a novel technique for improving the performance of turbo codes using orthogonal signalling,...
TRANSCRIPT
A Novel technique for Improving the Performance of Turbo Codes using Orthogonal signalling, Repetition and Puncturing
by Narushan Pillay
Supervisor: Prof. HongJun Xu
Slide 2 © CSIR 2006 www.csir.co.za
Breakdown of Presentation
• Structure of a new scheme Repeat-Punctured Superorthogonal Convolutional Turbo Code (RPSCTC) Encoder
• Procedure of Encoding – Parallel Concatenated Superorthogonal Recursive Convolutional Constituent Codes
• Structure of the RPSCTC Decoder• Procedure for Decoding utilizing the MAP algorithm
• Importance of Interleaving and its effects
Slide 3 © CSIR 2006 www.csir.co.za
Breakdown of Presentation
• A variation of the scheme – Dual-Repeat-Punctured Turbo Code (DRPSCTC)
• Performance Evaluation for RPSCTC using transfer function bounding techniques
• Simulation Results and Bounds for AWGN and Rayleigh Fading Channels – RPSCTC, DRPSCTC versus an existing scheme Superorthogonal Convolutional Turbo Codes (SCTC)
Slide 4 © CSIR 2006 www.csir.co.za
Encoding Structure of RPSCTC
• Structure that of a Parallel Concatenated Code (PCC)• Two constituent codes, two parity sequences• Separated by repeat and interleaver structures• Puncturing mechanism to control the code rate
S R C C 1
S R C C 2
kd
)(2,11,10,1 1...,Nmyyy
)(2,22,20,2 1...,LNmyyy
C o n s t itu e n t C o de 1
C o n s t itu e n t C o de 2
N )(2 1 Nm
)(2 1 LNmLN
N
LN
R ep ea t 'L '
LN
P u n c tu r e
Figure 1. Low-level Encoding structure of RPSCTC
Slide 5 © CSIR 2006 www.csir.co.za
Procedure of Encoding for RPSCTC
0a 1a 2a 3a
W als h - Had ar m ar d G en er a to r
F ir s t p ar itys eq u en c e
0a 1a 2a 3a
W als h - Had ar m ar d G en er a to r
R ep ea t 'L '
P u n c tu r e
S ec o n d p u n c tu r edp ar ity s eq u en c e
C o n s titu en t C o d e 1
C o n s titu en t C o d e 2
Nm 12
LNm 12
N
LN
LN
Nm 12
I n f o r m atio nS o u r c e
• More detailed structure of the RPSCTC encoder
• Identical constituent Superorthogonal Recursive Convolutional encoders
Figure 2. Detailed structure of encoder
Slide 6 © CSIR 2006 www.csir.co.za
Structure of RPSCTC Decoder
D ec o d er 1C o r r u p tedp ar ity s eq u en c e 1
ydL k |1
-Ad d d u m m y
b its
channelL
R ep eat 'L '
D ec o d er 2
ydL k |2
12eL
Ad d d u m m yb itsC o r r u p ted
p ar ity s eq u en c e 2
Av er ag e
-channelL
21eL
1
1 Av er ag e
LNea LL 12
21eL21ea LL
21ea LL
Nddd~
,...,~
,~
21
ydL k |2
ydL k |2
E s tim ate o fm es s ag e s eq u en c e
C o m p ar a to r
1
2
12
11 1,...,, p
N
ppmyyy
1
2
12
11 1,...,, p
LN
ppmyyy
LNeL 12
LNea LL 12
• Two constituent Maximum a-posteriori (MAP) decoders
• Mutual exchange of soft extrinsic
information• Cooperative network
Figure 3. Structure of the Decoder
Slide 7 © CSIR 2006 www.csir.co.za
Procedure for Decoding for RPSCTC
• The MAP algorithm• Somewhat like the Viterbi
algorithm but trellis traversed in two directions.
