Hybrid AE and Bi-LSTM-Aided Sparse Multipath Channel Estimation in OFDM Systems

OFDM is a powerful modulation technique that efficiently transmits high-speed digital data over frequency-selective fading channels. It divides the signal into multiple subcarriers, each experiencing flat fading, and compensates for independent noise using a simple one-tap equalizer. In wireless systems, coherent detection is often employed, and accurate channel knowledge is crucial for optimizing performance. Traditional channel estimation methods, such as Least-Squares (LS), utilize randomly spaced pilot patterns, which can compromise spectral efficiency. Significant losses in BER and MSE affect many such existing methods in detrimental manner. To counter such failure, this paper proposes a new hybrid deep learning-aided sparse multipath channel estimation approach for OFDM communication systems. Additionally, a hybrid deep learning HA-Bi-LSTM model is developed by combining Bidirectional Long Short-Term Memory (Bi-LSTM) and Auto-Encoder (AE) to enhance data communication performance. Further, an efficient Grasshopper Electric Fish Optimization Algorithm (G-EFOA) is developed to optimize the parameters of AE and Bi-LSTM, resulting in superior solutions. AE handles feature extraction, while Bi-LSTM performs estimation. The proposed model aims to minimize BER and MSE values. Results show about an average 77% improvement in BER across varied modulation schemes, along with a significant drop in MSE values by 84%. Also, a sizable drop-in simulation time proves the contribution of the algorithm. Finally, the proposed model’s effectiveness is demonstrated through performance evaluations.