Abstract:
Array antennas have a nonlinear, complex
relationship between the antenna beams generated and the
array input functions that generate the steerable beams. In this
paper we demonstrate the use of a simple, computationally less
intensive Perceptron Neural Network with non-linear sigmoid
activation function to do the synthesis of the desired antenna
beam. The single neuron is used, where its optimized weights
will yield the beam shape required. This paper presents a
successfully implemented Perceptron and discusses the error
between the desired and Perceptron generated beams The
successful beam control gives high accuracy in the maximum
radiation direction of the desired beam, as well as optimization
in the direction of null points. Moreover, a comparison between
the array antenna beams obtained using the Perceptron Single
Neuron Weight Optimization method (SNWOM) and the
optimized beams obtained using the Least Mean Square (LMS)
method, further demonstrates the reliability and accuracy of the
Perceptron based beamformer. The tests were performed for
two different desired antenna beams: one braod side beam and
the other with the antenna radiating in four different desired
directions. The Perceptron based antenna may be embedded in
the Arduino microcontroller used. It is also shown why it is not
possible to get a single beam, linear array antenna with the
Perceptron based array reported herein.
