Broadband Communication Signal Detection Using Deep Convolutional Autoencoder and Spectrum Center Network

Abstract

In the increasingly complex electromagnetic environment, wireless communication technology faces multiple challenges such as noise, interference, multi-path effects, and attenuation, which seriously threaten the reliability and efficiency of communication systems. Therefore, research was conducted on broadband communication signal detection based on deep convolutional autoencoders and spectral center networks. The experiment used a dataset of real-world signals and conducted comprehensive evaluations under various signal-to-noise ratios, signal-to-noise ratios, as well as different code lengths and colored noise backgrounds. The evaluation indicators included computational complexity indicators such as detection accuracy, F1 score, recall rate, specificity, and inference time. The experimental results showed that on the training set, the loss values of convolutional autoencoder networks 1 and 5 decreased rapidly, and their relatively stable loss function values were both around 10-5 . The loss value of Convolutional Autoencoder Network 10 decreased the slowest, with a minimum loss function value of about 10-6. Moreover, the method achieved detection accuracy, F1 score, and recall rate of 98.5%, 0.99, and 0.98, respectively, with an average inference time of 0.025 seconds. Compared with the existing stateof-the-art methods, it improved detection accuracy by 13.5% compared to energy detection and 5.3% compared to deep learning detection. The maximum improvement in F1 score and recall was 0.11 and 0.16, respectively. The research results indicated that the proposed method was significantly superior to existing methods in complex electromagnetic environments, with higher detection accuracy and robustness. This method offers insights for the design, technology, and solutions of future communication systems, which helps to promote the continuous development and progress of communication technology.