Averaging AtoD samples

[HankMcSpank]

In a different application I used a "weighted Average" where no one sample could mangle the running average.
This is the regurgitated conversion to PBP:

I initialize the values and add new value to the stack and subtract off the old average, giving the new value 1/weight of effect

It must be cos I suck at maths, but I've read that a few times & it's not sinking in. Can I use an example here to how that works in practise. Let's say six AtoD Samples are taken....

100
99
100
101
499
100

You said that your weighted average stops a spurious reading skewing the overall average...can you illustrate using those six samples how the 499 would be ignored?

Many thanks!

[Plcguy]

It must be cos I suck at maths, but I've read that a few times & it's not sinking in. Can I use an example here to how that works in practise. Let's say six AtoD Samples are taken....

100
99
100
101
499
100

You said that your weighted average stops a spurious reading skewing the overall average...can you illustrate using those six samples how the 499 would be ignored?

Many thanks!

It will affect it, but only at 1/sample rate.
so, for a 10 sample example, assume the running average = 100
100 *10 (sample rate) = 1000 = sum of the samples

new value 101
101+1000 - 100 = 1001
ave = 1001/sample rate = 100.1 0r 100 for PBP (rounds down)

new value 99
99+1001 - 100 = 1000
ave = 1000/sample rate = 100.0 0r 100 for PBP (rounds down)

new value 499
499+1000 - 100 = 1400
ave = 1400/sample rate = 140

new value 101
101+1400 - 140 = 1361
ave = 1361/sample rate = 136.1 0r 136 for PBP (rounds down)

far less than a (499 +100)/2 = 300

to filter the 1st bad sample add the lines (or similar)
if incoming > ave*10/9 AND if incoming < ave*10/9 then bad_count = 0
if (incoming < ave*10/9 OR if incoming > ave*10/9) AND bad_count = 0 then bad_count = 1
if (incoming < ave*10/9 OR if incoming > ave*10/9) AND bad_count = 1 then bad_count = 0

if bad_count = 0 then gosub average
if bad_count = 1 then Skip average routine

This (or similar) should skip the 1st out of whack signal

[Ioannis]

I played with Darrels 16 bit fast averaging routine and changing the values in HystLevel, AvgCount, FAspread, I got a very slow changing average if the sampling value was close to the average.

But if suddently the input was grounded or connected to Vdd, then the Value of the average was instatly zeroed or maximized (1024 or 255 according).

It is an amazing routine. Great job Darrel, once again.

So, do not be afraid of changing the User Options as Darrel marked them and experiment.

Ioannis

Test code:

include "Fast_Average.pbp"

lcdout com,1

while 1
    lcdout com,$80,#Value,"    "
    adcin channel,var
    Value=var
    gosub average
    pause 100
wend

[HankMcSpank]

I too played with Darrel's averaging & was very impressed, but for this specific need (throwing away a sprurious reading that arrives in every 10 or 12 samples), I think perhaps Melanie's number sorting is more appropriate.

Wrt Melanie's number sorting - I dabbled with a 'rolling window' (which just takes the next sample into an array & then sorts along with the previously stored samples already in the array) or a 'fill an empty array' type (which starts with an empty array....and waits untilt he array is filled before sorting, then empties & repeats). It might have been my cack approach, but I didn't see a whole lot of difference (results wise) between the two methods.

plcguy ...thanks for the extra explanation, but that 140 'spurious' count - though much slimmed down - would still cause chaos in the frequency detect circuit I'm intending this to go in.

I also mulled a "deduce which sample value appeared the most' in a group of samples ....which would be a handy utility to have, but at this stage of my programming life (early!), it's beyond me!

[Plcguy]

Mabe I'm missing part of your challenge.
How many samples will you throw before deciding it's a real change?
If you look at the filter I added you can set the # of samples to skip as well as the % above and % below to determine what level of input is out of range.

I would prob work on the spurious signal issue. good data acquisition is critical if the program is to be robust. That routine I added is really for a failed sensor check, rather than a continuous filter.

[HankMcSpank]

Mabe I'm missing part of your challenge.
How many samples will you throw before deciding it's a real change?
If you look at the filter I added you can set the # of samples to skip as well as the % above and % below to determine what level of input is out of range.

I would prob work on the spurious signal issue. good data acquisition is critical if the program is to be robust. That routine I added is really for a failed sensor check, rather than a continuous filter.

The challenge is that when detecting frequency using the internal PIC comparators......the resulkting count appears to be jittery. By that I mean for say 1Khz, you get 'counts' arriving like thus...

5000, 4999, 5000,4999, 5001, 5000, 4999,5001,3479, 5000,4999,5001


now for the life of me I can't get to the bottom of why the spurious count arrives! (which would be the preferred option), so I'm faced with getting rid of it in software. What Melanie's number sort routine does, is sort an array of samples into numerical order (which will put the spurious count at the beginning), and then averages only the middle of the array. It gets rid of the spurious count completely.

If the frequency changes, then the middle of the array (after sorting) should reflect the count.

So my challenge if you like, is to get rid of spurious counts (ie not 'average' with them at all) ...and also to do this as quickly as possible.

As I say Melanie's routine works....but I'm always open to other methods to rid spurious readings from the equation.

[Ioannis]

Have you tried Darrel's routine with values set high enough?

Ioannis