SIGNALS NUMBER ESTIMATION IN SMART ANTENNAS BASED ON NEURAL NETWORKS

Abstract

Incident signal number estimation is a vital problem in smart antenna systems to readjust the radiation pattern adaptively. The performance of traditional methods in this field is seriously degraded in some extreme conditions, such as the observation of a relatively short signal, a low signal/noise ratio (SNR), or a noise covariance structure. Recently, machine learning has been used to estimate the number of signals due to the advantages of not requiring subjective parameter settings and being based on data. However, it has as a fundamental deficiency the fluctuation in performance in complex environments, so a more significant number of training samples is necessary. Given this problem, a preprocessing scheme and a neuronal network are proposed to manifest the characteristics of the input data. The conceived scheme adjusts the input data range in real time to improve prediction capacity and reduce training samples. The proposed method allows the application of simpler algorithms for the classification process focused on signal detection. Experimental and simulation results verify that the proposed schemes can improve estimation performance while reducing the computational cost.