DMM6500 & DMM7510 vs, 34465A & 34470A AC Readings

[mawyatt]

Anyone know the relative merits of these DMMS regarding ability to accurately measure AC RMS waveforms like Sine, Triangle, Trapezoid and Square Waves up to ~100KHz from DMM low noise levels to 100VPP? All claim True RMS, and believe the KS use a computed method, but not sure how Keithley does this?

Also, any optinions on overall performance from those that actually have worked with these DMMs is appreciated.

Best,

[HKJ]

I did measure the frequency response of some DMM's, including there:
https://lygte-info.dk/info/DMMFrequencyResponse%20UK.html

All is with sinus at 1V

[bdunham7]

accurately measure AC RMS waveforms like Sine, Triangle, Trapezoid and Square Waves up to ~100KHz from DMM low noise levels to 100VPP?

Before you look at the differences between analog TRMS converters, digital subsampling and so on, you should consider what the bandwidth of those signals would be if they have a periodic rate of 100kHz.  EEVBlogger HKJ wrote a nice article that gives BW measurements to 500kHz for a variety of multimeters, including all four that you have listed.  The Keysights hit brick walls at 300kHz, the Keithleys actually have a bit of a peak at HF, which is a feature of the compensation network and adjustment.

https://lygte-info.dk/info/DMMFrequencyResponse%20UK.html

Edit:  I see HKJ himself has beaten me to it!

And I should add that each range needs to be compensated individually, so there are typically slightly different responses in each range.  And there are dV/dt (V*Hz) limits--a +/-50V square wave at 100kHz may be excessive and result in an erroneous reading or even meter damage.

[mawyatt]

Before you look at the differences between analog TRMS converters, digital subsampling and so on, you should consider what the bandwidth of those signals would be if they have a periodic rate of 100kHz.  EEVBlogger HKJ wrote a nice article that gives BW measurements to 500kHz for a variety of multimeters, including all four that you have listed.  The Keysights hit brick walls at 300kHz, the Keithleys actually have a bit of a peak at HF, which is a feature of the compensation network and adjustment.

https://lygte-info.dk/info/DMMFrequencyResponse%20UK.html

Edit:  I see HKJ himself has beaten me to it!

And I should add that each range needs to be compensated individually, so there are typically slightly different responses in each range.  And there are dV/dt (V*Hz) limits--a +/-50V square wave at 100kHz may be excessive and result in an erroneous reading or even meter damage.

Already looked into the BW and slew rate long before posting, the values of 100KHz and 100VPP are the ultimate extremes. Realistic measurements will likely center around 1KHz and under 20VPP, but voltage may venture to 100VPP but not at 100KHz for the squarewave.

Yes, I noted HKJ review which is very helpful

Best,

[2N3055]

Keithley DMM6500 is sampling and calculating RMS. That is certain and confirmed by Keithley engineer. It does not have RMS converter chip inside.
Like others said, it's frequency response is not well defined and varies between ranges.
I believe user HKJ did some measurements but not full characterization, if I remember correctly.
It was also mentioned few more times by some other users, but I forgot where exactly.

[mawyatt]

Thanks, do you know if the DMM7510 uses the same technique? In the teardown it seems it was showing a RMS converter chip.

Best,

[E-Design]

Thanks, do you know if the DMM7510 uses the same technique? In the teardown it seems it was showing a RMS converter chip.

Best,

I also worked on the DMM7510 - this design still uses the RMS converter chip.

[mawyatt]

Thanks, do you know if the DMM7510 uses the same technique? In the teardown it seems it was showing a RMS converter chip.

Best,

I also worked on the DMM7510 - this design still uses the RMS converter chip.

Thanks for the note. Any idea why they aren't using the RMS technique used in the DMM6500?

Best,

[mawyatt]

Here's a thought

Since E-Design indicates the DMM7500 uses a RMS converter chip and not a computational approach like the DMM6500. Could one use the Hi resolution and speed digitizer in the DMM7500 as a means to create a computational RMS approach and achieve a reasonable RMS result & speed? Don't know what the algorithms are utilized in the DMM6500 or the Keysight True RMS, but maybe this computational method could work at a higher accuracy/resolution than the RMS chip??

Best,

[E-Design]

Thanks, do you know if the DMM7510 uses the same technique? In the teardown it seems it was showing a RMS converter chip.

Best,

I also worked on the DMM7510 - this design still uses the RMS converter chip.

Thanks for the note. Any idea why they aren't using the RMS technique used in the DMM6500?

Best,

Yes, the 7510 is an older model where the design was leveraged from an older model in order to complete development faster.
In the 65xx case, the RMS converter IC was eliminated to reduce cost.

