Fronthaul Compression for Cloud Radio Access Networks O. Simeone - - PowerPoint PPT Presentation

Fronthaul Compression for Cloud Radio Access Networks

• O. Simeone

New Jersey Institute of Technology (NJIT)

Joint work with S.-H. Park1, O. Sahin2 and S. Shamai3

1 2 3

Cloud Radio Access Networks

• Base stations operate as radio units
• Baseband processing takes place in the “cloud”
• Fronthaul links carry complex (IQ)

baseband signals

Cloud Radio Access Networks

Advantages:

• Low-cost BSs
• Effective interference mitigation via joint baseband

processing Key challenge: Effective transfer of the IQ signals on the fronthaul links

Cloud Radio Access Networks

• CPRI standard based on ADC/DAC

… Need for fronthaul compression

• Point-to-point fronthaul compression:

– Algorithms [Segel and Weldon] [Samardzija et al ‘12] [Nieman and Evans ’13] – Testbed results [Irmer et al ’11] [Vosoughi et al ‘12]

State of the Art

State of the Art

• Multiterminal fronthaul compression:

– Uplink: Distributed source coding coding [Sanderovich et al ’09] [del Coso and Simoens ’09] [Zhou and Yu ’11] [Marsch and Fettweis ’11] – Downlink: Multivariate compression [Park et al ’13]

• Compute-and-forward:

– Uplink [Nazer et al ’09] [Hong and Caire ’11] – Downlink [Hong and Caire ‘12]

Overview

• Uplink

– Multiterminal compression – Compute-and-forward

• Downlink

– Multiterminal compression – Compute-and-forward

• Performance evaluation
• Extensions and conclusions

Overview

• Uplink

– Multiterminal compression – Compute-and-forward

• Downlink

– Multiterminal compression – Compute-and-forward

• Performance evaluation
• Extensions and conclusions

System Model

MS MS 1 RU1 RU RU

M

N

i

B

N

i

H

1

H

B

N

H

CU

1

C

i

C

B

N

C

1

y

i

y

B

N

y

1

ˆ y ˆ i y ˆ

B

N

y

Single-cluster single-hop fronthaul topology

RU 1 Decompressor Decoder RU NR

ul 1

y

ul 2

y

ul

R

N

y

Fronthaul

R

N

C

1

C

ul 1

ˆ y

ul 2

ˆ y

ul

ˆ

R

N

y

Control Unit Decompressor Decompressor

Point-to-Point Fronthaul Compression

Compressor RU 2

Fronthaul

2

C

Compressor

Fronthaul

Compressor

Decoder

ul (1) 

y

ul (2) 

y

ul ( )

R

N 

y

ul (1)

ˆ  y

ul (2)

ˆ  y

ul ( )

ˆ

R

N 

y

Control Unit Decompressor WZ Decompressor WZ Decompressor

Joint Fronthaul Decompression

RU 1 RU NR

Fronthaul

R

N

C

1

C

Compressor RU 2

Fronthaul

2

C

WZ Compressor

Fronthaul

WZ Compressor [Sanderovich et al ’09] [del Coso and Simoens ’09] [Zhou and Yu ’11] [Park et al ’13]

Joint Fronthaul Decompression

Point-to-point compression

ul

y

000 001 010 100

… …

101

ul

ˆ y

Joint Fronthaul Decompression

… Coset coding at the RU and channel decoding at the CU [Pradhan and Ramchandran ’03]

WZ compression

ul

y

ul

ˆ y

000 001 010 100 101

… …

Compute-and-Forward

[Nazer et al ’09] [Hong and Caire ’11]

• The MSs use (nested) lattice codes:

[B. Nazer]

Decoder Control Unit

Compute-and-Forward

[Nazer et al ’09] [Hong and Caire ’11]

RU 1 RU NR

ul 1

y

ul 2

y

ul

R

N

y

Fronthaul

R

N

C

1

C

Integer Decoder RU 2

Fronthaul

2

C

Fronthaul

Integer Decoder Integer Decoder

• Three-cell SISO circular Wyner model

   

Numerical Results



CU

C C C

Numerical Results

3 bit/s/Hz and =0.4 C  

5 10 15 20 25 30 1 1.5 2 2.5 3 MS transmit power [dB] per-cell sum-rate [bits/s/Hz] Cut-set upper bound Point-to-point compression Single-cell processing

Numerical Results

3 bit/s/Hz and =0.4 C  

5 10 15 20 25 30 1 1.5 2 2.5 3 MS transmit power [dB] per-cell sum-rate [bits/s/Hz] Cut-set upper bound Joint decompression Point-to-point compression Single-cell processing

