Diagnosis of Array Failure in Impulsive Noise Environment Using Unsupervised Support Vector Regression Method

Fast and accurate diagnosis of array failure is important for the maintenance of array antennas. Locations of failing elements are usually detected using near field data, which may be polluted by noises. Most existing diagnosis methods assume non-impulsive noise in the measurement data. However, the practical measurement environment may contain impulsive noises, which has not been considered in existing methods. This work proposes to impose a penalty function in the residual between the measured and recovered near field. The impulsive noise in the near field data can then be suppressed by using an appropriate function as the penalty function. Furthermore, minimum ℓ_p-norm is imposed on the excitation vector. The condition imposed on the noise and the minimum ℓ_p-norm constraint on the excitation vector constitutes an optimization problem, which is solved using the unsupervised support vector regression. The proposed method is more accurate than existing methods when impulsive noise presents in the near field data, and it is able to deal with a wide range of number of failing elements by adjusting the value of p. Numerical results are presented to show the advantages of the proposed method and to study the choice of p.