Direction of arrival estimation using artificial neural networks

Radio engineering


Efimov E. N.*, Shevgunov T. Y.**

Moscow Aviation Institute (National Research University), 4, Volokolamskoe shosse, Moscow, А-80, GSP-3, 125993, Russia



This paper introduces artificial neural network (ANN) based approach to the problem of direction of arrival estimation. The optimal solution to the given problem is based on estimation using maximization of the likelihood function (known as ML method), which depends on the estimated parameter, i.e. the direction of arrival, and the data calculated based on the received signals. This method allows one to obtain estimation with high accuracy, but it gives no closed form of the estimator owing. Consequently, it has to be calculated via numerical solution of optimization problem which requires a vast amount of computational power. The alternative approach based on multilayer perceptron ANN with a special neuron in its output layer is presented in this paper. This network learned with deterministic learning approach is able to estimate consuming significantly lesser computational power at cost of a small accuracy reduction.

The paper includes theoretical explanation of the proposed approach and the structure of ANN used to obtain the estimator. In order to investigate the efficiency of the proposed approach, a set of numerical simulations was performed using the known model of passive radar system with ring antenna arrays. It is shown that the time required by the ANN estimator to perform a single estimation is 10 times less than time required by ML estimator, while the accuracy reduction is kept lower than 10% within the signal-to-noise ratio from −8 dB to 18 dB. The results also indicate the absence of the significantly important dependency between the accuracy of the estimation and the true value for the ANN estimator.


artificial neural network, multilayer perceptron, signals and systems, model simulation, passive radar systems, Cramer-Rao lower bound


  1. Dubrovin A.V., Sosulin Yu.G. Radiotekhnika i elektronika, 2004, Vol. 49, no. 2, pp. 156-170.

  2. Dubrovin A.V., Sosulin Yu.G. Radiotekhnika i elektronika, 2007, Vol. 51, no. 4, pp. 441-457.

  3. Samarasinghe S. Neural Networks for Applied Sciences and Engineering: From Fundamentals to Complex Pattern Recognition, Auerbach Publications, 2006, p. 570.

  4. Haykin S. Neural Networks: A Comprehensive Foundation, 2nd Edition, Prentice-Hall, 1999.

  5. Setiono R., Feedforward Neural Network; Construction Algorithm; Maximum Likelihood Function; Quasi-newton Methods, Connection Science, № 7:2, 1995.

  6. Baum, E., Supervised Learning of Probability Distributions by Neural Networks, American Institute of Physics, 1988.

  7. Efimov E.N., Shevgunov T.Y. Zhurnal radioelektroniki, 2012, Vol. 8 :

  8. Shevgunov T.Y., Efimov E.N. Elektronnyi zhurnal «Trudy MAI», 2012, no. 51, available at: (accessed 26.03.2012)

  9. Battiti R. First- and Second-Order Methods for Learning: Between Steepest Descent and Newton’s Method, Neural Computation, 1992, Vol. 4, № 2, Pp. 141 — 166.

  10. Levenberg K.A method for the solution of certain problems in least squares, Q. Appl.Math, 1944, Vol. 2, pp. 164–168.

  11. Marquardt D.W. An algorithm for least-squares estimation of non-linear parameters, J. Soc. Ind. Appl. Math, 1963, Vol. 11, pp. 431–441.

  12. Dubrovin A.V. Radiotekhnika i elektronika, 2006, Vol. 51, no. 13, pp. 268-270.

Download — informational site MAI

Copyright © 2000-2020 by MAI