Identification of Cellular Measurements: A Neural Network Approach

The efficient utilization of the wireless spectrum is essential to fulfill the rising demand of scarce bandwidth resources. Identifying the cellular signal types occupying the spectrum allows for usage optimization. Neural networks (NNs) are a promising approach for the signals' identification problems. This article proposes a hybrid convolutional and feedforward NN (HCFNN) that classifies the cellular signals from the power spectral density (PSD) of real measurements into their corresponding types: global system for mobile communications (GSM), universal mobile telecommunications service (UMTS), and long-term evolution (LTE). The measured dataset is collected based on two acquisition modes: multiple-band (MB) and in-band (IB) PSD acquisition modes. In the MB model, the data are collected, trained, and tested from various frequency bands, while in the IB model, the data are collected, trained, and tested from a single frequency band. The accuracy and the precision-recall (PR) metrics are used to evaluate the performance of the proposed HCFNN model. Moreover, the complexity analysis of the model is derived in terms of the number of real additions, real multiplications, and parameters. The extensive assessments of the over-the-air measurements show that the proposed HCFNN model accurately identifies the cellular signal types in all studied scenarios.