DetecSHUN: Detection of Signals Hidden Under the Noise

Precise detection of the presence of a signal in low signal-to-noise ratio (SNR) environments has significance in many wireless applications, such as primary user detection in a dynamic spectrum access (DSA) model, localization of signals, or operating in long range scenarios. Existing methods rely largely on traditional signal processing-based techniques for detection, and fail to perform reliably under high noise. To alleviate this problem, this paper develops a new convolutional neural network-based model to detect signals, specifically for high noise environments. Our proposed method outperforms existing deep-learning techniques by ~5 dB at accuracy levels above 90% for signals down to -18 dB SNR, utilizing network sizes within 10% of existing techniques and without any signal pre-processing. The model is generalized for varying sequence lengths and timing offsets, making it ready for adoption in realistic scenarios.