usb-power-profiling

Make the data from USB power meters usable in the Firefox Profiler.

Quick start

The following instructions will start a server on localhost:2121.

git clone https://github.com/fqueze/usb-power-profiling.git
cd usb-power-profiling
npm i
node usb-power-profiling.js

Then open the Firefox Profiler UI to load the data from the meter.

External power profiling in the Firefox Profiler

In Firefox 121 or later:

• in about:config, set the devtools.performance.recording.power.external-url preference to http://localhost:2121/power.
• use the 'power' preset (or any configuration that uses the 'power' feature) when starting the profiler.
• When capturing the profile, the Firefox Profiler will automatically fetch additional power tracks and add them to the profile.

Seeing a profile containing only the data from the USB power meter

Load http://localhost:2121/profile in the Firefox Profiler.

HTTP API

• GET /profile will return a profile containing all the data since the script has started. You can view it by loading it in the Firefox Profiler.
• GET /power?start=<start timestamp in ms>&end=<end timestamp in ms> returns only a power track to be added into a profile from the Gecko Profiler. The start timestamp should be profile.meta.startTime + profile.meta.profilingStartTime from the profile and the end timestamp should be profile.meta.startTime + profile.meta.profilingEndTime.

Supported devices

Power meters known to work

The example profiles are taken using a USB light, first keeping the light off for a while to record noise from the power meter, then turning the light on at different levels of brightness for about 5s, and finally turning the light off again.

Brand	Model	Example profile	Min interval between samples	Notes
ChargerLab Power-Z	FL001 Super	https://share.firefox.dev/4714rQQ	32ms
ChargerLab Power-Z	KM001Pro	https://share.firefox.dev/4ag8xqN	2ms
ChargerLab Power-Z	KT002	https://share.firefox.dev/3RkPsvf	1ms	Samples contain timestamps in µs, and sampling is driven by the power meter, making the sampling rate very consistent (no degradation of the data when the USB bus is busy)
ChargerLab Power-Z	KM003C	https://share.firefox.dev/3Rg6z15	1ms	Sampling driven by the computer, causing overhead on the computer and relying on the USB communication being smooth.
Shizuku YK-Lab	YK001			See Power-Z KT002. Alternative names: AVHzY CT-3, Power-Z KT002, or ATORCH UT18.
Shizuku YK-Lab	YK003C			See AVHzY C3.
AVHzY	CT-3			See Power-Z KT002.
AVHzY	C3	https://share.firefox.dev/41BVhcf	1ms	Samples contain timestamps in µs, and sampling is driven by the power meter, making the sampling rate very consistent (no degradation of the data when the USB bus is busy)
AVHzY	TC66C (RD)			See RuiDeng TC66C
FNIRSI	C1	https://share.firefox.dev/4asQhLh	10ms	Significant power use changes are smoothed over 500ms.
FNIRSI	FNB48S	https://share.firefox.dev/3RjtVTl	10ms	Significant power use changes are smoothed over 120ms.
RuiDeng	TC66C	https://share.firefox.dev/3v4AvFV	80ms	Sampling rate depends on how much data is displayed on the power meter's screen. With the full display, 100ms is the minimum interval between samples. With only the main 3 values displayed, 90ms is in the minimum between samples, with a static screen (eg. settings) the minimum interval is 80ms. Significant power changes take up to 500ms to stabilize (with a few samples showing only a part of the change).
WITRN	C5	https://share.firefox.dev/41nqAaQ	10ms	Samples contain timestamps in ms.
ATORCH	ACD15P	https://share.firefox.dev/3SIWpGS	1s	Very low sampling rate.
YZXStudio	1280E	https://share.firefox.dev/3Wr9HeW	250ms	Low sampling rate. Significant power changes take up to 500ms (2 samples) to stabilize. Seems to have a low level of noise, making it possible to see a difference between low power values ( <10mW) and 0.

Power meters likely to work

Compatibility with these devices has not been verified, but they are likely to either "just work", or work with a trivial adjustment to the code (eg. tweak a USB product id).

Brand	Model	Notes
ChargerLab Power-Z	KM002C	Same protocol as the KM003C.
FNIRSI	FNB48	Expected to use the same protocol as the FNIRSI C1.
FNIRSI	FNB48P	Expected to be the same as the FNIRSI FNB48S in a different package.
FNIRSI	FNB58	Expected to use the same protocol as the FNIRSI FNB48S.
RuiDeng	TC66	Expected to use the same protocol as the TC66C.
WITRN	A2	Expected to use the same protocol as the C5.
WITRN	A2L	Expected to use the same protocol as the C5.
WITRN	A2C	Expected to use the same protocol as the C5.
WITRN	U3	Expected to use the same protocol as the C5.
WITRN	U3L	Expected to use the same protocol as the C5.
WITRN	C4 / C4L	Expected to be the same as the C5 with lower data precision.
ATORCH	C13P	Expected to be the same as the ACD15P.