JMIR Publications

ABSTRACT

Background:

Long-term Physical Activity (PA) is crucial for preventing and managing a wide range of health disorders, while also improving quality of life. Internet of Medical Things (IoMT) devices such as wearables offer a practical mean to track PA but their widespread use raises sustainability concerns. To estimate the efficiency and sustainability of wearables, established methodologies such as Health Technology Assessment (HTA) and Life Cycle Assessment (LCA) are typically applied at very advanced stages of development, when substantial knowledge of the final design and operating conditions of these devices is available, but unfortunately, when little or no room for modification remains.

Objective:

This article addresses this methodological paradox in the context of the early design and evaluation of wrist-worm step counters, for which recent research would suggest overdimensioned electronic design. Specifically, we seek to reveal the optimal resources-performance tradeoffs of critical electronic components of frugal smartbands, based on the amount of data they can collect in the long-term, and the information content they can preserve for step counting, while accounting for the variability in users’ gait speeds and the uncertainty in the energy consumption of final devices.

Methods:

We conducted a secondary analysis on an existing accelerometer dataset characterizing the wrist motion of two healthy subjects walking at different speeds for two minutes. The preliminary sampling rate of the movement signals along the axes x, y and z was reduced through cubic spline interpolation and discretized to quantify the preservation of information contained in the modified signals, first as a function of the frequency variations, and second with respect to the increase of motion velocity and the decrease of frequency. Based on this analysis, we constructed and evaluated four design archetypes for frugal smartbands, according to the coupling between energy consumption and sampling frequency of four well-used accelerometers, and the critical capabilities of the essential electronic components required to implement a suboptimal asynchronous First-Input-First-Output (FIFO) algorithm operating under different sampling rates.

Results:

Between 70% and 90% of the information is lost in signals modified at very low frequencies (between 2 Hz and 5 Hz), whereas moderate losses below 24% are observed from 20 Hz onwards. Additionally, a significant loss of Information in the modified signals is observed when subjects walk vigorously or jog (at speeds of 8 km/h or higher); or when they engage in normal walking at speeds below 8 km/h and the accelerometer records data at sampling frequencies below 7Hz or above 25Hz. On the other hand, relative high sampling rates rapidly saturate FIFO buffers and increase energy overheads, particularly when they are implemented fully in memory-rich components that handle both data transfer and data processing tasks.

Conclusions:

Our results reinforce the thesis that sampling frequency and gait speed are both critical for step counting, particularly when individuals walk, although no conclusions could be drawn when individuals jog or run. On the other hand, our worst-case analysis based on our design archetypes showed that frequency, storage and processing capacities of electronic components are all important for (1) evaluating uncertainty at early design and (2) distributing the computation load of final devices.