Dynamic sample rate adaptation for long-term IoT sensing applications

Adaptive Sensing Algorithm for IoT Applications with Data and Temporal Accuracy Requirements

“…6). As stated in [9], the variation of the buffer size n used in dyn vd has no significant impact on the error…”

“…Kulau et al [9] uses the Bollinger Bands (BB) to reduce the sampling rate. They adapt the BB with the vertical distances dynamic estimation: dyn vd .…”

“…They adapt the BB with the vertical distances dynamic estimation: dyn vd . BB performances were already evaluated in [8], [10] and in [9] with a small real-world time series (1 day and ≈ 1 month) at a high sample rate (0.3Hz). For n acquired samples, dyn vd (t) is the mean of differences between raw values x(i) and estimated values x(i) using a linear approximation x:…”

“…Therefore, we set n to 8 to provided the best performances of dyn vd as evaluated in [9]. This selected setting allows to use the same storage space of the IoT device as in our approach (in our prediction |W |=8).…”

“…We present an extensive evaluation conducted to assess the performance of the proposed prediction, outliers removal and the benefit of temporal windows. We compare the performance of the proposed algorithm with Kulau et al [9] in terms of accuracy and transmissions to satisfy the application requirements. We also provide insights into the minimal energy saved for different IoT technologies.…”