Electronics > Projects, Designs, and Technical Stuff
Kelvin Varley Divider [and Precision Voltage Source]
amspire:
--- Quote from: fmaimon on November 05, 2011, 10:17:57 pm ---
--- Quote from: amspire on November 05, 2011, 02:37:21 am ---The OPA735 is pretty good, but Linear Technology have IC's that are an order of magnitude better:
http://www.linear.com/products/zero_drift_amplifiers
--- End quote ---
What, other than offset voltage, drift and voltage gain, do you usually watch for when looking for a chopper stabilized amp?
--- End quote ---
Many of the chopper stabilized opamps have a low frequency response, but that is fine for a DC reference. So you are looking at offset, drift and input current.
--- Quote ---
--- Quote from: amspire ---If you are after accuracy, you do have to invest in good resistors - the more powerful the better as they will heat up less. The last time I was balancing a bridge-type circuit built on 0.1% 25ppm SMD resistors, just the 10V I had across a 10K 10:1 divider was enough power to see the resistors drift due to heating. In a 10:1 divider one resistor gets 9 times the power of the other resistor, so one resistor heats more then the other. Because of the drift, best accuracy I could get was probably 0.005% even though I had the resolution to adjust within 0.000001%. I would love to have some of the unbelievable Vishay metal foil resistors, but their prices really hurt.
--- End quote ---
I'm looking for accuracy, but not that much. I'll use through hole resistors, as these have better thermal dissipation. Did you try making the traces and pads wider, so it acts as a heat sink?
It's already expensive buying low ppm resistors (~$.6-$1 each), but those metal foil are crazy! :o
--- End quote ---
The stand alone resistors will have a much lower temperature rise then SMD's. Wider traces may have an effect but probably doesn't matter. Work really hard not to have any mechanical stress on the resistors, so bend the legs to match the mounting exactly before you solder. Stress on a component is one of the factors that cn result in a slow long term drift.
What you can also do is in the first decade, use 2 or 3 resistors in series to make up each divider resistor. It gives you more combinations to end up with 10 exactly matched divider resistors, spreads the thermal load, and gives you a higher input voltage level that the divider can take. So if you got 3k3 resistors, the first decade can be 3 in series to make up one divider resistor and all the other decades use one of the 3k3. If you want to save money, you can go to cheaper resistors after the 3rd decade. What I have found is that the more you can match resistors rather then have adjustment pots, the better the final long term stability. If you do have adjustment pots, it is better if it only has to adjust over +/- 0.01% rather then +/- 1%. I haven't looked at the fluke, but I would think its adjustment on the first stage resistors is probably something like +/- 0.0005%
Richard
Conrad Hoffman:
Yes, pots are trouble! When Julie Research did their KVDs, they matched the resistors as best they could (which was very good) but then soldered a short length of resistance wire to one end of each resistor to get the final value. You'll also find that regular metal films will change value when you solder them, so use long leads, heat clips and work fast. Read the old article a good number of times because I think it mentions all sorts of things to watch for. BTW, it was 15 years ago, so I don't remember as much as you might think.
See if you can download the manual for an Analogic 8200 (same as Data Precision 8200) voltage source. That will show you how to use analog switches, but basically if you're feeding a high impedance opamp from a divider, having 200 ohms in series with the input doesn't matter, and whatever pickoff point is connected swamps out the leakage of the other ones. It's all about ratio of impedances. Then, the decade amps are all summed by another amp, using different value summing resistors so each decade has the correct weight in the final answer. The 8200 doesn't actually use decades, but octal, since it's no problem for a processor to display it.
Conrad
fmaimon:
--- Quote from: amspire on November 06, 2011, 12:05:04 am ---Many of the chopper stabilized opamps have a low frequency response, but that is fine for a DC reference. So you are looking at offset, drift and input current.
--- End quote ---
That's pretty much what I thought. Thank you.
