REVIEW OF DEEP LEARNING MODEL FOR SPECTRUM SENSING IN UHF BANDS FOR 5G-TVWS NETWORK
Conventional methods of spectrum sensing, especially energy detection, though simple to design, have certain limitations including high sensitivity to noise uncertainty, which affect the efficacy of spectrum sensing in practical applications. Renowned deep learning network models, including Convolutional Neural Network (CNN), Support Vector Machine (SVM), Long Short-Term Memory (LSTM), Short-Time Fourier Transform (STFT), and Artificial Neural Networks (ANN) have been utilized for extracting different characteristics of signals for spectrum-sensing tasks. Although these models have demonstrated superior performances compared to conventional spectrum-sensing methods, many of the deep learning approaches lack complete awareness of the underlying source signal structure. Most of the conventional and deep learning-based spectrum sensing algorithms seem to have high perceptual performance under high-SNR environments, but their performances often fall short of expectations under low-SNR environments. Due to difficulties in obtaining large training datasets from the field, several deep learning methods depend on computer-generated datasets to train the models, which cannot reflect the practical application scenarios. This paper reviews a deep learning spectrum sensing model for spectrum sensing in UHF bands for 5G-TVWS network using a hybrid CNN-STFT-LSTM network to obtain local information for a single node spectrum sensing. While the CNN effectively extracts the spatial features from the frequency domain representations produced by STFT, the LSTM captures temporal dependencies (patterns and anomalies) in the time series data. The STFT plays a crucial role of transforming time-domain signals into time-frequency representation (spectrogram), which serves as an input dataset to CNN and LSTM, enabling the extraction of spectral features that are essential for detecting and classifying signals. Hybridizing CNN, STFT and LSTM networks can enhance the model's ability to learn complex, non-linear relationships in data, which is particularly useful in the challenging environments of signal detection. This multi-feature combination can lead to improved accuracy in identifying the presence of signals in noisy environment and ultimately enhancing overall performance in spectrum sensing tasks. The proposed model can be integrated into 5G-TVWS network for delivery of broadband internet services to rural and underserved areas. The 5G-TVWS network is suitable for rural deployment of broadband internet services due to its long-distance transmission capabilities, obstacle penetration, large coverage area, and relatively little infrastructure requirement.
Authors : Pallam, S.W., Thuku, I.T., Luka, M.K. and Ibrahim, V.M.
Category : Open Access Volume (Issue) : 11(1) Date Uploaded : 7th May 2025