ON THE ANALYSIS AND APPLICATION OF

LDPC CODES

OLGICA MILENKOVIC
UNIVERSITY OF COLORADO, BOULDER

A joint work with:

VIDYA KUMAR (Ph.D)

STEFAN LAENDNER (Ph.D)
DAVID LEYBA (Ph.D)
VIJAY NAGARAJAN (MS)
KIRAN PRAKASH (MS)
 OUTLINE
 A brief introduction to codes on graphs
 An overview of known results: random-like codes for standard
channel models
 Code structures amenable for practical implementation: structured
LDPC codes
 Codes amenable for implementation with good error-floor properties
 Code design for non-standard channels
 Channels with memory
 Asymmetric channels
 Applying the turbo-decoding principle to classical algebraic codes:
Reed-Solomon, Reed-Muller, BCH…
 Applying the turbo-decoding principle to systems with unequal
error-protection requirements
 THE ERROR-CONTROL PARADIGM
Noisy channels give rise to data errors: transmission or storage systems
Need powerful error-control coding (ECC) schemes: linear or non-linear
Linear EC Codes: Generated through simple generator or parity-check
matrix
1 0 0 0 1 0 1
   1 1 1 0 1 0 0
0 1 0 0 1 1 0  
Binary linear codes: G H  0 1 1 1 0 1 0 
0 0 1 0 1 1 1
   1 0 1 1 0 0 1 
0 0 0 1 0 1 1 

Binary information vector u  (u1 , u2 , u3 , u4 ) (length k)

T
Code vector (word): x  uG, Hx  0 (length n)
Key property: “Minimum distance of the code”, dmin , smallest separation
between two codewords
Rate of the code R= k/n
 LDPC CODES
More than 40 years of research (1948-1994) centered around dmin
Weights of errors that a code is guaranteed to correct  ( dmin  1)/2 

“Bounded distance decoding” cannot achieve Shannon limit

Trade-off minimum distance for efficient decoding
Low-Density Parity-Check (LDPC) Codes

Gallager 1963, Tanner 1984, MacKay 1996

1. Linear block codes with sparse (small fraction of ones) parity-check matrix
2. Have natural representation in terms of bipartite graphs
3. Simple and efficient iterative decoding in the form of belief propagation
(Pearl, 1980-1990)
 THE CODE GRAPH AND ITERATIVE DECODING

Most important consequence of graphical description:

efficient iterative decoding

Message passing:
Variable nodes: communicate to check nodes their
reliability (log-likelihoods)
Check nodes: decide which variables are not reliable
Variable nodes Check nodes and “suppress” their inputs
(Irregular degrees/codes)
Small number of edges in graph = low complexity
 1 1 1 0 1 0 0
  Nodes on left/right with constant degree: regular code
H  0 1 1 1 0 1 0 
 1 0 1 1 0 0 1  Otherwise, codes termed irregular
Can adjust “degree distribution” of variables/checks
Best performance over standard channels: long, irregular, random-like LDPC codes
Have dmin proportional to length of code, but correct many more errors
 DECODING OF LDPC CODES
Iterative decoding optimal only if the code graph has no cycles
Vardy et.al. 1997: All “good codes” must have cycles
• What are desirable code properties?
• Large girth (smallest cycle length): for sharp transition to waterfall
region; large minimum distance;
• Number of cycles of short length as small as possible;
• Very low error-floors (girth-optimized graphs are not a very good choice
Richardson 2003);
• Make the performance capacity approaching: irregular degree
distribution (depends on the channel, “safe” gap to capacity about 1dB);
(Density Evolution, Richardson and Urbanke, 2001)
• Practical applications: mathematical formula for positions of ones in H
 HOW CAN ONE SATISFY THESE
CONSTRAINTS?

For “standard channels” (symmetric, binary-input) the questions are

mostly well-understood
BEC – considered to be a “closed case”
(Capacity achieving degree distributions, complexity of decoding,
stopping sets, have simple EXIT chart analysis…)

Error-floor greatest remaining unsolved problem!

