Using op amps, in the usual inverting configuration and the non inverting configuration.
What is important to achieve a best gain stability.
I am thinking of:
The resistors with low temperature coefficient.
Thermal matching of resistors.
What else, should I care ?
The gain value does not need high accuracy, there will be calibration to take care of this aspect.
The important thing is the need for a perfectly stable value.
What is important to achieve a best gain stability.
I am thinking of:
The resistors with low temperature coefficient.
Thermal matching of resistors.
What else, should I care ?
The gain value does not need high accuracy, there will be calibration to take care of this aspect.
The important thing is the need for a perfectly stable value.
You might want to look at this paper from TI: https://www.ti.com/lit/slyt374
You'll have to use the "wayback" machine, but there were articles written decades ago showing the use of a transistor array to maintain a constant temperature for log and anti-log converters.
It's an issue with simple low noise NFET amplifiers -- Groner kept the Vds of his very low noise BF862 amplifier low to avoid self heating the devices.
You'll have to use the "wayback" machine, but there were articles written decades ago showing the use of a transistor array to maintain a constant temperature for log and anti-log converters.
It's an issue with simple low noise NFET amplifiers -- Groner kept the Vds of his very low noise BF862 amplifier low to avoid self heating the devices.
Resistors with a power rating far above their dissipation, so there is hardly any self-heating. Strings of equal-valued resistors that conduct the same signal current, so their self-heating matches. Cross-quadding or other common-centroid layouts of those equal-valued resistors, so temperature gradients due to an external source of heat affect both feedback resistors to approximately the same extent. Large distances between the resistors and anything that gets hot.
Like Jack suggested, you could indeed put the feedback resistors in an oven, that is, put them in a thermally isolated enclosure with a power resistor and a temperature sensor of some kind, and use the power resistor and sensor to bring the temperature to some constant value just above the maximum expected ambient temperature. Ideally you want the same distance between the power resistor and each of the feedback resistors, so you don't get any temperature differences.
Like Jack suggested, you could indeed put the feedback resistors in an oven, that is, put them in a thermally isolated enclosure with a power resistor and a temperature sensor of some kind, and use the power resistor and sensor to bring the temperature to some constant value just above the maximum expected ambient temperature. Ideally you want the same distance between the power resistor and each of the feedback resistors, so you don't get any temperature differences.
For those not familiar with common-centroid layouts: the trick is to try to put the "centre of gravity" of two strings of devices that need to match at the same place. Examples:
For a 1:1 ratio, you could make an ABBA layout:
A B B A
For a 3:1 ratio, this changes into:
A A A B B A A A
For a 1:1 ratio, you can also choose a cross-quad:
A B
B A
This is a popular one for 8:1 ratios:
A A A
A B A
A A A
It also works with 24:1:
A A A A A
A A A A A
A A B A A
A A A A A
A A A A A
These techniques are quite common in analogue IC layouts. In IC layouts, there are usually dummy devices around the whole thing to ensure that each unit device sees the same surroundings, but in this case, I can't see any advantage in putting dummy feedback resistors around the real feedback resistors.
Besides matching of the feedback components and the PCB track resistances that are in series with them, you will also need a large loop gain. You could consider using a conditionally stable composite op-amp (Samuel Groner style, Samuel Groner's super opamp ) instead of just a single op-amp, or a conditionally stable, huge open-loop gain op-amp on one chip, like Opamp with open loop gain of 1,000,000,000,000,000.
.
For a 1:1 ratio, you could make an ABBA layout:
A B B A
For a 3:1 ratio, this changes into:
A A A B B A A A
For a 1:1 ratio, you can also choose a cross-quad:
A B
B A
This is a popular one for 8:1 ratios:
A A A
A B A
A A A
It also works with 24:1:
A A A A A
A A A A A
A A B A A
A A A A A
A A A A A
These techniques are quite common in analogue IC layouts. In IC layouts, there are usually dummy devices around the whole thing to ensure that each unit device sees the same surroundings, but in this case, I can't see any advantage in putting dummy feedback resistors around the real feedback resistors.
Besides matching of the feedback components and the PCB track resistances that are in series with them, you will also need a large loop gain. You could consider using a conditionally stable composite op-amp (Samuel Groner style, Samuel Groner's super opamp ) instead of just a single op-amp, or a conditionally stable, huge open-loop gain op-amp on one chip, like Opamp with open loop gain of 1,000,000,000,000,000.
.
Last edited:
I rather not use an oven. May be last over all tricks at room temperature.
I will look into the centroid technique.
Gain values.
The input stage will convert from symmetric to assymmetic with -14 dB gain.
