Efficient Power Prediction for Intersatellite Optical Wireless Communication System Using Artificial Neural Network

Abstract

Intersatellite optical wireless communication (IsOWC) system has garnered global attention for facilitating high-speed data transfer between two space-based satellites. However, accurately predicting received or output signal power in a lower earth orbit trajectory is challenging due to factors such as background light, scintillation, pointing error, and optical crosstalk. To overcome this problem, a technique based on artificial neural networks (ANN) is proposed to enhance the efficiency of received signal power in the IsOWC system. The input features for an IsOWC system include propagation distance, scintillation attenuation, wavelength, pointing error, and input power, ranging from 1 to 25 km, 0 to 6 dB, 800 to 1600 nm, 0 to 1 µradian, and 0 to 4.77 dBm, respectively. The output feature i.e., received signal power, ranges from − 100 to 34.99 dBm. Before training, exploratory data analysis is performed on 2100 datasets generated by 16-quadrature amplitude modulation based IsOWC system. Furthermore, an ANN model is trained, resulting in a low mean squared error (MSE) of 4.8 × 10^− 6 compared to other machine learning model. The impact of hyperparameter tuning on the MSE curve is rigorously discussed. Additionally, the scatter plot between true power and ANN power prediction, along with an error density plot analysis are thoroughly explored. The proposed technique is intended to efficiently predict the received signal power and find applications in terrestrial communication, military operations, 5G beyond communication, underwater communication, and more for global internet connectivity.