Author Topic: In-circuit component testing  (Read 2686 times)

0 Members and 1 Guest are viewing this topic.

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3518
  • Country: us
In-circuit component testing
« on: May 18, 2020, 11:35:42 pm »
I've recently had an adventure with trying to diagnose and repair an instrument for which no service data is available.  Sadly, that is the future for much electronic repair work.

In this instance, I was concerned with testing capacitors in circuit.  After failing with the Arduino LCR tester based BSIDE ESR02 Pro and the Atlas Peak ESR70, I bought an EDS-88A II which arrived today.

I pulled a bunch of boards from the junk box and set to testing caps.  None of these instruments is able to reliably test capacitors in circuit. If there are significant parallel low impedance paths the simple tests being used fail.

On careful consideration, I think that a vector network analysis approach might succeed.  What I have in mind is a 3 receiver VNA equipped with shielded probes.  Touching the probe tips together sets the cable delay term.  If the probe tips are then placed at the leads of a particular part, all other parts in parallel will appear at larger phase delays.  Those delays are easily solved which in turn allows separating capacitor leakage from a parallel resistor or inductor.  So far as I know, no current tester does that.

If you have the skills to be able to identify the delays between arrival times  to much less than the resolution in the time domain, i.e understand that this is done by applying the shift theorem and linear fits in the frequency domain, I'd like to discuss the matter.  Either publicly as an OSSW/OSHW project, or if you prefer, privately as a commercial venture.  I'm not concerned with making money from this. I'm retired.  My concern is to make repairs easier or in some cases simply possible.

I'd like to add that what I am considering is *much* more sophisticated than an "octopus".  what I'm after is the same answer as you would get if you removed the part and put it on an LCR meter.

The nanoVNA design seems as if it would be adequate HW with some changes and more CPU.  A Pi CM3+ should be more than enough CPU capacity.

Have Fun!
Reg
 

Offline Jay_Diddy_B

  • Super Contributor
  • ***
  • Posts: 2766
  • Country: ca
Re: In-circuit component testing
« Reply #1 on: May 19, 2020, 12:00:21 am »
Hi,
IMHO this is not possible.

Consider this simple test:



Two capacitors in parallel, one is 4.7uF and the other is 470nF

This the impedance plots from a VNA:



The bold trace is just the 4.7uF capacitor, the other trace is the parallel combination.

On the left side of the self resonance the impedance changes by 10%
On the right side of the resonance the inductance changes by factor of two. It is actually dropping from 1nH to 500pH

You tell with the VNA if the 470nF is fitted.

This example is just two capacitors …

The technique used is described in this thread:

https://www.eevblog.com/forum/projects/high-bandwidth-current-injector-for-impedance-measurements/msg3026052/#msg3026052

Regards,
Jay_Diddy_B
 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3518
  • Country: us
Re: In-circuit component testing
« Reply #2 on: May 19, 2020, 12:14:40 am »
Thanks for the reply, but the mathematics of the VNA software are inadequate for what I want to discuss.

If you look at a TDR of a pair of caps in parallel shunt across a transmission line, the closest cap will make a bump in the TDR before the farthest cap.  "Farthest" can be a few millimeters or less with adequate phase resolution.  High BW is not needed.

I've got a long term thread on TDR testing of RF connectors which shows every little inductance or capacitance change  in the impedance along a transmission line at fractional mm resolution.

https://www.eevblog.com/forum/rf-microwave/testing-rf-connectors-and-cables/

In the frequency domain your example is  two vectors at different quasi-linear phase shifts.  It's linear if the parts are ideal.  If the phase response of the parts is not linear, then it gets slightly more messy, but not impossible.

Of course, if the VNA probes are equidistant from the parts, you can't solve it.  I'm presuming that the probes are at a particular part.

Have Fun!
Reg
« Last Edit: May 19, 2020, 12:19:39 am by rhb »
 

Offline Jay_Diddy_B

  • Super Contributor
  • ***
  • Posts: 2766
  • Country: ca
Re: In-circuit component testing
« Reply #3 on: May 19, 2020, 01:42:13 am »
Thanks for the reply, but the mathematics of the VNA software are inadequate for what I want to discuss.

If you look at a TDR of a pair of caps in parallel shunt across a transmission line, the closest cap will make a bump in the TDR before the farthest cap.  "Farthest" can be a few millimeters or less with adequate phase resolution.  High BW is not needed.


Snip ..
Have Fun!
Reg

I believe that this is wrong. You need very short very fast pulses to do high resolution TDR. Short fast pulses are high BW.

This means that you are trying to measure 'normal' components at microwave frequencies.

TDR can be modelled in LTspice:



I have the components 1ns apart, which is about 30cm.
I am using 1 \$\Omega\$ transmission line, because power distribution is normally done with low impedance planes.

This is the result of running that model:



The result with the open circuit and the electrolytic are the same, implying that the Electrolytic is not there at microwave frequencies.

Regards,
Jay_Diddy_B

* tdr model.JPG (53.98 kB. 747x673 - viewed 1385 times.)
 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3518
  • Country: us
Re: In-circuit component testing
« Reply #4 on: May 19, 2020, 03:44:13 am »
Please show the BW term in the shift theorem.
 

Offline Jay_Diddy_B

  • Super Contributor
  • ***
  • Posts: 2766
  • Country: ca
Re: In-circuit component testing
« Reply #5 on: May 19, 2020, 05:12:32 am »
Hi,

Let me try this one:

]

I have an open transmission line 100ps long (corresponding to 3cm)

The time for a signal to be reflected by the open end is twice that. The signal goes to the end and bounces back.

If I step the risetime of the source 20p, 100p, 200p and 1n I get the following result:



I can't see the length of the transmission line in the TDR measurement at all if the risetime is greater than twice the electrical length of the line.

To measure the length of the line accurately the rise time needs to be much shorter the electrical length of the line.

Fast rise time in the time domain is high bandwidth in the frequency domain.

This why you need a 20GHz scope SD24 to see the discontinuity in the connectors you tested.

This picture is from your own measurements:



The discontinuity is around 100ps wide. A 1GHz scope has risetime of around 350ps. How you going to measure this without BW?

Jay_Diddy_B

 
The following users thanked this post: Someone

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3518
  • Country: us
Re: In-circuit component testing
« Reply #6 on: May 19, 2020, 01:05:16 pm »
The shift theorem states that a time delay constitutes multiplication by exp(i*pi*f*t) in the frequency domain.

Consider a trace which has two reflection events.  Let the input signal be an arbitrary function A(f).

The shift theorem says those two events are described by D =A*exp(i*pi*f*T1) and A*exp(i*pi*f*T2).

The exponents are linear functions, a + b*f,  of frequency  with a=0.  At each frequency D is the sum of two vectors which are rotating in the complex plane at different rates relative to each other.

There is no constraint on BW. In fact,  A could be a sine wave, the minimum possible BW.  All that is required to solve for T1 and T2 is two measurements.  The details are in the trig formulae in any math handbook.

If you are working strictly in the time domain you do need large BW.  But if you want to get microsecond timing from 70 Hz BW data, you do it in the frequency domain.  Another grad student  measured the multiplexer skew from channel to channel in such data for his dissertation project.  For his work he had to correct all the channels to the same time reference and the 81 microsecond channel skew had to be measured and corrected.  That data had a Nyquist of 125 Hz.

Once again, the limitation is *phase resolution*, not BW.




 

Offline Jay_Diddy_B

  • Super Contributor
  • ***
  • Posts: 2766
  • Country: ca
Re: In-circuit component testing
« Reply #7 on: May 19, 2020, 02:09:33 pm »
Hi,

This is a reasonable model of two decoupling capacitors 3.0cm apart on a PC.




This is a very simple test case.

Speed of light is 3x 108 metres/second

so 3.0cm in 0.1ns

(propagation in FR4 is going to 0.8 to 0.9 x speed of light)

Can you tell the difference between the transmission line model and the lumped model using low frequency signals?



When I run this simulation I get:




So there is an observable difference at a reasonable frequency.

Sensitivity

Effect of the length of the transmission line:



Impedance versus Td




Phase versus Td

Sensitivity to Transmission line impedance




Stepping the transmission line impedance 1, 2, 5 \$\Omega\$




Sensitivity to ESL

Model


Impedance versus ESL



Phase versus ESL




Observation


A difference can be seen between the lumped model and the transmission line model for two decoupling capacitors.

Given the sensitivity to transmission line impedance, the length of the transmission and the equivalent series inductance of the capacitors.
I am not sure how useful this observation is.

The modelling presented here was two decoupling capacitors, there is typically a large number.

Jay_Diddy_B


 
The following users thanked this post: Vovk_Z

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3518
  • Country: us
Re: In-circuit component testing
« Reply #8 on: May 19, 2020, 03:54:33 pm »
Thank you.  That's an excellent start at explaining the problem graphically.  To solve for the component values it is necessary to solve for all the things you varied in your simulations.  This is where an L1  basis pursuit comes in.  Of *all* the possible component and parameter combinations, it solves for the sparse set which best matches the data.  Time gating can be used to restrict the analysis to only components which are in close proximity to the test point.

Solving such problems typically requires many steps and transformation between time and frequency several times in order to place the problem in forms which can be solved numerically at each step.

While the problem of finding the optimal combination of components and parameters is NP-hard, David Donoho proved in 2004 that if a sparse L1 solution exists, it is the optimal L0 solution.  While a solution is not guaranteed, it has also been shown that there is an overwhelming probability that one does exist.

I received a copy of an IEEE paper from 2011 this morning on "blackbox macromodeling using scattering parameters".  It's very close to what I'm talking about.  It's just a bit more upscale than repair work.  But it confirms that what I've posited is correct and is already being done.  Just not in a repairman's tool.  It's being used in high end simulation software to model signal integrity problems and candidate solutions.

A mashup of a nanoVNA and an LCR/transistor tester using a Pi CM3+ instead of an Arduino would easily solve the problem of testing components in-circuit accurately.  It's not a weekend project, but it is possible to build such an instrument for under $100 BoM.

Reg

 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3518
  • Country: us
Re: In-circuit component testing
« Reply #9 on: May 23, 2020, 04:06:23 pm »
I wrote this for an ESR testing mailing list.  I thought it worth adding here.  It's a non-mathematical explanation of basis pursuit.

Consider a transmission line which has an inductive and a capacitive discontinuity of unknown magnitude and location. The TDR reflection response is a flat line with a bump up and a bump down. That's it. The location reveals the distance, and the magnitude of the bump, the value.
Here's an ASCII cartoon example. The proportional fonts mess it up a bit, but the idea should be clear.

The measurement:

------------------------^--------------v----------------------

So, let's compute a very large number of TDR traces for a capacitive discontinuity of a range of expected values and locations and a similar collection of TDR traces for the inductive discontinuity. For simplicity lets assume we compute 25,000 locations and values for the capacitance and the same number for the inductance. So we have 50,000 possible answers.

Some sample solutions from our set of 25,000 capacitance values and locations:

---------^----------------------------------------------------
---------------------^----------------------------------------
------------------------^-------------------------------------

and 25,000 inductance values and locations:

-----------------v---------------------------------------------
-----------------------------v---------------------------------
-----------------------------------------v---------------------

Then we look for the best match using 2 of the TDR traces we computed. As you can see the sum of the 3rd solutions in each set above is the best match for the measurement.

------------------------^-------------------------------------
-----------------------------------------v---------------------
------------------------^--------------v----------------------

This is simple to do visually, though it would be time consuming with so many traces to compare. In this case we have 50,000 unknowns and a single equation.

There can be any number of discontinuties, so long as there are not "too many". Understanding the exact definition of "too many" took me 3 years and in the end, it doesn't matter most of the time. Certainly not for in-circuit testing of passive components. We can simply cutoff the length of the TDR trace to eliminate the parts which are far away from the component we wish to test.

That's what a basis pursuit does. It selects those two traces from what is called the "dictionary" of possible solutions.

In practice the problem would be solved in the frequency domain, but that's just a programming detail which allows us to use the phase information at lower frequencies to determine the exact location of the discontinuities to picosecond (~0.2 mm) accuracy. So nanoVNA hardware combined with the Arduino based $20 LCR tester hardware and a Pi CM3+ compute module should allow separating the ESR of adjacent caps in parallel with smaller caps and resistors.

Have Fun!
Reg
 

Offline RoGeorge

  • Super Contributor
  • ***
  • Posts: 7012
  • Country: ro
Re: In-circuit component testing
« Reply #10 on: May 23, 2020, 07:08:25 pm »
The idea of measuring in circuit components based on reflectometry is very nice, and might come handy in some very narrow domains of application.   :-+

However, I don't think it will be practical for day to day measurements, and certainly not at $100 or so BOM. 



the limitation is *phase resolution*, not BW

Phase resolution is the same as time resolution.  Small time corresponds to high frequency because f=1/T, therefore high bandwidth is a must.

- in math: The method might appear to be working on paper, because in math the signal processing is made with infinite bandwidth (precision), and also without noise or other limitation imposed by the real world.
- in physics: For the real world, everything is also a filter (and since we don't know the characteristic of those filters, that implies loosing some information about the studied signal).
- in engineering: Measurements are altered by noise, sampling frequency are limited by technology, digital processiing (computing) can be done with a limited precision only.

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3518
  • Country: us
Re: In-circuit component testing
« Reply #11 on: May 23, 2020, 07:59:41 pm »

The time resolution is (frequency * 360) / phase_resolution_in degrees for a single frequency.  If you are measuring the phase delay by a linear fit to the phase in the frequency domain across multiple frequencies you can achieve better time resolution than the phase resolution at any one frequency or the Nyquist of the sampling in the time domain even in the presence of zero mean Gaussian noise.

The math is only infinite precision if you assume that.  One is quite at liberty to make  calculations under assumptions of finite precision.  That is often very necessary.

Noise is certainly an issue, but  I know a few ways to address that.  Probe design is probably the hard part.

*By design* the GPS signal at the antenna is around 56 dB below the noise floor of the analog front end.  Processing gain takes care of it.

BTW it's spelled "losing" not "loosing".  Pet peeve of mine.

Have Fun!
Reg
 
The following users thanked this post: WattsThat

Offline RoGeorge

  • Super Contributor
  • ***
  • Posts: 7012
  • Country: ro
Re: In-circuit component testing
« Reply #12 on: May 24, 2020, 12:43:29 am »
Thank you for pointing out the spelling error, I was not aware about "loosing" vs "losing".

Googled it, and the funny thing is Cambridge dictionary has the sentence "This would eliminate the risk of loosing important information." as an example https://dictionary.cambridge.org/dictionary/english/loosing .  That is almost identical with my "that implies loosing some information about the studied signal"   ;D , but I guess Cambridge example is about unintentionally broadcasting confidential information, while I was trying to say that some info will be lost.



Back to the measuring idea, there might be yet another limitation, at the physics level.

Let's assume we know all the parameters of the probing cable, and we have a very fast ADC.  At one side of the probe is a generator+ADC, at the other side is the component, let's say a capacitor we want to measure, soldered on a PCB.
  • 1.  We generate a signal, let's say a step voltage because it's easy to visualize it's evolution in time.
  • 2.  The step wavefront will start to propagate through the transmission line, and will arrive at the capacitor's terminals.
  • 3.  At this moment a fraction of the signal will travel back, and another fraction will continue to travel through the PCB, and will met other components.
  • 4.  When the wavefront in the PCB will hit the most nearby component next to our probed C, that component will reflect some energy back, just as the probed capacitor did, except this new reflection from the nearby component will be a little late and lower in amplitude.
  • 5.  To simplify calculation, let's say the velocity factor in the PCB is 1, and the next nearby PCB component is a SMD at 1.5mm distance from the probed capacitor.  The delay between the probed C and the next reflection will be about 10ps.
  • 6.  That means after about 10ps other reflections from other components will start polluting our signal reflected by the probed capacitor.  In other words, we need to keep only first 10ps and discard the rest.  Since the velocity factor will be less than 1, let's be generous and instead of only 10ps we will give it 100ps before the signal reflected by C will be flooded by other reflections from the other components soldered on the same PCB traces.
  • 7.  Even with a 100ps window of useful signal instead of 10ps, acquisition and conversion won't be easy.  Maybe a sampling head can deal with that.  Let's park this problem for a moment, because there is a much bigger one.
  • 8.  In the first 10...100ps we calculated above, there is not enough time for the signal to travel inside the probed capacitor's plates.  In other words, there is no capacitor yet to reflect back the wavefront.  The reflection will start to happen only after the capacitor's plates are traveled by the wavefront.
  • 9.  I didn't calculate the velocity factor inside the capacitor, but I suspect it will be much lower than the velocity factor in PCB traces.  Chances are nearby resistors will reflect faster than the wavefront will travel through the C plates and back.

Electrolytic capacitors will be even slower in reflecting the signal, because they are made from very long Al foil plates coiled as a cylinder.  This means we will never know if the reflected signal we see is from the probed capacitor or from some other components on the PCB.
« Last Edit: May 24, 2020, 12:53:20 am by RoGeorge »
 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3518
  • Country: us
Re: In-circuit component testing
« Reply #13 on: May 24, 2020, 01:43:38 am »
"That's a fine glory." - Humpty Dumpty

What does "loose"mean?  What does "lose" mean?

I have a BA in English lit.  The reference you cited  merely reflects the demise of competence with the English language.  It merely cites common incorrect usage collected by a web scraper.  Read the disclaimer.

"Loosing" is to "let loose,  let free,  unhinder etc"  *Very* different.  "Losing"  is to suffering a loss.

Oxford is the custodian of the English language in the form of the  Oxford English Dictionary aka the OED. 

I suggest the rest be left until I have the  time to make a circuit and subject it to the tender mercies of my 11801/SD-24. I am *not* suggesting that TDR alone can solve the problem.  What I'm saying is that the addition of TDR can.  Even if the reflections overlap, basis pursuit can resolve them.

Have Fun!
Reg
« Last Edit: May 24, 2020, 01:59:04 am by rhb »
 

Offline virtualparticles

  • Regular Contributor
  • *
  • Posts: 149
  • Country: us
Re: In-circuit component testing
« Reply #14 on: June 04, 2020, 08:47:15 pm »
You can do some interesting tests using a VNA. There are three modes which can be used for this. Simple shunt, Trans-shunt and series. Here is an article I wrote which describes these measurements.

https://www.signalintegrityjournal.com/blogs/8-for-good-measure/post/1344-using-a-vna-for-power-plane-impedance-analysis

It is possible to distinguish between the capacitor and a resistor if one applies a little logic and observes over frequency.

Thanks!

Brian
 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3518
  • Country: us
Re: In-circuit component testing
« Reply #15 on: June 04, 2020, 10:22:19 pm »
You can do some interesting tests using a VNA. There are three modes which can be used for this. Simple shunt, Trans-shunt and series. Here is an article I wrote which describes these measurements.

https://www.signalintegrityjournal.com/blogs/8-for-good-measure/post/1344-using-a-vna-for-power-plane-impedance-analysis

It is possible to distinguish between the capacitor and a resistor if one applies a little logic and observes over frequency.

Thanks!

Brian


Thank you for supplying prior art.

That's *exactly* what I have in mind.  When I get some free time I'm going to build a microstrip circuit with parts in parallel and hit it with my 11801.  That will show each part along the stripline separately.  And if you can do it in the time domain, you can do it in the frequency domain.

I'll then use a VNA and FFT to demonstrate doing it with a VNA using the same hardware.

Parts that can take 500+ V charges on capacitors are pricey ($12/each) so I may relax that back to 50-100 V.

Have Fun!
Reg
 

Offline hendorog

  • Super Contributor
  • ***
  • Posts: 1635
  • Country: nz
Re: In-circuit component testing
« Reply #16 on: June 08, 2020, 09:08:20 pm »
You can do some interesting tests using a VNA. There are three modes which can be used for this. Simple shunt, Trans-shunt and series. Here is an article I wrote which describes these measurements.

https://www.signalintegrityjournal.com/blogs/8-for-good-measure/post/1344-using-a-vna-for-power-plane-impedance-analysis

It is possible to distinguish between the capacitor and a resistor if one applies a little logic and observes over frequency.

Thanks!

Brian


Thank you for supplying prior art.

That's *exactly* what I have in mind.  When I get some free time I'm going to build a microstrip circuit with parts in parallel and hit it with my 11801.  That will show each part along the stripline separately.  And if you can do it in the time domain, you can do it in the frequency domain.

I'll then use a VNA and FFT to demonstrate doing it with a VNA using the same hardware.

Parts that can take 500+ V charges on capacitors are pricey ($12/each) so I may relax that back to 50-100 V.

Have Fun!
Reg

If I'm understanding correctly - big if.

Your comment about enhancing time domain resolution typically seen on a VNA - as it is really limited by the ability to resolve phase and not frequency - could be used with time domain gating to allow the probe effects to be removed.

And the same approach should work with the individual components - in the time domain the only interesting component is the one at the end of the probe. Anything earlier is fixture, and anything later is some other component. So you gate the time domain and then transform back to frequency.

 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3518
  • Country: us
Re: In-circuit component testing
« Reply #17 on: June 08, 2020, 10:01:06 pm »
That's a good summary of my point.  There is a significant time delay between the reflections from different parts and that can be used to distinguish the part at the probe tips from everything else.  Touching the probe tips together gives the delay time to the tips.  Gate out the distant parts and a capacitor between the probe tips has a series capacitance and resistance with a leakage resistance in parallel.

With a 16 bit ADC and a 500 kHz sweep, the phase resolution will allow excluding parts more than 3 mm away. One would need to sweep higher for modern SMD stuff, but for older thru hole gear even 250 kHz would suffice.

The low frequency end of the sweep gives a measure of the leakage resistance, the slope gives the capacitance and the high frequency end gives the ESR.  A perfect capacitor is a constant slope.  A flat section at low frequency is the leakage and a flat section at high frequency is the ESR.

If one precomputes the values as a function of frequency for a range of values, then a basis pursuit will find the optimal choice of the 3 values. As we are simply going for good/bad that doesn't need to be especially precise.  The parallel leakage resistance should be zero if the cap is good.  The ESR varies with working voltage and capacitance, but if the display shows ESR for measured capacitance as a function of working voltage with the measured ESR shown, a quick look at the cap or knowledge of the circuit will tell the user if it's good or bad.

Have Fun!
Reg
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf