Autonomous mobile low-cost sensor node for high-resolution noise sampling in urban environments

This paper outlines algorithm-level optimizations that allow a low-cost Mobile Acoustic Sensor Node (MASN) to achieve fully autonomous, continuous noise monitoring using hybrid solar and vibration energy harvesting. The core of this work is the algorithm-level optimization of signal processing, data transmission, and positioning tasks, combined with the efficient integration of hardware low-power modes. These optimizations were quantitatively evaluated and compared with non-optimized references and sensors from existing literature. The signal processing enhancements, specifically the computation of acoustic indicators via Power Spectral Density, resulted in a 61.4% power reduction while maintaining theoretical compliance with IEC standards (61672-1:2013, 61260-1:2014). Furthermore, a comparative analysis with a Class 1 Sound Level Meter experimentally validated the accuracy of the results. For the positioning task, a server-side interpolation procedure reduced power consumption and data volume by enabling longer update periods. Compared to 1-second sampling, extending the positioning period to 16 s reduced power consumption by 81.7%, resulting in an average positioning error of about 4.7 m. Finally, for the transmission task, the combination of data coding and delta compression yielded a 76.5% reduction in the required data payload compared to an ASCII representation. The cumulative effect of these algorithmic optimizations resulted in a total power consumption of 23.5 mW, enabling the node to operate fully autonomously with an optimal data transmission periodicity of approximately three minutes. • Autonomous mobile noise sampling is enabled by efficient algorithms. • PSD-based signal processing reduces consumption by 61.4% while meeting IEC standards. • Longer positioning periods with interpolation reduces power consumption and payload. • Bit-packing and delta-coding cut transmission payload by 76.5%. • Sensor’s power consumption reduced to 23.5 mW by algorithms and low-power hardware.