Spectrum sensing based on deep learning with fading channels

Abstract

With the development of fifth generation (5G) and sixth generation (6G), more devices will be accessed in the channel, spectrum resources will be continuously occupied, and there is currently almost no unallocated spectrum. However, spectrum is a limited resource, and due to technical limitations, it is difficult to expand spectrum, leading to the current spectrum shortage. Nevertheless, the utilization rate of existing spectrum is not high and there are often gaps in the spectrum.

To achieve high spectrum utilization, the capacity to reliably detect the presence of primary users (PU) in the spectrum, or to detect the presence of holes in the spectrum, is critical for secondary users (SU) to make efficient use of spectrum resources. To achieve this, typically, we employ spectrum sensing, which is one of cognitive radio techniques. However, the traditional spectrum sensing approaches, such as energy detection and cyclic smooth feature detection, are ineffective and computationally complex.

With the development of artificial intelligence (AI), deep learning is gradually used in communication technology, this dissertation uses MATLAB and Python platform, based on previous research, proposed a ResNet and BiLSTM combination of deep learning model. The model does not need to pre-extract the depth characteristics of the signal. The signal pre-processing is relatively simple but, in the presence of fading channels, still achieves a better detection effect. In this dissertation, we compare its effect with some classical deep learning networks. The experiments show that our network is more effective.