Received signal strength indicator prediction for mesh networks in a real urban environment using machine learning

Mesh networks are self-managing wireless systems with dynamic topology. These networks differ from broadcast and mobile networks because their mesh nodes can directly exchange information without the intervention of any other infrastructure. However, the radio propagation environment in urban regions, characterized by dense building clusters and human-made structures, influences signal attenuation and path loss. Therefore, deploying these networks brings distinct challenges from the more intensively studied indoor or rural scenarios. In line with this, predicting radio signal propagation attenuation is crucial for planning and deploying reliable networks. The literature on received signal strength indicator (RSSI) prediction for mesh networks in urban areas is scarce. This paper proposes machine learning-based RSSI prediction models for highly urbanized areas. We highlight the most influential features, including the distance between the transmitter and receiver, obstruction details in the first Fresnel zone, and terrain variability measures. Considering data from two mesh networks in the Metropolitan Region of São Paulo, Brazil, owned by a power utility company, we trained a Random Forest and a Support Vector Regression model for the RSSI prediction task. Comparative analysis indicates an improvement of up to 66% in the RSSI prediction error using the Random Forest approach in comparison with classical and empirical models.

Uncontrolled Keywords

feature importance; machine learning; mesh networks; network planning; RSSI prediction