TURBO-LIKE CODES FOR 4Gdsplab.hanyang.ac.kr/.../cheun/Turbo-like_codes_for_4G_4.pdf · 2006. 5. 6. · Parallel Concatenated Turbo Codes? Pros? Proven scheme (3G)? Acceptable performance?

TURBOTURBO--LIKE CODES FOR 4GLIKE CODES FOR 4G

Prof. Kyungwhoon Cheun (전 경 훈)

Communication Systems and Circuits Design (CSC) Labhttp://csc.postech.ac.kr

Pohang University of Science and Technology(POSTECH)

Work performed in part for Samsung Electronicsand in part for HY-SDR center

2003-4-21 The 5th Smart Antenna /SDR Workshop 2

Talk Outline

? Code Requirements

? State of the Art

? Code Options for 4G : Pros and Cons? Turbo, LDPC, RA

? Trade-off Decisions

? Design Example ? RP–Turbo Code

? Conclusions


Code RequirementsCode Requirements? Performance? Performance approaching channel capacity

? Allow for high speed decoding approaching Gbps rates

? Complexity? Reasonable encoder/decoder gate count, memory

requirements and power consumption

? Flexibility ? Rate, Code length, Puncturing, Modulation, MA scheme

? ARQ? Allow for effective hybrid ARQ


Code RequirementsCode Requirements

? The FEC design issue? Design a code that is an acceptable

compromise among the above 4 factors

Performance

Complexity Flexibility

ARQ


State of the ArtState of the Art

Eb/No in dB

? So,… , where are we?

Mariner1969

OdenwalderConvolutionalCodes 1976

Turbo Code19930.7dB

N=65,536

Galileo:BVD1992Galileo:LGA

1996

Pioneer1968-72

Voyager1977

0 1 2 3 4 5 6 7 8 9 10-1-2

0.5

1.0 BPSK Capacity Boun

d

Cod

e R

ate

r

Shan

non

Cap

acity

Bou

nd

UncodedBPSK

Globalstar1999

Iridium1998

510 ??bP

Spe

ctra

l Effi

cien

cy

Irregular RA0.98dB

N=10,000

Irregular LDPC

0.95dBN=10,000

Vector LDPC ~1dB

N=8192

Irregular RA0.50dB

N=100,000

Irregular LDPC

0.48dBN=100,000


Code Options : TurboCode Options : Turbo

1px

sx

RSCEncoder 1

Interleaver

RSCEncoder 2

N

N

N

n

N

N

2px

* RSC: Recursive Systematic Code* RSC: Recursive Systematic Code

* Invented by C. Berrou, et al., “Near Shannon limit error-correcting coding and decoding: Turbo codes,”Proc. 1993 Int. Conf. Commun., pp 1064-1070, 1993.

* Invented by C. Berrou, et al., “Near Shannon limit error-correcting coding and decoding: Turbo codes,”Proc. 1993 Int. Conf. Commun., pp 1064-1070, 1993.

Puncturer

N


Code Options : TurboCode Options : Turbo? Parallel Concatenated Turbo Codes? Pros

? Proven scheme (3G)? Acceptable performance? Excellent flexibility – (trellis based / puncturing)

? Cons? There are now better performing codes? Rather complex decoding? Not suitable for very high data rates? Performance highly dependent on frame size? Not suitable for short frame sizes

? Possible Improvements?? Irregular parallel concatenated Turbo codes? Repeat-Puncture Turbo code (RP-TC)? We will later show an example of this scheme


Code Options : LDPCCode Options : LDPCc1

m vcm cv

c2 c3 c4

v1 v2 v3 v4 v5 v6 v7

Checknodes

Variablenodes

mv

??

???

???

? ??

1,0,

}\{'

)('

)(ll

ll ifmmifm

mcCc

vcv

v

vc

v

Regular - Invented by R. G. Gallager (1962), “Low density parity check codes,” IRE Trans. Inform. Theory, vol. IT-8, pp. 21-28, Jan. 1962.

Irregular - Invented by T. J. Richardson, M. A. Shokrollahi, and R. L. Urbanke (2001), “Design of Capacity-Approaching Irregular Low-Density Parity-Check Codes,” IEEE Trans. Inform. Theory, vol. 47, no. 2, pp. 619-637, Feb. 2001.

? ? ? ?

? ? ? ???

??

??

??

??

??

???

????

