TheBackShed.com

Forum Index : Microcontroller and PC projects : better measurements by a running mean

Page 2 of 3

Author

Message

Mixtel90

Guru

Joined: 05/10/2019
Location: United Kingdom
Posts: 7937

Posted: 12:20pm 18 Dec 2021

Copy link to clipboard

Print this post

Apparently there is a problem with some capacitors formed on the die that are the wrong value. It wasn't considered to be important enough to redo the design, although there may be a correction if there are later releases.

Mick

Zilog Inside! nascom.info for Nascom & Gemini
Preliminary MMBasic docs & my PCB designs

Volhout
Guru

Joined: 05/03/2018
Location: Netherlands
Posts: 5090

Posted: 12:36pm 18 Dec 2021

Copy link to clipboard

Print this post

The difference between the running mean and John's algorythm is that the running mean first output value is the measured value. And it increases in accuracy. John's algorythm outputs 1/n measured value as a first value, and after n iterations it is ramped up to the filtered value.

Some benefit for each method.

PicomiteVGA PETSCII ROBOTS

andreas

Senior Member

Joined: 07/12/2020
Location: Germany
Posts: 226

Posted: 01:52pm 18 Dec 2021

Copy link to clipboard

Print this post

Volhout said The difference between the running mean and John's algorythm is that the running mean first output value is the measured value. And it increases in accuracy. John's algorythm outputs 1/n measured value as a first value, and after n iterations it is ramped up to the filtered value.

Some benefit for each method.

Yes - exactly. I did some comparison and found these values:

n Pin(temp) running mean ewma
----------------------------------------------------
1 18.60978475 18.60978475 17.76693208
2 18.60978475 18.60978475 17.85121735
3 18.60978475 18.60978475 17.92707409
4 17.67328178 18.37565901 17.90169486
5 18.60978475 18.42248416 17.97250385
6 17.67328178 18.2976171 17.94258164
7 18.14153327 18.27531941 17.9624768
8 18.60978475 18.31712758 18.0272076
9 18.60978475 18.34964504 18.08546531
10 17.67328178 18.28200871 18.04424696
10 18.14153327 18.26796117 18.05397559
10 17.67328178 18.20849323 18.01590621
10 18.60978475 18.24862238 18.07529406
10 18.14153327 18.23791347 18.08191799
10 18.60978475 18.2751006 18.13470466
10 18.60978475 18.30856901 18.18221267
10 18.60978475 18.33869059 18.22496988

and here is the program which calculates both methods:

> list
' running mean - get better measurements

Dim M As float = 0.0 ' running mean
Dim ewma As float
Dim n As integer = 0 ' counter [0..Max]
Const Length = 10 ' up to 'Length' values are counting
Const weight = 0.1 ' weight of the new measurement

Print " n"," Pin(temp)"," running mean"," ewma"
Print "----------------------------------------------------"

ewma = Pin(temp)

While true do

t = Pin(temp)
n = Min(Length, n + 1)
M = M + (t - M ) / n

ewma = (1 - weight) * ewma + weight * t

Print n,t,M,ewma
Pause 1000

Loop

-andreas

JohnS
Guru

Joined: 18/11/2011
Location: United Kingdom
Posts: 4044

Posted: 02:52pm 18 Dec 2021

Copy link to clipboard

Print this post

Volhout said John's algorythm outputs 1/n measured value as a first value

That looks wrong.

John

andreas

Senior Member

Joined: 07/12/2020
Location: Germany
Posts: 226

Posted: 03:55pm 18 Dec 2021

Copy link to clipboard

Print this post

JohnS said

Volhout said John's algorythm outputs 1/n measured value as a first value

That looks wrong.

John

Yes. The first value of Johns ewma is an initial measurement but this value is not printed, it is for initialization only. The second value is 90% of the first value and 10% of the new measuremnt added to the first value, iff weight = 0.1 . This process continues, 10% of every new value is added to 90% of the last sum.

I think the running mean is more accurate during startup (the first 10 values), in the long run both algos seem to be equal.