mk),0(
1mb
k
),1(1
mbk
),0(,01
mbk
),1(,11
mbk
1k k
mk
),0(1
mfk
),1(1
mfk
mk
,0
mk,1
k 1k
),1(,11
),1(1
),0(,01
),0(1
mbk
mbk
mbk
mbk
mk
),1(1
,1),0(1
,0 mfk
mk
mfk
mk
mk
mikk
ikk
ik
mik vyux ,, exp
Figure 4. The MAP algorithm
Slide 8 © CSIR 2006 www.csir.co.za
Procedure for Decoding for RPSCTC
Starting with the log-likelihood ratio (LLR)
m
mk
m
mkkdL ,0,1 loglog)ˆ(
With the likelihood ratios defined by
m
mfk
mk
mk
m
mk
),1(1
,1,1
m
mfk
mk
mk
m
mk
),0(1
,0,0
…………….........……………..(1)
…………….........……………..(2)
…………….........……………..(3)
Slide 9 © CSIR 2006 www.csir.co.za
Procedure for Decoding for RPSCTC
Extrinsic information from decoder 1 given by
)()()()( 12221 kekckdecoderke dLxLdLdL
Extrinsic information from decoder 2 given by
)()()()( 112 kakckdecoderke dLxLdLdL
……………...………..(4)
……………..….…….…..(5)
Slide 10 © CSIR 2006 www.csir.co.za
Interleaving
B = S m all
B = L ar g e
B =
Mul
tiplic
ity
D is tan c e
• Conventional turbo coding frame length set equal to interleaver size.
• Increase in frame length – performance increase
• Limit on frame length:• Transmission delay• Decoding delay• Hardware delay
• Use repeat block prior to interleaving
• Frame length constant – larger interleaver size – better performance
Figure 5. Effect of Interleaving
Slide 11 © CSIR 2006 www.csir.co.za
Interleaving
S R C C 1
S R C C 2
kd
)(2,11,10,1 1...,Nmyyy
)(2,22,20,2 1...,LNmyyy
C o n s t itu e n t C o de 1
C o n s t itu e n t C o de 2
N )(2 1 Nm
)(2 1 LNmLN
N
LN
R ep ea t 'L '
LN
P u n c tu r e
Figure 6. Effect of Interleaving
• Repeat ‘L’ block allows for the use of a larger interleaver.
• Spectral thinning• Weight one sequence – weight two sequence –
high weight turbo codeword.
Slide 12 © CSIR 2006 www.csir.co.za
Interleaving
• Weight two sequence – weight four sequence – still yields a high weight turbo codeword.
Slide 13 © CSIR 2006 www.csir.co.za
Performance Evaluation of RPSCTC
Obtain the state transition matrix from state diagram using equation (6)
jz
dilz
dil
jdildil
DILDIL
DILDIL
DILA
,0,
,00,0
),,(
Transfer function can be expressed as equation (7):
……………..…..(6)
…………….........……..(7)
0 0 0
),,(),,(l i d
dil diltDILDILT
Slide 14 © CSIR 2006 www.csir.co.za
Performance Evaluation of RPSCTC
Since the transfer function is defined by equation (8),
and,
132 )(... AIAAAI
…………….........………..(8)
……………...........…………..(9)
mmDILAIDILT0,0
1])),,([(),,(
Slide 15 © CSIR 2006 www.csir.co.za
Performance Evaluation of RPSCTC
Then the probability of producing a codeword fragment of weight d given a random input sequence of weight i is given by equation (10) for component encoder 1 and equation (11) for component encoder 2.
.),,(
),,(
),,()|( 1
1
11
1
i
N
diNt
diNt
diNtidp
d
.),,(
)|( 22
Li
LN
dLiLNtidp
………..........(10)
……………..............……..(11)
Slide 16 © CSIR 2006 www.csir.co.za
Performance Evaluation of RPSCTC
Equation (12) is used to achieve the union bound
n
dd i d dbit dpidpidp
i
N
N
ip
min 1 2221 )()|()|(
where,
)/2()(2 ob NdREQdp
for an AWGN channel.
……..(12)
……………...........……..(13)
Slide 17 © CSIR 2006 www.csir.co.za
Performance Evaluation of RPSCTC
and p2(d) is given by (14) for side-information (SI) or equation (15) for no side-information (NSI) for a Rayleigh fading channel,
d
os
SI
NEdp
/1
1
2
1)(2
d
NSI edp
1/21
1/21.
