Pasi T Mustalahti wrote:

> PTM: Q&D solution:
> TEMP = (TEMP + NW)/2
> This needs 8 bit + carry to handle 0..255 values.
> This function works if the sloshing makes small differences to both sides
> of the right one and the time period is rather short, about 1/20 sek.
>
> If this still give odd values, you can make even:
> TEMP = (TEMP + NV - TEMP/D)
> VAL =TEMP / D
> There   TEMP is a 16 bit register for the summ
>         NV = nev measured value (0..255)
>         D is 2,4,8,16... (2^n) static
>         VAL is the value that is shown in the display.
> This gives you totally wrong values from the first D*5 times (I tried
> this) but after that you get every time a bit better approximation.
> D is selected as a 2^n to get the program easier.
>
> You can even make a try with QBasic:
>
> '--------------------------------------------------
> 'PTMUSTA@UTU.FI
> 'Demonstrating running averages
> 'shareware ;)
> '--------------------------------------------------
> CLS
> NV = 3
> D = 4
> FOR i = 1 TO D * 6
>     'here we simulate random slosh
>     NVU = NV + (RND(1) - .5) / 10
>     'running average value
>     TEMP = (TEMP + NVU - TEMP / D)
>     DISPLAY = TEMP / D
>     PRINT i;"New=";NVU;"Displ=";DISPLAY;"err=";(DISPLAY-NV)/NV;"%"
> NEXT

I'm planning on using something similar to this running average algorithm in
my current project which is to read current drawn by a 40HP AC motor.  The
value read fluctuates wildly sometimes (nature of the machinery) and needs to
be damped.  I don't know much about PICs but the algorithm can be easily
implemented on 68HC11 because of 16 bit arithmetic instructions.  Maybe
somebody could implement it in PIC code.

What I would do in your case would be:-

LEVEL = 0     ; tank level value read by sensor (0 - 255)
N = 16        ; number of readings to average over (2, 4, 8, 16,.... 256)
RAVG = 0      ; running average value of tank level (0 - 255) over N readings
RTOTAL = 0    ; running total after N readings
:LOOP
    LEVEL = read new tank level value
    RAVG = (RTOTAL + LEVEL) / N
    RTOTAL = RTOTAL + LEVEL - RAVG
END LOOP

I get the running average first then the running total, I save on one divide
operation.

N value is chosen based on the frequency of the fluctuation in relation to
your sampling rate.  Choosing N=256 would really damp slow fluctuations and
has one additional bonus in that you only have to take the hi order byte of
the 16 bit value for (RTOTAL + LEVEL) / N to get RAVG.  For any other N values
you would have to do some shift right operations.  And for N=256, RTOTAL has
max of (2^16) - 1.

I've not implemented this Q&D solution (as you put it) yet.  I'm not sure
whether it's going to work.

--------------
seetho