Deep Learning-Based Signal-to-Noise Ratio Estimation for Underwater Optical Wireless Communication

A Deep Learning (DL)-based signal-to-noise ratio (SNR) estimation technique for Underwater Optical Wireless Communication (UOWC) systems is proposed in this work. By learning from the constellation diagrams, the proposed scheme can achieve accurate predictions of the SNRs of UOWC system. Experimental results show that high accuracies of 99.7%, 98% and 94.7% for 2-QAM, 4-QAM and 8-QAM can be obtained accordingly. Our study demonstrates the potential of DL for system performance estimation and monitoring of UOWC.

I. Introduction

The ocean contains rich resources, which have a significant impact on the economic and social development and thus there is a great significance for ocean exploration. With the advantages of high efficiency, high modulation bandwidth and high data rate, the underwater optical wireless communication (UOWC) substantially meets the demands of ocean exploration and real-time subsea monitoring [1], [2]. However, UOWC is severely affected by the turbulences, pressure, temperature, suspended particles and so on [3], [4]. Therefore, the channel measurements and modeling of the UOWC is of prime importance. A channel modeling and characterization study of UOWC was proposed for investigating the effects of shadowing and blockage, which is based on ray-tracing Zemax Tool [5]. Monte Carlo method was used for the verification of the calculation of the optical path loss in UOWC [6]. Deep learning (DL) is utilized for channel estimation, channel classification, and signal detection, which can identify the water types and improve the performance of UOWC [7]. However, most of the above works are based on the simulation, where there is a gap with practical applications. Moreover, various characteristics of UOWC need to be estimated for system performance improvement.