Neural-net based receiver structures for single- and multi-amplitude signals in interference channels

Abstract

This paper presents analysis and performance evaluation results for several neural-net based receiver structures which effectively combat additive channel interference, such as co-channel interference (CCI) and adjacent channel interference (ACI). Although the idea of employing neural net based receivers for interference channels is not new, the novel technical contributions of the authors' paper can be summarized as follows. (i) Propose, analyze and evaluate a training algorithm for Nyquist filtered single- and multi-amplitude signals which is based upon a novel non-uniform signal sampling technique. (ii) Propose and evaluate neural net structures employing a novel non-linear activation function for the detection of multi-amplitude signals. (iii) Present novel bit error rate (BER) performance evaluation results for coherent and noncoherent single- and multi-amplitude signals, including binary phase shift keying (BPSK), quadrature phase shift keying (QPSK) and quadrature amplitude modulation (QAM), operated in generalized CCI and ACI channels. The authors' research has demonstrated that, as compared to more conventional detection techniques, the proposed neural net receivers provide significant performance improvements in CCI and/or ACI channels. Their tolerance for inaccuracies in symbol timing synchronization also makes them good candidates for practical modem implementation. >