reading from an Analogue to Digital Convertor

[Derek Brabrook]

Hi,
Bear with me on this I have to fill in some background information
so that you can understand the type of help I need.

I am about to embark on a project, which will do the following


A plasma deposition oven will coat a material with a compound
such as Silicon Dioxide and for arguments sake the material is
metal which has a high refractive index.

the material will be scanned by a laser interferometer
which will produce a diminishing sine wave output as the metal/material
gets coated (in other words it starts to become more and more opaque)
which in turn it's output will be fed to an Analogue to Digital convertor.
this AD convertor will be read by a PC104 computers (small footprint)
serial port which in turn will monitor the process and send a stop signal once
the process is complete.

before the process starts it is critical to log the start point of the plasma
deposition cycle and to log one complete sine wave.

bearing in mind that the start point could be negative or positive
depending on the refractive index of the material, so we could be
starting at plus or minus zero on the sine wave or indeed zero, but
that would be considered the "start point"

this would then have to be monitored and at the 90 degree point ignored
(half sine wave) until reaching the 180 degree point (full sine wave)
and then an allowance of say 10% (or whatever you set it to) of the total
sine wave time/distance added.

the extra bit on the end is to make sure that the whole surface is
uniformly covered with the coating.

the sine wave would complete it's 180 degrees in a minimum of 30 seconds
or at a maximum of 2 to 3 hours, depending on the coating and
the material to be coated.

What makes this project interesting is that the PC104 doesn't have
disk space but is 100% IBM compatible, so magnetic file data storage
is out of the window and although
it emulates disk space by using flash memory
I would prefer to process all this in it's RAM as flash memory is
finite in the amount of times you can flash it, and this would have
to run reliably for years at a time.

Now that you have some background, I would like to ask for some help
in getting this started on the right foot in the first place
I intend to develop it in PB 3.5 for DOS, as a Windows environment
is out of the question because of stability and reliablity.

Any pointers to some ready made code/routines that could be modified
would be helpful if only to get the project off the starting blocks,
and of course any sage advice on PB 3.5 and comm port usage would be helpful
let's see what you guys can come back with !!!

if you would prefer to email me directly or have any further
questions to clarify the project then please feel free
using the following email address

[email protected]


Regards

Derek Brabrook

[Mel Bishop]

Originally posted by Derek Brabrook:
I intend to develop it in PB 3.5 for DOS, as a Windows environment
is out of the question because of stability and reliablity.

That is an extremely intelligent decision on somebodys part. Stability
and reliability is not one of windows strong points.

It sounds to me that your main problem is determinining the "starting
point" value of the process.

Since the physical starting point on the sine wave can be anywhere
along the curve, you may want to consider getting the average value
of the first sine wave and then build in a "sliding scale" to make
adjustments.

For example, suppose the starting point is -5. Adjust it to zero, get
the average value of the first sine wave then add 5 to it.

I hope this makes sense. It sounds perfect to me but...

------------------


[This message has been edited by Mel Bishop (edited June 23, 2002).]

[Karl Lessmann]

"this would then have to be monitored
and at the 90 degree point ignored
(half sine wave) until reaching the 180
degree point (full sine wave) ..."
* * *

Unless I'm missing something here 180
degrees would be the half sine wave
point in a true sinewave ... 360 degrees
would be the full-cycle point.
-Karl

------------------

[Derek Brabrook]

Yes sorry your right, late night last night
180 degrees would be the half way point
360 degrees being the full sine wave
........whoops

thanks for pointing it out

Regards
Derek



Unless I'm missing something here 180
degrees would be the half sine wave
point in a true sinewave ... 360 degrees
would be the full-cycle point.
-Karl

[/B][/QUOTE]



------------------

[Derek Brabrook]

Why thank you for that vote of confidence on reliability <grin>

Unfortunately we can't get an average value on the first sine
wave because that would mean that the first lump of material
(actually it's for a silicon wafer fab plant) would be wasted and if the
coating took 2 - 3 hours then that would be 2 - 3 hours blown away
along with the coated wafer, so the wastage would be unacceptable
unfortunately it would have to be done in real time rather than an
average, mainly because each wafer going through the plasma deposition
oven would be different from the last.

