Design of on-line arithmetic networks for signal processing applications : a color filter implementation

by Sinky, Mohammed H.

Networks of on-line modules for Digital Signal Processing (DSP) applications

have been designed and analyzed in many papers. Various arguments have

been made for on-line arithmetic over other bit-serial and parallel approaches.

However, the majority, if not all of the on-line network implementations presented

throughout the literature do not take into account the input and interfacing issues

involved when dealing with real applications. Most studies of on-line network implementations

are limited to common components of DSP algorithms but do not

consider the impact they have on the whole system.

This thesis investigates the use of on-line arithmetic in implementing a

color image filtering algorithm. A complete analysis of the algorithm takes into

account input value ranges, memory requirements, error analysis, and processing

throughput. An equal emphasis is placed on what the end-user is to expect.

The realization of the image processing algorithm is presented using conventional

on-line modules. A comparison to the estimates of parallel and other

traditional bit-serial approaches is done. The results show that the on-line implementation saves the designer an overwhelming amount of area as opposed to

a strictly parallel implementation and reduces complexity when compared to an

implementation using traditional bit-serial modules, while still providing a significant

throughput.