AC gain drift: 0.001 dB
Noise and Bandwith: Good for audio line level.
Here, I do not talk about a couple of resistors in series, there would be 5, 7, 8. I am more concerned about layout, parasitics, rf pick up, rather than cost.
I feel 8 resistors in series is a nice antenna or a nice distributed line.
Yes this is outstanding technology.
Vos 3μV is great for instrumentation, asking for tremendous care in the layout and knowledge about minute effects like Seebeck.
Fortunately, I do not need DC accuracy here, I have no need for ultra low Vos, Ios, Ibias.
AC gain drift: 0.001 dB
Noise and Bandwith: Good for audio line level.
Here, I do not talk about a couple of resistors in series, there would be 5, 7, 8. I am more concerned about layout, parasitics, rf pick up, rather than cost.
I feel 8 resistors in series is a nice antenna or a nice distributed line.
As an IC designer I'm used to see dozens of unit resistors connected in series, but of course those are physically much smaller. Whether parasitics are a problem is hard to say since the value and the physical size of the resistors is not yet known, nor the frequency at which the loop gain drops to 0 dB. You can solve RF pick-up with a shielded box around the whole thing.
Anyway, series connections and common-centroid layouts only help when all resistors have similar temperature coefficients. That seems likely when they are unit resistors from the same batch, but it is not guaranteed.
0.001 dB is 115 ppm, so with +/- 25 ppm/K resistors (the usual temperature coefficient for 0.1 % resistors), in the worst case when one feedback resistor is at +25 ppm/K and the other at -25 ppm/K, it requires the temperature to stay constant to well within 2.3 K (well within, because there will be other error sources, such as loop gain variations). Or actually, since you have two stages, it's probably 1.15 K. Without an oven, can you keep the temperature constant within 1 K during the experiment?
These 0207 through-hole metal film resistors https://asset.conrad.com/media10/ad...-56-k-axiaal-bedraad-0207-06-w-01-1-stuks.pdf have a thermal resistance of 140 K/W, so 7 mW of dissipation will cause a self-heating of 1 K.
I do not think the oven would help.
Here we need resistor pairs to be at the same temperature within 1℃.
The resistors will be at some ambient temperature, the one, inside the oven, I see no difference with being at some ambient room temperature.
A regulated ambient temperature makes no difference with an unregulated ambient temperature about the addressed issue.
Am I missing something ?
I think thermal bounding resistor pairs could be the way to go. I have never seen that, may be a wacky idea of mine, what do you think ?
Each pair ( I need 4 pairs ) would sit in their own enclosure filled with thermal conducting stuff.
Resistors are not supposed to contact each other, why not, I checked the cases are isolated.
Capacitive coupling, may be an issue ?
Self heating flows out via the resistor leads and PCB tracks, to make all this more mind boggling.
Here we need resistor pairs to be at the same temperature within 1℃.
The resistors will be at some ambient temperature, the one, inside the oven, I see no difference with being at some ambient room temperature.
A regulated ambient temperature makes no difference with an unregulated ambient temperature about the addressed issue.
Am I missing something ?
I think thermal bounding resistor pairs could be the way to go. I have never seen that, may be a wacky idea of mine, what do you think ?
Each pair ( I need 4 pairs ) would sit in their own enclosure filled with thermal conducting stuff.
Resistors are not supposed to contact each other, why not, I checked the cases are isolated.
Capacitive coupling, may be an issue ?
Self heating flows out via the resistor leads and PCB tracks, to make all this more mind boggling.
I think there are two cases to distinguish:
1. All resistors have essentially the same temperature coefficient (in ppm/K that is)
2. The resistors have different temperature coefficients, randomly distributed between the specified minimum and maximum value
In case 1, tricks to keep the resistors at the same temperature will help, and ambient temperature changes don't matter because they affect all resistors equally. If you use unit resistors from the same batch, you will hopefully end up in case 1, although no-one guarantees that.
In case 2, the temperature of each individual resistor has to stay close enough to the temperature it had during calibration. Thermal coupling and common-centroid techniques then won't help. You then need very low self-heating and a constant ambient temperature. You could use an oven if the ambient temperature varies too much.
