Linear Least-Squares Based Source Localization for New DRSS Model

Localization by the received signal Strength (RSS) measurement is inexpensive and has low computational complexity, thus extending the lifetime of the sensors in the wireless sensor network. The conventional propagation model for RSS has a log-normal distribution and since the probability density function is known, the best estimator for localization is Maximum Likelihood Estimator (MLE). The ML estimator is nonlinear and nonconvex and Gauss-Newton and convex optimization methods are presented in the papers. These methods impose a lot of complexity on the system and reduce the energy of the battery. In this paper, a two-step linear estimator is employed to solve the nonlinear ML estimator. In the first step, a new DRSS model is presented and nonlinear terms of ML cost function are replaced with linear variables. Also, in contrast to the estimators based on the conventional DRSS model, the performance of this estimator doesn't reduce by the random selection of the number 1 reference node. In the second step, the error of approximation of the first step is minimized, thus increasing the accuracy of the location estimation. Simulations show that in both the first and second steps, the accuracy is improved and the average error root error is reduced by up to 13% compared to the existing estimators.