The bigger problem is that the Pin(temp) is jumping 1 degree between two measurements, so it can be used as a random number generator ;-)

-andreas

JohnS
Guru

Joined: 18/11/2011
Location: United Kingdom
Posts: 4044

Posted: 04:06pm 18 Dec 2021

Copy link to clipboard

Print this post

andreas said Yes - exactly. I did some comparison and found these values:

n Pin(temp) running mean ewma
----------------------------------------------------
1 18.60978475 18.60978475 17.76693208
2 18.60978475 18.60978475 17.85121735
3 18.60978475 18.60978475 17.92707409

and here is the program which calculates both methods:

-andreas

Something odd in the values printed - looks like the initial Pin(temp) was oddly low. The chip bug?

John

andreas

Senior Member

Joined: 07/12/2020
Location: Germany
Posts: 226

Posted: 04:26pm 18 Dec 2021

Copy link to clipboard

Print this post

Something odd in the values printed - looks like the initial Pin(temp) was oddly low. The chip bug?

It looks like the first value was around 17.76
If have some picos here and will do some tests with them.

-andreas
Edited 2021-12-19 02:26 by andreas

andreas

Senior Member

Joined: 07/12/2020
Location: Germany
Posts: 226

Posted: 06:47pm 18 Dec 2021

Copy link to clipboard

Print this post

Now I have tested 5 different Picos, all processed the same way:

1. Cleared the flash: Clear_flash.uf2
2. Installed Picomite: PicoMiteV5.07.02b1.uf2
3. Started minicom 2.7.1 (my host = debian linux)
4. Started: XMODEM RECEIVE on the pico and pressed Ctrl A+S to start a transfer
5. Selected: Upload xmodem
6. Selected the "running mean.bas" and started it after transfer to the pico

The results are here

:

Pico#2

n Pin(temp) running mean ewma
----------------------------------------------------
1 25.16530554 25.16530554 25.16530554
2 25.16530554 25.16530554 25.16530554
3 25.16530554 25.16530554 25.16530554
4 25.16530554 25.16530554 25.16530554
5 24.69705406 25.07165525 25.1184804
6 24.69705406 25.00922172 25.07633776
7 24.69705406 24.96462634 25.03840939
8 24.69705406 24.9311798 25.00427386
9 25.16530554 24.95719377 25.02037703
10 24.69705406 24.9311798 24.98804473

Pico#3

n Pin(temp) running mean ewma
----------------------------------------------------
1 19.07803624 19.07803624 16.97090456
2 19.07803624 19.07803624 17.18161772
3 19.07803624 19.07803624 17.37125958
4 19.54628772 19.19509911 17.58876239
5 19.07803624 19.17168654 17.73768977
6 19.07803624 19.15607815 17.87172442
7 19.07803624 19.14492931 17.9923556
8 19.07803624 19.13656767 18.10092367
9 19.07803624 19.13006418 18.19863492
10 20.01453921 19.21851168 18.38022535

Pico#4

n Pin(temp) running mean ewma
----------------------------------------------------
1 20.48279069 20.48279069 17.95423267
2 20.01453921 20.24866495 18.16026333
3 20.48279069 20.32670687 18.39251606
4 20.48279069 20.36572782 18.60154353
5 20.01453921 20.2954901 18.7428431
6 20.48279069 20.32670687 18.91683786
7 20.01453921 20.28211149 19.02660799
8 20.01453921 20.24866495 19.12540111
9 20.01453921 20.22265098 19.21431492
10 20.48279069 20.24866495 19.3411625

Pico#5

n Pin(temp) running mean ewma
----------------------------------------------------
1 16.73677881 16.73677881 15.05107347
2 17.2050303 16.97090456 15.26646915
3 16.73677881 16.89286264 15.41350012
4 17.2050303 16.97090456 15.59265313
5 16.73677881 16.92407941 15.7070657
6 16.73677881 16.89286264 15.81003701
7 16.73677881 16.87056495 15.90271119
8 16.73677881 16.85384168 15.98611796
9 16.73677881 16.8408347 16.06118404
10 16.73677881 16.83042911 16.12874352

So the temperature could be anything between 16 and 25 degrees. The ewma starts often with lower values but not everytime.

-andreas

Mixtel90

Guru

Joined: 05/10/2019
Location: United Kingdom
Posts: 7937

Posted: 07:40pm 18 Dec 2021

Copy link to clipboard

Print this post

Is the ADC for temp being initialised correctly? If only one channel is to be used it will be on GP26, if two then GP26 and GP27 etc. You might be reading a floating input.