[bdunham7]

Here's a thought

Since E-Design indicates the DMM7500 uses a RMS converter chip and not a computational approach like the DMM6500. Could one use the Hi resolution and speed digitizer in the DMM7500 as a means to create a computational RMS approach and achieve a reasonable RMS result & speed? Don't know what the algorithms are utilized in the DMM6500 or the Keysight True RMS, but maybe this computational method could work at a higher accuracy/resolution than the RMS chip??

I think you could make it work, but would it be better?  Look at the specs.  At 10V 10kHz the DMM7510 is 900ppm and the 34470A is 700ppm, not a big difference even with a highly optimized digital sampling system.  Even the HP 3458A with its three AC modes uses the analog TRMS converter for the best accuracy of any signal that isn't perfectly repetitive, and that's 300ppm best-case.  The only thing that I know of that is going to be much better is a thermal transfer system.

[Kleinstein]

The spec limits reflect the estimates at the time when the meters were new, not necessary the actual performance. The analog RMS chip performance depends on the atual signal in a quite complicated way. One weakness there is ther performance with higher frequency at low amplitude (this is already the rectifier part usually used) and also the waveform can make a difference. Typically the BW of the analog RMS chips goes down with amplitude, which can complicate the interpretation with higher frequency components in the signal.

Besides the accuracy an advantage of the digital RMS is that the reaction is fast (it ususally uses a kind of FIR fitler). Analog RMS converter chips need quiet some time for settling of the filter. It can take seconds until the result settles to the full accuracy. So even if not getting better accuracy digital RMS would be a nice option for the DMM7510. The math for digital RMS is relatively simple, except if additional compensation for amplifer effects is needed.

The 34470 uses the same ADC (though at a much higher conversion speed) for the digital RMS. The high speed mode of the ADC may not have the same accuracy. This also applies to the 3458: in the high speed mode the INL is no longer that good and is known to be more non-linear, though details are rarely shown. For the high speed the separate ADC chip as used in the DMM6500 and DMM7510 has a chance to perform better than pushing the multislope ADC to the highest speed.

For the AC performance it is not only the ADC / RMS converter that matters. A big part is also the divider / amplifier compensation, especially for the higher frequencies.

[alm]

Another place where digital sampled AC can perform much better than analog conversion is for very low level AC signals: https://www.eevblog.com/forum/metrology/low-ac-voltage-measurement-issues/

I think you could make it work, but would it be better?  Look at the specs.  At 10V 10kHz the DMM7510 is 900ppm and the 34470A is 700ppm, not a big difference even with a highly optimized digital sampling system.  Even the HP 3458A with its three AC modes uses the analog TRMS converter for the best accuracy of any signal that isn't perfectly repetitive, and that's 300ppm best-case.  The only thing that I know of that is going to be much better is a thermal transfer system.

Swerlein's algorithm claims an accuracy of about 10 ppm, but is limited to (repetitive) sine waves up to 1 kHz. For arbitrary signals, I'm indeed not sure if sampling can be that much better.

[mawyatt]

Yes, the 7510 is an older model where the design was leveraged from an older model in order to complete development faster.
In the 65xx case, the RMS converter IC was eliminated to reduce cost.

Interesting, thanks for the details. Do you think the computational method used in the DMM6500 is any better than the RMS chip method used in the DMM7510? It seems Keysight moved away from the RMS chip some time ago for the higher end DMMs, would guess this was for better results with the computational method since it's utilized in the 3458A? Of course all this assumes you have a high enough resolution, accurate and speed digitizer to support the computational method.

Best,

[Kleinstein]

The analog RMS chips have there limitations, especially poor linerarity at the low end, amplitude dependent BW and the settling time from the filter.
The digital RMS gets around these 3 points. I would expect the digital method with the ADC from the 6500 to out-compete the analog RMS chips in most aspects.  The 2nd, fast ADC could be worth it for the AC part alone to save on the analog RMS chip.

The performance still depends on the front end. The compensation of the divider and amplfier can still be the main error source for both. For the higher frequency part there is also the principle point of loading the source: with usually some 1 M and maybe 100 pF in parallel the impedance is not that high compared to 50 ohms as a typical impedance for higher frequency circuits.  With an active input circuit the impedance can differ from the simpel R + C model. The data for the frequency response of the dmm6500 digitizer mode do not look that good. Chances are the digitizer part has a seprate and better tuned path for the AC part.
The DMM7510 may not have the separate path from the AC signal to the fast ADC. So a later added digital RMS may have lower performance.

Digital RMS could even offer options like a limited BW to reduce the noise or get more specific results (e.g. for noise measurements).
The slightly tricky part is getting the aperture time / filtering right with lower frequency AC, if not mains related. This can add a bit to the math and a little to the settling time.

