A low-complexity deep residual CNN for channel estimation in OFDM wireless sensor networks

Abstract Accurate channel estimation is essential for reliable communication in OFDM-based wireless sensor networks (WSNs), where resource constraints and dynamic environments pose significant challenges. In this paper, we propose a low-complexity deep residual convolutional neural network (CNN) for channel estimation, using pilot-domain least-squares (LS) estimates to reconstruct the full channel response. The proposed architecture integrates residual connections with convolutional layers to enhance noise suppression and feature representation, while maintaining a compact structure with significantly fewer trainable parameters than existing deep learning approaches. Simulation results based on the 3GPP ETU channel model indicate that the proposed method consistently improves upon LS with interpolation across a broad SNR range. It achieves performance comparable to, and in certain SNR regimes slightly better than, prior deep learning models such as ReEsNet, despite requiring more than an order of magnitude fewer parameters. Compared with ChannelNet and ReEsNet, the proposed architecture reduces the number of trainable parameters by factors of approximately 135 and 10, respectively, and also lowers computational complexity in terms of multiply–accumulate operations. These findings suggest that the proposed network provides a favorable trade-off between estimation accuracy and computational efficiency, making it a promising candidate for resource-constrained WSN implementations.