A new IIR algorithm based on the overlap-save approach

In this work, a new IIR algorithm is presented. It is based on a block implementation method for IIR filters, and has two stages. First, a circular convolution is used to transform the input into intermediate results. Second, a correction circuit transforms the intermediate result into the correct output. Because the correction circuit uses part of the previous output to correct the current intermediate result, our approach is an overlap-save based algorithm. The multiplicative complexity of this algorithm is shown to be 2 log (kn)+[8 log(n)+12/k]+1 real multiplications per output point, for a block of size kn, where n is the order of the filter and k/spl ges/2, is a constant. This complexity is much less (by more than 50%) than other known algorithms. Another advantage of this algorithm is that it is easily parallelizable.