Application of multi-zero artificial neural network to the design of an m-valued digital multiplier

Abstract

An M-ary digital multiplier using artificial multi-zero neural networks and elementary analog arithmetic units has been derived. This multiplier should be accurate because its main arithmetic process is digital, while the speed should be very high because it is a free-running, parallel, and M-ary operation. The multi-zero neural network is a feedback artificial neural system consisting of N neurons. Each neuron is a nonlinear amplifier with input-output response function equal to a polynomial function containing 2M+1 real zeros. A unique property possessed by this nonlinear feedback system is that if the connection matrix is programmed correctly, any N-bit analog input vector will always be converged to an N-bit M-valued digital vector at the output. This output will be locked-in in place (or it can be memorized) even when the input is removed.<>