Where to Improve? A Systematic Evaluation of Sampling Parameters for Pressure Sensor Matrix

The pressure sensor matrix shows strong potential in ubiquitous computing, such as human-computer interaction and human activity recognition. However, a considerable gap remains between the diverse and often casually configured laboratory prototypes currently in use and practical systems designed for cost-effectiveness. A critical factor underlying this gap is the lack of theoretical guidance for selecting system sampling parameters, namely, sampling frequency, spatial resolution, and bit depth. In the absence of such guidance, parameter selection tends to be arbitrary. To address this issue, we systematically select representative datasets in the field of pressure sensing, and propose a general multi-level data quality assessment framework based on downsampling. Experiments on 6 datasets across 5 device types show that downstream task performance can be effectively modeled using a logistic growth function, and reveal the following insights: (1) Spatial resolution is the primary bottleneck in most systems, with an ideal value exceeding 200 sensors/m. This highlights a critical direction for improving sensor and electronics design. (2) Bit depth is sufficient, with a recommended value of 8 bit or higher. This suggests that data compression could serve as a promising direction for reducing bandwidth and computational costs in future systems. (3) Sampling frequency appears to be relatively less influential, possibly due to the limited consideration of dynamic activities in existing datasets. This finding underscores the need for the research community to develop datasets that specifically target such activities. In summary, our study provides preliminary theoretical guidance for selecting sampling parameters for the pressure sensor matrix and identifies clear directions for future improvements, helping pave the way from laboratory prototypes to practical applications.