ESR Meter Adapter Design and Construction

[dannyf]

In addition to the sine wave you also need a square wave that is in phase with the sine wave

To generate a precisely in phase square wave from a sine wave, you will need a good zero-cross circuit - not easy.

It is far easier to generate a sine wave from a square wave.

The AD9838 would be ...

Check out 5932/5933 - they are meant for this type of work (true LCR meters).

fast 32b micro's to process the signals (phase /amplitude)

People have done that with 8-bit mcus (8051), 16-bit mcus (msp430) and 32-bit mcus (stm32). The ones with 8-bit / 32-bit mcus are good to 0.4% full range.

I prototyped one on PIC24F with minimum out-board hardware (fixed frequency, onboard adc) but it required some pretty fancy math - ieee 1057 3-parameter sine wave fitting.

[dannyf]

You arent for example going to be measuring phase differences down to a couple 100nS with a 4MHz PIC for example at least not accurately.

Depends on how it is implemented. Most people would get the bulk of the work done in the analog work and the mcus would be doing signal generation, adc, and display controls (and sometimes calibration). A 4Mhz PIC would be an overkill for those type of works.

If you are, however, doing some serious math - what I did with the pic24 for example, it takes some hard work to get it going.

[BravoV]

Just an academical question, what is the advantage using sine wave instead of square wave while still using this method in measuring ESR, again "ONLY" ESR, not other impedance attributes like in LCR meter.

[PA4TIM]

For impedance a sinewave will do better. I build something like that to  measure femtoFarrads

I think the meter from this topic is good as it is now. It is foolproof to build without critical components and because he made a pcb layout impossible to screw up using the wrong pcb desing (my meter is less fool proof because I do not have a pcb design. A friend builded it and he used bad sample caps that he mounted next to each other so they influenced each other and the trace holding the measured result routed all around the pcb picking up a lot of garbage. It still was usefull as an indicator but performed not even half as good)

 I measure phase difference between voltage and current in an other project with an arduino with Atmel 328 ( I only use the arduino IDE and a USBasp to program the 328 using ISP) and an analog circuit. Two LM311 comparators make a square from the AC signals. Then a 74LS86 and the output from that is feed to the 16 MHz atmel. It measures first the high and then the low time.

But for my ESR meter all is done analog
http://www.pa4tim.nl/?p=1728 the final version is at the bottom of the page. This one measures real ESR and not impedance (I can measure caps smaller as 100 nF) It has an adjustable frequency and uses 4 cheap IC's and some resistors. I use a panel volmeter module for display but you can use a DMM or microprocessor and LCD display or like I did at first an analog meter. The latter has the advantage that you see problems better. If the meterreading  is not steady it could be a shorted cap (they have the ultime low ESR  )

I spend a lot of time measuring and exploring ESR on every possible way. I also collect bad caps. I solder them on documented pcb strips for research purpose, like a sort of bad-standard caps )

I used IV-meter, VNA , LCR meters, C bridges, squarwave and scope, several homebuild meters, a few commercial ones and the result is, I allmost never measure ESR since then. In my opinion the an over rated parameter and most times measured the wrong way (but by accident often the right way because the datasheets spec impedance at 100 kHz not ESR and 90% of the meters measure impedance).

Besides that without knowing the ESR specs you still do not know anything and measuring in situ can mask a bad cap if it is has a good one parallel, and often there are more caps between a rail and ground. I have measured many caps first in situ and then desoldered them and test them again. The difference is often rather big and a complete bad cap can measure good in situ, A DC shorted cap can measure low ESR too.
The most cap failures are: DC shortage, loss of capacitance, high ESR.

When I started experimenting on ESR a few years ago I thought it was great. I used it a lot for testing. I use a scope as main trouble shooter if possible and I noticed a lot of caps that looked bad in ESR did not give problems in the circuit and that was strange. I have a rather big collection LCR bridges and other component testers so I started analysing the "bad"caps. For instance I repaired a sat tuner and it had 34 bad caps. 2 had real high ESR, most had degraded capacitance  and several leaked DC. But the good thing about an impedance "ESR" meter is that the degraded capacitance shows as high ESR.

So my ESR meter measures ESR and that makes it less usefull if you want an ESR meter to repair consumer gear in situ. Better use the impedance versions that have troubles measuring ESR under 10 uF because reactance becomes to high.

Today I sometimes use the ESR meter in situ as a first impression while restoring long not used gear (and as part of the proces of analysing caps for other people).
Then I desolder one leg, reform the cap and do a DC leakage test and after that I measure capacitance  and measure D. Most times at 1 kHz. D tells you realy if a cap is bad. If the caps do not need reforming and I can power up the instrument I use my scope to check everything and I find every  bad cap this way.

http://www.pa4tim.nl/?p=1728 measuring bad caps
http://www.pa4tim.nl/?p=3775 about ESR and several testing methods
http://www.pa4tim.nl/?p=1385 About DC leakage

[kripton2035]

in elektor esr meter (sept 2002) , they use the same principle as the one described here with a square test signal and 4 4066 gates and a lf412 as difference amplifier. but they said that they had to use a VHC 4066 to minimize at best the unwanted reactance. nowhere did they tell about sine wave.

[BravoV]

I think the meter from this topic is good as it is now.

Agree, this circuit is more than enough for it's purpose, really appreciate JDB's effort, and for sure I'm going to build it.

For impedance a sinewave will do better. I build something like that to  measure femtoFarrads .... <snip> ...

How exactly ? Any chance to explain further on how sinwave is better ?

Looking at your tools, skills and experiences above, you must be very expert and pretty sure you able to explain it to noob like me, please.

[PA4TIM]

A cap has a certain impedance at a certain frequency. In that case the signal must be a sinewave by nature of the principle and math. And impedance is tghe "resistance" for AC signals. Just like you measure a resistance while using a known current or voltage or a ratio measurement (an unknown in series with a well known resistor) you can use a siewave current or voltage to measure impedance for a resistor, coil or capacitor.

A square wave in not "1" frequency like the frequency in the reactance formula. So it will have a lower reactance for the fast components in the squarwave. Impedance |Z| is the absolute value of real R + imaginair R, and split it is Z=R+jX where jX is negative for a cap.

[oldway]

ESR schematic:
C8 = 470pF
R7 = 5K6
R21 = 56K

[codeboy2k]

since PA4TIM posted the method... I'm working on an LCR meter that calculates impedance exactly like that.. |Z| = R+jX. It uses a 4-wire measurement, putting a 600mV p-p up to 1.5V p-p peak sine wave through the DUT, measure voltage and current and determine if the phase is leading or lagging...I'm using an MSP430 and its onboard 16-bit ADC. I am not sure if I want to use another processor yet, or if I want to use an external ADC, but the MSP430 seems ok for the job right now.

I only have the front end prototyped up, but I am not sure of the opamps I've chosen, and I may change them out still, but they seem ok so far. Just I haven't finished testing all the ranges. And since I didn't want to use $11.00 450Mhz opamps, I chose the cheaper ones with GBW up to 50Mhz, these are just between $3.50 and $4.50 on digikey for duals. I also tried cheap CFB triple video amps in one device (3 for less than $3.00), and they had great bandwidth out to 200Mhz or more, but I couldn't get them to go below 100Hz, and I was shooting for 20Hz to 1 Mhz flat, which I am able to get now with the amps I am using, just with some finicky bits.

The sine wave passes though the DUT and I differentially measure both the voltage across the DUT and the current across switched sense resistors. 

The frequency response is flat from 20Hz up to 1Mhz, that took me a while to get it flat. It's literally .058 dB down at 1Mhz using those parts, but I've found it susceptible to stray capacitances causing peaking at 10Mhz or so, which causes small gain errors at 1Mhz. That may be unavoidable without moving up to a 100Mhz or higher part. I'm trying to keep this a CHEAP high-performance swept LCR meter from 20Hz to 1Mhz with reasonable performance comparable to the $4000 meters.

I have found that I can measure the voltage across the DUT and currents through the DUT to within .01% accuracy, from 20Hz to 1Mhz, but I haven't completed testing with all the ranges.  Maybe a final build will end up with 0.1% guaranteed accuracy as an LCR meter.  I am hoping for that at least.

Cheers!

[PA4TIM]

Codeboy, that sounds as an very interesting project. Almost a vna like system. Very impressive results. I know how hard it is to get such a flat responce over that range. I have not tested CFB opamps. I use LT1028 and OPA37. The GBW is OK but I want to try opamps with a faster slewrate (I used good IC feed so it is not much work to change opamps in a later stage.(I need arounsd 10 of those so that is a bit expensive. I want to try a very fast LT opamp a friend tested, but those are around 8 euro a piece so I do not know yet is the extra investment makes it so more usable as it is now (I can measure power from DC to 500 kHz and 0V to 300V and 0- 10 A. RMS AC+DC , PF, phase, current, voltage, shunt temp, Watt and VA and I will use it 90% of the time to monitor 230V and current to the instruments I'm repairing or building)

I just bought a launchpad but have not used it much yet because I only have 5V displays here and I'm rather new in digital. I now use the arduino ide and a tried energy but I am busy learning normal C.

Are you planning to make it sweeping ? A sawtooth X out to the scope and a two DAC's to make the Y signals. Or a small graphic lcd display (i'm learning C to  couple analog design with digital, not to replace it. More like 1+1=3) My first project was a 2GHz counter and RF powermeter added as readout and powercontrol to an analog project I just finnished (a 25MHz to 2000MHz sweeping signal generator)

In the powermeter I'm working on, I use a LTC2400 as ADC. Not needed it but I had a testboard in my junkbox for some time. That is the first time I use it but it works great (thanks to code I found and use as base and the SPI header)

[dannyf]

I can measure the voltage across the DUT and currents through the DUT to within .01% accuracy

That would be an impressive (almost impossible I think) result on a production unit. It is fairly doable on a one-off unit, with calibration.

[codeboy2k]

Codeboy, that sounds as an very interesting project. Almost a vna like system. Very impressive results. I know how hard it is to get such a flat responce over that range. I have not tested CFB opamps. I use LT1028 and OPA37. The GBW is OK but I want to try opamps with a faster slewrate (I used good IC feed so it is not much work to change opamps in a later stage.(I need arounsd 10 of those so that is a bit expensive. I want to try a very fast LT opamp a friend tested, but those are around 8 euro a piece so I do not know yet is the extra investment makes it so more usable as it is now (I can measure power from DC to 500 kHz and 0V to 300V and 0- 10 A. RMS AC+DC , PF, phase, current, voltage, shunt temp, Watt and VA and I will use it 90% of the time to monitor 230V and current to the instruments I'm repairing or building)

Thanks. Yes, I wanted to have a different LCR meter, that wasn't just a few fixed frequencies, but I could set any test frequency from 20Hz to 1Mhz  or sweep over the entire range and plot the output.

Yes, it's very hard to get the signal chain flat from 20Hz to 1Mhz, and in the end if I build it up on a PCB I might not be able to get it that good anyways; I will probably have to include some code to compensate; probably do a baseline sweep and use that to compensate for offset and gain errors in the signal chain.  So maybe it's trying so hard to get it flat is not worth it after all and just  a few dB down at the top and bottom end can always be compensated for in software, but I took a hardware approach and did my best to get it at flat as possible.    I'm using two OPA727 and an LT6231 in a standard instrumentation amplifier configuration.  I'm using the OPA727 (GBW=20Mhz, SR=30V/us) as gain=2 buffer, at the top and bottom of the DUT, followed by a LT6231 (GBW=215MHZ, SR=70V/us).  There are two LT6231's, one is gain=2.5 (total gain=5) and the other is gain=2, for a total gain=10).  I am probably going to have to add 2 more LT6231's in cascaded x2, x5 gains to get another gain x10, and reach a total gain of 100 (or a single LT6231-10 in gain=10, depending on price).  The LT6231 has higher input offset voltages, but at the ends of the gain stage it is less and less of the error budget.

I like the high common-mode input impedance of the LT1028, it would make a nice buffer, but I don't think it's slew rate is high enough for signals larger than 1V @ 1Mhz. 

