The Turbo Decoding Principle Tutorial Introduction and State of ... - PowerPoint

Document Sample

```					                      Combined Multiuser Reception
and Channel Decoding
for TDMA Cellular Systems
48th Annual Vehicular Technology Conference

Matthew Valenti and Brian D. Woerner
Mobile and Portable Radio Research Group
Virginia Tech
VIRGINIA POLYTECHNIC INSTITUTE & STATE UNIVERSITY
Blacksburg, Virginia         Virginia
Tech
VIRGINIA POLYTECHNIC INSTITUTE

MPRG                                                                  1872   AND STATE UNIVERSITY

MOBILE & PORTABLE RADIO RESEARCH GROUP
Introduction
   Performance of multiple access systems can
be improved by multiuser detection (MUD).
 Verdu,  Trans. Info. Theory ‘86.
Introduction

 Viterbi algorithm, complexity O(2K).

   MUD for CDMA systems.
detect signals from the same cell.
 Jointly
 Optimal MUD is too complex for large K.

   MUD for TDMA systems.
 Jointlydetect signal from within the cell plus
5/21/98         one or two strong interferers from other cells.
MUD for Coded TDMA
   TDMA systems use error correction coding.
   Soft-decision decoding outperforms hard-
decision decoding (2-2.5dB).
However, the optimal MUD passes hard-
Introduction



decisions to the channel decoder!
   Don’t use optimal MUD if loss due to hard-
decision decoding is greater than gain due to
multiuser detection.
   Alternatively, the interface between MUD
and channel decoder could be improved.
5/21/98
Outline of Talk
   System Model.
asynchronous.
 bit
 Generalized for both TDMA and CDMA.

   MUD for TDMA.
Introduction

 Turbo   processing.
   Simulation results
 RSC   coded system.
 1 strong interferer.
 SOVA decoders.
5/21/98
System Model
K
r (t )   sk (t )  n(t )
k 1
L
sk (t )  Pk  bk (i )ak (t  iT   k )e jk
System Model

i 1

   For TDMA:                          1 for 0  t  T
ak (t )  
0 elsewhere
   Matched Filter Output:


yk (i)    r (t )ak (t  iT   k )e  jk dt 


5/21/98
Optimal Multiuser Detection
   Place y and b into vectors:
y   y1 (1),, yK (1),, y1 ( L),, yK ( L) 
b  b1 (1),, bK (1),, b1 ( L),, bK ( L) 
   Compute cross-correlation matrix:
 1                      

 cos(i  j   j )  ai  j (t   i  j )a j (t   j  T )dt,             if i  j  K
MUD

 T
Gij                         

1
 cos(
i  j  K   j )  ai  j  K (t   i  j  K )a j (t   j )dt ,   if i  j  K
T
                        

   For the TDMA case the above reduces to:
cos(   )   ,
  1
                         if i  j  K
i j     j    i j     j
Gij           T
 cos(i  j  K   j )T   j   i  j  K , if i  j  K
1
5/21/98
T
Optimal MUD (Continued)
    Run Viterbi algorithm with branch metric:
                     K 1                                
i (b)  N o log p(bi )  bi P (i ) 2 yi  bi P (i )  2 bi  j P (i  j ) GK  j , (i ) 
                     j 1                                
i mod K if (i mod K )  0
 where           (i)  
MUD

 K      if (i mod K )  0

    Note that the p(b) term is usually dropped.
 The       channel decoder will provide this value.
    The algorithm produces hard bit decisions.
 Not  suitable for soft-decision channel
5/21/98
decoding.
Soft-Output MUD
   Several algorithms can be used to produce
soft-output.
   Trellis-based.
 MAP   algorithm
MUD

 Log-MAP,Robertson et al, ICC ‘95
 OSOME, Hafeez & Stark, VTC ‘97

 SOVA   algorithm
 Hagenauer   & Hoeher, Globecom ‘89
   Non-trellis-based.
5/21/98        Suboptimal,   reduced complexity.