For “non-standard” channels still a lot to work on

CODE CONSTRUCTION AND
IMPLEMENTATION FOR
STANDARD CHANNELS
 STRUCTURED LDPC CODES WITH GOOD
PROPERTIES
 Structured LDPC: amenable for practical VLSI implementations
 Construction issues: minimum distance, girth/small number of
short cycles, error-floor properties

Cycle-related constraints

Trapping-set related constrained

 THE WATERFALL AND ERROR-FLOOR REGION

Encyclopedia of
Sparse Codes on
Graphs:
by D. MacKay

http://www.inference.phy.cam.ac.uk/mackay/codes/data.html

 Waterfall: Optimization of minimum distance and cycle length

distribution (regular/irregular codes)
 Waterfall: Optimization of degree distribution and code construction
 Error floor: Optimization of trapping sets or … maybe a different
decoding algorithm?
Laendner/Leyba/Milenkovic/Prakash: Allerton 2oo3, 2004, ICC 2004, IT 2004, IT
2005;

Tanner 2000; Vasic and Milenkovic 2001; Kim, Prepelitsa, Pless 2002; Fossorier 2004
 P1,1 P1, 2 ... P1,n  0 1i i
0 ... 0  i i
 P1 P2 P3 ... P s 
   is 
 P P2, 2 ... P2,n  0 0 1 0 ...
P P i1 P i2 ... P i s 1 
H 
2 ,1  P  H 
 ... ... ... ...     ... ... ... ... ... 
  0 0 ... 0 1 
 i s m 2 
 Pn k ,1 Pn k , 2 ... Pn k ,n  1
 0 0 ... 0   P P i s m 3
P i s m 4
... P i s m1 

Pi , j  P ( i 1)( j 1) P
0
P P 0 0 0 P 0
0 0 P P P 0 P

P P 0 P P 0 0

0 Irregular
I I  I  Masking… 
0 0 P 0 0 P P

P Codes
I P  P q 1  0 P P P 0 P 0 0
 
  P 0 0 0 P 0 P P
   
   P i1 0 0 0 0 P i6 P7 P8
i i
0 
I P
r 1
 P q 1r 1   i 
P 9

P i1
i
0 0 0 P5
i
i5
i
P i6 P 7
i6
P i8 
i 
Codes
 0 P9 P i1 0 0 0 P P P7
H  0 P i8 P9
i
P i1 0 0 0 P5
i
P i6  with large
 i 
girth
i
( ri 1)( c j 1)  0 0 P i8 P9 P i1 0 0 0 P 5
Pi , j  P  0 0 0 P i8
P9
i
P i1 0 0 0 
 i8 i9 i1

 0 0 0 0 P P P 0 0 

 I I I I 0 0 P2 P2 0 0 
P P P 
 I P 2
0 0 I I I 0 0 P2


H  0 I 0 P 2
I P 2
0 I I 0 
 
P 2 0 I 0 0 I 0 I P2 I 
 
0 0 P2 I P2 0 I 0 I I 
P P P
 MATHEMATICAL CONSTRUCTIONS OF
EXPONENTS
P P P P r1 LOW TO MODERATE RATE CODES:
Fan, 2000: Cycles exist if sum of exponents in
P P P P r2 closed path is zero modulo size (P)=q
( r1  r2 )ci1  ( r2  r3 )ci2  ...  ( rl  r1 )cil  0 mod q
P P P P r3
Proper Array Codes (PAC): Row-Labels {ri}
P P P P r4 form an Arithmetic Progression with common
mean ri+1-ri=a;
Improper Array Codes (IAC): Row-Labels {ri}
ci cj ck cl do not form an Arithmetic Progression;

