Firefox OS/Performance/Power/Sample Delay Calibration of COTS Ammeter: Difference between revisions
Line 14: | Line 14: | ||
{| border="1" | {| border="1" | ||
| Mean: || 31.76 | | Mean: || 31.76 ms | ||
|- | |- | ||
| Std Dev: || 6.20 | | Std Dev: || 6.20 ms | ||
|- | |- | ||
| Variance: || 38.47 | | Variance: || 38.47 ms | ||
|} | |} | ||
I think we can assume a 32 ms delay when time correlating the power usage samples with profiling data. |
Revision as of 23:13, 17 September 2013
Calibrating for Initial Sample Dealy of COTS Ammeter
The Yoctopuce Ammeter we're using for the initial baselining of power consumption of Firefox OS has a small delay before we receive the first sample over the USB connection. This page documents the method used to measure the delay so that we can use it to time-corelate the power usage data with other profiling data for analysis.
Setup and Methodology
I used a Goodfet digital I/O peripheral to toggle on a digital I/O pin and then measure the delay until the ammeter measures current. I wrote some custom firmware for the Goodfet and a python client that interfaced with the yoctopuce ammeter to measure the delay between the goodfet pin going high and the ammeter seeing a measurable increase in current.
The code work is in the fxos branch of my fork of the goodfet repo.
Data and Results
To get a good idea of what the expected delay to first sample will be for subsequent power consumption test runs, I ran 4 trials, each taking 100 measurements. I merged the data into File:Ammeter delay measurements.xls. The important numbers to come out of this measurement are:
Mean: | 31.76 ms |
Std Dev: | 6.20 ms |
Variance: | 38.47 ms |
I think we can assume a 32 ms delay when time correlating the power usage samples with profiling data.