CNN-Based Spectrum Sensing with Asymmetric Weighting in LPD Communication System
Jae-Hyeon Lee, Soyeon Jeon, Eui-Rim Jeong · 2025
In this paper, we propose CNN (Convolutional Neural Network)-based spectrum sensing method with asymmetric weighting to enhance false alarm performance in LPD (Low Probability of Detection) communication systems. The proposed method utilizes WBCE (Weighted Binary Cross Entropy) as the loss function, in which different weights are assigned to the signal-present and idle classes to reflect their importance better. This approach extends the conventional BCE (Binary Cross Entropy) by emphasizing the idle class, where no signal is present. Through simulation, the performance of the proposed method was evaluated with an FFT (Fast Fourier Transform) size of 256 and an observation length of 128. The results show that the average FAR (False Alarm Rate) of CNN trained with standard BCE was approximately 1.5% across all SNR (Signal to Noise Ratio) levels. In contrast, when applying WBCE with a weighting factor of 3 to the idle class, the FAR significantly decreased to approximately 0.05%, demonstrating the effectiveness of the proposed method.