The performance prediction of an optical communications link over maritime environments has been extensively researched over the last two decades. The various atmospheric phenomena and turbulence effects have been thoroughly explored, and long-term measurements have allowed for the construction of simple empirical models. The aim of this work is to demonstrate the prediction accuracy of various machine learning (ML) algorithms for a free-space optical communication (FSO) link performance, with respect to real time, non-linear atmospheric conditions. A large data set of received signal strength indicators (RSSI) for a laser communications link has been collected and analyzed against seven local atmospheric parameters (i.e., wind speed, pressure, temperature, humidity, dew point, solar flux and air-sea temperature difference). The k-nearest-neighbors (KNN), tree-based methods-decision trees, random forest and gradient boosting- and artificial neural networks (ANN) have been employed and compared among each other using the root mean square error (RMSE) and the coefficient of determination (R²) of each model as the primary performance indices. The regression analysis revealed an excellent fit for all ML models, indicative of their ability to offer a significant improvement in FSO performance modeling as compared to traditional regression models. The best-performing R² model found to be the ANN approach (0.94867), while random forests achieved the most optimal RMSE result (7.37).

Free space optical communications (FSO), which make use of the visible and infrared spectrum for data transmission, offer significant advantages such as a very high data rate, security and immunity, low cost of installation and ease of use without any license restrictions. However, a significant challenge for FSO systems is their inherent constraints due to environmental conditions and especially atmospheric turbulence. This paper focuses on the experimental performance analysis of a real FSO system in a maritime environment. We propose a new model which allows an FSO link performance estimation over sea and depends upon point measurements of environmental parameters. The Received Signal Strength Indicator (RSSI) has been measured and a second-order polynomial has been constructed using regression modeling to quantify its relation with macroscopic environmental parameters collected by a weather station. This model has then been validated against real meteorological data over different period of times and exhibited a reasonably strong correlation. Atmospheric turbulence has been determined using bulk estimates of the structure index parameter extracted from the same meteorological data, and thus allowed for a statistical correlation between turbulence and RSSI. In the second part of the paper, the probability distribution of the RSSI data has been investigated and the Kullback-Leibler (KL) divergence has been used to investigate the difference between probability distributions over the same variable. As an illustrative example of the process, the Weibull, Lognormal and Gamma distributions have been evaluated against the RSSI data probability distribution and the latter has proved to exhibit the best fit.

A laser beam propagating in the free space suffers numerous degradation effects. In the context of free space optical communications (FSOCs), this results in reduced availability of the link. This study provides a comprehensive comparison between six machine learning (ML) regression algorithms for modeling the refractive index structure parameter ($C_n^2)$. A single neural network (ANN), a random forest (RF), a decision tree (DT), a gradient boosting regressor (GBR), a k-nearest neighbors (KNN) and a deep neural network (DNN) model are applied to estimate $C_n^2$ from experimentally measured macroscopic meteorological parameters obtained from several devices installed at the Naval Postgraduate School (NPS) campus over a period of 11 months. The data set was divided into four quarters and the performance of each algorithm in every quarter was determined based on the R² and the RMSE metric. The corresponding RMSE were 0.091 for ANN, 0.064 for RF, 0.075 for GBR, 0.073 for KNN, 0.083 for DT and 0.085 for DNN. The second part of the study investigated the influence of atmospheric turbulence in the availability of a notional FSOC link, by calculating the outage probability (P_out) assuming a gamma gamma (GG) modeled turbulent channel. A threshold value of 99% availability was assumed for the link to be functional. A DNN classification algorithm was then developed to model the link status (On-Off) based on the previously mentioned meteorological parameters.

Efforts to develop high-energy laser (HEL) weapons that are capable of being integrated and operated aboard naval platforms have gained an increased interest, partially due to the proliferation of various kinds of unmanned systems that pose a critical asymmetric threat to them, both operationally and financially. HEL weapons allow for an unconstrained depth of magazine and cost exchange ratio, both of which are essential characteristics to effectively oppose small unmanned systems, compared to their kinetic weapons counterparts. However, HEL performance is heavily affected by atmospheric conditions between the weapon and the target; therefore, the more precise and accurate the atmospheric characterization, the more accurate the performance estimation of the HEL weapon. To that end, the Directed Energy Group of the Naval Postgraduate School (NPS) is conducting experimental, theoretical and computational research on the effects of atmospheric conditions on HEL weapon efficacy. This paper proposes a new approach to the NPS laser performance code scheme, which leverages artificial neural networks (ANNs) for the prediction of optical turbulence strength. This improvement could allow for near real-time and location-independent HEL weapon performance estimation. Two experimental datasets, which were obtained from the NPS facilities, were utilized to perform regression modeling using an ANN, which achieved a decent fit (R² = 0.75 for the first dataset and R² = 0.78 for the second dataset).

The goal of this study was to characterize optical turbulence in the near-coastal environment. Measurements to obtain the refractive index structure parameter and other meteorological data were taken over the course of a month along the shore of Monterey Bay. The results were compared to a new version of the Navy Vertical Surface Layer Model (NAVSLaM), a model of turbulence originally developed for maritime environments but now extended to terrestrial environments. The new version has not been previously validated by comparisons to experiments, particularly in a complex environment such as near the coastline. Our experimental results showed generally good agreement between measured and modeled levels of turbulence. Specifically, the differences between experimental and modeled values of the refractive index structure parameter were less than an order of magnitude in most conditions and followed the same diurnal trend. There were some greater differences during near-neutral conditions, but this is a known limitation of the model. Overall, this extended model appears to do a good job of predicting turbulence in this environment for the observed time period.

The Hellenic Naval Academy (HNA) reports the latest results from a medium-range, near-maritime, free-space laser-communications-testing facility, between the lighthouse of Psitalia Island and the academy’s laboratory building. The FSO link is established within the premises of Piraeus port, with a path length of 2958 m and an average altitude of 35 m, mainly above water. Recently, the facility was upgraded through the addition of a BLS450 scintillometer, which is co-located with the MRV TS5000/155 FSO system and a WS-2000 weather station. This paper presents the preliminary optical turbulence measurements, collected from 24 to 31 of May 2022, alongside the macroscopic meteorological parameters. Four machine-learning algorithms (random forest (RF), gradient boosting regressor (GBR), single layer (ANN), and deep neural network (DNN)) were utilized for refractive-index-structural-parameter regression modeling. Additionally, another DNN was used to classify the strength level of the optical turbulence, as either strong or weak. The results showed very good prediction accuracy for all the models. Specifically, the ANN algorithm resulted in an R-squared of 0.896 and a mean square error (MSE) of 0.0834; the RF algorithm also gave a highly acceptable R-squared of 0.865 and a root mean square error (RMSE) of 0.241. The Gradient Boosting Regressor (GBR) resulted in an R-squared of 0.851 and a RMSE of 0.252 and, finally, the DNN algorithm resulted in an R-squared of 0.79 and a RMSE of 0.088. The DNN-turbulence-strength-classification model exhibited a very acceptable classification performance, given the highly variability of our target value ($C_n^2$), since we observed a predictive accuracy of 87% with the model.