Attention-Based LSTM Autoencoder for Noise-Resilient Semantic Communication for IoT Devices in 6G Networks

Semantic communication (SC) has emerged as a key enabler for 6G networks to overcome the Shannon limit. SC aims to transmit meaning rather than raw data. In this context, many researchers proposed the architecture for SC. However, most of the suggested models overlooks the influence of noise in the semantic channel. This limitation reduces the robustness of the models in the real world. In this context, we propose an Attention-Based LSTM Autoencoder for Noise-Resilient SC for IoT devices in 6G Networks. The proposed model combines LSTM with self-attention to preserve contextual meaning and reconstruct messages under additive Gaussian noise. Compared to baselines for LSTM, GRU and transformers, the proposed method achieves the lowest MSE (52.43), MAE (54.88) and MAPE (75.40), along with improved sMAPE (93.53).