Author(s)
Blessing Andrew Okoro, Petko Bogdanov, and Mariya Zheleva
Abstract
We present a novel framework for joint transmitter characterization across time, frequency, and space, from multi-vantage point traces. Traditional spectrum analytics focus on time-frequency characterization or localization, and thus have limited applicability in complex real-world scenarios where both time-frequency and spatial awareness are critical for spectrum sharing and enforcement. Our approach processes multi-sensor spectrum measurements as a three-way tensor, where each tensor slice represents Power Spectral Density (PSD) data from individual sensors, and the complete tensor captures spatial diversity across the sensor network. Through a dictionary-based tensor decomposition framework, we simultaneously extract time-frequency occupancy patterns, and proxy Received Signal Strength Indicator (RSSI) for each sensor measurement enabling joint detection and localization in a unified optimization. We evaluate our framework on synthetic dataset and preliminary results demonstrate the superior performance of our framework compared to existing baselines in multi-transmitter detection and characterization tasks. Significantly, our data-driven dictionary-estimated proxy RSSI consistently achieves higher localization accuracy than sensor-measured RSSI, indicating that the structured signal representation learned through tensor decomposition provides more reliable power estimates for spatial inference.