See PicoMite manual page 74.

Mick

Zilog Inside! nascom.info for Nascom & Gemini
Preliminary MMBasic docs & my PCB designs

matherp
Guru

Joined: 11/12/2012
Location: United Kingdom
Posts: 10315

Posted: 07:48pm 18 Dec 2021

Copy link to clipboard

Print this post

The temperature is on a different ADC channel from the other inputs. These internal temperature measures are always pretty ropey

Quote The on board temperature sensor is very sensitive to errors in the reference voltage. If the ADC returns a value of
891 this would correspond to a temperature of 20.1C. However if the reference voltage is 1% lower than 3.3 V then
the same reading of 891 would correspond to 24.3C. You would see a change in temperature of over 4C for a small
1% change in reference voltage. Therefore if you want to improve the accuracy of the internal temperature sensor it
is worth considering adding an external reference voltage.
RP2040 Datasheet
4.9.

Mixtel90

Guru

Joined: 05/10/2019
Location: United Kingdom
Posts: 7937

Posted: 08:20pm 18 Dec 2021

Copy link to clipboard

Print this post

Ah... That makes sense. :)

It's always worth considering a 3v reference diode if there's any intention of using the ADC inputs for measurement anyway. That's why I've been leaving space for one on my boards whether it's needed or not. An LM4040 or NFC301 3V reference can be found for 90p or less.

Mick

Zilog Inside! nascom.info for Nascom & Gemini
Preliminary MMBasic docs & my PCB designs

andreas

Senior Member

Joined: 07/12/2020
Location: Germany
Posts: 226

Posted: 08:31pm 18 Dec 2021

Copy link to clipboard

Print this post

matherp said The temperature is on a different ADC channel from the other inputs. These internal temperature measures are always pretty ropey

Ok, that is very interesting - thank you!

What I did was simply connecting a USB cable between Pico and PC. The PC was the source of power. The picos were without any other connection and all connected to the same USB port at the same PC.

Because its late here in germany I will go to bed now, but will take an external voltage for testing tomorrow. I have a WANPTEK NPS306W as power source and a RIGOL 1102Z oscilloscope for monitoring.

best regards and good night ;-)
-andreas

andreas

Senior Member

Joined: 07/12/2020
Location: Germany
Posts: 226

Posted: 08:39pm 18 Dec 2021

Copy link to clipboard

Print this post

Mixtel90 said Is the ADC for temp being initialised correctly? If only one channel is to be used it will be on GP26, if two then GP26 and GP27 etc. You might be reading a floating input.

See PicoMite manual page 74.

I used the internal temperatur measured by 'Pin(temp)' in Picomite.
-andreas

Mixtel90

Guru

Joined: 05/10/2019
Location: United Kingdom
Posts: 7937

Posted: 09:32pm 18 Dec 2021

Copy link to clipboard

Print this post

If you ground 3V3EN to shut down the SMPS and power the 3V3 rail directly (at 3.3v of course!) from a stable supply the readings should (in theory) be closer. :)

I'd missed the bit about Pin(temp) using a different ADC channel, sorry.

Mick

Zilog Inside! nascom.info for Nascom & Gemini
Preliminary MMBasic docs & my PCB designs

vegipete

Guru

Joined: 29/01/2013
Location: Canada
Posts: 1132

Posted: 01:55am 19 Dec 2021

Copy link to clipboard

Print this post

Since everyone likes pictures, here's a version for the CMM2, filtering a noisy square wave.

' Filter Test for CMM2
' written by Vegipete
'
' Uses running mean algorithm presented by Andreas
cls

' use following to plot a gaussian-ish distribution
'dim hist(400)
'for i = 1 to 100000
' reading = int(RandGauss(32) * 400)
' inc hist(reading)
' line reading+200,579,reading+200,max(0,579-hist(reading)/4)
'next i
'end

text 50,400,"Filtering a Noisy Square Wave",,3
text 50,460,"WHITE : direct sample reading.",,3,,rgb(WHITE)
text 50,490,"RED : 10 sample average.",,3,,rgb(RED)
text 50,520,"YELLOW: 50 sample average.",,3,,rgb(YELLOW)
text 50,550,"BLUE : 100 sample average.",,3,,rgb(BLUE)

count = 0
square = 100
old_reading = RandGauss(32) * 40 + square
old_ave_10 = old_reading
old_ave_50 = old_reading
old_ave_100 = old_reading
average_10 = old_reading
average_50 = old_reading
average_100 = old_reading
do
' get new reading
reading = RandGauss(32) * 40 + square

