ML Classification of BLE Packets Using Restrain Receiver QoS Metrics

The pervasiveness of technology in people’s lives is creating real difficulties for wireless protocols to cope with the constant increase in communications. Avoidance mechanisms exist, but they are no longer robust enough to limit collisions. A renewal of countermeasure techniques is needed to optimise spectrum management in a dynamic spectral context. In this paper, we propose an improvement of the radio receiver AGC computation based on machine learning (ML). We studied four ML algorithms for classifying BLE packets interfered by Wi-Fi: Support Vector Machine (SVM), K-Nearest Neighbour (KNN), bagged tree (BT), and random subspace using KNN. The analysis suggests that the receiver can use signal quality metrics (RSSI, SNR, LQI) to classify packets, to determined an optimized AGC index and to detect non-ideal reception conditions.