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

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 sensor measurements into log-likelihood ratios (LLRs), which are fed as a fixed-resolution image into a neural network. The use of LLRs provides robustness to the effects of noise and occupancy threshold, allowing a system to be trained for a nominal number of sensors, threshold and noise level, and still operate well at 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. It is shown that even a sparse sensor distribution can render a high-resolution occupancy map. 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. Finally, we show that the use of noisy sensors with low dynamic range in this system can still yield good performance.