Noninvasive sensor arrays: separate evaluation of inputs for adaptive noise reduction

Abstract

An algorithm is described which optimally reduces additive noise from a biological signal, given one primary input and multiple reference inputs, by the method of least squares. The algorithm separately compares two parallel cost functions for each reference input, in gain space, as a means of updating the global weight vector. At the global minimum of the cost function, optimal noise reduction is achieved. The algorithm requires minimal code, and rapidly converges even with the signal to noise ratio <<1. On an IBM PC-XT, the algorithm adaptively processed four analog inputs for noise reduction (400 Hz sampling) to provide an output optimized in real time.<>