Deep Learning for channel estimation in FSO communication system

Perfect channel estimation is a complex task with high power consumption and cost; in addition, requiring pilot transmission reduces the data rate. So, it is not favourable especially in mobile communication systems. The aim of this paper is to design (a new, low cost and low complexity) deep learning based channel estimator for free space optical (FSO) communication. In order to have a better understanding, this paper goes deeper through the problem, and presents different new deep learning based FSO systems, in which deep learning is used as detector, joint constellation shaper and detector, channel estimator, joint channel estimator and detector, joint constellation shaper and channel estimator and detector. For comparison with conventional systems, the outstanding QAM modulation, perfect channel estimation and maximum likelihood detection is applied. Considering wide range of atmospheric turbulences, from weak to strong by Gamma–Gamma model, symbol error rate performance of the proposed structures is investigated. Results indicate that the proposed deep learning based channel estimation technique, despite its less complexity, cost and power consumption provides close enough performance to the perfect channel estimation. It should be noted that the proposed structure does not need pilot sequence, hence, it has higher data rate than perfect channel estimation. The performance of the proposed deep learning based structures does not change with atmospheric turbulence variation. Furthermore, they are low cost, low complexity, with favourable performance. Accordingly, they could be good choices especially for mobile communication systems. Because the transceiver of these systems is a small mobile phone that should have low cost, complexity, and power consuming.