Anything really good is $10, $15 or more.

Are you planning to make it sweeping ? A sawtooth X out to the scope and a two DAC's to make the Y signals. Or a small graphic lcd display (i'm learning C to  couple analog design with digital, not to replace it. More like 1+1=3) My first project was a 2GHz counter and RF powermeter added as readout and powercontrol to an analog project I just finished (a 25MHz to 2000MHz sweeping signal generator)

Yes, I will make it settable test frequencies and sweeping across the range.  I wasn't planning on scope output, but that sounds interesting.
Actually, I wanted to avoid a front panel at first, so I've just considered it to be USB controlled, and software on the PC. Then sweep it and graph the result on the PC like a VNA.  I'll probably start like that.

In the powermeter I'm working on, I use a LTC2400 as ADC. Not needed it but I had a testboard in my junkbox for some time. That is the first time I use it but it works great (thanks to code I found and use as base and the SPI header)

Yes, I'm also considering whether I will use an off chip ADC like the LTC2400 too.  This might improve the noise level as well, since I hear the horror stories of using a high-resolution ADC that's on the same substrate as the digitally switching micro... And with an off board ADC then I can use a (cheaper?) micro with a USB built in and no concern for it's ADC capabilities.  As it stands, it's hard to find a reasonable cost micro with 16 or 24bit ADC and USB. It's easy to find a micro with USB and a 12-bit ADC, or a micro with no USB and a 16-bit or 24bit ADC.   And I can always use one of the USB to Serial chips from FTDI, but I don't want to go that way if I can use a micro with built in USB.

Right now I'm at the early stages, and just seeing what performance I can get and where I can go with it.

That would be an impressive (almost impossible I think) result on a production unit. It is fairly doable on a one-off unit, with calibration.

Yes, I think so too.  I can get it with my hand-tweaking on the bench, but I want to be able to specify say, a 2.2pf feedback bypass cap for gain limiting to reduce the peaking, and not be way off because the tolerance of the cap actually changes the gain peak so much that I get gains errors anyways (and thus can't read the voltage across the DUT accurately anymore).

So ultimately, I think all my effort to get a flat response like this might not be worth it, and in the end I'll have to run a calibration cycle anyways to get the baseline as built, and use that baseline to compensate and flatten the signal chain.

[Jay_Diddy_B]

Hi Group,

I have been experimenting with the effects of sine versus square wave excitation and the interaction with the synchronous detector.

Here is the first LTspice mode.



I have a made a difficult circuit to analyze consisting of a 2.2uF capacitor, a 0.1 Ohm resistor and a 1uF capacitor.

I have implemented the synchronous detector with a circuit that has a gain of +1 for one half cycle and -1 for the other half cycle. This is feed to an averaging filter. This circuit gives that magnitude of the in-phase component, the ESR. The current source is implemented as relatively high voltage and a resistor, this is close enough to an ideal current source for the experiment.

One of the models has a sine wave source and the other model has a square wave source.


These are the voltage waveforms of the circuits under test:



The shapes are very different. The sine wave case results in a sine wave. The square wave case has pulse which come from V=L di/dt from the edges. and then dv/dt = I /C for the triangular portion.



These are the waveforms at the output of the phase detector, before averaging. If the timing is perfect, the components from inductance and the capacitance average to zero and you are left with the real component.

Here is the output from the filter. It is the same for both the sine wave and the square wave case.




Introduce a small delay.

Have added a small delay, only 50ns, to the synchronous detector timing in this model:



The output of the synchronous detector changes to these wave forms:



After averaging, there is little change to the sine wave circuit and a relatively large change to the square wave circuit.




This effect is only really sensitive to inductance. The effect of the capacitor is much less.

I have attached the LTspice models.

Jay_Diddy_B

[oldway]

Very intesting: it show that choosing sine or square wave does not matter if there is no or little delay in the synchronous detector.
Using a 4013B to switch the synchronous detector should reach this goal.
It proove also that the option of 2 frequencies (100Khz and 10Khz) is far better, as the changes will be 10 times reduced with 10Khz instead of 100Khz.

