A neural network pre-processor for multi-tone detection and estimation

Abstract

A parallel bank of neural networks each trained in a specific band of the spectrum is proposed as a pre-processor for the detection and estimation of multiple sinusoids at low SNRs. A feedforward neural network model in the autoassociative mode, trained using the backpropagation algorithm, is used to construct this sectionized spectrum analyzer. The key concept behind this scheme is that, the network when trained for a certain spectral band, serves as an excellent filter with sharp transition and near complete attenuation in stopband, even at low SNRs. Simulation results to support the advantages of the proposed scheme are presented. Statistical measurements to determine its degree of reliability in detection have been made.<>