Abstract
In this paper, we propose efficient importance sampling estimators to evaluate the outage probability of maximum ratio combining receivers over turbulence-induced fadings in underwater wireless optical channels. We consider two fading models: exponential-lognormal, and exponential-generalized Gamma. The cross-entropy optimization method is used to determine the optimal biased distribution. We show by simulations that the number of samples required by importance sampling estimator is much less compared to naive Monte Carlo for the same accuracy requirement.
Introduction
Underwater Wireless Optical Communication (UWOC) is a high-speed alternative to traditional acoustic links. However, its performance is significantly degraded by underwater turbulence, which causes signal fading. To ensure system reliability, engineers must calculate the outage probability—the probability that the signal-to-noise ratio (SNR) drops below a specific threshold ($\gamma_{th}$). In high-reliability systems, these outages are "rare events" (e.g., $10^{-9}$ or lower).
Standard "naive" Monte Carlo (MC) simulations are inefficient for estimating such low probabilities, as they require a number of samples inversely proportional to the event probability. This paper introduces an efficient simulation framework using Importance Sampling (IS) and the Cross-Entropy (CE) method to address this challenge, specifically for bimodal fading models that accurately represent underwater environments.
Methodology
The core of the proposed approach is an Importance Sampling (IS) estimator. IS shifts the sampling process from the original fading distribution to a biased distribution where outages occur more frequently. The resulting estimates are then weighted to ensure unbiasedness relative to the original distribution.
To optimize the biased distribution, the authors utilize the Cross-Entropy (CE) method. This iterative algorithm minimizes the Kullback-Leibler divergence between the optimal (but unknown) zero-variance distribution and a chosen parametric family. Specifically, an exponential distribution is used for biasing because its "heavy left tail" is well-suited for capturing the low-SNR events that trigger outages in bimodal models like Exponential-Lognormal (EXP-LN) and Exponential-Generalized Gamma (EXP-GG).
Results and Discussion
The proposed IS scheme was evaluated against naive MC simulations across various scenarios. A key finding is the "Deep Reach" of the IS method. While naive MC simulations become computationally infeasible for outage probabilities below $10^{-6}$ or $10^{-7}$, the proposed method maintains accuracy down to $10^{-11}$.
The efficiency gain is most evident when comparing the required number of simulation runs. For a system with $L=4$ receiver branches, achieving a 5% relative error with naive MC would require approximately $10^{15}$ samples. In contrast, the proposed IS method achieves the same precision with only $10^4$ samples—a billion-fold reduction in computational effort.
The results also suggest that the proposed estimator possesses "bounded relative error," meaning the required number of samples does not explode as the outage probability approaches zero.
Conclusion
This research demonstrates that combining Importance Sampling with the Cross-Entropy method allows for the efficient and accurate estimation of rare outage events in underwater optical communication systems. By leveraging bimodal fading models and automated biasing, the framework provides a powerful tool for designing robust UWOC links. Future work will focus on extending this methodology to non-linear combining schemes and experimental validation in hardware-in-the-loop settings.
Summary updated