World Journal of Engineering Research and Technology

Andrew Adagbor Okwoche*, Lateef Adewale Fatoki, Tawo Godwin Ajuo and Etim Eyo Bassey

ABSTRACT

This study presents an improved channel estimation technique for MIMO-OFDM systems using a Hybrid Neural Network (HNN) architecture. The proposed model combines dense layers with ReLU activation functions to effectively learn and predict channel state information (CSI) from pilot-assisted input data. Simulations were conducted in MATLAB to benchmark the HNN against traditional estimators Least Squares (LS) and Minimum Mean Square Error (MMSE). Results demonstrated a significant reduction in Mean Square Error (MSE), from 0.012 (LS) and 0.0058 (MMSE) to just 0.0021 at an SNR of 20 dB. Similarly, the Bit Error Rate (BER) drops from 0.16 at 0 dB to 1.2×10⁻⁵ at 40 dB with the HNN. In terms of throughput, the hybrid approach Throughput increases from 0.7 bps/Hz at 0 dB to 6.6 bps/Hz at 40 dB LS and MMSE. These improvements highlight the potential of neural network-based estimators to deliver more robust and efficient performance in modern wireless communication systems, especially under challenging channel conditions like fading and noise.

[Full Text Article] [Download Certificate]