Energy-Efficient Neuromorphic RF Localization Using Analog Spiking Neurons

As the Internet of Things (IoT) expands with more connected devices and complex communications, the demand for precise and energy-efficient localization has intensified. Although advanced artificial intelligence (AI) techniques provide high accuracy in radio-frequency (RF) localization, they often come at the cost of increased complexity and power consumption. To achieve energy-efficient AI-based RF localization, this paper presents an analog-based neuromorphic system. The system accurately identifies IoT transmitter positions within a 360-degree range on a two-dimensional plane, with a 10-degree angular resolution. At its core, the analog-based neural network incorporates the post-layout analog characteristics of spiking neurons into the learning process. The network is trained and tested using noisy datasets from MATLAB simulations and anechoic chamber measurements, both developed in this paper. The proposed system achieves high accuracies of 96.7% with simulated data and 93.9% with measured data at a 20 dB signal-to-noise ratio (SNR). Even in the worst-case scenario of 0 dB SNR, it achieves high accuracies of 90.7% with simulated data and 86.5% with measured data for angular estimation. The obtained results, with a power consumption of only 233 nW, highlight the system’s robust performance and its suitability for hardware implementation.