Autoencoder-driven MIMO architecture for underwater optical wireless communications systems

Abstract A deep autoencoder-based multiple-input multiple-output (AE-MIMO) architecture is proposed for underwater wireless optical communication (UWOC) systems operating in low-noise, turbulence-affected channels. A low-noise regime is assumed (normalized noise variance <m:math xmlns:m="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <m:msubsup> <m:mrow> <m:mi>σ</m:mi> </m:mrow> <m:mrow> <m:mi>n</m:mi> </m:mrow> <m:mrow> <m:mn>2</m:mn> </m:mrow> </m:msubsup> <m:mo>∈</m:mo> <m:mrow> <m:mo stretchy="false">[</m:mo> <m:mrow> <m:mn>1</m:mn> <m:msup> <m:mrow> <m:mn>0</m:mn> </m:mrow> <m:mrow> <m:mo>−</m:mo> <m:mn>4</m:mn> </m:mrow> </m:msup> <m:mo>,</m:mo> <m:mn>1</m:mn> <m:msup> <m:mrow> <m:mn>0</m:mn> </m:mrow> <m:mrow> <m:mo>−</m:mo> <m:mn>2</m:mn> </m:mrow> </m:msup> </m:mrow> <m:mo stretchy="false">]</m:mo> </m:mrow> </m:math> ${\sigma }_{n}^{2}\in [1{0}^{-4},1{0}^{-2}]$ ). Unlike conventional schemes, the model jointly learns transmitter and receiver mappings via a differentiable neural pipeline informed by Beer–Lambert attenuation, log-normal turbulence, and additive Gaussian noise. Evaluations across SISO, SIMO, MISO, and MIMO configurations demonstrate that AE-aided schemes consistently outperform conventional baselines by achieving 3–4 dB lower signal-to-noise ratio (SNR) requirements at symbol error rate (SER) <m:math xmlns:m="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <m:mo>≤</m:mo> <m:mn>1</m:mn> <m:msup> <m:mrow> <m:mn>0</m:mn> </m:mrow> <m:mrow> <m:mo>−</m:mo> <m:mn>3</m:mn> </m:mrow> </m:msup> </m:math> ${\le} 1{0}^{-3}$ . In particular, the proposed 3 × 2 AE-MIMO achieves a 22 % increase in effective 3 dB bandwidth, a fifty-three-fold reduction in SER at moderate SNR, 56 % faster convergence, and requires 7 dB less SNR to maintain SER <m:math xmlns:m="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <m:mo>≤</m:mo> <m:mn>1</m:mn> <m:msup> <m:mrow> <m:mn>0</m:mn> </m:mrow> <m:mrow> <m:mo>−</m:mo> <m:mn>3</m:mn> </m:mrow> </m:msup> </m:math> ${\le} 1{0}^{-3}$ in turbid water. These gains align with diversity gain and capacity scaling predictions, confirming the model’s physical validity. The framework exhibits strong robustness against attenuation, scattering, and fading, making it suitable for high-throughput, low-latency underwater applications such as autonomous vehicle coordination, IoT sensor networks, and real-time inspection.