 2i+j-k-2l=0
i+j-2k=0 mod q (CW3)  i+j-k-l=0
i+j-2k=0 and i+2j-3k=0 mod q (CW4)
 2k+i-3j=0
 2j-i-k=0
 ANALYTICAL RESULTS ON ACHIEVABLE
RATES
 Number theoretic bounds for contructions:
 For PACs with girth g=8:  C q log log q / log q
 For IACs with girth g=10:  2 q
 Results based on work of Bosznay, Erdos,Turan, Szemeredi, Bourgain
1/ 3
 3q 2 
q  1  (1 /(2i  1))  , or  
1 i  2 ( q 1)

 2( q  1) 

( d  1) k  2  1  log q 
 , k 
k  log(dD  1) 

s=1 0,1,5,14,25,57,88,122,198,257,280,...
s=1 0,1,3,4,9,10,12,13,27,28,30,38,...
s=2 0,2,7,18,37,65,99,151,220,233,545,...
s=2 0,2,3,5,9,11,12,14,27,29,30,39,...
s=3 0,3,7,18,31,50,105,145,186,230,289,...
s=3 0,3,4,7,9,12,13,16,27,30,35,36,...

Mathematical constructions: Joint work with N. Kashyap, Queen’s ,

University, AMS Meeting, Chicago 2004
 CYCLE-INVARIANT DIFFERENCE SETS
 Definition:V an additive Abelian group, order v.
 (v,c,1) difference set Ω in V is a c-subset of V with exactly one ordered pair
(x,y) in Ω s.t. x-y=g, for a given g inV.
 Examples: {1,2,4} mod 7, {0,1,3,9} mod 13.
 “Incomplete” difference sets – not every element covered
 CIDS: Definition
 Elements of Ω arranged in ascending order.
 C i operator:cyclically shifts sequence i positions to the right
 For i = 1,…, m, i  C i    mod v (component-wise)
are ordered difference sets as well
Ω is an (m+1)-fold cycle invariant difference set over V
 Example: V=Z7 and Ω ={1,2,4}, m=2
1  {4,1,2}  {1,2,4}  {3,6,5}  3{1,2,4} mod 7
 CIDS – GENERALIZATION OF BOSE METHOD
S = i1 , i2 ,..., is  constructed as {a : 0  i  q 2  1,  a    GF (q)}

where  is the primitive element of GF(q2) , q odd prime

 P i1 P i2 P i3 ... P i s 
 is 
P P i1 P i2 ... P i s1 
H 
 ... ... ... ... ... 
 i s m 2 i s  m 1 
 P P i s  m 3 P i s m 4 ... P 

GF ( q m ) : {a : 0  i  q m  1,  a    GF ( q)}

Excellent performance for very low rate codes

HIGH-RATE CODES: LATIN SQUARE
CONSTRUCTION

I I I I 0 0 P2 P2 0 0
 
I P2 0 0 I I I 0 0 P2 
H 0 I 0 P2 I P2 0 I I 0
 2 
P 0 I 0 0 I 0 I P2 I 
0 0 P2 I P2 0 I 0 I I 

 ERROR FLOOR ANALYSIS
 Near codewords (Neal/MacKay, 2003)
 Trapping sets (Richardson, 2003)
 Considered to be the most outstanding problem in the area!

Inherent property of both the decoding algorithm and code structure

APPROACH: Try to predict error floor (Richardson, 2003, 2004), or

Try to change the decoding algorithm

Observations leading to new decoding algorithm:

1) Different computer codes give very different solutions; quantization has a
very large influence on the error floor
2) Certain variables in the code graph show extremely large and uncontrolled
jumps in likelihood values
3) Strange things happen when message values are very close to +1/-1
OBJECT OF INVESTIGATION: MARGULIS CODE
 Elementary trapping set (ETS): small subset of variable nodes for which
induced subgraph has very few checks of odd degree
 INTUITION: channel noise configuration such that all variables in ETS are
incorrect after certain number of iterations; small number of odd degree
checks indicates that errors will not be easily corrected

