PerSurvivor Processing PSP

1
Per-Survivor Processing (PSP)

• Bahador Amiri
• Fall 2005

2
Introduction

• High data rate transmission ? Inter-symbol
  Interference (ISI)
• Equalization
• Linear Equalization (LE)? noise enhancement
• Decision Feedback Equalization (DFE) ?negative
  affected by error propagation
• Maximum Likelihood Sequence Estimation (MLSE)
• Viterbi Algorithm (VA) Trellis-coded modulation
  (TCM)
• improve coding gain and reducing the complexity
  of MLSE
• Incorporating VA with DFE
• Reduce the complexity by truncation of channel
  impulse response
• Incorporating DFE within the Viterbi instead of
  externally

3
Typical Communication System Model

4
Conventional MLSE Receiver

• Unknown channel
• carrier phase
• WMF estimation
• timing epoch used in symbol rate sampling
• discrete time overall channel
• Using data-aided estimation techniques
  (decision-directed mode)

5
Conventional MLSE Receiver (cont.)
6
Per-Survivor Processing
7
Reminder Branch Metric
8
Per-Survivor Processing (Cont.)
9
PSP Block Diagram

10
Channel updating in PSP fashion

11
Joint MLSE and MSE Channel Identification

• The channel estimates and the updated based on
  the classic gradient algorithm
• ß is selected as a compromise between speed of
  convergence and stability
• a is a encoded symbol vector
• e( ) is the error at k-th epoch for all
  possible transitions
• Associated with each survivor, besides a metric
  and a data sequence typical of the Viterbi
  algorithm, there are estimate of the encoded
  symbol sequence and estimate of the channel
  response

12
Numerical result for joint MLSE MSE

• 16 QAM
• ß 0.01
• Discrete time channel response

13
Joint ML estimation of channel sequence

• Maximizing likelihood function over and
  
• RLS algorithm
• Kalman gain vector
• Inverse of correlation matrices
• Channel impulse response

14
Joint ML estimation of channel sequence (cont.)

• The maximization of the likelihood function with
  respect to the channel vector for a given
  survivor, performed by Recursive Least Square
  (RLS) algorithm
• ? is the weight factor and the stability and
  convergence rate of the RLS is critically
  affected by it
• a is a encoded symbol vector
• e( ) is the error at k-th epoch for all
  possible transitions
• Associated with each survivor now there are a
  metric, a survivor sequence, a channel vector, a
  Kalman gain vector, and an estimate of the
  inverse of the correlation matrix

15
Numerical Results for joint ML estimation of
channel sequence

• 4 PAM ?0.999
  4 QAM