Residual learning based convolution neural network for improved channel estimation for VehA channel

Accurate channel estimation is crucial for reliable communication in orthogonal frequency division multiplexing (OFDM) systems, especially in high-mobility scenarios. Traditional channel estimation techniques, such as least squares (LS) and linear minimum mean square error (LMMSE), face limitations in terms of estimation accuracy and computational complexity. To address these challenges, this paper proposes a novel convolutional neural network (CNN)-based channel estimation framework utilizing residual learning and iterative refinement. The residual learning approach mitigates vanishing gradient issues and enhances convergence speed, while iterative refinement progressively improves the channel estimates. Comprehensive simulations using the VehicularA (VehA) channel model demonstrate that the proposed method achieves up to 30% lower mean squared error (MSE) compared to LS estimation and 15% lower MSE compared to ChannelNet at an SNR of 12 dB. Furthermore, the proposed framework shows an MSE reduction of over 20% compared to FSRCNN at low pilot densities, while maintaining robustness across a wide range of SNRs. With its reduced computational complexity and superior performance, the proposed CNN-based framework is suitable for real-time implementation in next-generation wireless systems, including 5G and beyond.