Regards

Derek

Originally posted by Mel Bishop:
That is an extremely intelligent decision on somebodys part. Stability
and reliability is not one of windows strong points.

It sounds to me that your main problem is determinining the "starting
point" value of the process.

Since the physical starting point on the sine wave can be anywhere
along the curve, you may want to consider getting the average value
of the first sine wave and then build in a "sliding scale" to make
adjustments.

For example, suppose the starting point is -5. Adjust it to zero, get
the average value of the first sine wave then add 5 to it.

I hope this makes sense. It sounds perfect to me but...

------------------

[Jim Martin]

Derek -

This is a fun problem. But the fun part is not in coding, it's in
selecting and tuning your algorithm.

There are 2 items you first need to consider: how fast is the A/D
running, and how much noise is present on the data?

And, as a 3rd consideration, is it possible to input a reasonable guess
as to the expected process duration? 2 orders of magnitude variation is a
lot to have to work blindly with.

Assuming a decent SNR and a fast A/D process, I'd be tempted to work as
follows:

1) Prefilter the data. As an example: upon receipt of a data value, use
it as the latest update in a boxcar integrator. Use the timing guess to
set the length of the integration window.

2) Using the filter output,look either for a max or min.

3) If either a max or a min has been found, look for the other.

4) If both have been found, wait until you hit the original value.

5) Now, in theory, you've made a complete 360 degrees. You can compute
how many samples you need for your extra 10%.

Depending on how noisy your data is, any number of complications may
present themselves, like

a) how do you deal with a start point that is at 90 or 270 degrees?

b) step 4 is tricky, since it implicitly assumes the data forms a
pure sine wave, rather than a decaying one. My guess would be
that, after detecting the 2nd max/min, you would have to do some
sort of correlation analysis to determine an approximate decay
factor (such as an inverse exponential), and factor that in. That
is, after detecting the 2nd max/min, you've determined the period
of the nominal sine wave. You can assume various decay factors and
see which one gives the best fit to your observed data. But you have
to be damned sure your data is good, or any simple algorithm will
wander off into hyperspace under some conditions. Murphy's Law
applies rigorously.

c) You _really_ need some sort of estimate on each run's duration, in order
to properly set up your prefilter. If you have too wide a bandwidth (in
the case of a boxcar integrator, too short a summation sequence) you get
too much noise with resulting false peaks. Too narrow a bandwidth and your
peak-to-peak amplitude gets hammered, and you can't accurately find the
peaks.

d) Nonlinear processes like max and min can really get you in trouble. You
might do well to set you max and min conditionally, to be overridden if
another, better value is detected near the first.

All of these difficulties I leave as an exercise to the reader. The heart of the
process is almost trivial, but there are these undefined areas like the prefilter
and the correlation process. Keep in mind the Gary Larson cartoon which
shows a typical Gary Larson science lab with a big blackboard. At the top and
bottom of the blackboard are these masses of symbols, and in the middle is a line
that says, "And then a miracle occurs".

Do enjoy yourself.

[Michael Mattias]

I'm not so sure I'd do a blanket buy-in to the "DOS is more stable than Windows" argument <U>if</U> you limit Windows to Win/NT or Windows 2000.

I use 98 here, and that has some problems; but at all my client sites I work on either NT or 2K and none of those machines have ever hiccuped, leave alone burped.

Just another point of view...

MCM

[Derek Brabrook]

[QUOTE]Originally posted by Jim Martin:
[B]Derek -

This is a fun problem. But the fun part is not in coding, it's in
selecting and tuning your algorithm.


I totally agree Jim, it is a bit of a stinker because the data although
pretty clean and fast coming from the ADC, it is never the same
so I am going to have to put a lot of thought into it

Regards

Derek


------------------

[ARTHUR KOHL]

Hy,
Try using a data filter.
For instance, read in 5 units of data, see if they are within
a narrow percentage of each others value, discard the data that
is out of range, sum the valid units and devide by the number
of units accepted as valid.

I use an algorithim sismilar to this one in a laser scanner,
with exelent results.

