I. Introduction

A. Purpose of DSP.

General purpose digital signal processors are by far the most commonly used method for digital filter implementation, particularly at audio bandwidths. These systems possess architectures well suited to digital filtering, as well as other digital signal processing algorithms.

B. Development of DSP

The generic microprocessors of 1980's were ill suited for the implementation of DSP algorithms, due to the lack of hardware support for numerical algorithms of significant complexity in those architectures. The primary requirement for digital signal processing implementation was identified to be hardware support for multiplication, due to the large number of multiply-accumulate operations in digital signal processing algorithms and their large contribution to computational delays.

C. Classification of DSP

The processors are best classified in two categories, fixed point processors and floating point processors. In both cases, these architectures are commonly based on a single arithmetic unit shared amongst all computations, which leads to constraints on the sampling rates that may be attained.

1. Fixed Point Processors

Fixed point processors exhibit extremely high performance in terms of maximum throughput as compared to their floating point counterparts. In addition, fixed point processors are typically inexpensive as compared to floating point options, due to the smaller integrated circuit die area occupied by fixed point processing blocks. A major difficulty encountered in implementing filters on fixed point processors is that overflow and underflow need to be prevented by careful attention to scaling, and roundoff effects may be significant

2. Floating Point Processors

Floating point processors, on the other hand, are significantly easier to program, particularly in the case of complex algorithms, at the cost of lower performance and larger die area. Given the regular structure of most digital filtering algorithms and computer-aided design support for filters based on limited precision arithmetic, fixed point implementations may be the more cost effective option for this type of algorithm.

D. Abstract

The main abstract of this research is to understand one of the famous signal processing applications, Mu-Law Companding, and floating point algorithm. To reach the goal, C code compander will be compiled and run with Texas Instrument's floating point digital signal processor, TMS320C30.