Machine Learning for Isotropic Antenna Design

This research presents novel isotropic antenna designed by applying machine learning algorithm. Fitness proportionate selection algorithm resulted a design that has a total gain variation of 0.35 dB. This is the best design isotropy for an antenna with non-zero impedance so far reported in the literature. The design was optimized over parameters of isotropy, impedance, structure complexity and standing wave ratio (SWR). After compensating the imaginary part (jX Omega) of impedance, the resulting wire antenna showed an input resistance of 48.3 Omega at an operating frequency of 107 MHz. SWR is 1.03 reference to 50 Omega transmission line. For VSWR <= 2, the bandwidth is 1 MHz. Isotropic property of learned antenna resulted to be independent of frequencydimension scaling as well as independent from compensation of jX Omega. All the simulations were performed in Numerical Electromagnetics Code (NEC) for evaluations. The measured variation in total gain for the fabricated antenna was 1.37 dB with an input impedance of 61.7 Omega. The minimum SWR of 2.84 was observed at a slightly shifted frequency.