On 11/9/24 9:39 AM, Jonathan Cameron wrote: > On Thu, 7 Nov 2024 10:47:52 -0600 > David Lechner <dlechner@xxxxxxxxxxxx> wrote: > >> On 11/7/24 10:13 AM, David Lechner wrote: >>> On 11/7/24 4:51 AM, Miclaus, Antoniu wrote: >> >> >>>>> I'm pretty sure that calibscale and calibbias also need to take into >>>>> account if resolution boost is enabled or not. >>>> >>>> Can you please detail a bit on this topic? I am not sure what I should do. >>>> >>> >>> We haven't implemented oversampling yet in ad4695 yet, so I don't know >>> exactly what we need to do either. ;-) >>> >>> But this is how I would test it to see if it is working correctly or >>> not. We will need to test this with a 20-bit chip since that is the >>> only one that will change the _scale attribute when oversampling is >>> enabled. >>> >>> First, with oversampling disabled (_oversampling_ratio = 1), generate >>> a constant voltage of 1V for the input. Read the _raw attribute. Let's >>> call this value raw0. Read the _scale attribute, call it scale0 and >>> the _offset attribute, call it offset0. >>> >>> Then we should have (raw0 + offset0) * scale0 = 1000 mV (+/- some >>> noise). >>> >>> Then change the offset calibrate to 100 mV. To do this, we reverse >>> the calculation 100 mV / scale0 = calibbias (raw units). Write the >>> raw value to the _calibbias attribute. Then read the _raw >>> attribute again, call it raw0_with_calibbias. >>> >>> This time, we should have (raw0_with_calibbias + offset0) * scale0 >>> = 1100 mV (+/- some noise). >>> >>> Then set _calibbias back to 0 and repeat the above by setting the >>> _calibscale attribute to 0.90909 (this is 1 / 1.1, which should >> After a bit more testing, I realized I was testing with a differential channel, this math only applies to that. For a single ended channel, applying a calibscale of 1.1 with a generated signal of 1V will cause the measured value to change from 1V to 1.1V as one might expect. >> Now that I have written this, this has me second-guessing if I >> implemented calibscale correctly on ad4695. It would seem more >> logical that if we have an actual input voltage of 1 V and a >> calibscale of 1.1, then the resulting processed value we read >> should be 1100 mV. >> >> Jonathan, can you set me straight? The sysfs ABI docs aren't >> clear on this point. > > Deliberately vague in this case. calibbias is kind of the wild west > of ABI. Often we have no meaningful information on what the tweak > register settings actually do beyond 'up vs down'. In some cases > the datasheets even refer to them as taps up or taps down. > > I don't think we've ever said if it should be consistent as you > change other parameters. If you care about calibration you probably > need to redo it for your new settings anyway and tweak the calibbias > /calibscale till it gives the right values. > > Obviously that is easier to do if you have a consistent scheme for > a given device and if possible allow calibrating at just one setting > but I don't think we can apply general rules. > Thanks for the clarification. > Jonathan > >> >>> add 10% to the measured raw value). Read, the _raw attribute again, >>> call it raw0_with_caliscale. >>> >>> This time, we should have (raw0_with_caliscale + offset0) * scale0 >>> = 1100 mV (+/- some noise). >>> >>> Set _calibscale back to 0. Then set _oversampling_ratio to 2. Read >>> _scale and _offset again, call these scale1 and offset1. >>> >>> Then repeat the steps above using scale1 and offset1 in the >>> calculations. The raw values will be different but the resulting >>> processed values (mV) should all be the same if the attributes >>> are implemented correctly. >>> >