Numerical Results

3 bit/s/Hz and =0.4 C  

5 10 15 20 25 30 1 1.5 2 2.5 3 MS transmit power [dB] per-cell sum-rate [bits/s/Hz] Cut-set upper bound Joint decompression Point-to-point compression Single-cell processing Compute-and-forward

Overview

• Uplink

– Multiterminal compression – Compute-and-forward

• Downlink

– Multiterminal compression – Compute-and-forward

• Performance evaluation
• Extensions and conclusions

System Model

MS MS 1 RU1 RU RU

M

N

i

B

N

1

H

M

N

H

CU

1

C

i

C

B

N

C

Single-cluster single-hop fronthaul topology

1,...,

M

N

M M

Point-to-Point Fronthaul Compression

1

C

1

M

Channel encoder 1 Precoding RU 1 Control Unit

M

N

M

Channel encoder NM

1

s

M

N

s

Compressor 1

1

x 

B

N

x 

B

N

C

  

1

x

RU

   B

N

x

B

N

B

N

Compressor

Joint Fronthaul Compression

[Park et al ’13]

1

C

1

M

Channel encoder 1 Precoding RU 1 Control Unit

M

N

M

Channel encoder NM

1

s

M

N

s

1

x 

B

N

x 

B

N

C

  

1

x

RU

   B

N

x

B

N

Joint compression

• Multivariate compression produces compressed signals

with correlated quantization noises

Joint Fronthaul Compression

can be reduced by controlling

1,2 2,1 H

 Ω Ω

RU 1 RU 2 MS

1 1 1 H

  x E As q

2 2 2 H

  x E As q

1,1

H

1,2

H

1

z

1 1 1 1 1 2

         q H q y H As z

1,1 1,2 1 1 2,1 2,2

,

H

              Ω Ω 0 H H Ω Ω CN

CU

1

C

2

C

Joint Fronthaul Compression

2

x

1

x

Point-to-point compression

Joint Fronthaul Compression

2

x

1

x

Point-to-point compression

Joint Fronthaul Compression

2

x

1

x

Multivariate compression

• Successive estimation-compression implementation [Park et

al ’13]:

(1) 

x 

Compressor Compressor

(2) 

x 

Compressor

( )

B

N 

x 

RU π(1) RU π(2) RU π(NB)

(1) 

x

( )

B

N 

x

(2) 

x

MMSE estimation

(2)

ˆ  x

MMSE estimation

( )

ˆ

B

N 

x

        

Joint Fronthaul Compression

• Reverse compute-and-forward (RCoF) [Hong and Caire

‘12]

Compute-and-Forward

1

C

1

M

Channel encoder 1 Integer precoding RU 1 Control Unit

M

N

M

Channel encoder NM

1

s

M

N

s

1

x 

B

N

x 

B

N

C

  

1

x

RU

   B

N

x

B

N

• Three-cell SISO circular Wyner model

Numerical Results

    

CU

C C C

2 4 6 8 10 12 14 1 2 3 4 5 6 C [bits/s/Hz] per-cell sum-rate [bits/s/Hz] Cut-set upper bound Joint compression Point-to-point compression Linear precoding Single-cell processing

• Three-cell SISO circular Wyner model ( and )

Numerical Results

20 dB P 

0.5  

• Three-cell SISO circular Wyner model ( and )

Numerical Results

20 dB P 

0.5  

2 4 6 8 10 12 14 1 2 3 4 5 6 C [bits/s/Hz] per-cell sum-rate [bits/s/Hz] Cut-set upper bound Joint compression Point-to-point compression DPC precoding Linear precoding Single-cell processing

• Three-cell SISO circular Wyner model ( and )

Numerical Results

20 dB P 

0.5  

2 4 6 8 10 12 14 1 2 3 4 5 6 C [bits/s/Hz] per-cell sum-rate [bits/s/Hz] Cut-set upper bound Joint compression Point-to-point compression DPC precoding Compute-and-forward Linear precoding Single-cell processing

Overview

• Uplink

– Multiterminal compression – Compute-and-forward

• Downlink

– Multiterminal compression – Compute-and-forward

• Performance evaluation
• Extensions and conclusions

• In each macro-cell, pico-BSs and MSs are uniformly

distributed.