[E-Design]

In addition to Kleinsteins points, when using a "digitizer" to perform AC RMS calcs, the input pathway tends to be optimized to support higher bandwidth --- and that invariably brings in more noise which plagues low amplitude input signals. Additionally, the input pathway frequency "flatness" (over a higher BW) becomes another problem that requires careful design and/or correction.

The other consideration for the end products is even if one can get an improved accuracy, it can only be specified for as good as one can verify it and make it traceable.

Do you have an application in mind that can make use of higher accuracy?

[mawyatt]

In addition to Kleinsteins points, when using a "digitizer" to perform AC RMS calcs, the input pathway tends to be optimized to support higher bandwidth --- and that invariably brings in more noise which plagues low amplitude input signals. Additionally, the input pathway frequency "flatness" (over a higher BW) becomes another problem that requires careful design and/or correction.

The other consideration for the end products is even if one can get an improved accuracy, it can only be specified for as good as one can verify it and make it traceable.

Do you have an application in mind that can make use of higher accuracy?

Thanks for the reply and information, much appreciated.

Yes we have an application, it's a new development project for a new type Phased Array System. The controller, basically from what I can say, it's like a 128 Independent Channel (will grow) AWG capable of 160VPP outputs with ~14 Bit Precision and each output can be Bi-Phase or Offset Phase modulated and must maintain millivolts levels of average offset over all waveform types. The thought was to use a better than what we already have DMM like the 34470 or DMM7510 as an AC baseline reference to use to compare the other DMMs with. We have a KS34465A and DMM6500, plus a SDM3065X and a pair of 34401As. We don't use the 34401As or SDM3065X much for this because of the slightly differing readings (they all use the RMS chip), however the KS34465A and DMM6500 do tend to agree well with various waveforms while watching things in real time, can't say much more at this point tho.

We thought about getting a next level up DMM as reference to compare with the KS3465A and DMM6500, but sounds as if we should get another KS3465A or DMM6500 or two with the computational RMS capability and use them as a "Voting" reference.

One of the reasons for the concern on my part about the RMS performance was not knowing how well things stand up to the various arbitrary waveforms which we obviously can't describe here.

Think you, Kleinstein and others have convinced me to not consider the DMM7510, or the KS34470A and stay with a KS34465A or DMM6500. We have a critical pair of demonstrations next month and in the middle of getting the hardware and first revision user manual documentation prepared (along with the Theory of Operation), so no time to fiddle around with multiple options and such.

Anyway, thanks for the help.

Best,

[Kleinstein]

For the AC performance there should be no big difference between the 34465 and 34470. The main difference between the 2 is the reference. The AC measurements are in a range where the LM399 is good enough.

For critical AC performance a high resolution scope (e.g. some pico-scope ones) may be worth a look too. The ADC resolution may not be as good, but chances are the amplifier is better at higher frequencies (e.g. > 100 kHz). A scope has a better defined input impedance.

[mawyatt]

Knew about the better references (LTZ1000) in the 34470, and with the DMM7510 (they evidently use a special Fluke LT reference). Agree the LM399 should be "good enough", especially the pre-aged versions used in the 34465, not sure what is used in the DMM6500.

We've been "looking" at the higher resolution scopes for some time now, and already contacted LeCroy and others last year. Also watched with keen interest the new Siglent SDS6000 scope and will be evaluating it soon, although it's the 8 bit ADC version and likely not good enough

The PicoScope also looks attractive, but prefer an full fledged instrument rather than one that requires a laptop.

Even looking into using a "known good reference" to generate some of the Arbitrary Waveforms and make measurements with the DMMs and "infer" results based upon mathematically computed results. The HP App note based upon Swerleins Algorithm that alm referred has a note on how NIST used a precise and stable 8 bit DAC to create accurate sinusoids with remarkable precision, so our thought was to use this method for some Arbitrary Waveforms. We have even considering for some time making a special custom calibrated source for these waveforms. Even looking into a precision AWG (like the upcoming SDG6000) for such but all this effort will be after the upcoming demos next month.

If the project moves on to the custom IC design/fabrication phase then we'll need a means to evaluate the test chip performance, and preparing for this.   

So yes we've been looking into this for some time now.

Best,

[tautech]

Even looking into a precision AWG (like the upcoming SDG6000) for such but all this effort will be after the upcoming demos next month.

SDG7000A. 
SDG6000X is an existing range.

[2N3055]

Knew about the better references (LTZ1000) in the 34470, and with the DMM7510 (they evidently use a special Fluke LT reference). Agree the LM399 should be "good enough", especially the pre-aged versions used in the 34465, not sure what is used in the DMM6500.

We've been "looking" at the higher resolution scopes for some time now, and already contacted LeCroy and others last year. Also watched with keen interest the new Siglent SDS6000 scope and will be evaluating it soon, although it's the 8 bit ADC version and likely not good enough

The PicoScope also looks attractive, but prefer an full fledged instrument rather than one that requires a laptop.

