Narrowband IoT Signal Identification in LTE Networks Using Convolutional Neural Networks

Abstract

Narrowband Internet of Things (NB-IoT) is an emerging standard serving massive wireless communications devices. It is implemented based on the legacy long-term evolution (LTE) technology and thus shares many system configurations with it. In fact, NB-IoT is deployed on some of the spectrum allocated to LTE. This sometimes introduces interference to the frequency bands with LTE transmission. Signal identification has been found as a critical and effective method for spectrum awareness and the improvement of resource allocation performance thus mitigating potential interference. While not been studied in previous work, this article fills the hole in identifying NB-IoT transmissions in different communications scenarios. In particular, we develop a convolutional neural network (CNN)-based signal identification method to distinguish NB-IoT signals from other cellular signals. The performance of the signal identification method is tested with different CNN training setups. Our experiments demonstrate that the proposed model can successfully identify the existence of NB-IoT signal with 98.13% accuracy in the Rayleigh fading channel where the signal-to-noise ratio (SNR) = 10 dB. Our results also show that the proposed signal identification method is able to provide promising identification accuracy under various and unknown SNR environments in both additive white Gaussian noise (AWGN) and Rayleigh fading channels.