Arthur Kohl
[email protected] www.moldaco.com

------------------

[Jim Martin]

Howdy, Derek -

Well, a little mature reflection can certainly help, particularly
when one has over-complicated a problem.

My steps 1 - 3 stand.

However, once you've identified the max and min (nominally 90 and
270 degrees), you now know the duration of 180 degrees. Multiply
by 2.2 for total duration and Bob's your uncle. No correlation or
curve fitting. (If you think about it, the problem of handling
a start point at either critical point falls out).

Just be careful to get clean peaks.

Jim

------------------

[Derek Brabrook]

Well, a little mature reflection can certainly help, particularly
when one has over-complicated a problem.

My steps 1 - 3 stand.

However, once you've identified the max and min (nominally 90 and
270 degrees), you now know the duration of 180 degrees. Multiply
by 2.2 for total duration and Bob's your uncle. No correlation or
curve fitting. (If you think about it, the problem of handling
a start point at either critical point falls out).


Points taken on board BUT, not only does the sine wave diminish over time
but also the timebase between peaks can become either longer or shorter
depending on the process being run and the refractive index of the material
so your first half sine wave maybe say 1 hour and the second half of the sine wave
could be 45 minutes, so not only do I have to cope with a diminishing signal
but a non linear time base, so using a calculation of 2.2 on the 180 degree mark
would be fatal to the process because it could either pass through
the full sine wave before or after that, so the process has to be monitored
from start to finish and once it see's a full sine wave add a user variable
percentage and then send the stop signal to the PLC. I have pondered your
method some weeks ago but found it flawed and unable to take into account
the true non linear progression of this sine wave, The other alternative
is for me to do some wave shaping and quantitisation on the signal
to make it linear and make up the lost area's of the sine wave and timebase
but that requires more hardware, the most elegant way would be to monitor it
realtime and make decisions in logic at that point,

ie/ is it going up or down, if it's going up wait until it is
coming down, watch it go through the start point (at 180 Deg) and wait for it to
go up again etc etc, I could push the variables into an array
I suppose and make decisions by reading the array, an array also
appeals because I can make some pretty graphs on the screen showing
the sine wave's progress and estimate it's endpoint by say a timing loop
which will sample the data and re-calculate the endpoint in realtime

I also think some fuzzy logic is going to have to be implemented
to make an informed decision as to where the start point, mid point and end point
are bearing in mind that the AD convertor probably won't give me
a truly accurate picture with only 12 bits of resolution


Keep it coming many minds are better than one !!!
once I get the AD hardware here along with the pretty
HP Signal Generator, I can dump some data up from
the AD convertor for those interested


Regards

Derek


------------------

[Derek Brabrook]

I'm not so sure I'd do a blanket buy-in to the "DOS is more stable than Windows" argument <U>if</U> you limit Windows to Win/NT or Windows 2000.
I use 98 here, and that has some problems; but at all my client sites I work on either NT or 2K and none of those machines have ever hiccuped, leave alone burped.
Just another point of view...

Windows is pretty, but thats about it.

this program will be running on a Transmeta PC 3x3 inches
at 600Mhz so booting windows of whatever flavour would
just take too long and life is short.

I would much rather flash the compiled basic program
into eeprom and boot up and be running in about 100 msecs
much better than waiting for 5 minutes don't ya think ?.....hehe

And Windows is about as stable as California isn't going
to be shaken to bits and fall into the sea, nobody knows when
or how big the next GPF is, but they know it's coming and soon !!

did you know that Windows 98 and ME
has a bug that if it runs for more than 49 days it crashes
Microsofts response to this bug was "Well who runs their pc
for longer than 49 days most of our users aren't effected by
this bug !!!"

and that Windows 2000 has 635,000 KNOWN bugs service pack 2
went some way to fixing a few thousand but probably introduced
another couple of thousand in the process.

Don't get me wrong Windoze has it uses but not for ANYTHING
critical, if you put a Windoze PC in a critical area of PLC
you would have a disaster waiting to happen.

Just an observation and another point of view......<grin>

it's no use starting a flame war I have asbestos underwear on
and a kevlar coated keyboard......<grin>


Regards

Derek


------------------