LSTM Guided Modulation Classification and Experimental Validation for Sub-Nyquist Rate Wideband Spectrum Sensing

Future generation wireless radios (WRs) are expected to dynamically tune their transmission parameters like transmitting frequency and modulation scheme to enhance the spectrum utilization and data rate. To perform this, the base station needs to identify the suitable spectrum resources and determine the modulation schemes of legacy users which are transmitting in the neighboring vacant spectral bands. Since the determination of modulation scheme indicates the presence of a signal, in this paper, we perform automatic modulation classification (AMC) based wideband spectrum sensing (WSS) to reduce the additional overhead caused by WSS. To avoid the need of power and area hungry Nyquist rate analog to digital converter (ADC), we exploit the sparse nature of the wideband spectrum to perform digitization at the sub-Nyquist rate. The proposed sub-Nyquist sampling based AMC (SNS-AMC) employs a long short-term memory neural network to classify various modulation schemes. The performance of the proposed SNSAMC is analyzed for different tapped delay line (TDL) channel models and is also validated on USRP based hardware testbed. It has been shown that the classification accuracy of the proposed SNS-AMC outperforms machine learning SNS-AMC and the accuracy of SNS-AMC approaches to the accuracy of Nyquist sampling based AMC with an increase in signal to noise ratio.