Simulation Set-up

N K

• Frequency reuse pattern with reuse factor for 1-cell cluster

[Wang and Yeh ’11]

1/ 3 F 

Simulation Set-up

• Cell-edge throughput versus average spectral efficiency

–

macro pico max

Uplink, 1-cell cluster, 3 pico-BS, 5 MSs, ( , ) (9,3)bps/Hz, 10, 0.5, 1/ 3 N K C C T F       

0.85 0.9 0.95 1 1.05 1.1 1800 2000 2200 2400 2600 2800 3000 3200 3400 3600 spectral efficiency [bps/Hz] 5%-ile rate (cell-edge throughput) [kbps] Point-to-point compression Multiterminal compression

=2.0 =1.0 =0.5 =0.25

1.6x

Numerical Results

• Cell-edge throughput versus average spectral efficiency

–

Numerical Results

macro pico max

Downlink, 1-cell cluster, 1 pico-BS, 4 MSs, ( , ) (3,1)bps/Hz, 5, 0.5, 1/ 3 N K C C T F       

0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2 1.25 500 1000 1500 2000 2500 3000 3500 4000 spectral efficiency [bps/Hz] 5%-ile rate (cell-edge throughput) [kbps] Point-to-point compression Multiterminal compression

=1.5 =0.5 =0.25

2x

Overview

• Uplink

– Multiterminal compression – Compute-and-forward

• Downlink

– Multiterminal compression – Compute-and-forward

• Performance evaluation
• Extensions and conclusions

Multiterminal Compression with Imperfect CSI

[Park et al ‘13]

Multi-Hop Fronthaul Topology

[Park et al ‘14]

RU RU

1

y

/2 N

y

RU RU

/2 1 N

y

 N

y

RU RU RU

1 N

y

 2 N

y

 3 N

y



CU

1, 1 N

C

 1, 2 N

C

 /2, 1 N N

C

 /2, 2 N N

C

 /2 1, 2 N N

C

  , 2 N N

C

 , 3 N N

C

 /2 1, 3 N N

C

  1, 4 N N

C

  2, 4 N N

C

  3, 4 N N

C

 

1 / 2 N

/ 2 1 N  N 1 N  2 N  3 N 

Inter-Cluster Multivariate Compression Design

[Park et al ‘14]

CU 1

1,1

C

CU 2 RU (1,1) RU (1,2)

1,2

C

2,1

C

RU (2,1) RU (2,2)

2,2

C

MS (1,1) MS (1,2) MS (2,1) MS (2,2)

1,1 1,2

, M M

2,1 2,2

, M M

Inter-cluster interference

1,1

ˆ M

1,2

ˆ M

2,1

ˆ M

2,2

ˆ M

Joint Compression of Data and CSI

[Kang et al ‘13]

2 4 6 8 10 12 14 1 2 3 4 5 6 Backhaul capacity C [bits/s/Hz] Ergodic sum−rate [bits/s/Hz] Cut−set Bound Non−coherent Joint Adaptive Joint Separate CFE Semi−coherent w 1 bit CSI Semi−coherent w/o 1 bit CSI

• Survey of fronthaul designs inspired by network information theory
• Multiterminal compression (joint decompression, joint compression)
• Compute-and-forward

Concluding Remarks

Simulation Set-up

Parameters Assumptions

System bandwidth 10 MHz Path-loss (macro-BS - MS) Path-loss (pico-BS - MS) Antenna pattern for sectorized macro-BS antennas Lognormal shadowing (macro-BS - MS) 10 dB standard deviation Lognormal shadowing (pico-BS - MS) 6 dB standard deviation Antenna gain after cable loss (macro-BS) 15 dBi Antenna gain after cable loss (pico-BS, MS) 0 dBi Noise figure 5 dB (macro-BS), 6 dB (pico-BS), 9 dB (MS) Transmit power 46 dBm (macro-BS), 24 dBm (pico-BS), 23 dBm (MS) Small-scale fading model Rayleigh-fading Synchronization Perfect synchronization Inter-site distance (site: macro-BS) 750 m Frequency reuse factor F=1/3 Number of antennas

Single antenna at each macro/pico-BS and MS

Channel state information (CSI)

Full CSI at control units about BSs in the cluster

10

PL(dB) 128.1 37.6log ( in km) R R  

10

PL(dB) 38 30log ( in m) R R  

2 3dB 3dB

( ) min[12( / ) , ] ( 65 , 20 dB)

m m

A A A         

• LTE rate model [3GPP-TR-136942]

Simulation Set-up

min attenuate min max max max

0, if ( ) ( ), if , if

k k k k k k

R S R                     

1 2 max max attenuate attenuate max min

: SINR at MS ; ( ) log (1 ); ( / ); : attenuation factor representing implementation losses; : Maximum and minimum throughput of the codeset, bps/Hz; : Minimum SINR of the codeset.

