A unified framework for finite-memory detection

Gianluigi Ferrari

doi:10.1109/JSAC.2005.853812

2005-09-01

PRE

Gianluigi Ferrari

Giulio Colavolpe

R. Raheli

DOI：10.1109/JSAC.2005.853812

原文链接

原文求助

摘要

En 中文

In this paper, we present a general approach to finite-memory detection. From a semi-tutorial perspective, a number of previous results are rederived and new insights are gained within a unified framework. A probabilistic derivation of the well-known Viterbi algorithm, forward-backward, and sum-product algorithms, shows that a basic metric emerges naturally under very general causality and finite-memory conditions. This result implies that detection solutions based on one algorithm can be systematically extended to other algorithms. For stochastic channels described by a suitable parametric model, a conditional Markov property is shown to imply this finite-memory condition. This conditional Markov property, although seldom met exactly in practice, is shown to represent a reasonable and useful approximation in all considered cases. We consider, as examples, linear predictive and noncoherent detection schemes. While good performance for increasing complexity can often be achieved with a finite-memory detection strategy, key issues in the design of detection algorithms are the computational efficiency and the performance for limited complexity.

Keyword:

adaptive detection

finite-memory detection

forward-backward (FB) algorithm

graph-based detection

iterative detection

maximum a posteriori (MAP) sequence/symbol detection

sum-product (SP) algorithm

Viterbi algorithm (VA)