Remote battery state estimation for IoT devices via radio fingerprinting

Sammanfattning

Low-power wireless IoT devices powered by small form-factor batteries or capacitors often lack mechanisms to monitor or report their own State-of-Charge (SoC). This thesis investigates whether RF features extracted from Bluetooth Low Energy (BLE) packets can be used for remote estimation of battery State-of-Charge (SoC). The hypothesis is that transmitter hardware components are sensitive to supply voltage variations, which can introduce measurable changes in the observed RF features. Prior work in RF fingerprinting has shown that hardware imperfections produce traceable features in wireless signals that can be captured using resource-constrained embedded platforms. Experiments were conducted under controlled supply voltage levels and real battery discharge scenarios using nRF52-series boards. Carrier Frequency Offset (CFO) and intra-packet frequency-based features were extracted from received BLE packets and evaluated using Random Forest machine learning models. The results show that CFO exhibits measurable voltage-dependent behavior across all experimental phases. Controlled voltage experiments achieved 44% classification accuracy across ten voltage levels, while leave-one-out cross-validation on battery discharge data from the same physical battery achieved SoC prediction errors of approximately 13-21 percentage points. The results demonstrate the feasibility of passive RF-based battery-state estimation as a proof-of-concept.