k

k S S R R       



   Parameter UL DL Notes 2.0 4.4

Based on 16-QAM 3/4 (UL) & 64-QAM 4/5 (DL)

• 10 dB
• 10 dB

Based on QPSK with 1/5 (UL) & 1/8 (DL)

0.4 0.6

Representing implementation losses

max

R

[3GPP-TR-136942, Annex A]

min



attenuate



where

• Proportional fairness metric [Tse]

– At each time , the rate is updated as

Simulation Set-up

sum-PF 1

( ) ( )

K k k k

R t R t R 





: fairness constant; ( ): instantaneous rate for MS at time ; : historical data rate for MS until time 1.

k k

R t k t R k t  

where

(P1) 

(1 ) ( )

k k k

R R R t     

t

k

R

where

[0,1]: the forgetting factor.  

• An achievable rate with JDD was derived in [Sanderovich et

al][Lim et al] as

Joint Decompression and Decoding

 

sum

ˆ ˆ min ( ; | ) ( ; ) ,

j j j j

R C I I

 

        



y y x x y

B

S S N S

• The problem of maximizing the sum-rate is a Difference of

Convex (DC) problem

Joint Decompression and Decoding

• 30
• 25
• 20
• 15
• 10
• 5

7.5 8 8.5 9 9.5 inter-cell channel gain  [dB] average per-cell sum-rate [bit/c.u.] cutset upper bound JDD w/ MM algorithm SDD w/ exhaustive ordering SDD w/ greedy ordering separate decompression and decodinng joint decompression and decoding

RU RU

1

y

/2 N

y

RU RU

/2 1 N

y

 N

y

RU RU RU

1 N

y

 2 N

y

 3 N

y



CU

1, 1 N

C

 1, 2 N

C

 /2, 1 N N

C

 /2, 2 N N

C

 /2 1, 2 N N

C

  , 2 N N

C

 , 3 N N

C

 /2 1, 3 N N

C

  1, 4 N N

C

  2, 4 N N

C

  3, 4 N N

C

 

1

{1, , } N   V

2

{ 1, , 3} N N     V

3

{ 4} N   V

Layer 1 Layer 2 Layer 3

1 / 2 N

/ 2 1 N  N 1 N  2 N  3 N 

System Model

Compression

i

y

ˆ i y

RU i

MUX

( )

I i

 ( )

O i



From RUs in previous layers To RUs and CU in next layers

Forwarding

( )

O i



Compression  

( )

I

e e i 

u

i

y

i

Compression Decompression Linear Processing

( )

I i



RU

From RUs in previous layers To RUs and CU in next layers

In-Network Processing

[Park et al ‘13]

4 MSs, average received per-antenna SNR of 20 dB

2 4 6 8 10 12 14 16 18 20 22 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 Number N of RUs in layer 1 Average sum-rate [bits/s/Hz] In-network processing Routing C=4 bits/s/Hz C=3 bits/s/Hz C=2 bits/s/Hz

Numerical Results

Multivariate Compression Lemma

 

 

| , for all {1, , }

i i i i

h X h X X R M

 

  

 

 

S S S

S

1,

,

M

 C C

1 1

( , , , ) ( ) ( , , | )

M M

p x x x p x p x x x       i.i.d. joint typicality wrt

Multivariate Compression Lemma

(contrapolymatroid)

• Linear precoding (DPC treated in a similar way)
• Gaussian test channel:
• The compressed signal is given as

with and

Multivariate Compression

,

, ~ ( , ),

i i i i i i

i    x x q q 0 Ω 

B

CN N

,   x As q

1 ,

,

B

H H H N

     x x x 

1 ,

, ~ ( , )

B

H H H N

     q q q 0 Ω  CN

1,1 1,2 1, 2,1 2,2 2, ,1 ,2 ,

B B B B B B

N N N N N N

               Ω Ω Ω Ω Ω Ω Ω Ω Ω Ω       

• Weighted sum-rate maximization

where

• DC problem: Local optimum via MM algorithm

Optimization

   

 

, 1 ,

maximize , s.t. , , for all , tr , for all .

M

N k k k i B i H i i i i i B

w f g C P i

  

    

 

A Ω 0

A Ω A Ω E AAE Ω

S S

S N N

   

 

     

   

,

, ; logdet ( ) logdet , , | logdet logdet .

k k k H H H H k k k l l k l k i i H H H i i i i i i i

f I g h h C

   

                        

   

A Ω s y I H AA Ω H I H A A Ω H A Ω x x x E AA E Ω E ΩE 

S S S S S S S