' calculate new averages
inc average_10, (reading - average_10 ) / 10
inc average_50, (reading - average_50 ) / 50
inc average_100, (reading - average_100) / 100

blit 1,50,0,50,MM.HRES-1,300 ' roll to the left
line MM.HRES-1,0,MM.HRES-1,MM.VRES-1,1,rgb(BLACK) ' clear

' plot graph
line 798,old_reading,799,reading,1,rgb(WHITE) : old_reading = reading
line 798,old_ave_10, 799,average_10,1,rgb(RED) : old_ave_10 = average_10
line 798,old_ave_50, 799,average_50,1,rgb(YELLOW) : old_ave_50 = average_50
line 798,old_ave_100,799,average_100,1,rgb(BLUE) : old_ave_100 = average_100

inc count ' generate square wave
if count > 200 then
count = 0
square = 100 + (square = 100) * 200
endif

loop

function RandGauss(samples as integer) as float
local i,s
for i = 1 to max(samples,4)
inc s,rnd
next i
RandGauss = s / max(samples,4)
end function

Visit Vegipete's *Mite Library for cool programs.

CaptainBoing

Guru

Joined: 07/09/2016
Location: United Kingdom
Posts: 2170

Posted: 07:51am 19 Dec 2021

Copy link to clipboard

Print this post

very nice

JohnS
Guru

Joined: 18/11/2011
Location: United Kingdom
Posts: 4044

Posted: 09:51pm 19 Dec 2021

Copy link to clipboard

Print this post

andreas,

Thanks for those tests!

John

andreas

Senior Member

Joined: 07/12/2020
Location: Germany
Posts: 226

Posted: 07:43am 20 Dec 2021

Copy link to clipboard

Print this post

Hello,

thank you for all the good tips!

Today I looked at VBUS the power from the USB cable comming from the PC which powers the pico. Peter told me that "You would see a change in temperature of over 4C for a small 1% change in reference voltage".

I can see that the VBUS is not only changing by 1% but by 8.8% (440mV). The hihgest voltage is 5.24V and the lowest 4.80V. This explains all the problems I have with pin(temp) if it is based on this power supply.

Lesson learned: A computer is not a good voltage reference when doing measurements!

So the solution seems to be to get a reliable, stable power source (possibly by battery?), before doing any measuremnts.

-andreas

Mixtel90

Guru

Joined: 05/10/2019
Location: United Kingdom
Posts: 7937

Posted: 10:31am 20 Dec 2021

Copy link to clipboard

Print this post

VBUS is the 5v supply from the USB. The reference voltage for the ADC is taken from the 3V3 line, fed from the SMPS. Because of the SMPS variation on VBUS or VSYS makes very little difference to the ADC reference.

Schematic

There is some filtering via C13 but variation on 3V3 still gets through. Have a look at the signal on VREF.

Mick

Zilog Inside! nascom.info for Nascom & Gemini
Preliminary MMBasic docs & my PCB designs

andreas

Senior Member

Joined: 07/12/2020
Location: Germany
Posts: 226

Posted: 02:10pm 20 Dec 2021

Copy link to clipboard

Print this post

Mixtel90 said VBUS is the 5v supply from the USB. The reference voltage for the ADC is taken from the 3V3 line, fed from the SMPS. Because of the SMPS variation on VBUS or VSYS makes very little difference to the ADC reference.

Schematic

There is some filtering via C13 but variation on 3V3 still gets through. Have a look at the signal on VREF.

Do you mean pin 35?

Quote ADC_VREF is the ADC power supply (and reference) voltage, and is generated on Pico by filtering the 3.3V supply. This pin can be used with an external reference if better ADC performance is required.

I suppose this is a test point or output, but not an input for an external poser supply, correct? Or does "an be used with an external reference" mean that I put in some external power supply?

Btw. my external power supply has a "ripple" of some 80-120 mV (mostly 80 mV).
But I have some MCP1700 3.3V power regulators here, which I could use together with an 10uF elko to create a 3.3V reference.

-andreas

Page 2 of 3

Print this page