Table 1. Fault Detection in Antenna Array with Soft Computing Techniques
Antenna arrays consist of more than one patch element and are used in various wireless applications. If a single patch element is defective then degraded radiation pattern is obtained which affect the performance of an antenna array. So it has become a critical issue to detect faulty patch in antenna array. In this paper, ANN and FFT based approach has been used to detect number and position of defective patch in designed arrays. Several optimization methods have been applied to detect partial and complete fault in linear and planar antenna arrays. These techniques include Firefly Algorithm (FA), Bat Algorithm, Bacteria Foraging Optimization (BFO) and Cuckoo Search Algorithm (CSA). A cost function has been developed for error between faulty power patterns and estimated one. Minimum value of this cost function will detect the location of defective patch in array. A virtual instrument based model had also been developed for fault analysis in a fractal array.
Antenna arrays are most crucial components in any wireless communication system. These arrays are key components in various applications including mobile, satellite and radar systems because of flexibility and control over radiation pattern (Balanis, 2005; Malloux, 1994). A single radiating patch is not sufficient for high gain and directivity. Therefore, in antenna array a number of patch elements are aggregated in a linear or planar manner to achieve high gain and directive pattern (Praveena & Ponnapalli, 2019). All radiating elements are arranged in such a way so that desired radiation characteristics can be obtained (Kumar et al., 2014). The geometrical arrangement of patch elements is capable of providing maximum radiation in desired direction and minimum in undesired direction (Barison et al., 2014; Pal et al., 2014). According to desired application, antenna arrays can be designed with circular, square and concentric circular shape radiating patch elements. All radiating patches are Different calibration systems (Leoni et al., 2018) can be used for detection of the faulty patch in active antennas but it leads to increase in weight, cost and size of active array systems. In the past, traditional methods were used to detect the fault in antenna array which were very time consuming. Therefore, new approaches were introduced for fault diagnosis in antenna arrays. In artificial neural networks (ANN), damaged radiation patterns were used as input and position of faulty elements as output. So neural networks were able to detect fault in array antenna by mapping the input and output variables. Due to flexibility of ANNs it is easier to compute and generate radiation pattern near to desired power pattern (Choudhury et al., 2009; Patnaik & Christodoulou, 2006; Patnaik et al., 2007; Vakula & Sarma, 2009). Discrete mean field neural nets (Castaldi et al., 2000) had been proven very effective to locate the faulty element position in large antenna array system. The Genetic algorithm (Rodriguez et al., 2000; Yeo & Lu, 1999) is a powerful technique for fault diagnosis in antenna arrays. To achieve desired pattern different machine learning methods have been described.
In this paper, an overview of various soft computing techniques being used for detection of fault present in linear or planar antenna array. These techniques comprise of BFO, BAT, FA and Cuckoo Search Algorithm along with optimization techniques. Fast Fourier Transform (FFT) and Artificial Neural Network (ANN) have also been described briefly. In addition to this, a Virtual instrument model has been also designed for fault finding in antenna arrays.
An array consists of various radiating elements. If one element is damaged then distorted radiation pattern is obtained which affects the performance of antenna array. Therefore to detect faulty patch is an important issue. Neural network and various optimization techniques have also been used for fault diagnosis of antenna array. This paper also proposes a virtual instrument model to identify fault in fractal antenna array.
A planar antenna array had been designed with patch elements which were arranged in rows and columns. In designed array one, two and three fault elements had been detected with Probabilistic Neural Network (PNN) and Radial Basis Function (RBF). Radiation patterns associated with amplitude and phase deviation determines the location of faulty elements. If there is one faulty patch in antenna array then location has been found from deviation of phase but for two or three faulty elements the distance between elements had been determined from amplitude deviation. For given distance between elements the location of faulty elements had been found from phase deviation. In RBF learning network, transfer of information from input to output layer is done by radial basis function associated with non-linear characteristics and that from hidden to output layer had been done with linear weights. In this work, training has been done with RBF network having fixed centres, self-organized centres and PNN networks. In a designed planar array 32 samples of pattern from -900 to 900 had been taken. There were 25 possible faulty patterns for one element fault, 300 for two element fault and 2300 for three elements fault had been considered. For training of network with RBF and PNN, 32 input patterns had been applied at input layer. Neural network had been trained with a spread constant of 3 and 1.5 for RBF and PNN respectively. After training it had been observed that efficiency with PNN is better in comparison to RBF network for detecting faulty element in antenna array (Vakula & Sarma, 2010).
