All about Keithley DMM7510. Bugs and features, recipes, advice, notes.

[essele]

The reduced resolution is really odd. There may be  reason to use reduced resolution for intermediate data to save on the memory - though still odd, as 1 µV resolution would be more than 24 bits and 32 bits would give enough resolution.

Yes, I had assumed this originally, but the internal data is absoutlely fine (that's what I used to do the excel graph) and you can zoom in and see the full resolution, so it's purely some unnecessary (in my view) rounding ... may be for performance, but I can't really see it.

There are so many horrible (but easy to fix) things with this meter .. just looking at the above screenshot ... why would you show min, max, and avg to 10uV resolution? .. it just makes them useless, I think I'd rather they weren't there at all!

There is another not so nice feature that was noted before: the grid lines are 1.1 µV apart, which is an odd choice. At least it looks like exactly 1.1 with no extra rounding error at the labels.

Don't get me started on the auto-scaling!  I keep meaning to capture screenshots of all the bizarre ways this manifests itself. I can understand the 1.1uV from a pure "best efforts" scaling perspective, but we should have the ability to set a min and max ... being only able to set to factors of 10 per division is awful (and again I don't understand this, their autoscaling can set it to whatever it wants) but we can only do 1uV, 10uV, 100uV etc. Ugh!

The graphing has the potential to be so much better than others (like the 34470 for example), but the implementation just lets it down all over the place. Still, I guess you can always export the data.

[MegaVolt]

print(display.lightstate) always returns display.STATE_LCD_OFF after power-up

At the same time, the device screen is turned on.
This variable starts to work normally only after writing any value to it.

[MegaVolt]

The device allows you to set NPLC with an accuracy of 1 ns, for example 10.001 μs. Does that make any sense?

[Kleinstein]

It could make sense to set the integration time to a specific value with rather high resolution: if there is some extra low frequency signal floating around not related to mains, one could set the time to get good suppression of this frequency.  No real need for 1ns resolution, but it does not really hurt. It is more like odd how they implement this. I could understand 1 µs or 100 ns as the clock used.

[MegaVolt]

It is more like odd how they implement this. I could understand 1 µs or 100 ns as the clock used.

I also wonder what the real step is. We’ll have to put on experiments and try to understand how the device works

[MegaVolt]

It is more like odd how they implement this. I could understand 1 µs or 100 ns as the clock used.

Everything turned out to be much simpler. A lot of decimal places is just math. In reality, the integration time directly depends on the 48 MHz oscillator according to the formula N = 300 + k * 400 clock cycles of the oscillator.
This is true for a network frequency of 50 Hz. Does this formula change for the network from 60 Hz I can not check

The consequences:
1 NPLC is not exactly equal to 0.02c it can be or 0.0199979167 s
or 0.0200062500 s

And I also understood why the generator is 48 MHz and not 50. 50 MHz is badly divided by 60

[MegaVolt]

I hurried. The formula 300 + k * 400 applies only to the numbers on the screen. Real accumulation time has a different formula with the same step. Those. what is shown on the screen does not match what is set physically 

 I will continue the investigation. Students who programmed this device should be fired

[MegaVolt]

Here is a table of what happens in the device. It's hard for me to explain the logic of what is happening. The selected integration time is not even in the middle of the range.

But eat and plus. If you set the integration time as a multiple of 8.33 (3) ms, then the displayed and real time will coincide.

[Kleinstein]

The 8.3 µs make some sense as the period of the run-up phase feedback.  One is not limited to these steps, but it makes things easier.   So setting the integration time in ns steps is a bit dishonest.  Only 8.3 µs steps could be slightly limiting in some cases, but is not that bad.

[MegaVolt]

The 8.3 µs make some sense as the period of the run-up phase feedback.  One is not limited to these steps, but it makes things easier.   So setting the integration time in ns steps is a bit dishonest.  Only 8.3 µs steps could be slightly limiting in some cases, but is not that bad.

