Sample and Hold IC - Peak Current Measurement

[fourierpwn]

Hi all,

Looking for a crash course/guidance to help speed things up for a project I have in mind.

I want to measure (fast(?)) pulsed current accurately, in a somewhat of an automated manner. Pulse widths on the order of single digit microseconds, potentially hundreds of nanoseconds. Ideally, a fast ADC (amongst other things) would do the job and allow to re-create the entire pulse waveform for post processing but I'm looking at alternatives as a shortcut as I can live with only sampling the peak current value once every pulse.

Sample and hold ICs with an aperture time of 1us or less are appealing; HA5351 (pricey), AD783/1 etc. I'm not sure how to integrate this into my project though - never dealt with such chips before. Pulsed current specs at the moment are ~10V (would like to have room for expansion to +28V max for future use) and around 5A max... Current sense resistor -> current sense amp (possibly an integrated current shunt monitor solution?) -> sample and hold IC -> 6.5 digit DMM with GPIB control to post process DMM reading and convert to current. Would this be the way to go? Not sure how to interface the sample and hold IC with the DMM though. Any input would be greatly appreciated.

Thanks!

[Benta]

Check out the LF398 page at TI.

https://www.ti.com/product/LF398-N

It probably doesn't have the specs you need, but there are some excellent application notes describing specifications and pitfalls which are really helpful.

[fcb]

Do you know when the sample is coming, i.e. do you have a triggering system in-place already?

Are you trying to build 1 or 100 or 10,000?  You indicated that some parts were expensive - does that mean you have a budget constraint?  If you are considering an external DMM does indicate low volume. A fast ADC feeding a fast micro or FPGA would be the simplest from a system POV if you're building many of them, else the HA5351 looks like quite a good solution (if you need the bandwidth) if you have a DMM. 

The droop on the S/H on the HA5351 (40MHz) is around 0.3uV/us, so you'll want to sample the voltage pretty quickly, the AD783JRZ (15MHz) is 0.02uV/us - a slower part, but slower to loose precision. Perhaps use a AD783 feeding a 18-24bit ADC, you could almost drive the ADC with an Arduino.

34465A (6.5digit) has an ext trig input and an Ethernet port - pretty easy to interface to, but you won't be sampling at 6.5digit resolution fast enough for droop not to introduce significant errors.

[Doctorandus_P]

You want to do a 6.5 digit measurement over a shunt resistor in a < 1us time frame?

What is the actual signal bandwidth of your current signal?
Just A 5A shunt that gives 6.5 digit accurate results at those frequencies would already be quite difficult I think, just because of the inductance of the shunt itself.

[mawyatt]

If I understand your needs correctly and you want to use a precision 6.5 digit DMM as the digitizer, consider two sequential S&Hs. First is fast with higher droop rate, followed by slower S&H with low droop rate. Small delay between 2nd and 1st trigger samples. Output of 1st feeds the input of the 2nd, and 2nd S&H output feeds the DMM.

Edit: As mentioned above, the shunt inductance equates to ~20pH for 100ppm error due to inductive peaking at 1us, assuming a shunt of 0.2 ohms (1V/5A).

Interesting project, sounds like fun, keep us posted on your progress.

Best,

[fourierpwn]

Do you know when the sample is coming, i.e. do you have a triggering system in-place already?

Not quite - triggering system is a bit ad-hoc at the moment, though I imagine that the sample would need to have some sort of delay on it to ensure that the pulse has settled etc. There will probably be a bit of debug work here, depending on how well the current pulse holds up with different load conditions.

Are you trying to build 1 or 100 or 10,000?  You indicated that some parts were expensive - does that mean you have a budget constraint?  If you are considering an external DMM does indicate low volume. A fast ADC feeding a fast micro or FPGA would be the simplest from a system POV if you're building many of them, else the HA5351 looks like quite a good solution (if you need the bandwidth) if you have a DMM. 

Just one system for now. Possibly a handful (5 max.) if there is the need. No real budget constraint although I don't intend on spending tens of thousands here, would just buy a commercial system to do the job. Fast ADC feeding a fast micro or FPGA would be future work, i.e., if sampling the pulse once isn't enough and I need to re-create the waveform - this is exactly what commercial systems (pulsed SMUs - DiVA, Auriga, AMCAD etc) do.

You want to do a 6.5 digit measurement over a shunt resistor in a < 1us time frame?