Margulis code: based on Cayley graphs (algebraic construction), has good girth
Frame error rate curve has floor at 10E-7, SNR=2.4dB, AWGN
Iteration number:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 ….
182 142 108 73 66 55 38 29 20 16 15 14 14 14 14 14 14 14 14 ….
Number of erroneous bits:

At “freezing” point, all values of variable messages are +1/-1 and significant
oscillations happen +1 → -1
Frozen activity lasts for 13 iterations: then “bad messages” propagate through
the whole graph; can have jumps from -0.33 to 49.67
 ALGORITHM FOR ELIMINATING THE ERROR FLOOR
 “Oscillations” reminiscent to problems encountered in theory of
divergent series (G.H. Hardy, Divergent Series)
 Trick: use a low-complexity “extra-step” in sum-product algorithm
based on result by Hardy
 Parameters tunable: when do you “start” using “extra-step”, what
“numerical biases” should one use
 Can do density evolution analysis for optimization purposes – no loss
expected in waterfall region (nor one observed)
(Paper with S. Laendner, in preparation)

1000 good frames: Standard and modified message passing both take 7.4
iterations on average for correct decoding
Modified message passing never takes more than 16 iterations to converge
20 Bad frames (in MATLAB): Corrected 2/3 fraction of errors after 35, 41, 26, 119,
38, 98… iterations (standard failed even after 10000 iterations)
Can use slightly more complicated result from G.H.H to reduce number of iteration
to 20-30
 VARIABLE AND CHECK NODE ARCHITECTURE
(JOINT WORK WITH N. JAYAKUMAR, S. KHATRI)
Globecom 2004, TCAS 2005

 Check node  Variable node

- Adder PLA From check 1 -
PLA
Adder

Adder
PLA From
Array Adder
channel Array To checks
To variable
nodes
Adder
Adder PLA
PLA
- From check 2
-
Log(dv+1)+1 Stages of Two-Input
log(tanh) and arctanh PLAs Manchester Adders, Rabaey 2000

LDPC ENCODING: SHIFT AND ADDITION UNITS FOR QUASI-CYCLIC CODES

 ADVANTAGES OF NETWORK OF PLA
ARCHITECTURE
 Logic function in 2-level form: good delay characteristics as long as
size of each PLA is bounded (Khatri, Brayton, Sangoivani-Vincentelli,
2000)
 Standard cell implementation: considerable delay in traversing the
different levels (i.e. gates)
 Local wiring collapsed into compact 2-level core: crosstalk-immunity
local wiring delays are reduced
 Devices in PLA core are minimum-sized: compact layouts
 NMOS devices used exclusively: devices placed extremely close
together
 In a standard cell layout, both PMOS and NMOS devices are present
in each cell: PMOS-to-NMOS diffusion spacing requirement results
in loss of layout density
 PLAs dynamic: faster than static standard cell implementation
 Can easily accommodate re-configurability requirement
 S1 Bank
C/V Clusters

S2 S3

Check Node

Variable Nodes
S4 Bank
Ring of C/V Node Clusters

Clocking and Logic Control

HILBERT SPACE FILLING CURVES

 P i1 P i2 P i3 P i4 P i5 P i6 0 0 0 0 0 0
 i6 i1 i2 i3 i4 i5 
P P P P P P 0 0 0 0 0 0
0 0 0 0 0 0 P i1
P i2
P i3
P i4
P i5 P i6 
 
0 0 0 0 0 0 P i6 P i1 P i2 P i3 P i4 P i5 
H 
0 0 0 P i1 P i2 P i3 P i4 P i5 P i6 0 0 0
 
0 0 0 P i6 P i1 P i2 P i1 P i4 P i5 0 0 0
 P i4 P i5 P i6 0 0 0 0 0 0 P i1 P i2 P i3 
 
