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

Area of Science:

• Wireless Optical Communication
• Signal Processing
• Machine Learning Applications

Background:

• Ultraviolet Communication (UVC) suffers from high Bit Error Rate (BER) due to atmospheric scattering.
• Existing channel equalization methods struggle with complex nonlinearities in UVC channels.
• Advancements in wireless optical communication highlight the need for improved equalization techniques.

Purpose of the Study:

• To propose a novel Long Short-Term Memory - Deep Neural Network (LSTM-DNN) based channel equalization approach for UVC.
• To enhance signal recovery accuracy and transmission quality in UVC systems.
• To address the limitations of traditional equalization methods in nonlinear UVC channels.

Main Methods:

• Developed an LSTM-DNN model integrating LSTM for temporal dependencies and DNN for nonlinear feature extraction.
• Applied the LSTM-DNN model for channel equalization in UVC systems.
• Compared the performance of LSTM-DNN against conventional methods like LMS, RLS, PSO, SVM, and MMSE.

Main Results:

• The LSTM-DNN model significantly reduced Bit Error Rate (BER) and Mean Squared Error (MSE) compared to traditional methods.
• At 0 dB SNR, LSTM-DNN achieved a BER of 0.135, outperforming LMS (0.45) and MMSE (0.20).
• At 20 dB SNR, LSTM-DNN's BER dropped to 0.015, demonstrating robust performance and an average reduction of ~67.8% in BER and ~70.8% in MSE.

Conclusions:

• The LSTM-DNN model offers superior performance for UVC channel equalization, enhancing signal recovery accuracy and transmission quality.
• This approach effectively mitigates signal attenuation issues in UVC, showing high precision and stability.
• The proposed LSTM-DNN method holds significant theoretical value and practical applicability for advanced UVC systems.