Format for manual measurement

[bennop]

Thought I'd contribute by measuring one - maybe later some - of my bulbs.
As I don't have one of the supported devices I did a manual measurement of a Lidl/Livarno GU10 CCT/Color device.

Looking at the provided files I'm not quite clear on what granularity is required. So here some questions before I proceed:

For CCT I find pretty good linearity of wattage vs. brightness and almost no variation with color temperature so I'd think brightness values of 1 and 255 and mireds of 153 and 500 would make up a minimal table (2x2) from which all the other values can be interpolated.

For color it seems the R, G, and B are linear in themselves so I would do that with some more intermediate points.
At what hue values should the values be taken (actually how is hue encoded - seems like a 16bit integer)?
How many sat values are needed?

Benno

PS: Interestingly, I also find that power consumption is lower in low intensity color setting than in standby - suggesting to keep the light on for power savings. Do others see that, too?

[bramstroker]

Thought I'd contribute by measuring one - maybe later some - of my bulbs. As I don't have one of the supported devices I did a manual measurement of a Lidl/Livarno GU10 CCT/Color device.

There are multiple people using the LIDL lights, so it would be very nice to get them supported for the LUT mode.

Looking at the provided files I'm not quite clear on what granularity is required. So here some questions before I proceed:

There are no really strict rules about the granularity but with the default settings in the automatic measure script it will take the following amount of measurements. It's recommended to use this amount of measurements, because we try to be as precise as possible. It's only a one time effort to create the LUT files.

• brightness: 255
• color_temp: 2958
• hs: 24787

For CCT I find pretty good linearity of wattage vs. brightness and almost no variation with color temperature so I'd think brightness values of 1 and 255 and mireds of 153 and 500 would make up a minimal table (2x2) from which all the other values can be interpolated.

Almost all of the lights we measured didn't have really linear characteristics. I have seen some graphs of @nepozs proving this. Maybe he can share here.
So I am not sure if we should allow lut files which are done by only a few manual measurements to the repository.
Don't you have a power meter in house which you can connect to HA? The measure tool is very simple to use, so when you set this up you don't have to bother doing any manual measurements.
Here is all the information about that: https://github.com/bramstroker/homeassistant-powercalc/wiki/Contributing-new-lights

For color it seems the R, G, and B are linear in themselves so I would do that with some more intermediate points. At what hue values should the values be taken (actually how is hue encoded - seems like a 16bit integer)? How many sat values are needed?

See this chapter https://github.com/bramstroker/homeassistant-powercalc#lut-data-files, there is an explanation about the hue and sat ranges.

Benno

PS: Interestingly, I also find that power consumption is lower in low intensity color setting than in standby - suggesting to keep the light on for power savings. Do others see that, too?

Which power meter did you use to measure the power? Most cheap power meters are very inaccurate or even incapable to measure very low current. So maybe you have inaccurate readings. Are you able to reproduce this when taking multiple measurements?
I would be highly surprised if the standby power is less than when on at the lowest brightness.
Would be interesting if this is the case for this light. Than you can leave it on for power savings lol.

[bennop]

I have made the following plots from my CT measurements a 1, 10, 20, 30, ..., 100 and rescaled to 0 (off) .. 255:

The line shows the mean of the reading for mireds={153, 350, 500}, SD is less or equal to 0.02W across the three measurements so the error bars would not extend beyond the dots in the plot.

The hue measurements are a bit more "bumpy" and maybe not quite as linear over the full range:

(sorry - dropped the x-axis labels, should be the same as before (linear 0..100%))

Tis is the first such measurement I did, but to me that looks pretty linear to me (I also checked differences between adjacent points, that confirmed the visual appearance).

I have an old powermeter where I have a resolution of 1W and no connectivity whatsoever but I used this device:

https://de.aliexpress.com/item/1005002487021327.html?spm=a2g0o.store_pc_groupList.8148356.15.3c004391ZbGCv2

While it has built-in BlueTooth (and I can connect my phone through the manufacturer's app), I have not yet even tried to connect it with HA (any hints on how to do that would, of course, be appreciated)
The Watt readings offer two decimals and, thus, a nominal resolution of 10mW with adjustment for the power factor (see last [big] line of display).
I do get fluctuations on the order of ±30mW with some outliers (usually within ±50mW) over a period of 10 to 20s. I then take what you may call a visual average to find the mean of the fluctioans - I would think that turns out to be accurate to around ±10mW.
Looking the the nice linearity of the curves I feel that I’m not too far off with this.

I am in the process of building an ESP32-based meter that that will take some more time to realisation.

