Deep learning based signal detection for low signal-to-noise ratio system

• المؤلف الرئيسي: Tang, Kirk Ji Wei
• مؤلفون آخرون: Teh Kah Chan
• التنسيق: Final Year Project
• اللغة: English
• منشور في: Nanyang Technological University 2022
• الموضوعات: Engineering::Electrical and electronic engineering::Wireless communication systems; Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• الوصول للمادة أونلاين: https://hdl.handle.net/10356/163390

For cognitive radios, it is quintessential for systems to accurately detect the presence of primary users’ (PU) signal in licensed spectrum, allowing secondary users (SU) to opportunistically utilize the idle spectrum. Traditional energy detection method is widely used due to its simplicity and effectiveness of blind signal detection, but suffers from the phenomenon of signal-to-noise ratio (SNR) wall due to noise uncertainty. To overcome this problem, we dive into deep learn- ing methods for signal detection, which learns patterns and trends from the signal’s modulation structure. Deep learning methods have shown significant improvements as compared to energy detection, while requiring no prior information about background noise and channel conditions of the system. Further investigation of the impact of modulation schemes on deep learning performance suggests that some modulation schemes (frequency-shift keying) have more distinct structures as compared to others, and is more suitable to be detected by deeper and complex deep neural networks (DNN). Our proposed ensemble model of ResNet 5 layers + Long Short- Term Memory (LSTM) achieved the best performance in detecting Gaussian Frequency Shift Keying (GSFK) signals. On the other hand, when detecting modulated signals with less distinct structures (phase-shift keying and amplitude modulation), or a mixture of signals with varied modulation schemes, a simple Convolutional Neural Network (CNN) works the best. Finally, impacts of sample length on detection performance are also investigated. Keywords: Spectrum Sensing, SNR-wall, Deep Learning