Most of all, I don’t like the fact that I have to get these data and numbers myself. I cannot read them in user manual.

[Kleinstein]

The 8.3 µs make some sense as the period of the run-up phase feedback.  One is not limited to these steps, but it makes things easier.   So setting the integration time in ns steps is a bit dishonest.  Only 8.3 µs steps could be slightly limiting in some cases, but is not that bad.

Most of all, I don’t like the fact that I have to get these data and numbers myself. I cannot read them in user manual.

I absolutely agree with that.

With modern instruments the manuals get increasingly confusing and the specs get unclear, missing details or the relevant part to compare different products. For the DMM7510 I can absolute understand that they don't want to show the 100 PLC noise. So the specs tend to leave out the weak points.

In part this is because of the flood of new features and options and planed additions to the software.  So they have to release a manual before the software is really ready (if it ever gets  ). Fixing the bugs is usually slow and updating the manuals is usually even slower. When writing instructions one tends to focus on the new / special features, so one may miss one the classical important part.

The manual is already quite long and is still missing quite a bit on app programming.

[MegaVolt]

Yes that's right

At the expense of noise in my opinion the data is outdated. I'm busy measuring noise right now. Soon I will share the results.

[MegaVolt]

I found that DCV noise in the 0.1V range changes little until the integration time is 0.007s
It became interesting to me how this noise looks and it looks like a certain periodic signal

Aperture = 0,000475 s

[MegaVolt]

During the night I collected more data and there is a similar generation for long integration times. For example, here is a picture for NPLC = 8.5. And similar pictures for all times of integration of a kind 5.5; 6.5; 7.5;
It looks like a network frequency.

[Kleinstein]

Using things like 7.5 PLC makes the meter sensitive to mains hum (50 Hz, but not 100 Hz). Due to the time needed for rundown, there is a beat frequency and this would be visible. So the amplitude seen here is likely from the 50 Hz hum. This is why the preferred modes are integer PLC numbers.

For the much faster test in the digitizing mode one may see residues of ripple from an chopper OP too.

[MegaVolt]

Using things like 7.5 PLC makes the meter sensitive to mains hum (50 Hz, but not 100 Hz). Due to the time needed for rundown, there is a beat frequency and this would be visible. So the amplitude seen here is likely from the 50 Hz hum. This is why the preferred modes are integer PLC numbers.

Yes, I understand this feature of integrating ADCs. Everyone has it and it seems even 3458a judging by the picture in the application.

For the much faster test in the digitizing mode one may see residues of ripple from an chopper OP too.

You express your thoughts very briefly. And I often cannot understand because the translator often loses some of the meaning. Could you write a little more about the idea of this test. If we talk about the noise that I see with DigiV, then it does not look like 50Hz. I see 14KHz noises I wrote about them above.

[MegaVolt]

Here are the noise graphs. For the rms value and for the constant component (Vm0).

For the DC component and the 0.1 V range, there are curious points of 0.1; 0.35; 0.6 NPLC they give a big shift but in fact there is a periodic process that is longer than 1000 samples. Therefore, we see the error of expectation because we consider it for part of the period.

I attached a picture taken for 100 thousand counts.

[MegaVolt]

Looking at the similarity of the 1 V and 100 V graphs, there is a suspicion that they divide 100V by 1000 and then process it as a 1V signal.

For channels 1V and 0.1V, 5 NPLC is really the best, followed by 1 and 2 NPLC. I believe in these ranges software averaging of samples with 1 NPLC will give good results.

I'll check it a little later.

[Kleinstein]

Using a divide by 100 and than use the 1 V range is the normal path for the 100 V range in many meters, so no surprise here.

The extra periodic background looks strange with steps in frequency. The frequency is likely some beat frequency with mains hum. With some extra time for signal processing / run-down 10 readings at 0.1 PLC would likely take a little longer than 20 ms and thus some phase shift with every reading. The mains frequency will also have an effect, but I doubt the more discrete frequency ranges would be due to the mains frequency.
The periodic signal would also cause the plateau at low PLCs for the 0.1 V range, as different PLC settings would mainly effect the frequency and less the amplitude.

