Fundamentals of Statistical Signal Processing, Volume III

Chapter 6. Noise Model Selection

6.1. Introduction

We complete our discussion of model selection in this chapter by focusing on noise models. Most of the introductory comments made for signal model selection given in Section 5.1 apply here as well. Rather than repeat the discussion, we just summarize the salient points.

1. A noise model is only an approximation to reality. It need not be perfect, as it is only intended to model the important features of the noise. Those features are the ones that affect the signal processing algorithm performance. Consider, for example, the modeling of the spectral characteristics of noise over bands that are not likely to contain a signal. These bands are irrelevant to the algorithm performance, and so any ...