AC-BiLSTM: A Spatial Bidirectional LSTM With Multi-Head Self-Attention for SNR Estimation

The Signal-to-Noise Ratio (SNR) in wireless communication is an important indicator that effectively reflects the quality information of communication signals. Traditional SNR estimation methods have problems such as insufficient estimation accuracy and limited application range of signal modulation types. To solve the above problems, this paper proposes a novel architecture based on deep learning. We significantly enhanced the model's capacity for processing sequential data by integrating multi-head self-attention mechanisms into a spatial bidirectional LSTM (BiLSTM). Furthermore, we introduced convolutional layer to further optimize the extraction of input data features and enhance the model's ability to capture and represent critical information. The proposed architecture is called attention-based BiLSTM with convolution layer (AC-BiLSTM). We used simulation to construct a communication signal dataset containing information such as different SNR and modulation methods. Experiments on this dataset showed that compared with the baseline methods, the performance and robustness of our proposed method are better, and the modulation types are more applicable.