Faster way to median filter 3 values in PASM?

lonesock

Hi, All.

I'm thinking of adding a simple median-of-3 filter to EasyADC to get rid of some noise spikes. What I have so far sums the 3 values, and tracks the max and min values, as the median would be sum - max - min. I would appreciate it if anyone has a faster way, as I'm trying to keep the sampling speed up.

Here's the code to date:

              ' median filter
              mov       filter_max, A
              max       filter_max, oldA1
              max       filter_max, oldA2
              
              mov       filter_min, A
              min       filter_min, oldA1
              min       filter_min, oldA2
                           
              mov       medA, A
              add       medA, oldA1
              add       medA, oldA2
              sub       medA, filter_max
              sub       medA, filter_min

              mov       oldA2, oldA1 
              mov       oldA1, A              
              mov       A, medA

Thanks,
Jonathan

Dave Hein

This uses 4 fewer instructions.

        mov temp0, A
        mov temp1, oldA1

        max temp0, oldA1
        max temp1, oldA2
        max oldA2, A

        min temp0, oldA2
        min temp0, temp1

        mov oldA2, oldA1
        mov oldA1, A
        mov A, temp0

lonesock

Ah, a sort instead of tracking min and max and sum! I'll need some time to wrap my brain around it.

Thanks!

Jonathan

Andrey Demenev

                    mov     medA, A
                    cmp     maxA, A             wc, wz
            if_b    mov     medA, maxA
            if_b    mov     maxA, A
            if_ae   cmp     minA, A             wc, wz
            if_a    mov     medA, minA
            if_a    mov     minA, A
                    mov     A, medA

Dave Hein

lonesock wrote: »

Ah, a sort instead of tracking min and max and sum! I'll need some time to wrap my brain around it.

With 3 numbers you can compute the minimums for each of the 3 possible pairings. The maximum of the 3 minimums will be the median value.

lonesock

@ Andrey: Thanks! I think that's kind of a sliding median, where the last min/median/max values are kept for each update...I'd have to add the 3 most recent points each time to allow the median filter to respond quickly to changes in input. That brings up an interesting idea, though...what if I kept the median value from the last round and only added in the 2 most recent samples? I'm sure it would delay the response a bit, but it's at least worth simulating with a bunch of different waveforms + noise.

@ Dave: You are the man! Thanks for the explanation!

Jonathan

P.S. To make this a real challenge, might be nice to have a 5-point median filter as well....just saying [8^)

lonesock

I was able to shave one instruction off of Dave's code:

              ' Store the original value of A
              mov       filter0, A   
              ' Get the minimum of each of the 3 possible pairs ( A & oldA1, A & oldA2, oldA1 & oldA2 )              
              mov       filter1, oldA2
              max       oldA2, A                ' this modifies oldA2, but it will be overwritten soon anyway
              max       filter1, oldA1
              max       A, oldA1                ' A is modified (will be the median filtered value)
              ' find the maximum of those 3 minima               
              min       A, oldA2
              min       A, filter1
              ' update my tracking variables               
              mov       oldA2, oldA1
              mov       oldA1, filter0

Anyone see a way to minimize this further?

Jonathan