Even looking into using a "known good reference" to generate some of the Arbitrary Waveforms and make measurements with the DMMs and "infer" results based upon mathematically computed results. The HP App note based upon Swerleins Algorithm that alm referred has a note on how NIST used a precise and stable 8 bit DAC to create accurate sinusoids with remarkable precision, so our thought was to use this method for some Arbitrary Waveforms. We have even considering for some time making a special custom calibrated source for these waveforms. Even looking into a precision AWG (like the upcoming SDG6000) for such but all this effort will be after the upcoming demos next month.

If the project moves on to the custom IC design/fabrication phase then we'll need a means to evaluate the test chip performance, and preparing for this.   

So yes we've been looking into this for some time now.

Best,

Creating precise AWG waveforms is not impossible. Amplifying and scaling them to 160V p-p with good pulse response and good DC accuracy is the fancy part. Especially into capacitive load, and if you need to do that fast..


Like Tautech says, new Siglent AWG is SDS7000A. Up to 1 GHz for sinewaves, 14 bit res, 24Vp-p output wit +-12V (24V total) offset range for total of +-24V output range, differential outputs, digital pattern generation and all kinds of goodies...
That will be interesting one..

[mawyatt]

Even looking into a precision AWG (like the upcoming SDG6000) for such but all this effort will be after the upcoming demos next month.

SDG7000A. 
SDG6000X is an existing range.

Yep, SDG7000, my bad

Best,

[mawyatt]

Knew about the better references (LTZ1000) in the 34470, and with the DMM7510 (they evidently use a special Fluke LT reference). Agree the LM399 should be "good enough", especially the pre-aged versions used in the 34465, not sure what is used in the DMM6500.

We've been "looking" at the higher resolution scopes for some time now, and already contacted LeCroy and others last year. Also watched with keen interest the new Siglent SDS6000 scope and will be evaluating it soon, although it's the 8 bit ADC version and likely not good enough

The PicoScope also looks attractive, but prefer an full fledged instrument rather than one that requires a laptop.

Even looking into using a "known good reference" to generate some of the Arbitrary Waveforms and make measurements with the DMMs and "infer" results based upon mathematically computed results. The HP App note based upon Swerleins Algorithm that alm referred has a note on how NIST used a precise and stable 8 bit DAC to create accurate sinusoids with remarkable precision, so our thought was to use this method for some Arbitrary Waveforms. We have even considering for some time making a special custom calibrated source for these waveforms. Even looking into a precision AWG (like the upcoming SDG6000) for such but all this effort will be after the upcoming demos next month.

If the project moves on to the custom IC design/fabrication phase then we'll need a means to evaluate the test chip performance, and preparing for this.   

So yes we've been looking into this for some time now.

Best,

Creating precise AWG waveforms is not impossible. Amplifying and scaling them to 160V p-p with good pulse response and good DC accuracy is the fancy part. Especially into capacitive load, and if you need to do that fast..


Like Tautech says, new Siglent AWG is SDS7000A. Up to 1 GHz for sinewaves, 14 bit res, 24Vp-p output wit +-12V (24V total) offset range for total of +-24V output range, differential outputs, digital pattern generation and all kinds of goodies...
That will be interesting one..

Agree, the HV scaling with good DC performance as well as capacitive loads and such is difficult indeed, which we've conquered 

Had thought the new SDG7000 series was going to be 16 bit DACs, but as you mention only 14 bit resolution. It's also going to be priced in the Keysight range, so likely well out of our projected budget.

So now the SDG6000X series is looking as a possible candidate. Can these be expanded to higher speeds with firmware like the SDG2000X series, where the hardware is all the same between various models?

My thinking at this point is to create a routine where the AWG output levels are recorded with a good DMM (6500 or 34465A) over the entire 16 bit range, like the NIST approach mentioned in the HP App note based upon Swerleins Algorithm mentioned earlier. This data would be used as a correction factor with the actual desired waveform under test.

Here's an image of 64 channels in operation, a quick single value magnitude calibration was done but not the offset. The offset is reading ~15mv @ 140VPP which is below what we believe we'll need of ~18mv worst case, but with an offset calibration this should drop to a few millivolts. Since we don't know how sensitive the "System" is to these things we've over designed, but do know that a small amount of long term offset can destroy the sensor, folks already found this out

You can see the KS34465A and DMM6500 readings at ~70Vrms, both are using slow filters, smoothing and scaling, and both read ~130uv SD.

Anyway, getting way ahead on myself since this all hinges upon the upcoming demos next month. Now back to writing/editing the Operations and Theory Manuals.

Edit: Update, let the DMM6500 and KS34465A run for a couple hours with a 25Vrms input from the "System", they both agree within 5ppm 

Best,