Weak Signal Detection Methods under Strong Background Noise Based on Physics-Informed Neural Networks

Weak signal detection is a long-standing research topic in signal processing, playing an essential role in many fields such as communications, radar, and biomedicine. Unfortunately, conventional identification methods have long been plagued by problems such as insufficient identification accuracy and poor adaptability. Especially when facing a poor signal-to-interference level environment, these limitations are more prominent. To overcome this challenge, this paper proposes a new approach, namely a faint signal identification method founded on physical data artificial brain systems (PINNs). The core of this method is to intelligently incorporate the physical characteristics of the signal into the learning process of the artificial brain and restrict the network output to align with the physical laws of the signal, thereby considerably enhancing the identification performance and achieving good interpretability. During the evaluation and experimental tests, it was found that under weak signal-to-interference ratio circumstances, the identification precision of the PINN method surpasses that of conventional energy identification (such as: based on convolution-based neural network (CNN)). What is more noteworthy is that the PINN method also demonstrates excellent generalization capabilities and robust adaptability to signal and channel variations. This paper offers a fresh viewpoint for resolving the traditional challenge of faint signal identification, and also creates broad opportunities for the implementation of physical information learning in the domain of signal processing.