Real time deconvolution using the conjugate gradient algorithm

Abstract

A solution to the deconvolution problem is presented using iterative methods, which has been implemented in real time on the AT&T DSP-32 processor. Given the output of a linear system, and its transfer function, a procedure is shown to compute the input time function. Since there are, in general, infinitely many inputs which will produce the same output time function, the problem cannot be solved directly. However, using an iterative approach it is possible to converge to one solution. The technique uses the conjugate gradient algorithm to obtain a candidate input function and post processes the function based on its periodicity in the frequency domain to enhance resolution.<>