I had looked at the documentation but get the feeling it describes what is found in the files but not necessarily how to generate them. Color temperatur I did in mireds as described. But especially for hue I have to assume what the 0..65535 range maps to. I'll go with the usual 0..360 range for red-orange-yellow-...-violet-red, so 0=red, 10923(65535/6)=yellow, 21845(65535/3)=green, ...
Is this correct?

I'm thus a bit puzzled by the grid step you mention for hs not being an obvious fraction of 65536

I also see that for the color setting I get higher readings for mixed colours (eg, yellow as red + green) as two (or more) color LEDs are driven simultaneously. Therefore, I would recommend a finer resolution of 10922 (65535/6) for hs.

[bramstroker]

I don't have much time right now to answer extensively but a quick headsup on the hue range.
It's just:
hue = int(normal_hue / 360 * 65535) to map the HA hue 0-360 to the range Philips uses.

[bennop]

great - that's what I assumed and is all I think I need now ;-)

But then I would reiterate my suggestion for adjusted hue grid values

[bramstroker]

The measure tool uses steps / grid size of 2000 for hue. The last step will be from 64001 to 65535. See https://github.com/bramstroker/homeassistant-powercalc/blob/master/utils/measure/measure.py#L291. When calculating power consumption the component will pick the the lower closest matching hue value from the CSV, so when you have set your light to 64350 it will pick the CSV row with 64001. Hope this answers your question.

[bennop]

One more comment regarding the drop for low intensities: I did notice that the power factor changes from 1.0 to somewhere near 0.5 - that could be an explanation for the drop seen at low intensities.

[bramstroker]

you mean the resolution of your power meter jumps from 0.5 to 1.0 and back?

[bennop]

The resolution is always at 10mW

I guess that is related to voltage and current no longer being in phase - have no way of measuring that beyond the shown device ... power factor is 1 fo sine on resistive load, see, e.g., https://www.rapidtables.com/electric/Power_Factor.html)

[bramstroker]

The resolution is always at 10mW

I guess that is related to voltage and current no longer being in phase - have no way of measuring that beyond the shown device ... power factor is 1 fo sine on resistive load, see, e.g., https://www.rapidtables.com/electric/Power_Factor.html)

I don't have enough knowledge about the hardware side of things, and the exact science behind electricity. Glad you have. Software development is more my expertise ;-).
Will have a read when I have some time on hand, maybe I learn a thing or two.

[nepozs]

@bennop
Lidl/Livarno GU10 CCT (if you mean white version with variable color temperature) is probably a twin model of Mueller-Licht "tint GU10 Spot 350lm"
https://github.com/bramstroker/homeassistant-powercalc/tree/master/custom_components/powercalc/data/mueller-licht/ZBT-ColorTemperature
all of them are Tuya based and very low quality.

(I've had only RGBW Livarno models in the past, but all of them I returned to the shop and it was before homeassistant-powercalc project started, so there are no measurements)

Some words about linearity - most smart-bulbs work non-linear (in reality most of that non-linear characteristic is really "artificial" non-linearity created in design phase to form desired light characteristic, so it can have shape of polygonal chain, because it is created from line segments, but changing power factor can also mold it, so resultant shape is just non-linear).

BUT in case of Tuya-based lights it could be almost linear, as Tuya based products are designed to be cheapest in production.
BTW You can check characteristic of measured Mueller-Licht.

example of non-linearity: #124

I have an old powermeter where I have a resolution of 1W

Maybe it is declared accuracy, because on the picture I see, that yours power meter has resolution 0,01W

@bramstroker
In my opinion for brightness 100 steps are really sufficient.

I've also thought about of possibility to linear estimate characteristics using polygonal chain - 5-6 line segments for brightness should be sufficient for any existing model (but it is so simple for brightness only, because CCT and RGB/RGBW lights should use correction factors as well, because efficiency of different monochromatic LEDs are different, and efficiency of cold-white and warm-white also are different).
BUT before estimation like above every model MUST be measured as accurately as possible.

Some words about manual measure - in my opinion it is really possible only for brightness-only models (as 100 points is doable - I've measured "by hand" some models in that way).

[nepozs]

Better example of artificial non-linearity
#197
@bennop it would be nice if you can move this thread to "ideas".

[bennop]

I did add the plots for my measurements - and looked at the deltas - they are very close to linear for the CT and pretty close for the color mode.
I plan to have some more across the CT axis and a finer one along the hue axis to see in how far a simple interpolation can yield reasonably close values to submit - once I find the time again ...

Not sure about the thread classification but I followed your advice.