 P P i2 
i3
P i4 P i5 0 0 0 0 0 0 P i6 P i1
S1 , S2 , S3 , S4

Group variable/check nodes based on distance in code graph

and place on adjacent fields in HP curve: proximity preserving
CHIP PARAMETERS: 0.1 MICRON PROCESS
Throughput (Gbps) Side of chip Power
(mm) (W)
Flat-out (max. duty
cycle of 77.14%) 1479.4 11.0923 104.5185

50% Duty cycle 958.9158 11.0923 83.6372

Lower clock for practical

applications 2 11.0923 13.3214

Estimates for n=28800

Throughput (Gbps) Side of chip (mm) Power

(W)

Flat out (max. duty 369.8537 5.5461 30.4711

cycle of 77.14%)

50% duty cycle (max. 239.7289 5.5461 20.9093

performance)

Lower clock for 2 5.5461 3.4406

practical applications

Estimates for n=7200

 LDPC CODES FOR
NON-STANDARD CHANNELS
 PRIOR WORK
 Partial analysis of the Gilbert-Elliott’s (GE) channel: (Eckford,
Kschischang, Pasupathy, 2003-2004)
 Polya’s Urn (PC) Channel Model (Alajaji and Fuja, 1994):
Simpler description, more accurate in terms of autocorrelation
function for many fading channels

B G State transition probabilities:

p(B,G), p(G,B)
Binary symmetric channel
transition probability
1-p(G)
p(G), p(B)

1-p(G)
 FINITE MEMORY POLYA URN CHANNEL
 Binary additive channel with memory
 Occurrence of error increases probability of future errors
 Output sample at time i (Yi) is modulo-two sum of input Xi and noise sequence Zi ,
where {Xi} and {Zi} are independent sequences
 The noise process is formed based on the following scheme:

Noise samples are outcomes of consecutive draws from

urn, and Zi is one if ith ball was red and zero if black

 Stationary, ergodic channel with memory M

 Three parameters: (σ, δ, M), where ρ=R/T < ½ , σ = 1- ρ = S/T and δ = Δ/T
 M=1 PC CHANNEL
(DEMONSTRATIVE EXAMPLE)

 Reduced 2-parameter space

 a = (σ + δ)/(1+ δ) and b = (1- σ + δ)/( 1+ δ)
 A two-state Markov chain with states S0 and S1 transition
probabilities 1-a and 1-b
 Channel transition matrix is
 a 1  a

1  b b 

 State S0 corresponds to a noise sample of value zero, and state S1

to one

 Observation: this type of channel is not included in the GE

analysis by Eckford et. al.
Nagarajan and Milenkovic, IEEE CCEC, 2004
 JOINT DECODING/CHANNEL STATE
ESTIMATION
 Use extrinsic information from the decoder to improve the quality of
the estimate of the channel state and vice-versa
 BCJR EQUATIONS FOR PC CHANNEL
ESTIMATION

 Forward message vector 

   (1),  (2)

 (1)  P { S (i )  S0 ;Y1i ;  }    (1) aq(  )    (2)(1  b)q(  )

i 
(1)(1  a)1  q(  )    (2) b1  q(  )
 Backward message vector (2)  P {S (i )  S1 ;Y1 ;  }  

  (  (1),  (2))

 (1)  P {Yin | S (i )  S0 ,  }    (1) aq(  )    (2)(1  a)(1  q(  ))

 Channel information passed to decoder
 (2)  P {Yin | S (i )  S1 ,  }    (1)(1  b) q(  )    (2) b(1  q(  ))


  (1)   1 a    2 1  b 
LLR  ( 1)Yi log
 

  (2)   1 1  a     2 b
 
CODE DESIGN CONSTRAINTS: ESTIMATOR
AND DECODER

 Avoid short cycles from

channel state estimator

 Avoid short cycles from LDPC

graph

 Optimize degree distribution

 SIMULATIONS

 Performance of CIDS-based structured code in PC channel with

channel estimation as compared to random codes
 δ = 0.2 and σ = 0.95-1 to generate pairs of (a,b)
ASYMMETRIC CHANNELS: DISTRIBUTION SHAPING
Gallager 1968: Linear codes can achieve capacity on any Discrete Memoryless
Channel (DMC) when combined with “distribution shapers”

Mapping not 1-1

Rate loss?