--- Quote from: amspire ---The stand alone resistors will have a much lower temperature rise then SMD's. Wider traces may have an effect but probably doesn't matter. Work really hard not to have any mechanical stress on the resistors, so bend the legs to match the mounting exactly before you solder. Stress on a component is one of the factors that cn result in a slow long term drift.
What you can also do is in the first decade, use 2 or 3 resistors in series to make up each divider resistor. It gives you more combinations to end up with 10 exactly matched divider resistors, spreads the thermal load, and gives you a higher input voltage level that the divider can take. So if you got 3k3 resistors, the first decade can be 3 in series to make up one divider resistor and all the other decades use one of the 3k3. If you want to save money, you can go to cheaper resistors after the 3rd decade. What I have found is that the more you can match resistors rather then have adjustment pots, the better the final long term stability. If you do have adjustment pots, it is better if it only has to adjust over +/- 0.01% rather then +/- 1%. I haven't looked at the fluke, but I would think its adjustment on the first stage resistors is probably something like +/- 0.0005%
--- End quote ---
Using multiple resistors is a pretty good idea. That should also help matching sets, as you can combine them anyway you like.
Note taken about the pots. I'll try to match the resistor as best I can and make the adjustments the pots make as small as possible.
The fluke looks like it has lots of pots! Check the attached schematic. :o
--- Quote from: Conrad Hoffman on November 06, 2011, 02:33:55 am ---Yes, pots are trouble! When Julie Research did their KVDs, they matched the resistors as best they could (which was very good) but then soldered a short length of resistance wire to one end of each resistor to get the final value. You'll also find that regular metal films will change value when you solder them, so use long leads, heat clips and work fast. Read the old article a good number of times because I think it mentions all sorts of things to watch for. BTW, it was 15 years ago, so I don't remember as much as you might think.
--- End quote ---
I'm reading the article a lot. :)
--- Quote from: Conrad Hoffman ---See if you can download the manual for an Analogic 8200 (same as Data Precision 8200) voltage source. That will show you how to use analog switches, but basically if you're feeding a high impedance opamp from a divider, having 200 ohms in series with the input doesn't matter, and whatever pickoff point is connected swamps out the leakage of the other ones. It's all about ratio of impedances. Then, the decade amps are all summed by another amp, using different value summing resistors so each decade has the correct weight in the final answer. The 8200 doesn't actually use decades, but octal, since it's no problem for a processor to display it.
--- End quote ---
That's one thing google can't find... Or maybe my googlefu is not that good... :(
The only manual I found is an users manual describing the functions. It didn't have a schematic.
I think I got the idea. One amplifier in a follower configuration (gain of 1) for each decade and another for summing everything. But that brings the problem of matching ratios of resistors in the summing amplifier. Maybe the Hamon idea can be used in here...
Can you explain better how the octal thing works? As it's easier to find analog switches in multiples of 2, it should be easier to construct.
Thank you,
Felipe Maimon
[edit] added the comment about the schematic.
fmaimon:
I think I may found out how the octal thingy works!
Instead of 10 resistors for each decade, use 8 (octal, remember :) ). Then you just have to work in a octal base. For example, the ratio of 0.3125 is 0o24 in octal (I've put the 0o in front of any octal numbers). Then the first "octal decade" (is there a name for this?) is set at the second tap and the second "octal decade" is set on the fourth tap.
Now it's possible to use an easy to find 8 channel analog multiplexer, like the 4051 or similar, but more decades are needed to match the same resolution. Doing the math, you need (log 10 / log 8 ) = about 1.1073 more decades. And now there is another problem of the summing amplifier needing resistors in ratios of multiples of 8, but some modifications of the hamon divider can do it.
[edit] fixed an incorrect smiley
fmaimon:
Just had an idea. Instead of lots of different resistors for each deacade in the summing amplifiers, use a ladder in similar fashion to the r-2r ladder. If my math is correct, it should look like the schematic below. Now I just need 3 different resistor values. Vout is followed by a buffer, of course.
Lol. This thing is looking more like a very big and complex DAC :D
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version