Adaptive System Identification using Markov Chain Monte Carlo

TL;DR

This paper explores the application of Markov Chain Monte Carlo (MCMC) methods to adaptive system identification, aiming to leverage MCMC's low computational complexity for improved impulse response estimation.

Contribution

It provides a comprehensive analysis of MCMC techniques specifically tailored for adaptive system identification, filling a gap in existing research.

Findings

MCMC offers a promising low-complexity alternative for adaptive filtering.

The paper demonstrates the effectiveness of MCMC in reducing identification error.

Analysis of MCMC properties in the context of adaptive filtering is presented.

Abstract

One of the major problems in adaptive filtering is the problem of system identification. It has been studied extensively due to its immense practical importance in a variety of fields. The underlying goal is to identify the impulse response of an unknown system. This is accomplished by placing a known system in parallel and feeding both systems with the same input. Due to initial disparity in their impulse responses, an error is generated between their outputs. This error is set to tune the impulse response of known system in a way that every change in impulse response reduces the magnitude of prospective error. This process is repeated until the error becomes negligible and the responses of both systems match. To specifically minimize the error, numerous adaptive algorithms are available. They are noteworthy either for their low computational complexity or high convergence speed. Recently, a method, known as Markov Chain Monte Carlo (MCMC), has gained much attention due to its remarkably low computational complexity. But despite this colossal advantage, properties of MCMC method have not been investigated for adaptive system identification problem. This article bridges this gap by providing a complete treatment of MCMC method in the aforementioned context.