2
1)( /
2
12 ob NRE /
where
and
……………………..….….(14)
……………....…....(15)
…….……...…...(16)
Slide 18 © CSIR 2006 www.csir.co.za
Another scheme
I n f o r m atio nS o u r c e
R ep eat 'L ' S R C C 1
S R C C 2
P u n c tu r e
P u n c tu r e
1
2
N
LN
LNm 12
LNm 12
Nm 12
Nm 12
Figure 7. DRPSCTC Encoder
• Structure of the Dual-Repeat-Punctured Turbo Code (DRPSCTC) encoder
• Dual repetition prior to encoding• Puncturing at both output branches
Slide 19 © CSIR 2006 www.csir.co.za
Another scheme
D ec o d er 1
D ec o d er 2
C o r r u p tedp ar ity s eq u en c e o n ew ith d u m m y b itsa t p u n c tu r e p o s it io n s
11 2
C o r r u p tedp ar ity s eq u en c e tw ow ith d u m m y b itsa t p u n c tu r e p o s it io n s 1
Av er ag e
12
12
Av er ag e
R ep ea t 'L '
Av er ag e
R ep ea t 'L '
S o f t D ec is io n s
LNm 12
LNm 12
N
-
channelL
-channelL
C o m p ar a to r
ydL kLN |1 LN
eL 12 LNeL 12
LNeL 12
LNeL 12
LNea LL 12
LNea LL 12
12eL ydL k
LN |2
LNeL 21
LNeL 21
LNeL 21
21eL
LNeL 21
LNea LL 21
LNea LL 21
ydL kLN |2 ydL k
LN |2 ydL k |2
• Decoder for DRPSCTC
•Slightly greater complexity
•Better performance
Figure 8. DRPSCTC Decoder
Slide 20 © CSIR 2006 www.csir.co.za
Simulation Results and Bounds
-1 0 1 2 3 4 510
-12
10-10
10-8
10-6
10-4
10-2
100
SNR
BE
R
SCTC and RPSCTC Simulation for N=200
N=200 SCTC m=4
bound N=200 SCTC m=4
N=200 RPSCTC m=4
bound N=200 RPSCTC m=4
-1 0 1 2 3 4 510
-12
10-10
10-8
10-6
10-4
10-2
100
SNR
BE
R
SCTC and RPSCTC Simulation for N=200
N=200 SCTC m=2
bound N=200 SCTC m=2
N=200 RPSCTC m=2
bound N=200 RPSCTC m=2
Figure 9. SCTC and RPSCTC simulation m=2, m=4 for the AWGN channel
Slide 21 © CSIR 2006 www.csir.co.za
Simulation Results and Bounds
-1 0 1 2 3 4 510
-12
10-10
10-8
10-6
10-4
10-2
100
SNR
BE
R
SCTC and RPSCTC Simulation for N=200
N=200 RPSCTC m=4
bound RPSCTC m=4
N=200 RPSCTC m=2
bound RPSCTC m=2
Figure 10. RPSCTC simulation m=2, m=4 in the AWGN channel
Slide 22 © CSIR 2006 www.csir.co.za
Simulation Results and Bounds
-1 0 1 2 3 4 510
-14
10-12
10-10
10-8
10-6
10-4
10-2
100
SNR
BE
R
RPSCTC N=200 R=1/15 m=4
DRPSCTC N=200 R=1/15 m=4
SCTC N=200 R=1/15 m=4SCTC bound m=4 N=200
RPSCTC bound m=4 N=200
DRPSCTC bound
Figure 11. SCTC, RPSCTC vs DRPSCTC m=4 AWGN channel
Li
LN
dLiLNtidp q
q
),,()|(
q =1,2 for component encoders for DRPSCTC
Slide 23 © CSIR 2006 www.csir.co.za
Simulation Results and Bounds
0 5 10 15 20 25 30 35 40 45 5010
-8
10-6
10-4
10-2
100
102
104
106
Codeword distance
Mul
tiplic
ity
SCTC distance spectrum N=100 R=1/15
RPSCTC distance spectrum N=100 R=1/15
DRPSCTC distance spectrum N=100 R=1/15
Figure 12. SCTC,RPSCTC, DRPSCTC distance spectrum N=100, R=1/15
Slide 24 © CSIR 2006 www.csir.co.za
Simulation Results and Bounds
2 3 4 5 6 7 8 9 10 1110
-5
10-4
10-3
10-2
10-1
SNR
BE
R
N=200 SCTC m=2
bound SCTC m=2
N=200 RPSCTC m=2
bound RPSCTC m=2
2 3 4 5 6 7 8 9 10 1110
-6
10-5
10-4
10-3
10-2
10-1
SNR
BE
R
Rayleigh fading simulation for N=200
N=200 SCTC m=4
bound SCTC m=4
N=200 RPSCTC m=4
bound RPSCTC m=4
Figure 13. SCTC and RPSCTC simulation m=2, m=4 flat Rayleigh fading channel