}{\'}{\'

}{\'}{\')(

)1('

)1('

)1('

)1('

11

11log

vVv

m

vVv

m

vVv

m

vVv

m

cv

c

cv

c

cv

c

cv

c

cv

ee

eem

ll

ll

l

Low Density


Code Options : LDPCCode Options : LDPC? LDPC (Low Density Parity Check) : Irregular? Pros

? Best known theoretical performance (0.0045dB)? But how close do we need to be within capacity?? These performance may only be obtained at the expense of very large

frame lengths and degree orders (complexity)

? Cons? Complex encoding? Moderate to complex decoding? Very low flexibility? Hmmm… How much of these are we willing to give up to get

a few tenth of a dB improvement? Flarion Vector-LDPCTM

? Approximately 2.3dB at 10-6 BER, N=1024? Not enough information for fair comparison.


Code Options : RACode Options : RA

Regular - Invented by D. Divsalar, H. Jin and R. J. McEliece (1998)“Coding theorems for ‘Turbo-like codes,” in Proc. 36th Allerton Conf. On Commun., Control and Computing, Sept. 1998, pp.201-210.

Irregular - Invented by H. Jin, A. Kandekkar and R. J. McEliece (2000)“Irregular repeat-accumulative codes,” in Proc. 2nd Int. Symp. On Turbo Codes, Sept. 2000, pp.1-8.


Code Options : RACode Options : RA

? RA (Repeat Accumulate) : Irregular? Pros? Performance very close to LDPC? Moderate encoding complexity

? Cons? Moderate to complex decoding? Flexibility better than LDPC but still low


TradeTrade--Off DecisionsOff Decisions

? Strategy A? Start from LDPC? Accept ??dB loss in performance to gain code

“structure”? ? Reduce encoder/decoder complexity? ? Gain code flexibility

? In many cases this approach will lead to an equivalent “structured” code such as the RA code



? Strategy B? Start from Codes with Structure, i.e., Turbo/RA? Accept some loss in structure to gain ? dB in

performance? Loss in “structure” in some sense means “irregularity”or

“randomization”of the code

? There are ways to get code graph “irregularity”or “randomness”without significantly losing desirable code structure? An example will be shown shortly



? Our Choice?? We think that the second approach is the answer

? What about other “codes” for 4G at the CSC Lab? Space-time codes? QO-STBC for any number of antennas

? MIMO/BLAST? Combined space-time and Layered Space-Time? Efficient decoding for layered space-time codes

? The final decision for the selection of the FEC shall take into account ALL components and aspects of the system


Design Example : RPDesign Example : RP--TCTC

? Repeat-Puncture Turbo Codes (RP-TC)? A simple improvement to traditional parallel

concatenated TCs? Enhanced performance due to increased

interleaver size for a fixed frame size? Enhanced performance due to resulting code

graph irregularity? All desirable characteristics of the parallel

concatenated Turbo code are retained, e.g, flexibility

? Slight increase in decoding complexity


Design Example : RPDesign Example : RP--TCTC? Code Structure

RSCEncoder 1

Interleaver

Repeat‘L’

RSCEncoder 2

LN

N

N

LN

N

n

LN

Puncturer

N

1px

sx

2px


Design Example : RPDesign Example : RP--TCTC? Weight Distribution

? Code rate=1/3

? Generator=(1+D2)/(1+D+D2)

? Information frame length N=128

? Uniform interleaver

??

???

????

??

freedd

bdb N

EdRQBP

0

2


Design Example : RPDesign Example : RP--TCTC? Union Bound:

Packet Error Rate? Uniform interleaver

? Packet length:1024 Bytes

? Information frame length: N bits

? Code rate=1/3

★ Multiple code frames in a packet


Design Example : RPDesign Example : RP--TCTC? Code Graph? Decoding by BCJR and Message Passing Algorithm (MPA)

Interleaver

RSC1

RSC2

Variable nodes forsystematicinformation

bits

Tail bits for terminationPunctured parity bit positions

yt2,1 yt2,m

yt1,1 yt1,mys,1 ys,2 ys,N

yp2,1 yp2,L+1 ytp2,1 ytp2,m

yp1,1 yp1,2 yp1,N ytp1,1 ytp1,m

Lbranches Variable node

Hidden variable node

Factor node

yp2,2 yp2,L+2 yp2,(L-1)N+2yp2,(L-1)N+1

← BCJR

← BCJR


Design Example : RPDesign Example : RP--TCTC? Simulation :




? Code rate=1/3

? 30 decoder iterations



ConclusionsConclusions

? No clear winner yet, but? Will clearly require codes outperforming current Turbo

codes

? Will require at least as much or more flexibility than 3G

? Different requirements for different environments? Mobile

? Nomadic

? MIMO

? The coming couple of years will be exciting

Documents

TURBO-LIKE CODES FOR 4Gdsplab.hanyang.ac.kr/.../cheun/Turbo-like_codes_for_4G_4.pdf · 2006. 5. 6. · Parallel Concatenated Turbo Codes? Pros? Proven scheme (3G)? Acceptable performance?