IMPLEMENTATION OF FIXED POINT NOISE COMPENSATION IN IIR FILTERS USING ERROR FEEDBACK

Abstract

Infinite Impulse Response filters (IIR) with different approximations was first realized in 19th century. IIR filters have gained enormous popularity in the recent years due to its inherent ability to filter out the signals depending upon the approximation being used. A lot of work has been done on IIR filters since then, however concept of feedback compensation has been partially neglected. This thesis involves investigating a procedure to build an effective IIR filter with one of its trademark specifications. Different approximations have also been discussed to make IIR filter. Comparisons have also been made between different approximations with pros and cons of these approximations have also been stated. The second part of thesis represents an implementation and tradeoff analysis of compensated IIR filter with Error feedback (EF). Hardware analysis is made so that a better comparison can be made between the two techniques i.e. IIR filter with error feedback and without feedback. A simple method is devised to reduce the quantization error of the biquadratic section implementation by means of EF approach. The compensation method including the EF application utilized the fact that filter with an EF is more resistant to noise then filter without EF. Then an iterative method can be used afterwards to reduce the silicon area. First a standard recursive IIR filter is developed with basic quantization properties. Then a more advanced EFcompensated biquadratic section is designed. The compensated portion of the filter involves the calculation of feedback coefficients. Different methods have been stated for calculation of feedback coefficients. The quantization error of both the solutions is compared numerically and graphically. Although EF implementation results in more usage of resources and more complexity but its results are less error prone at the output of the quantizer. The designs are implemented using Xilinx ACCELDSP 10.1 and graphs have been developed using Matlab.