This paper demonstrates a Convolutional Neural Network (CNN) based mechanism for identification of low power wide area network (LPWAN) technologies such as LoRA, Sigfox, and IEEE 802.15.4g. Since the technologies operate in unlicensed bands and can interfere with each other, it becomes essential to identify technologies (or interference in general) so that the impact of interference can be minimized by better managing the spectrum. Contrary to the traditional rule-based identification mechanisms, we use Convolutional Neural Networks (CNNs) for identification, which do not require any domain expertise. We demonstrate two types of CNN based classifiers: (i) CNN based on raw IQ samples, and (ii) CNN based on Fast Fourier Transform (FFT), which give classification accuracies close to 95% and 98%, respectively. In addition, an online video is created for demonstrating the process .
Shahid Adnan, Fontaine Jaron, Haxhibeqiri Jetmir, Saelens Martijn, Khan Zaheer, Moerman Ingrid, De Poorter Eli
Place of publication:
IEEE INFOCOM 2019 – IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), 29 April-2 May 2019, Paris, France
A. Shahid et al., “Demo Abstract: Identification of LPWAN Technologies using Convolutional Neural Networks,” IEEE INFOCOM 2019 – IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), Paris, France, 2019, pp. 991-992, https://doi.org/10.1109/INFCOMW.2019.8845259
Read the publication here: