SMT resistor Nonlinearity

[James Meyer]

On Mon, 7 Mar 2005 15:09:02 +0000 (UTC), kensmith@green.rahul.net (Ken Smith)
wroth:

Now I'm curious as heck what the spectrum looks like, but have no idea how to
do a Fourier transform. )-;

I can think of four methods:

(1) For each frequency, multiply the signal by SIN(WT) and integrate to
get the SIN part. Do the same for the COS part.

(2) Use the kit for doing FFT in speadsheets.

(3) Make the curve in LTSpice and use the built in spectrum.

(4) Get a copy of the code for some FFT program and compile it.
--

I think the analog method has the best chance of working. We're talking
about distortion products that will probably be 50 to 80 db down from the
fundamental. How many bits would you need to do a good job digitally?

Jim

 

[Terry Given]

James Meyer wrote:

On Mon, 7 Mar 2005 15:09:02 +0000 (UTC), kensmith@green.rahul.net (Ken Smith)
wroth:



Now I'm curious as heck what the spectrum looks like, but have no idea how to
do a Fourier transform. )-;

I can think of four methods:

(1) For each frequency, multiply the signal by SIN(WT) and integrate to
get the SIN part. Do the same for the COS part.

(2) Use the kit for doing FFT in speadsheets.

(3) Make the curve in LTSpice and use the built in spectrum.

(4) Get a copy of the code for some FFT program and compile it.
--


I think the analog method has the best chance of working. We're talking
about distortion products that will probably be 50 to 80 db down from the
fundamental. How many bits would you need to do a good job digitally?

Jim

reminds me of a hilarious argument we had with Parkside Laborotories in
Christchurch, NZ once. They reckoned our AC drive drew DC current, which
we reckoned was bullshit for a rectifier-capacitor filter. How did these
idiots measure it? Fluke 87 on peak hold, measuring current one way then
the other. Subtract one reading from the other, and voila - DC current
"appears". Despite numerous written protestations, they refused to
certify our product. BTW the current was 16Arms, they "measured" about
2mA "DC". Which needs AT LEAST 16*1.414/0.002 = 11,314 counts just to
see it. 10% accuracy needed *at least* 17 bits of ADC precision, without
considering accuracy.

We went to a different certification lab, and of course passed with
flying colours. Why did we use those idiots at all? Our parent company
owned them.....

Cheers
Terry

 

[Ken Smith]

In article <dXriDJIUZHLCFw50@jmwa.demon.co.uk>,
John Woodgate <noone@yuk.yuk> wrote:

I read in sci.electronics.design that Ken Smith
kensmith@green.rahul.net> wrote (in <d0hqqe$fk8$1@blue.rahul.net&gt
about 'SMT resistor Nonlinearity', on Mon, 7 Mar 2005:

It is those others that may spell the downfall of this method of
specifying the distortion. Two resistors can look equally bad under
this test with one being much worse than the other. A flat spot as you
pass through zero can make the same amplitude 3rd harmonic as a smooth
curve.

It is possible but not very likely. For most passive non-linearities,
the third harmonic amplitude relative to the fundamental is a reliable
quality indicator. The harmonic spectra tend to be roughly monotonic,
decreasing in amplitude inversely proportional to order.

What about how they track with amplitude. Do they tend to maintain the
same % distortion over a wide amplitude range? I didn't check for this
when I encountered the problem. I just kicked the parts out of the
design.


--
--
kensmith@rahul.net forging knowledge

 

[Ken Smith]

In article <sdtp2156m1g31fotb24m62ueineqr0ml3c@4ax.com>,
James Meyer <the.hand> wrote:

On Mon, 7 Mar 2005 15:09:02 +0000 (UTC), kensmith@green.rahul.net (Ken Smith)
wroth:



Now I'm curious as heck what the spectrum looks like, but have no idea how to
do a Fourier transform. )-;

I can think of four methods:

(1) For each frequency, multiply the signal by SIN(WT) and integrate to
get the SIN part. Do the same for the COS part.

(2) Use the kit for doing FFT in speadsheets.

(3) Make the curve in LTSpice and use the built in spectrum.

(4) Get a copy of the code for some FFT program and compile it.
--

I think the analog method has the best chance of working. We're talking
about distortion products that will probably be 50 to 80 db down from the
fundamental. How many bits would you need to do a good job digitally?

It takes a surprisingly low number of bits per word if you are
specifically coding for finding the harmonics.

Using 16 bit numbers you can get almost 90dB down with fairly normal
processing. If you know the fundamental frequency and can control the
sample rate, you can do even better than this.

The trick is to find the amplitude of the fundamental first. You then can
remove most of it from the data before doing the next step.

If you have set you ADC sample rate at a frequency that is "N/M" times the
fundamental. Where "N" and "M" share no common factors, the quantization
noise from the ADC will usually be down around 1/3rd of an LSB and be
spread around enough that it doesn't fool the harmonic detection.

