Caire, "Queue-Aware Beam Scheduling for Half-Duplex mmWave Relay Networks." (ISIT 2020) [2] X. Song, Yahya H. Ezzeldin, et al. "Joint Topology Simplification and Beam Scheduling for Half-Duplex mmWave Relay Networks." (online available soon).

CommI

Communications and Information Theory Chair