McEliece, 2001: Attempts to use this scheme with Repeat-Accumulate or LDPC

codes failed
Biggest obstacle: cannot do “soft” de-mapping

Solution: Use soft Huffman encoding

(extension of idea by Hagenauer et.al. for joint source-channel coding)

In preparation: Leyba, Milenkovic 2005

ASYMMETRIC CHANNELS: DISTRIBUTION SHAPING

Completion of codeword
“Interior bits”
BCJR Equations for trellis
Bit
Time

Largest achievable probability bias

Symbol Time
 ITERATIVE DECODING OF
CLASSICAL ALGEBRAIC CODES
APPLICATIONS FOR UNEQUAL
ERROR PROTECTION
 GRAPHICAL REPRESENTATIONS OF
ALGEBEBRAIC CODES
 Consider a [7,4,3] Hamming code
0 0 0 1 1 1 1
H  0 1 1 0 0 1 1
 
1 0 1 0 1 0 1

Idea put forward by Yedidia, Fossorier

 Support Sets: S1={4,5,6,7}; S2={2,3,6,7}; Intersection Set  ={6,7}

 Set entries corresponding to  in these two rows to zero; insert new

column with non-zero entries in these two rows; call this an auxiliary
variable VB
 Insert a new row with non-zero entries at positions indexed by   VB

0 0 0 1 1 0 0 1
0 1 1 0 0 0 0 1
H1  
1 0 1 0 1 0 1 0
 
0 0 0 0 0 1 1 1
 MAXIMUM CYCLE NUMBER (MC)
ALGORITHM
w 1 Hw  H
nw  n mw  m
 Approach by Sundararajan
and Vasic, 2004 (SVS): use
Determine the coordinates i, j of the (non -diagonal)
 4 2 2 largest element in H w H wT
HH  2 4 2
T

2 2 4 Is Yes

( H w H wT ) ij  1
to identify all cycles
No attempt to minimize number
No
of Auxiliary variables
Determine  . For x  set, ( H w ) ix  0, ( H w ) jx  0.
 Key Ideas: Always select the two
rows i, j involved in maximum Set, w  w  1, m w  m w1  1, n w  n w1  1 and
( H w ) inw  1, ( H w ) jnw  1, ( H w ) mw x  1, ( H w ) mwnw  1.
number of cycles
 4 2 3
Determine the rows l with   S l. Set ( H w ) lx  0, ( H w ) l nw  1.
HH   2 4 5
T
 
 6 2 4

Output Hw
Kumar, Milenkovic: CISS’2005, IEEE CTL 2005
 CYCLE OVERLAP STRATEGY (COS)

i0  1, j 0  2,
 COS is a variant of MCS D0  0
algorithm
Determine B. Construct submatrix HB of H w
 Consider two rows i, j with with columns indexed by B.
intersection set  of their
support sets Compute s the cardinality of the set
{(l , m) | l  m and ( H B H BT ) lm  ( H w H wT ) ij }
 Let s be number of rows with
a support set S l such that
  Sl Compute D=  ( HH ) ij  s 
T
.
 2  2 
  
 Choose indices that
maximize If D  D0 then, D0  D, i  i0 , j  j 0 .
Update the values of j o . i0 sequentially.
 ( HH T ) ij  s 
  