Proposed System Architecture
p (bi )
Interleaver

SISO                       p (b j )
Multiuser   Deinterleaver
Detector
rbb (t )              yi                                                       SISO      ˆ
uk ( j )
Matched                                                            Channel
System Model

Filters                                Deinterleaver              Decoders

Channel                  2
Estimator

   Each user interleaves its coded bits prior to
transmission.
   Initialize p(bi) = 1/2
5/21/98
Simulation Parameters
   2 users
 Desired user
 1 co-channel interferer with 3 dB less power.
Example

   Recursive Systematic Convolutional codes
 Constraint     length 3.
 Rate    1/2.
   SOVA decoding.
 BothMUD and Channel decoder.
 Normalized outputs, Papke et al, ICC ‘96.
5/21/98
Simulation Details
   “Conservative” approach taken
 Only   the desired user is decoded.
 No   channel decoder for interferer.
 Only the APP of the systematic bits of the
Example

desired user is fed back to the MUD.
 The APP for the parity bits are not computed
or used.

5/21/98
Simulation Results:
Existing Methods
-1
10

   At BER=10-3
   MUD gain is 4.7 dB.
-2
10
   Coding gain is 6.7 dB.
   Gain using hard output
-3
10
MUD and coding, 4.6 dB.
   Therefore it does not
BER

make sense to use (hard-
-4
10
outut) MUD and channel
coding.
-5
10

matched filter, uncoded
multiuser detector (MUD), uncoded
MUD, hard-decision decoding
matched filter, soft-decision decoding
-6
10
4   6         8            10             12     14   16   18
E b /N o in dB
Simulation Results:
-2
New Method
10

   The proposed iterative
MUD / channel decoding
-3
10
strategy is used.
   At BER 10-5
   After 2 iterations,
proposed method shows
BER

-4
10
.4 dB improvement over
channel decoding alone.
   After 3 iterations, the
-5
10
   No measurable gain for
matched filter, soft-decision decoding
combined MUD/decoding, 2 iterations
combined MUD/decoding, 3 iterations
more than 3 iterations.
-6
10
9   9.5   10     10.5      11         11.5     12   12.5   13   13.5   14
E b /N o in dB
Conclusion
   Optimal MUD can be used for TDMA.
 However,   if channel coding is used then the
interface between MUD and decoder is critical.
Conclusion

   A strategy for iterative MUD/channel
decoding is proposed.
 Based   on the concept of turbo processing.
   Proposed strategy was illustrated by
simulation example.
 Modest gain by using proposed strategy over
5/21/98         channel coding alone.
Future Work
   More aggressive use of soft-information.
 Use parity information for RSC codes, or use
conventional convolutional coding.
Conclusion

 Decode the interfering user.

   Share information among base stations.
 Decode each user at the closest base station.
 Send the results to all the other base stations.

   Use MAP algorithm instead of SOVA.
   Fading, channel estimation, and equalization.
5/21/98
Future Work
    Combine MUD, decoding, and base station
diversity.
Conclusion

MUD       Ψ 1q )
(

y1     at
B.S. #1
Maximal
z   )q(   Bank of    Λ (q )
Ratio               K SISO
Combining             Channel    Ω(q )    ˆ
d (q)
Decoders
MUD      Ψ (q )
M
at
yM
B.S. #M

5/21/98
Simulation Results:
Diversity Combining
0
10
MUD and decoding only
MUD/decoding/diversity: One iteration
   2 users and 2 base
MUD/decoding/diversity: Two iterations
station.
-1
10
   At each B.S. closer user
is 3 dB stronger than
-2
10
more distant one.
BER

-3
10
   log-MAP decoder and
MUD.
-4
   K=3 r=1/2 conventional
10
convolutional code.
   4 dB gain after 1 iteration
-5
10
0   2   4   6
E b /N o in dB
8         10          12             14
   6 dB after 2 iterations.

```
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
 views: 3 posted: 10/3/2012 language: English pages: 17
How are you planning on using Docstoc?