Iterative Multi User DetectionDecoding and Applications in UTRA TDD

1
Iterative Multi User Detection/Decoding and
Applications in UTRA TDD

• Stefan Brück
• Global Wireless Systems Research
• Bell Labs, Lucent Technologies

2
Overview

• Motivation of the Problem
• Asymptotic Distance as Criterion for Code
  Construction
• Interleaver Design for Convolutional Codes
• Iteration between Multi User Detection and
  Decoding
• An UTRA TDD like System
• Simulation Results
• Conclusions

3
Uplink Transmission
u1
Coding/ Interleaving
h1(?,t)

• Performance Degradations
• Additive white gaussian noise
• Inter and intra cell interference
• Inter symbol interference

Source 1
n(t)
r(t)
Coding/ Interleaving
h2(?,t)
Source 2
u2
How must the channel codes be constructed for
intra cell interference?
4
Time Discrete Receiver Model
h1(?,t) h2(?,t) ?(?) ? AWGN only
colored noise
white noise
s1(-t)
r(t)
Whiten. Filter
s2(-t)
lt s1(t), s2(t) gt ?

E(n1n2) 0
5
Receiver Analysis
Vector Notation
y LT (b1,b2)T n y LT b n
ML receiver minimizes euclidian distance between
y and LT(b1,b2)T
E1,S E2,S
b2

• Matrix LT rotates signal space
• min. squared euclidian distance
• d2E,min min 4, 8(1-?) for E1,S E2,S
• 4(1 - ?2) for
  E2,S ?2 E1,S
• 
  (worst case)

4
(1,1)
(-1,1)
8(1-?)
b1
(1,-1)
(-1,-1)
6
Receiver Analysis - cont.
Asymptotic Distance dA 1/(4 E1,S)
min ?i1,...,N LT c1,i - LT c2,i2
(with respect to user 1)
C1 ? C2, E2,S
7
Receiver Analysis - cont.
Bounds df(1-?2) ? dA ? df
? Code construction according to
this distance!
8
Code Construction - Basic Idea
equal code books, E2,S ?2 E1,S
c1 ( 0 0 0 0 )
c1 ( 1 1 0 1 )
c2 ( 0 0 0 0 )
c2 ( 1 1 0 1 )
dA 3(1-?2)
different code books, E2,S ?2 E1,S
c1 ( 0 0 0 0 )
c1 ( 1 1 0 1 )
c2 ( 0 0 0 0 )
c2 ( 1 1 1 0 )
dA 2(1-?2) ?2 1
9
Main Theorem
Interleaving Let g be relatively prime to
the code length N. The interleaved
codeword is given by cg (cg,1, ? ,
cg,N) with cg,i cig mod N The
deinterleaver is given by g-1 with g g-1 1
mod N Theorem If the interleavers are chosen
properly, dA is lower bounded by df
(1- ?2) ? min df - ?2, df2 (1- ?2) ?
dA dA is always upper bounded
by dA ? df - ?2 lt df
10
Proof - Basic Idea
User 1 c1,g1g2-1 ( 1 0
0 0 1 0 0 0
1 )
User 2 c2

• Interleaver g1g2-1 shall stretch out minimal
  weight codewords of user 1
• Covering paths in trellis of user 2 shall have
  large weight ? active distances

11
Discussion

• Complete degradation by interference can be
  eliminated, because
• lim dA / df ? 1, if df ? ?
• Interleaving does not depend on matrix LT, i.e.
  it is independent on the channel
• The performance of Point-to-Point transmission
  may be achieved in a CDMA system
• Interleaving increases distance ? It is required
  even in memory less systems
• Derivation assumes ML receiver, i.e. joint multi
  user detection and decoding ? A suboptimal
  receiver may not fully exploit the increased
  distance

12
Iterative Receiver - General View
P(u1,k Y) ? P(u1,k)
P(u2,k Y) ? P(u2,k)
13
MAP Decoder
Output P(uk,jY) ? ?t(m) ?t(m)
P(ck,iY) ? ?t(m1)?t(m1,m2)?t1(m2)
14
Soft In/Soft Out Detector
Output p(yic1,i) ?c2,i
p(yici) P(c2,i) p(yic2,i) ?c1,i
p(yici) P(c1,i)
15
Comparison to Turbo Decoding

• Constituent codes of a Turbo code systematic and
  linear
• A priori knowledge, channel information and
  extrinsic information can be separated
• Only extrinsic information must be used for next
  decoder
• Extrinsic information is used for updating
  P(m1m2) the channel information is kept constant

• Intra cell interference
• inner non-systematic, non-linear code
• Channel information and extrinsic information can
  not be separated after detection
• A priori knowledge of user k is not included in
  detector output of user k
• P(m1m2) 1/2 in all iterations the channel
  information p(yick,i) is updated

16
Simulations - AWGN

• K 3 users
• (7,5) convolutional code
• code length N 364
• 0.6 ? lt si(t), sj(t) gt ? 0.8
• Optimal iterative receiver
• with FEC optimization
• Max. Likelihood receiver
• without FEC optimization

17
Supoptimal Iterative Receiver
y1,i c1,i n1,c,i ? p(y1,ic1,i)
MAP Dec.
Yc (y1,c,,yN,c) with yi,c R ci
ni,c (matched filter output)

R-1 ZF Filter
yi
Soft Bits C (c1,,cN)
MAP Dec.
y2,i c2,i n2,c,i ? p(y2,ic2,i)
Yc (y1,c,,yN,c) C (c1,,cN)
Parallel Interference Cancellation
Ynew (y1,new,,yN,new)
18
Parallel Interference Cancellation
Vector channel yi,c R ci ni,c
Channel matrix R D M, D
diagonal matrix, M interference matrix
Step 1 yi,new D-1(yi,c - M ci) ci
D-1 M (ci - ci) D-1 ni,c Step 2
yi yi,new ? corrected matched filter
output Step 3 calculate p(yj,icj,i)
as input for MAP decoder of user j
19
An UTRA TDD like System

• Frame duration 6 ms (10 ms)
• 12 time slots/frame (15)
• 4-/8-/16-PSK (QPSK)
• 48 symbols/time slot (122-2208)
• Interleaving over 4 time slots
• (inter-frame interleaving)
• Spreading length 14 (1...16)
• 2 Mcps (3.84 Mcps)
• 1.6 MHz (5 MHz)

P. Baier, P. Jung et. al.
User 1
Data
Data
Midamp.
?
134?s
168?s
30?s
168?s
User 8
20
Simulations - Typical Urban

• K 4, 5 users
• (23,35) convolutional code
• code length N 384
• 1/7 ? lt si(t), sj(t) gt ? 3/7
• Suboptimal iterative receiver
• with and without FEC
• optimization

21
Conclusions

• All results can be extended to K users
• The importance of FEC optimization in CDMA (and
  MIMO systems)
• has been shown
• Derivation of an optimal Soft In/Soft Out
  detector (see also M. Moher and M. Reed et. al.)
  for use in an iterative receiver
• Suboptimal iterative receiver with ZF filter and
  PIC for UTRA TDD
• BER is very close to Point-to-Point performance