Deep Multi-Emitter Spectrum Occupancy Mapping that is Robust to the Number of Sensors, Noise and Threshold

TL;DR

This paper introduces a neural network-based spectrum occupancy mapping system that is robust to variations in sensor count, noise, and thresholds, using aggregation of measurements into log-likelihood ratios.

Contribution

The novel approach aggregates sensor data into LLRs for neural network input, enabling robustness to sensor number, noise, and threshold variations without emitter count estimation.

Findings

Robust performance across different sensor counts and noise levels

Effective with low-resolution sensors

Operates without knowing the number of emitters

Abstract

One of the primary goals in spectrum occupancy mapping is to create a system that is robust to assumptions about the number of sensors, occupancy threshold (in dBm), sensor noise, number of emitters and the propagation environment. We show that such a system may be designed with neural networks using a process of aggregation to allow a variable number of sensors during training and testing. This process transforms the variable number of measurements into approximate log-likelihood ratios (LLRs), which are fed as a fixed-resolution image into a neural network. The use of LLR's provides robustness to the effects of noise and occupancy threshold. In other words, a system may be trained for a nominal number of sensors, threshold and noise levels, and still operate well at various other levels without retraining. Our system operates without knowledge of the number of emitters and does not explicitly attempt to estimate their number or power. Receiver operating curves with realistic propagation environments using topographic maps with commercial network design tools show how performance of the neural network varies with the environment. The use of very low-resolution sensors in this system can still yield good performance.