SWITCHED-BEAM LORA RSSI-BASED AOA ESTIMATION FOR IOT USING A BUTLER-MATRIX-FED ARRAY

Accurate and energy-efficient localization is an important requirement for long-range, low-power Internet of Things (IoT) applications. Conventional radio positioning systems often rely on phase-coherent I/Q sampling or computationally intensive Digital Signal Processing (DSP), which makes them difficult to use on resource-constrained gateways and end nodes. This thesis presents the design, implementation, and experimental evaluation of the building blocks of a LoRa-based localization system that combines Angle of Arrival (AoA) estimation and Round-Trip Time-of-Flight (RTToF) ranging. A compact 2 x 2 dipole array fed by a non-crossing modified Butler matrix and an SP4T switch provides four directional beams at 868 MHz, and the bearing of the transmitter is estimated from the Received Signal Strength Indicator (RSSI) measured sequentially on the four beam ports, using one LoRa packet per port. A low-power mobile node with an energy-harvesting power section was also designed. For AoA estimation, deterministic fingerprint-matching algorithms were compared with a Neural Network (NN) regression model on measurement sessions. In these tests the Neural Network achieved a Mean Absolute Error (MAE) of 16.78 deg, slightly lower than the 17.18 deg of the best fingerprint-matching configuration, and reduced the largest errors, although angular ambiguities remain in some regions; a model trained on a larger dataset reached 14.92 deg on a different session. The results indicate that the 1 dB RSSI quantization of the transceiver contributes to the ambiguities around the beam crossovers, and that a reference measured with the same hardware performs better than the laboratory radiation patterns. RTToF ranging was evaluated separately, on a different transceiver (Semtech LR2021), using frequency hopping: at short and medium distances the estimates showed a precision of about 1.2 m, but all campaigns showed a positive bias, which increased together with the dispersion at the longest distances. Since AoA and RTToF were validated on different hardware platforms, a complete 2D localization experiment combining angle and distance was not performed, and the AoA processing was executed offline. The integration of both functions on a single platform, the embedded implementation of the estimators, the LoRaWAN integration, and the experimental verification of the energy autonomy of the mobile node are left as future work. The results provide a proof of concept for low-complexity, RSSI-based direction finding in LoRa networks and a basis for these developments.