Channel equalization in ultraviolet communication based on LSTM-DNN hybrid model

Ultraviolet Communication (UVC) faces the challenge of increased Bit Error Rate (BER) due to signal attenuation caused by atmospheric scattering. In recent years, wireless optical communication technologies have made significant progress in both Ultraviolet (UV) and Visible Light (VL) communication domains. However, traditional channel equalization methods still exhibit limitations when handling complex nonlinear channels. This study proposes a Long Short-Term Memory - Deep Neural Network (LSTM-DNN)-based channel equalization approach to enhance signal recovery accuracy. The model leverages LSTM to process temporal dependencies and combines it with DNN for nonlinear feature extraction, thereby improving its adaptability to single-scattering channels. Experimental results demonstrate that the LSTM-DNN model shows significant advantages in improving signal recovery accuracy and transmission quality compared to conventional methods. These methods include Least Mean Squares (LMS), Recursive Least Squares (RLS), Particle Swarm Optimization (PSO), Support Vector Machine (SVM), and Minimum Mean Squared Error (MMSE). Specifically, the LSTM-DNN model outperforms traditional methods across key performance metrics such as BER and Mean Squared Error (MSE). When the Signal-to-Noise Ratio (SNR) is 0 dB, the LSTM-DNN model achieves a BER of 0.135, significantly lower than LMS (0.45), RLS (0.38), PSO (0.35), SVM (0.25), and MMSE (0.20). As SNR increases, the LSTM-DNN model's BER further decreases, demonstrating strong robustness. When the SNR is 20 dB, the BER of the LSTM-DNN model drops to 0.015, substantially outperforming conventional methods. Additionally, the LSTM-DNN model exhibits the smallest MSE values, with 0.035 at 0 dB SNR and decreasing to 0.004 with higher SNR. On average, the LSTM-DNN model reduces BER by approximately 67.8% and MSE by about 70.8% compared to traditional methods. These results confirm that the LSTM-DNN model significantly improves signal recovery accuracy and transmission quality in UVC systems. Overall, the LSTM-DNN model demonstrates superior performance in UVC applications compared to conventional methods, offering higher precision and stability. This study effectively addresses signal attenuation issues in UVC, significantly enhancing signal recovery accuracy and transmission quality, thus possessing important theoretical value and practical significance.