Bi-LSTM-Transformer hybrid network for V2V optical wireless ISAC systems

Abstract

With the growing integration of Integrated Sensing and Communication (ISAC) with Optical Wireless in intelligent transportation systems—particularly in Vehicle-to-Vehicle (V2V) communication for autonomous driving, traditional signal processing techniques encounter substantial challenges in detecting communication and LiDAR signals. This is particularly evident in high-noise environments, where these methods struggle to effectively suppress interference, leading to elevated bit error rates (BER) and diminished accuracy in ranging and velocity measurements. To tackle this issue, we introduce a hybrid neural network architecture that integrates Bidirectional Long Short-Term Memory (Bi-LSTM) and Transformer for signal detection. Simulation results demonstrate that, at a BER of 10^-3, the proposed algorithm achieves a Signal-to-Noise Ratio (SNR) gain of approximately 6.9 dB under the conditions of a 780 nm wavelength and moderate turbulence at the communication receiver. At the LiDAR receiver, under an SNR of 12 dB, the Root Mean Square Errors (RMSE) of distance and velocity are reduced by approximately 93.6% and 53.5%, respectively. Furthermore, the ambiguity function displays a more concentrated main lobe and superior suppression of side lobes, thus validating the algorithm's efficacy in enhancing distance resolution and Doppler tolerance.