Classification of BLE Packet According to AGC Index with Detection of Non-Ideal Reception Cases

The unlicensed 2.4 GHz band is popular among the personal wireless applications. Even if protocols using this band are implemented with countermeasures, they are no longer robust enough to maintain good quality of service (QoS). Interference between low data rate short-range communication protocols, such as Bluetooth Low Energy (BLE), and high data rate protocols, such as Wi-Fi, is symptomatic of the QoS loss in this band. We present a new way to allow the narrowband protocols receiver to adapt its internal states to the signal power by improving the adaptive gain control (AGC). As a preliminary work, we compare four machine learning algorithms (Decision Tree, Bagged Tree, Support Vector Machine (SVM) and K-Nearest Neighbour (KNN)) to classify BLE received packets subject to an interferer. The receiver robustness can then be improved by combining an optimised AGC index decision with the detection of non-ideal reception conditions. The best model in terms of its performance/complexity trade-off here is the KNN, which classifies the packets with an accuracy of 95.6 %.