 2  2
  
 Dimensions of four-cycle free parity-check
representations

Code Hamming Golay BCH RS_bin

Original (3,7) (5,31) (11,23) (8,15) (10,31) (33,63) (36,60)
SVS (4,8) (25,31) (29,41) (22,29) (57,78) (300,327) (158,182)
MCS (4,8) (10,36) (23,35) (16,23) (29,50) (166,193) (110,134)
COS (4,8) (10,36) (22,34) (16,23) (29,50) (157,184) (102,126)
EXTENSION: CODES OVER NON-BINARY
FIELDS
 Motivation
 Applications to Reed-Solomon (RS) codes
 Two Approaches:
 Binary image mapping: Kumar, Milenkovic, Vasic CTW 2004,
Milenkovic, Kumar ISIT 2004
0 1 0  0 
0 0 1  0 
 
M       
 
     
 0  1  2   m 1 

Choice of matrix field representation and primitive polynomial, initial

choice of parity-check matrix
followed by cycle elimination for the binary parity-check matrix
 Direct cycle elimination followed by vector message
passing
 DIRECT CYCLE ELIMINATION
 Define two types of cycles
 Type 1 cycle: 0 0 1
x y 
R   xh  RCF   0 0  h 
  yh 
  y
x
1 
If {v1 , v 2 , v3 }  GF (q ) are the values taken by the three variable nodes in R CF then,
v3   x v1   y v 2
 Type 2 cycle:
 x 0  y 
 x  y   
D z w
DCF   z  w 0 
   0 1 1 


If {v1 , v 2 , v3 }  GF (q) are the values taken by the three variable nodes in R CF then,

v3  v 2
 Type 2 cycle elimination introduces six-cycle while eliminating four-cycle
 Number of Type-1 cycles: (q  1) 3
 Number of Type-2 cycles: (q  2)(q  1) 3
 ALGEBRAIC CODES AND UNEQUAL
ERROR PROTECTION BASED ON
ITERATIVE DECODING
 Codes which provide UEP: practical applications for
 Non-uniform channels
 Joint coding for parallel channels
 Image transmission over wireless channels
 Page oriented optical memories

 Fekri 2002 – LDPC codes for UEP based on optimizing degree

distributions
No guarantee/tuning possibility for degree of protection
 Dumer 2000 - Decoding algorithm for RM codes based on their
Plotkin-type structure

 Decodes different parts of the codeword separately

 Inherently achieves UEP for different portions of the codeword
 PLOTKIN CONSTRUCTION

 Parity-check matrix of Depth-1 Plotkin-type code:

 H1 0  C = {|u|u+v |; u in C1; v in C2 }
H  
H 2 H 2 
 Global iterative decoder not a good option: H2 placed side by side, introduces
at least  ci 
i  2  four-cycles
 

 The UEP obtained for Plotkin-type codes is an inherent characteristic of

the proposed decoding algorithm

 As in other UEP schemes involving LDPC codes, one can vaguely associate
the UEP obtained here to the different degree distributions

 Encoding: Individual for all different component codes

 INTUITION

 |u|u+v|:
 Have “two copies” of u
 Only one copy of v, which is implicit

ADVANTAGES OF THE SCHEME:

 Reduced-complexity encoding
 Low-complexity soft-output decoders
 Flexibility of UEP code construction, levels of UEP error
protection and adaptability to channel conditions
 Excellent performance for best protected components
 MS  Multi-Stage
TMS  Threshold Multi-Stage
MR-MS  Multi-Round Multi-Stage

Kumar, Milenkovic
Globecom 2004, TCOMM 2004
 FLOWCHART OF MS, MS-MR DECODER
Channel
Information Channel
Information
Liy ' Liy ''
Liy ' Liy ''  comp
2
,1   ch /( j  1)
2
for a1  0,
 comp
2
,1   ch / j
4
for a1  1
v
L i
Lvi

Decode v to  comp
2
,1   ch /( j  1)
2
for a1  0,
v̂ Decode v to
v̂  comp
2
,1   ch
2
for a1  1

Lui  Liy '  vˆLiy ''

Lui  Liy '  vˆLiy ''
Equivalent noise
Decode v to
û variances
j times

Decode u to
Output û
| uˆ | uˆ  vˆ |
THANK YOU!