The output stage will convert from asymmetric to symmetric with +24 dB gain.
Input stage based on an OAP inverting with 5R,5R,R,R.
Output stage based on
an OAP inverting with R,8R.
Paralleled with
an OPA non inverting with R,7R
Or a variant, cascading two OPA, first one for gain, second one to give the symmetric signal.
I will look into the centroid technique.
Gain values.
The input stage will convert from symmetric to assymmetic with -14 dB gain.
The output stage will convert from asymmetric to symmetric with +24 dB gain.
Input stage based on an OAP inverting with 5R,5R,R,R.
Output stage based on
an OAP inverting with R,8R.
Paralleled with
an OPA non inverting with R,7R
Or a variant, cascading two OPA, first one for gain, second one to give the symmetric signal.
This is for AC gain. Audio bandwidth.
I don't care much about gain error, basically from resistors tolerance and open loop gain large but not infinite.
I do care about gain variation.
I presume this asks for high open loop gain and that gain stability.
I will read carefully the quoted, interesting TI paper.
Who knows about OAP open loop gain stability ?
This is usually not specified, unecessary in usual applications where one only needs a large guaranteed open loop.
I don't care much about gain error, basically from resistors tolerance and open loop gain large but not infinite.
I do care about gain variation.
I presume this asks for high open loop gain and that gain stability.
I will read carefully the quoted, interesting TI paper.
Who knows about OAP open loop gain stability ?
This is usually not specified, unecessary in usual applications where one only needs a large guaranteed open loop.
Last edited:
There's also the trick of using only one value of resistance for all of the feedback components,
with a series connection of identical resistors to get the higher values. Then the self-heating
drifts are all the same and cancel out, so their ratio (gain) is more stable.
If you need audio frequency gain stability, look for op amps with wide open loop gain,
so the amount of feedback in the range of interest will still be uniform and high.
If the open loop gain is high enough, there's no need to worry about its stability.
with a series connection of identical resistors to get the higher values. Then the self-heating
drifts are all the same and cancel out, so their ratio (gain) is more stable.
If you need audio frequency gain stability, look for op amps with wide open loop gain,
so the amount of feedback in the range of interest will still be uniform and high.
If the open loop gain is high enough, there's no need to worry about its stability.
Last edited:
Or these, around $50 each.
http://www.caddock.com/Online_catalog/Mrktg_Lit/TypeUSF.pdf
Or an array, some typical examples.
https://www.mouser.com/datasheet/2/427/vtfstd-1764257.pdf
http://www.caddock.com/Online_catalog/Mrktg_Lit/TypeUSF.pdf
Or an array, some typical examples.
https://www.mouser.com/datasheet/2/427/vtfstd-1764257.pdf
Last edited:
This also cancels out non-linearity (voltage coefficient), as voltage ratios are independent of current, assuming all resistors behave identically and the opamp input current is negligible.
Absolutely, that's why I mentioned that in post #3, but maybe phrased less clearly than you did.
Last edited:
Pendulum clocks routinely achieve 1/2 second per year drift. That's 16 parts per billion.
They do this by measuring the oscillation period of a fixed length pendulum. A pendulum whose length does not vary with temperature. A pendulum whose temperature-coefficient-of-expansion has been carefully compensated. If they can do it (in the 1930s!) then you can do it too.
They do this by measuring the oscillation period of a fixed length pendulum. A pendulum whose length does not vary with temperature. A pendulum whose temperature-coefficient-of-expansion has been carefully compensated. If they can do it (in the 1930s!) then you can do it too.
I think, first step is resistors with low temperature coefficient.
What resistors with lowest TC is available at reasonable cost ?
Second step is temperature matching.
Low current is a first way to go, limited by not too high resistor values because of noise.
Next, I think here, to play with thermal resistance to ambient.
Exemple R, 5R.
Under same current the self heating power is P, 5P.
With a over sized 5R, I can aim at 5 times lower Rth, to get the same temperature.
This is actually a variant of a serie of five resistors R.
Let's find data about resistor Rth versus power rating. I have seen something about SMD resistor size.
What resistors with lowest TC is available at reasonable cost ?
Second step is temperature matching.
Low current is a first way to go, limited by not too high resistor values because of noise.
Next, I think here, to play with thermal resistance to ambient.
Exemple R, 5R.
Under same current the self heating power is P, 5P.
With a over sized 5R, I can aim at 5 times lower Rth, to get the same temperature.
This is actually a variant of a serie of five resistors R.
Let's find data about resistor Rth versus power rating. I have seen something about SMD resistor size.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.