Equalizer Design: HBOA-DE-trained radial basis function neural networks

Abstract

Communication systems that rely on wireless technology require signal processing techniques to improve their channel performance. Wireless communications are susceptible to various signal distortions during transmission, including inter-symbol interference, adjacent channel interference, and co-channel interference. As a result, achieving error-free signal transmission in wireless communication is often challenging. To make sure the signal is recovered with a minimum bit error rate, equalizers are needed at the front end of the receiver. As an optimization algorithm, a nature-inspired hybrid algorithm is applied, namely BOA/DE, which is a combination of the Butterfly optimization algorithm (BOA) and differential evolution (DE). Based on a suitable network topology and transfer function, the presented work proposes an algorithm for training radial basis function neural networks (RBFNNs) that is applied to the problem of channel equalization. Both BOA and DE are advantageous in the proposed algorithm, which permits it to produce efficient results by balancing exploration and exploitation. Several methods have also been discussed in the literature that use optimization techniques to deal with the problem of equalization. The same problem is treated in this article as a classification issue. As a further step in the evaluation of the HBOA-DE-based RBFNN equalizer, three non-linear channels and adding different nonlinearities have been simulated. The proposed algorithm is compared with well-known algorithms in terms of Mean Square Error (MSE) and Bit Error Rate (BER). Additionally, the algorithm has been tested against a situation in burst error and evaluated via bit error probability (BEP) to establish its robustness and performance. Results showed that the method performed better in handling burst errors compared to others. It has been shown that the projected method outclasses other methods even in poor signal-to-noise ratio conditions, which is borne out by extensive simulation studies.