1. All resistors have essentially the same temperature coefficient (in ppm/K that is)
2. The resistors have different temperature coefficients, randomly distributed between the specified minimum and maximum value
In case 1, tricks to keep the resistors at the same temperature will help, and ambient temperature changes don't matter because they affect all resistors equally. If you use unit resistors from the same batch, you will hopefully end up in case 1, although no-one guarantees that.
In case 2, the temperature of each individual resistor has to stay close enough to the temperature it had during calibration. Thermal coupling and common-centroid techniques then won't help. You then need very low self-heating and a constant ambient temperature. You could use an oven if the ambient temperature varies too much.
Thanks to your answers, I realize, the gain will be signal level dependant.
Higher level, giving more self heating, giving higher or lower gain depending of relative resistor temperature coefficients.
A pretty annoying trouble.
Wait a minute. The resistors in a pair are thermally bound, they do share a common node.
Higher level, giving more self heating, giving higher or lower gain depending of relative resistor temperature coefficients.
A pretty annoying trouble.
Wait a minute. The resistors in a pair are thermally bound, they do share a common node.
OP, why do you even have these insane performance requirements in the first place? What is the system-level performance metric that necessitates them? CMRR or what?
I'm not sure why you are so worried about temperature. I would make sure on circuit level that you only need relative matching between resistors, then you can place identical value surface-mount parts onto a PCB in a smallish area (with critical matches right next to each other) and put them in a box (shield) or pot them if really necessary.
If a solution requires me to jump through major hoops, I would always double and triple check that its benefits are really worth it and that there isn't another, smarter solution that gets the same done a lot more elegantly. Sometimes you have no choice but often you do. As they say, if you've been digging yourself into a hole, stop digging.
I'm not sure why you are so worried about temperature. I would make sure on circuit level that you only need relative matching between resistors, then you can place identical value surface-mount parts onto a PCB in a smallish area (with critical matches right next to each other) and put them in a box (shield) or pot them if really necessary.
If a solution requires me to jump through major hoops, I would always double and triple check that its benefits are really worth it and that there isn't another, smarter solution that gets the same done a lot more elegantly. Sometimes you have no choice but often you do. As they say, if you've been digging yourself into a hole, stop digging.
I've never gone back to see how well the resistors did, but I had built a Kelvin-Varley divider and probably got it well under 100 ppm overall. A few years ago it was still accurate. One hint- use thru-hole and put a heat sink (hemostat) on the leads when you solder them. Resistor parameters can change permanently when you solder them.
I have to agree with the above post- why in the world do you need this? Want serious gain stability? Use a ratio transformer. There have been a bunch, like Gertsch and General Radio Corp. You might even wind your own. Step up or down, then buffer.
I have to agree with the above post- why in the world do you need this? Want serious gain stability? Use a ratio transformer. There have been a bunch, like Gertsch and General Radio Corp. You might even wind your own. Step up or down, then buffer.
That's only true when self heating is the problem. When the main problem is changing ambient temperature combined with mismatched temperature coefficients, those 5 ppm/K resistors are much better than the 25 ppm/K resistors.
I agree this 0.001dB requirement is madness.
This is for experimental studies, I know this is beyond ordinary audio, then let's find solutions and see how far one can go.
I found arrays of 8 same independent resistors.
Vishay NOMCA
TC absolute +- 25ppm
TC tracking +- 5ppm
This could implement
gain +7 with R, 6R
gain -7 whith R, 7R
gain -1 whith 4R,4R
gain 0.2 with 5R,R
What gain accuracy could I get ?
This is for experimental studies, I know this is beyond ordinary audio, then let's find solutions and see how far one can go.
I found arrays of 8 same independent resistors.
Vishay NOMCA
TC absolute +- 25ppm
TC tracking +- 5ppm
This could implement
gain +7 with R, 6R
gain -7 whith R, 7R
gain -1 whith 4R,4R
gain 0.2 with 5R,R
What gain accuracy could I get ?
AC response over audio frequencies is going to be a limit - check how many decades of margin is needed between amp corner frequencies and audio band limits for 100 ppm error
DC coupling would work for the lower corner but the upper limit is tough, gain intercept even 2 decades beyond 20 kHz and 100 dB excess gain @20k...
DC coupling would work for the lower corner but the upper limit is tough, gain intercept even 2 decades beyond 20 kHz and 100 dB excess gain @20k...
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.