Abstract:

Crowdsourced cellular coverage maps are built from measurements collected by end-user phones. The energy cost of the measurement application limits the availability of this data: an application that noticeably shortens battery life may be uninstalled. The standard remedy is adaptive sampling, in which the measurement rate depends on the detected motion state of the device; its effectiveness is well documented. This paper focuses on adding hysteresis to such an adaptive sampling algorithm, that is, a temporal mode-switching lockout measured from the last accepted change. We measured and evaluated what hysteresis contributes to such a policy. Six client configurations were examined on three Android phones, with five to six 45-minute runs per configuration along the same urban route. Consumption was measured with the battery charge counter and expressed relative to the result of the same phone in a reference configuration. Median charge consumption was similar with and without hysteresis (0.82 versus 0.83); no clear difference was detected. At similar median consumption, ADAPT+HYST covered 134 squares with 50 m sides, compared with 105 for ADAPT, or 28% more. The result indicates a favourable trade-off in this sample. Disabling the upload path with the same sampling policy reduced median consumption from 0.82 to 0.56, the largest difference among the ablation comparisons studied.