--
--
kensmith@rahul.net forging knowledge

 

[Ken Smith]

In article <B3qlbBBTXVLCFwYS@jmwa.demon.co.uk>,
John Woodgate <noone@yuk.yuk> wrote:

I read in sci.electronics.design that Ken Smith
[....]
What about how they track with amplitude. Do they tend to maintain the
same % distortion over a wide amplitude range? I didn't check for this
when I encountered the problem. I just kicked the parts out of the
design.


Well, as you might expect, in most cases relative distortion increases
with increasing amplitude.

So can I trust that I will just never see resistors that look kind of like
this:

----->!-----
! !
! !
------+ +---/\/\/\-------
! !
! !
-----!<-----

If I can trust that, we can simplify testing a bit.
--
--
kensmith@rahul.net forging knowledge

 

[John Woodgate]

I read in sci.electronics.design that Ken Smith
<kensmith@green.rahul.net> wrote (in <d0keh4$oev$2@blue.rahul.net&gt
about 'SMT resistor Nonlinearity', on Tue, 8 Mar 2005:

So can I trust that I will just never see resistors that look kind of
like this:

----->!-----
! !
! !
------+ +---/\/\/\-------
! !
! !
-----!<-----

With crossover distortion? I hope so, because that would indicate non-

ohmic contacts, with implications for reliability.

If I can trust that, we can simplify testing a bit.

That is almost always a worthwhile pursuit.
--
Regards, John Woodgate, OOO - Own Opinions Only.
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk

 

[Terry Given]

John Woodgate wrote:

I read in sci.electronics.design that Terry Given <my_name@ieee.org
wrote (in <kM6Xd.7652$1S4.822011@news.xtra.co.nz&gt about 'SMT resistor
Nonlinearity', on Tue, 8 Mar 2005:


reminds me of a hilarious argument we had with Parkside Laborotories in
Christchurch, NZ once. They reckoned our AC drive drew DC current, which
we reckoned was bullshit for a rectifier-capacitor filter. How did these
idiots measure it? Fluke 87 on peak hold, measuring current one way then
the other. Subtract one reading from the other, and voila - DC current
"appears". Despite numerous written protestations, they refused to
certify our product.


Why? Which standard has a limit below 2 mA for DC?

I forget now, but there is one....but I'd have to dig through 10 years
of emails to find it. That was before a standard for drives existed, so
we used a variety of different standards, cant recall all of them. This
was an AS/NZS standard.

BTW the current was 16Arms, they "measured" about
2mA "DC". Which needs AT LEAST 16*1.414/0.002 = 11,314 counts just to
see it. 10% accuracy needed *at least* 17 bits of ADC precision, without
considering accuracy.

We went to a different certification lab, and of course passed with
flying colours. Why did we use those idiots at all? Our parent company
owned them.....


Are they still in business?

yep. They may have learned how to take a meaningful measurement by now
though. To be fair, most of what they did was electrical testing (light
switches, power points etc), so they were pretty good at making
pneumatic fingers flip light switches millions of times.

The real problem of course was the particular idiot we dealt with on
that particular day. There was also a great deal of politics involved
too - the parent company made electrical gear, and thought we were a
bunch of elitist snobs because we only hired ME/PHDs, whereas they had
*no* degreed engineers at all (of which they were quite proud). Of
course they made thousands of things with 3 or 4 bits of plastic and
bent metal, whereas we made a couple of dozen products with thousands of
components (eg 1MW motor drives) all controlled with DSP.

This meant that escalating the "they dont know jack shit" complaint
upstairs had the *opposite* effect, with the parent companies
engineering manager (who had a technicians certificate) agreeing with
the idiot, *despite* our best technical arguments, which were never
refuted (hardly surprising, they were correct)

These are the same guys who initially built our (complex) plastic
injection mouldings. One part in particular, the user interface, was
quite tricky (moving bits etc). They made 10,000 or so, then about 9
months later we re-ordered, only to be told 2 weeks later that the
design was completely unmanufacturable. Despite the fact they had
already made 10,000 of them. Turns out they just "fiddled with the
knobs" on their injection moulder to get it right the first time, then
didnt write down the settings. Second time around, their random search
method failed to find a workable setting, so it "couldnt be done"

Our production manager flew down there and had a blazing row with their
production manager, the outcome of which was him uplifting the tool and
shipping it back to our factory. We got a local injection moulder
(Napier Tool & Die) to do the work, and had off-tool samples the next
day which were perfect.

The next project required extensive IM tooling, which we *had* to use
the parent company for. So we got NT&D to split the tooling cost into
(IIRC) ten bits, each of which fell below the budget approval threshold,
and ignored the parent company completely. Apparently there was hell to
pay when they found out.....but we got our plastics on-time, so bugger them.

And no, I am not exaggerating at all. We used to delight in comparing
our parent company with Scott Adams illustrations, the similarity was
remarkable.

Cheers
Terry