In this paper, a single neuron neural network beamformer is proposed. A perceptron model is designed to optimize the complex weights of a dipole array antenna to steer the beam to desired directions. The objective is to reduce the complexity by using a single neuron neural network and utilize it for adaptive beamforming in array antennas. The selection of nonlinear activation function plays the pivotal role in optimization depends on whether the weights are real or complex. We have appropriately proposed two types of activation functions for respective real and complex weight values. The optimized radiation patterns obtained from the single neuron neural network are compared with the respective optimized radiation patterns from the traditional Least Mean Square (LMS) method. Matlab is used to optimize the weights in neural network and LMS method as well as display the radiation patterns.
Keyword:Adaptive
INTRODUCTIONDue to its broad range of applications, adaptive array antenna is most popular in the present world. Present world applications require much faster beam steering that cannot be achieved using a mechanical systems. Hence it is required to use more consistent and much faster electronic beam steering techniques such as adaptive arrays. However, the beamforming withal most identical elements result lack of flexibility. On the other hand, adaptive beamforming methods by means of weight optimization are capable of managing the complexities of distinct elements. The adaptive array can detect, track and allocate narrow beams in the direction of the desired users while nulling unwanted sources of interferences. There are well known traditional techniques for adaptive beamforming in array antenna.Soft computing techniques namely Artificial Neural Networks (ANN), fuzzy logics, Genetic Algorithms (GAs), provide low cost solutions and robustness to different complex real world problems. ANN is a powerful information processing paradigm that tries to simulate the structure and functionalities of the biological nervous systems. The ANN is used to deal with many applications, and they have proved their effectiveness in several research areas such as image recognition, speech recognition, signal analysis, process control, and robotics. The true power of neural networks lies in their ability to represent both linear and nonlinear relationships. ANN, like human learning, learns by example. Training a neural network is, in most cases, an exercise in numerical optimization of a usually nonlinear function. Basic building block of every artificial neural network is an artificial neuron or perceptron that is a simple mathematical model.