LeNet with Bidirectional Long Short-Term Memory for Signal Detection in Underwater Optical Communication

Underwater optical signal detection performance suffers from occlusion and turbidity in degraded environments. Due to complex underwater environment the high directionality of light beam and vibration transceiver incident optical power usually spans a very wide dynamic range. This paper, proposed LeNet with Bidirectional Long Short-Term Memory (BiLSTM) technique performed efficiently in the signal detection in under water optical communication by effectively extracting spatial features and capturing temporal dependencies. The data collected for this research includes data bits of 72 vectors of transmitted signals through the Gulf seawater. The LeNet approach process the received optical communication signals and extract spatial features from convolutional layer, converting them into relevant signal to enhance the communication of underwater. The combination of LeNet feature extraction capabilities and Bi-LSTM sequential learning the signal then enhances the model ability leading to achieve better performance in signal detection. The performance of proposed method is evaluated with various metrics such as accuracy, precision, recall, f1-score and Bit Error Rate (BER) metrics based on sea water.