Performance evaluation of the particle swarm optimization algorithm to unambiguously estimate plasma parameters from incoherent scatter radar signals

Abstract

Abstract Simultaneously estimating plasma parameters of the ionosphere presents a problem for the incoherent scatter radar (ISR) technique at altitudes between ~ 130 and ~ 300 km. Different mixtures of ion concentrations and temperatures generate almost identical backscattered signals, hindering the discrimination between plasma parameters. This temperature–ion composition ambiguity problem is commonly solved either by using models of ionospheric parameters or by the addition of parameters determined from the plasma line of the radar. Some studies demonstrated that it is also possible to unambiguously estimate ISR signals with very low signal fluctuation using the most frequently used non-linear least squares (NLLS) fitting algorithm. In a previous study, the unambiguous estimation performance of the particle swarm optimization (PSO) algorithm was evaluated, outperforming the standard NLLS algorithm fitting signals with very small fluctuations. Nevertheless, this study considered a confined search range of plasma parameters assuming a priori knowledge of the behavior of the ion composition and signals with very large SNR obtained at the Arecibo Observatory, which are not commonly feasible at other ISR facilities worldwide. In the present study, we conduct Monte Carlo simulations of PSO fittings to evaluate the performance of this algorithm at different signal fluctuation levels. We also determine the effect of adding different combinations of parameters known from the plasma line, different search ranges, and internal configurations of PSO parameters. Results suggest that similar performances are obtained by PSO and NLLS algorithms, but PSO has much larger computational requirements. The PSO algorithm obtains much lower convergences when no a priori information is provided. The a priori knowledge of Ne and $${T}_{e}/{T}_{i}$$ T e / T i parameters shows better convergences and “correct” estimations. Also, results demonstrate that the addition of $${N}_{e}$$ N e and $${T}_{e}$$ T e parameters provides the most information to solve the ambiguity problem using both optimization algorithms.