Novel CNN and Hybrid CNN-LSTM Algorithms for UWB SNR Estimation

Ultra-wideband (UWB) technology is an excellent solution for short-range communication and localization systems in the next generation of cognitive radio (CR) paradigm. In such systems, received signal-to-noise-ratio (SNR), as an important channel state information (CSI) parameter, provides tremendous information about the communication environments. In this context, it is essential to have an accurate SNR estimation at the receiver end of the wireless networks. In this paper, two novel convolutional neural network (CNN) and hybrid CNN long-short-term-memory (LSTM) algorithms are proposed for UWB SNR estimation. The proposed models are capable of exploiting spatial and sequential information from the received signal where in contrast to traditional rule-based algorithms, suitable discrepancy feature for SNR estimation is extracted automatically. The performance of the proposed models is investigated against the IEEE 802.15.4a UWB standard model by evaluating mean absolute error (MAE), mean square error (MSE), and precision metrics.