[MegaVolt]

Using a divide by 100 and than use the 1 V range is the normal path for the 100 V range in many meters, so no surprise here.

I understand the essence of this decision. One divider instead of two. This saves 1 precision resistor. And this is expected in a budget multimeter. But in 7.5-digit, I thought there would be no such savings. Especially considering that the amplifier standing on the 1V input spoils the noise characteristics.

The extra periodic background looks strange with steps in frequency. The frequency is likely some beat frequency with mains hum. With some extra time for signal processing / run-down 10 readings at 0.1 PLC would likely take a little longer than 20 ms and thus some phase shift with every reading. The mains frequency will also have an effect, but I doubt the more discrete frequency ranges would be due to the mains frequency.
The periodic signal would also cause the plateau at low PLCs for the 0.1 V range, as different PLC settings would mainly effect the frequency and less the amplitude.

Forgot to add. If you disable AZ, then the cunning envelope disappears. And there remains just a certain frequency with a constant amplitude. The signal does not become smaller in amplitude but looks prettier.

[Kleinstein]

Normally the amplifier only add a little noise to the 1 V range. Much of the noise is often still from the ADC (and the references of cause). So the way with only one divider does not add that much noise.  The standard high voltage divider are this way, so even 8 digit meters like 3458, Datron 1281 and Keithley 2002 use only 1 divider, though here it may actually help to have 2 divider settings. Switching off the 1:10 divider may however need a relay for switching (or switch at the low side and this a variable impedance), and this can add more trouble than using just 1 divider.

Forgot to add. If you disable AZ, then the cunning envelope disappears. And there remains just a certain frequency with a constant amplitude. The signal does not become smaller in amplitude but looks prettier.

That is interesting: so the low frequency part is more like an aliasing of the higher frequency part.

The very fast readings well below 1 PLC are often used without auto zero, as this about speed and not so much about accuracy.

[MegaVolt]

New firmware released.

https://www.tek.com/digital-multimeter/dmm7510-software/model-dmm7510-firmware-revision-172-and-release-notes

[MegaVolt]

First impressions:

Fixed working with memory. For the first time I was able to allocate a buffer for 8.3 million samples.
But unfortunately, as before, it cannot be filled with data during digitization. Somewhere after 4 million there is a pause in filling the buffer

Kilo degrees left.

[MegaVolt]

I updated the new version and after ACAL I found an additional zero offset of 200 ... 300 nV.
It was in the region of 300-400 became 500-800 ....

Does anyone have a similar effect?

The only thing that I made a mistake and confused ACAL after flashing without turning off the device. That is, he rebooted himself, but I did not remove the power completely. Then I found that the device was not behaving quite correctly (for example, I could not switch the input impedance) and turned off the power. After turning on and re ACAL, I see the same shift

Perhaps I ruined something. Or in the new firmware there were some changes about which were not said in the description

[E-Design]

I updated the new version and after ACAL I found an additional zero offset of 200 ... 300 nV.
It was in the region of 300-400 became 500-800 ....

Does anyone have a similar effect?

The only thing that I made a mistake and confused ACAL after flashing without turning off the device. That is, he rebooted himself, but I did not remove the power completely. Then I found that the device was not behaving quite correctly (for example, I could not switch the input impedance) and turned off the power. After turning on and re ACAL, I see the same shift

Perhaps I ruined something. Or in the new firmware there were some changes about which were not said in the description

Hi Megavolt, do you have a low thermal short for when doing zero calibration? Thermal effects from metal of the short and environment can cause voltage gradients of 100+ nV How much thermal settling time? No airflow blowing on the instrument? These are some of the considerations for you. Even a brief power cycle is enough to upset the offset 100nV due to the internal temperature change and re-settling.

Also there will not be any undocumented firmware changes that you didnt see in the description in case you worry about it. If firmware makes a change to calibration or how the measurements are built, it will be described there.