Multi-Signal Detection Framework: A Deep Learning Based Carrier Frequency and Bandwidth Estimation.

Area of Science:

• Signal Processing
• Machine Learning
• Wireless Communications

Background:

• Multi-signal detection is crucial for applications like cognitive radio (CR), spectrum monitoring, and signal reconnaissance.
• Accurate detection and estimation of multiple signals' carrier frequencies and bandwidths are essential in these fields.

Purpose of the Study:

• To propose a deep learning-based framework, SigdetNet, for effective multi-signal detection.
• To enhance signal detection by localizing spectral locations using the power spectrum as input.

Main Methods:

• Welch's periodogram was used to reduce power spectral density (PSD) variance, followed by logarithmic transformation for signal enhancement.
• An encoder-decoder network with an embedding pyramid pooling module was developed to extract multi-scale features for signal detection.
• The impact of frequency resolution, network architecture, and loss function on detection performance was analyzed.

Main Results:

• SigdetNet demonstrated superior performance in multi-signal detection compared to benchmark schemes.
• The framework effectively localizes spectral signal locations using power spectrum inputs.
• Feature extraction using multi-scale analysis proved beneficial for detection accuracy.

Conclusions:

• The proposed SigdetNet framework offers a robust and high-performance solution for multi-signal detection.
• Deep learning approaches, particularly with encoder-decoder architectures and multi-scale feature extraction, show significant promise in signal processing applications.
• Further investigation into network parameters can optimize detection performance for various scenarios.