After a lot of research, a novel optimization algorithm such as Bacteria Foraging Optimization (BFO) has become popular to solve large and complex problems. In this work BFO algorithm has been used to find complete and partial fault present in a linear antenna array which consists of isotropic elements separated by a distance of half wavelength. To detect complete fault, amplitude excitation is taken as 0 and for partial fault, it is taken as half of the original amplitude excitation in array factor. Now array factor of defective antenna array has been compared with desired array factor of original antenna array and a cost function has been developed for BFO application. Few specified parameters for algorithm are number of bacteria, swimming length, number of iterations in chemotactic loop, number of reproduction, and probability of elimination and dispersal. The algorithm runs with all these parameters and minimization of cost function provides the best value of amplitude excitation of defective antenna array. This amplitude excitation is compared with original amplitude distribution. The deviation provides the position and level of fault (Acharya et al., 2011).
Khan et al. (2013) has discussed Firefly algorithm (FA) for detection of complete and partial fault in linear antenna array. It is a global optimized algorithm that works on flashing light actions of fireflies. FA algorithm is based on,
For fault diagnosis of antenna array, a fitness evaluation function has been developed that is based on error between damaged power patterns and estimated one. FA algorithm has been implemented by setting the values of parameters such as light absorption coefficient, attractiveness and randomization parameter. When number of iterations reaches the minimum value of fitness function with respect to weights of element, it gives the information about faulty element position within antenna array. It has been proved that a successful algorithm is needed for finding failure elements in a linear antenna array.
Element failure in an antenna array leads to enhancement in side lobe level along with ripples. In power pattern, it becomes necessary to detect the faulty element in antenna array. In the work by Muralidharan et al. (2014), array factor of defective antenna array has been determined by making amplitude excitation equal to zero for complete fault and half of amplitude excitation for partial fault. Then Fast Fourier Transform has been applied on this defective array factor to get current amplitude excitation directly. These defective amplitude excitations had been compared with amplitude excitation of original array to find element failure position in array antennas consisting of isotropic elements with equal space. In this paper, (Muralidharan et al., 2014) partial as well as complete fault has been discovered in less computational time. Therefore this technique can also be used for real time applications.
In the work by Grewal et al. (2015), fault diagnosis of symmetric antenna array has been done with bat algorithm (BA). This algorithm is associated with three interesting parameters such as pulse rate, frequency and loudness. For discovery of damaged element position for antenna array, degraded side lobe pattern has been observed and compared with estimated side lobe pattern. A cost function has been developed from the error between degraded and estimated side lobe pattern. BA has been applied with all parameters and it provides the excitation coefficients each time and computes the lowest value of objective function and detects the position of damaged patch in symmetric linear array. So, fault diagnosis with BA is an effective approach for antenna arrays.
Fault diagnosis with a cuckoo search algorithm (CSA) is a novel optimization technique to detect the position of faulty element in a Classical Dolph Chebyshev (CDC) array with equal space between elements. In this work, a power pattern for antenna array has been measured and it has been compared with power pattern for faulty antenna array. To detect complete fault in CDC array, excitation is made equal to zero and for partial fault it has been considered half (50%) of original excitation value. A cost function has been developed by comparing a desired pattern with faulty element array pattern. Minimization of fitness function computes the weights which generate power pattern close to desired pattern. This algorithm uses less parameter as compared to other algorithms. For detection of faulty element with CSA algorithm, population size, rate of discovery and step size, parameters are employed. The algorithm runs with all these parameters, minimizes the cost function and effectively discovers the location of damaged patch within antenna array (Khan et al., 2018).
Defective patch detection in array systems is a significant issue for researcher in antenna engineering. A virtual instrument based model has been designed to discover defective patch in a fractal antenna array consisting of circular shaped elements. Radiation pattern has been taken as input for designed model then a curve is fitted over original radiation pattern curve after curve fitting coefficients are generated. From these input coefficients and outputs for faulty elements, a neural network based model has been developed in MATLAB and trained with Levenberg Marquardt (LM) algorithm. Now neural code has been embedded in MATLAB script node of the designed virtual instrument model. Neural outputs are visible in response window and further compared with a predefined value. If output value is greater than preset value, comparator provides high output and corresponding LED will glow to indicate fault, otherwise LED does not glow. Therefore, it is a successful diagnostic system to detect fault present in antenna array (Rani & Sivia, 2020).
This paper reviewed a few papers on works already carried out in the domain of fault detection in array of antennas using neural networks.
Detection of defective patch in an array system from power pattern is a crucial task. Neural networks, FFT, ANN implementation with virtual instrument model and optimization algorithms have been successfully employed to solve this problem. All techniques have been summarized in Table 1.
Table 1. Fault Detection in Antenna Array with Soft Computing Techniques
This paper will pave way for future research in improving the the fault detection in antenna arrays using different elements of optimizing algorithms.
The element failure in a large antenna array system is a serious issue. If any one element is faulty then the side lobe level is increased and degraded for field pattern obtained. This problem has been solved effectively with neural networks and optimization algorithms. ANN implementation with virtual instrument has been also proved an effective diagnostic system in fractal antenna array. All these methods have been successfully solved in the problem of detecting the completely and partially defective patch within antenna array.