Fixed-Point Design of Generalized Comb Filters: A Statistical Approach

TL;DR

This paper presents a statistical approach to designing efficient fixed-point Generalized Comb Filters with improved selectivity and noise rejection, analyzing sensitivity to coefficient quantization and validating through simulations.

Contribution

It introduces a sensitivity analysis for GCF filters in a partial polyphase architecture and develops a fixed-point design methodology based on this analysis.

Findings

Sensitivity of filter response depends on decimation factor split

Fixed-point GCF filters show robust performance in simulations

Design method improves quantization noise rejection

Abstract

This paper is concerned with the problem of designing computationally efficient Generalized Comb Filters (GCF). Basically, GCF filters are anti-aliasing filters that guarantee superior performance in terms of selectivity and quantization noise rejection compared to classical comb filters, when used as decimation filters in multistage architectures. Upon employing a partial polyphase (PP) architecture proposed in a companion paper, we develop a sensitivity analysis in order to investigate the effects of the coefficients' quantization on the frequency response of the designed filters. We show that the sensitivity of the filter response to errors in the coefficients is dependent on the particular split of the decimation factor between the two sub-filters constituting the PP architecture. The sensitivity analysis is then used for developing a fixed-point implementation of a sample filter from the class of GCF filters, used as reference filter throughout the paper. Finally, we present computer simulations in order to evaluate the performance of the designed fixed-point filters.