Performance analysis of UAV-FSO links using deep learning under turbulent conditions

This article describes a way to use deep learning that will help unmanned Aerial vehicles (UAVs) flying with free space optics (FSO) deal with air movement troubles. The proposed model suggests that using convolutional neural network (CNN) and attention to figure out the channel better can really help a drone do much better when it faces fast and changing round-trip times (RTTs), giving it much higher accuracy. Adding the new model to the adaptive channel model (ACM) helps people talk better and easier in any type of weather condition. Tests up to 20 km showed that connecting a UAV to a ground station using FSO links at 1550 nm did work well and gave good results. The study says that the new system reduces the error of finding channels by up to 25%, speeds up how fast it adapts by 40% compared to gated recurrent unit (GRU) and long short-term memory (LSTM) systems and lets the network use about 30% more information per second. Our system maintains a bit of error rate (BER) below \(\:{10}^{-4}\) even under strong turbulence \(\:({C}_{n}^{2}={10}^{-13})\) , significantly outperforming traditional methods and fixed modulation schemes. The proposed system has a maximum speed of 10.2 gigabits per second at a signal-to-noise ratio (SNR) of 35dB. This strong performance in diverse UAV flight modes and turbulence conditions renders our system suitable with the real-world application in the rapidly evolving atmospheric channel.