Neural Model for Far-Field 1D Localization of Mobile Stochastic EM Sources with Partially Correlated Radiation

Abstract

Abstract:This paper considers a possibility of angle position determination of mobile stochastic sources with partially correlated radiation, where antenna array and multilayer perceptron neural network processing are used. It is shown that the neural model trained with samples from a system with uncorrelated source radiation cannot determine position of sources with a satisfactory accuracy when sources have some degree of correlation in radiation. That is why it is suggested training samples to be generated for different values of partial source correlation. This kind of generated neural network training may provide source position determination with satisfactory accuracy even when there is partial correlation, which in the paper is presented with an example of two sources that linearly move in azimuth plane

Direction of arrival estimation of mobile stochastic electromagnetic sources with variable radiation powers using hierarchical neural model

“…In all these references, neural models were applied for the DoA estimation of deterministic signals radiated from mobile electromagnetic (EM) sources. Multilayer perceptron (MLP) neural networks were also used for the DoA estimation of deterministic and stochastic EM sources . Regarding the stochastic scenario, in References uncorrelated mobile stochastic sources were considered and it was shown that only the values of elements of the first row of spatial correlation matrix at the receiver were sufficient to develop an accurate neural model.…”

“…All models presented in References were developed with an assumption that stochastic sources are with equal and constant radiation power during their movement. The architecture of the neural model that includes the possibility to perform the DoA estimation of signals emitted from movable stochastic sources with variable radiation powers was given in Reference .…”

“…It has allowed to model the variation of radiation power of stochastic source through the changeable radiation diagram of antenna elements used to represent the stochastic source. However, during the development, testing and estimation of performances for this model as well as for above mentioned models from References , the impact of noise, that exists more or less in real acquisition conditions, has not been taken into account.…”

“…1,[10][11][12] Artificial neural networks (ANNs) [13][14][15][16] represent an another direction of research for solving the DoA problems in real-time. [16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31][32][33][34] Beside ANNs, the more general methods in the field of artificial intelligence, such as genetic programming (GP) and support-vector machines (SVM), are used in today research for the same purpose, as demonstrated for the DoA estimation problem solving in Reference 35 and Reference 36, respectively. Although the GP and SVM-based DoA models have more generalization capabilities than the ANN models in modeling highlynonlinear dependences present in DoA problems, DoA ANN models still represent a good alternative.…”

“…Multilayer perceptron (MLP) neural networks were also used for the DoA estimation of deterministic and stochastic EM sources. [24][25][26][27][28][29][30][31][32] Regarding the stochastic scenario, in References 26-29 uncorrelated mobile stochastic sources were considered and it was shown that only the values of elements of the first row of spatial correlation matrix at the receiver were sufficient to develop an accurate neural model. In References 30 and 31, neural models were developed for the DoA estimation of partially correlated stochastic sources using not only the elements of the first row of the spatial correlation matrix, but also the elements from the upper triangle of this matrix.…”