Enhanced Channel Estimation for FBMC/OQAM Using M-IAM-LS-DNN Towards 6G

Filter bank multicarrier/offset quadrature amplitude modulation (FBMC/OQAM) is a multicarrier modulation technique projected to replace orthogonal frequency division multiplexing (OFDM) in upcoming sixth-generation (6G) networks. However, because its orthogonality is confined to real-valued symbols, FBMC/OQAM suffers from intrinsic imaginary interference, which complicates channel estimation (CE) tasks. CE plays a crucial role in wireless communication systems; therefore, accurate CE is essential for next-generation networks, especially those supporting low-latency and vehicular applications. While evaluating channel characteristics accurately is critical, conventional CE methods are often inefficient. Recently, feedforward deep neural networks (DNNs) have garnered attention for their impressive performance in enhancing CE techniques. In this study, we propose and investigate a CE scheme based on neural networks, specifically the M-IAM-LS-DNN (Modified Interference Approximation Method Least Squares) approach, for FBMC/OQAM systems. This method uses neural networks to correct noise errors in LS channel estimation. According to simulation data, the suggested M-IAM-LS-DNN surpasses the conventional M-IAM-LS in terms of accuracy, by reducing the NMSE of the proposed M-IAM-LS-DNN for the 64 subcarriers scenario by 10% and the BER by 33% across a wide range of SNR levels.