[Jay_Diddy_B]

Hi,

The main thing I see, if you can arrange the switches to come on for the central part of the square wave in each half cycle, the circuit will not see the effects of the caused by the inductance. I am trying to think of a simple way to generate the required signals.

Here is the model:



The switches S3 and S4 are turned on in the center of each half cycle. This is timing required for the switches:



The result:



The circuit is no longer sensitive to the inductor. The capacitive part is left, but it averages to zero.

I have attached the LTspice model.

Jay_Diddy_B

[oldway]

The main thing I see, if you can arrange the switches to come on for the central part of the square wave in each half cycle, the circuit will not see the effects of the caused by the inductance. I am trying to think of a simple way to generate the required signals.

A short delay of the on stage of the synchroneous rectifier switches would do the same and is easy to do.
But is it necessary?
Before to modificate a project, you should know if it's worth.

[Jay_Diddy_B]

Hi,
 A short delay is not sufficient. If you use a delay you will pick up the inductive pulse at the end of the half-cycle. You need to delay the start and turn the switch off before the end of the cycle.

This analysis is really academic. The design as published will find bad caps.  I works absolutely fine with capacitors 10uf and greater. A test lead inductance of 1uH causes about 0.1 Ohm the reading only if the 2-wire configuration is used.

I am conducting the analysis to reveal why some methods are better than others and what the deficiencies are. You can then use this information to predict how well other ESR meters will work.

Jay_Diddy_B

[BravoV]

My brain hurts just thinking about generating or orchestrating that pulses, a 4017 decade counter ?  Assuming ruling out using mcu, just a noob wild guess here. 

[codeboy2k]

My brain hurts just thinking about generating or orchestrating that pulses, a 4017 decade counter ?  Assuming ruling out using mcu, just a noob wild guess here. 

Those S3 S4 waveforms are similar to a synchronous buck converter top and bottom gate drives with dead time . I know it's overkill, but you could use a controller with external sync and lock it to your test frequency. The LT3838-1, for example, has an external PLL input and an internal VCO that can be phase locked to the external signal.  If you wire up the voltage and current feedback pins to reasonable values, the controller won't know any better

Unfortunately, it's not actually a viable solution. These controllers aren't designed to go much below 50KHz and the LT3838-1 in particular has an internal VCO with a lower limit of 140KHz.

But with a 50% duty cycle, the Top SW and Bottom SW would give the needed waveforms from Jay's spice run , syncronized to the source

If anything, perhaps the study of the internals of other synchronous controllers with dead time will give a (analog) solution.

[BravoV]

What I have in mind is, with the 10 counting pulses, as source by tapping output 0 and 5 is for main 50% pulse, while ouput 1 and 3 in pair with output 6 and 9 which have narrower pulse for both S3 and S4 respectively, cmiiw. 

[oldway]

My brain hurts just thinking about generating or orchestrating that pulses, a 4017 decade counter ?  Assuming ruling out using mcu, just a noob wild guess here. 

Very simple, have a look how i should do this using the triangular wave of the 555.
Schematic is only to explain my ideas, it has not been tested.

I modified several times the schematic, my apologies for that.

[BravoV]

oldway, thanks, learned something new here today ! 

[Jay_Diddy_B]

Hi,

I like the idea of using a 4017. I would actually use the 74Hc4017.

I would clock the part at 1MHz. The carry output, pin 12, would drive the analog switches in the source. This pin has a 100kHz 50% duty cycle.

I would use a 74HC4075 to combine outputs 1,2 & 3 for S3 and combine outputs 6,7 & 8 for S4.

I would use 74HC series because it is much faster than standard 4000 series CMOS.

Jay_Diddy_B

[BravoV]

Thanks JDB, just happy to hear the self taught and spent many days/weeks learning basic logic stuffs and Karnaugh map does mean something, even though just an idea.

Aware that your original version should be more than enough to rule out bad or good caps, its just I'm so fascinated and intrigued to build it, btw, is there an official name for such measuring technique ?

[oldway]

Offset compensation of the LM324 op-amp:
See schematic.