What is the actual signal bandwidth of your current signal?
Just A 5A shunt that gives 6.5 digit accurate results at those frequencies would already be quite difficult I think, just because of the inductance of the shunt itself.

I don't necessarily need the 6.5 digit resolution - would be nice for smaller currents (uA range) but the main application, for now, would only be concerned with mA resolution. I was avoiding having to write the code for the high speed ADC and micro/FPGA in order to save time and get some measurements happening ASAP. The idea of the DMM came to mind - whether it's practical or not, I don't know. Hence the thread.

Bottom line is I need to measure peak pulsed current with reasonable accuracy as soon as possible. The pulse train will pertain to the same conditions for its entirety, thus, I can live with only sampling one single pulse per bias condition. Bias condition will change (manually, not automated) and then a new pulse measurement will begin.

Not sure if any of this helps/makes sense - long day.

Thanks for the insights thus far! Pulsed measurements seem so simple on the surface but when you dig a little deeper they soon become tricky business!

[fcb]

Using a S/H would be the simplest for a very specific system.  But the devil will be in the triggering system:

If the pulses are uniform peak-amplitude then you could adopt a scheme like a sampling oscilloscope.  Ted Yapo did a very good TEDX style (it's on YT) talk about building one.  Essentially take a simple fast comparator and drive your S/H with a processor controlled variable delay.

If you need some-sort of single peak detect and the triggering isn't PERFECT then I'd stop now and develop a system based on a fast ADC and FPGA.

Another idea is a really fast ideal-diode rectifier, I haven't had much luck developing one for an application a few years ago (abandoned in-favour of fast sampling ADC/FPGA).

[David Hess]

A sample-and-hold is suitable if you know when the peak is.  What you are looking for is a peak-detector or track-and-hold.

The top half of a diode bridge can be used to make a simple high bandwidth peak-detector; replace the diodes with transistors for better accuracy.  Another high performance way is to use a fast comparator to compare the input to the output, and activate the track-and-hold as long as the output is less than the input.

6.5 accuracy is out of the question though.   That will require some type of active feedback which is unsuitable for a very fast detection time.  There are some newer very fast 16-bit accurate operational amplifiers which might do it, but that is still only 5 digits.

[fcb]

Actually David Hess's comments lead to another idea.

If you don't know exactly when the pulse peak is BUT can rely on multiple pulses being the same height, then there is pretty simple way to do this.

You could use a fast (precise) comparator (LT1016?) connected to your signal AND a precision DAC - simply build a successive approximation ADC.

[fourierpwn]

A sample-and-hold is suitable if you know when the peak is.  What you are looking for is a peak-detector or track-and-hold.

The top half of a diode bridge can be used to make a simple high bandwidth peak-detector; replace the diodes with transistors for better accuracy.  Another high performance way is to use a fast comparator to compare the input to the output, and activate the track-and-hold as long as the output is less than the input.

6.5 accuracy is out of the question though.   That will require some type of active feedback which is unsuitable for a very fast detection time.  There are some newer very fast 16-bit accurate operational amplifiers which might do it, but that is still only 5 digits.

Thanks for the reply. A concern I have with peak detectors, which I did consider, is overshoot. Will the peak detector take the overshot value of the pulse (if there is any present) as the detected peak and report a current value higher than what the settled value is? Granted that the peak detector is fast enough to detect such pulses.

The pulse train is set by an arb function gen., there may be some minor variation/delays between input and output but I still have reasonable control over when the pulses are coming.

If you don't know exactly when the pulse peak is BUT can rely on multiple pulses being the same height, then there is pretty simple way to do this.

You could use a fast (precise) comparator (LT1016?) connected to your signal AND a precision DAC - simply build a successive approximation ADC.

Interesting idea. For some circumstances this will work quite well. However, for other instances the load will be driven into compression and so the (pulse) current will vary.

[fcb]

Probably need more infomation on the application, pulse rate, pulse length etc.. (basically all the proprietary stuff  ) to comment much further.
This thread has a bunch of good ideas, you'll have to filter them through your knowledge of the application.

[David Hess]

A concern I have with peak detectors, which I did consider, is overshoot. Will the peak detector take the overshot value of the pulse (if there is any present) as the detected peak and report a current value higher than what the settled value is? Granted that the peak detector is fast enough to detect such pulses.

Of course the peak detector will detect any overshoot if present within its bandwidth and slew rate capabilities.