Products > Test Equipment
Some old school instruments showing how it's done (HP 3325A and Fluke 8506a)
SilverSolder:
--- Quote from: garrettm on February 05, 2021, 09:56:24 pm ---
--- Quote from: SilverSolder on February 05, 2021, 07:53:16 pm ---Perhaps just turning off the Filter will have the desired (noise) effect?
The filter will suppress any signal over 7.5Hz or 75Hz, respectively, so it is unclear how much effect adding a signal would have unless sample periods get very long?
The noise may need to be added later in the chain, perhaps in the R2 converter itself...
--- End quote ---
Good point.
One interesting feature of the R2 ADC design is that each reading from the ADC is 6.5 digits. So with S0 sampling, that’s 250 6.5 digit readings per second. To see if there really are missing codes and how the filters might affect the results, it could be worthwhile to do the averaging remotely on a PC than at the 8505/6A. Simply wait for 1024 samples before computing the mean and place that value into a new time series (which would be equivalent to an average mode reading direct from the DMM). This way we eliminate any limitations with the precision of the internal data types and can quantify the noise of the ADC from the raw samples. This also lets us compute 7.5 digits with any of the three filter settings.
--- End quote ---
Exactly, I also suspect we may be seeing an artifact of the math doing the averaging, rather than missing codes from the A/D converter.
--- Quote from: garrettm on February 05, 2021, 09:56:24 pm ---Unlike integrating ADCs, it makes more sense to use the fastest sample rate possible from a R2 ADC and perform the appropriate conditioning at the PC, rather than on the DMM. At least for these ancient boat anchors. With an auto zero circuit and more advanced controller and display, the 8506A could have been a much more impressive instrument. The "analyze" features on the DMM4050 (trend plot, stats and histogram) are be perfectly suited for the 8505/6A and its high speed, high resolution sampling. And with finer sample size control, speed vs noise could be better optimized for a given measurement. Then there is the possibility of using digital filtering to avoid the need for analog filters.
--- End quote ---
Agree, the 8505/6A coupled with a PC is the way to go. I have all of them on a GPIB bus, and you can do stuff like trigger two (or more) of them to read at the exact same time, for example. Then, as you say, you can do a ton of math on the numbers afterwards. - unfortunately, no matter how much math you do, if the last few digits are pure noise, they remain as pure noise afterwards too! :D
Averaging works really well if the numbers are high quality in the first place, which they obviously are from these instruments if you dot the i's and cross the t's.
SilverSolder:
--- Quote from: joeqsmith on February 05, 2021, 10:35:32 pm ---
--- Quote from: SilverSolder on February 05, 2021, 06:36:43 pm ---
--- Quote from: joeqsmith on February 05, 2021, 06:29:41 pm ---Outside of a short note about the there being a fast (Y) and slow (Z) filter and them being a 3-pole Bessel with different cutoffs, they seem to provide very little detail about them. Maybe the schematic IS the document.
With Dave's UEI meter, I would feed a chirp into it and record the filters response. Seems like you could do something like this to test out the hardware filters.
--- End quote ---
I ran the filters from the diagram in LTSpice and came up with the figures in red, in the Active Filter circuit diagram I posted earlier (i.e. 7.5Hz and 75Hz, 18dB/octave for the slow and fast sections respectively).
I think those numbers are unrelated to the 50ms and 550ms "timeouts" as they are called in the manual. My theory is that when one of the "timeouts" is activated, the meter simply waits that number of ms after a trigger before performing the reading (to guarantee the filter has settled). Should be easy to verify on GPIB (still not set up here!)
--- End quote ---
Odd they wouldn't have documented the filters better and that you had to run SPICE to sort it out.
Log sweep from 100mHz to 10Hz, attempting to collect the data as fast as the meter will allow it (about 20Hz). Showing the rolloff with the slow filter enabled.
--- End quote ---
Agreed, we shouldn't have to jump through hoops to learn what the response of these filters are. They obviously impact measurements at low frequencies so you need to take them into account.
Awesome, -6dBV at 7.5Hz.... exactly what SPICE predicted - and now confirmed experimentally! :D
The 75Hz filter is very close to the SPICE result as well.
SilverSolder:
--- Quote from: garrettm on February 05, 2021, 09:56:24 pm ---[...] With an auto zero circuit and more advanced controller and display, the 8506A could have been a much more impressive instrument. [...]
--- End quote ---
I have done the "auto zero" externally by GPIB - i.e. have a GPIB controlled relay regularly short the input, take a reading, set the zero, then switch back to what you are measuring. This approach works surprisingly well, but obviously has pitfalls (thermal EMF, etc.) which means you have to go slow and steady... but hey, these meters wouldn't be so cheap if they were easy to use as well as being accurate! :D
joeqsmith:
The fastest I can pull down data is about 65ms a sample. That's with S0 (4ms), filter off and internal triggering. The software is just pulling out data when it's available and stitching it together. Another process checks for a valid frame and pulls it out to be processed. The serial port only supports up to 9600 BAUD. They also have a minimum of 2 stop bits. A message is 14 bytes (using ASCII), or 14.6ms to send. I tried using asynch which brought it down to 54ms but it seems like it's slower than it should be. Maybe the binary format would improve it.
What are you able to achieve when using GPIB?
SilverSolder:
I use a serial-to-GPIB interface, with the serial side running at 38.4bps and I can get to about 15ms-20ms per "transaction".
The manual claims 500 readings per second for these Flukes, which is only 2ms per reading... so it looks like the serial baud rate is the limit in both cases!
If you have a fast GPIB host controller, perhaps you should take up @garrettm on his offer for his spare GPIB interface?
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version