Predicting Antenna Radiation Patterns and Types From Voxlated Measurements Using Neuro-Memristive 3D Crossbars

Abstract

This paper proposes a non-invasive way to detect the antenna type from its radiation patterns to cross-validate its proper functioning. Here, the radiation pattern of three types of antennas namely: a) Dipole Antenna, b) Monopole Antenna, and c) Patch Antenna are used for the study. The feature formation from radiation patterns is performed using pixel sampling. Hardware implementation of a $128\times 128$ pixel array layout is performed using the SkyWater 130 PDK. The cross-validation of the antenna radiation pattern is performed using a 3D Memristive Convolutional Neural Network (3D-CNN). The simulations of the 3D-CNN are done based on Skywater 130 PDK, and the results are analysed. Here, due to the flexibility of concurrent reading and writing, the area, power and latency for the classification is getting reduced. The accuracy and robustness of AI/ML models are used for predicting the antenna type and are tested under various additive noise, such as a) Gaussian, b) White, c) Pink, d) Speckle and e) Salt and Pepper. The AI/ML models like a) Convolutional Neural Network (CNN) b) YOLOv8, c) VG-19 Net, d) Decision Tree, e) Naive Bayes, f) Random Forest and g) K-Nearest Neighbours (KNN) are used for the performance evaluation.