Distortion in Resistors is real
Distortion in Resistors is real. I would like to share some of mine events.
When I was designing the ATS2 (first all SMD Audio analyzer @ AP) I started to test / debug the first PCB Assy. The generator DAC output test was 0.0003% @ 1 kHz but 0.01% @ 10Hz. The DAC current to voltage stage feedback resistor 0805 was running about 62% rated power. The parts were 25ppm type. The resistor was heating and cooling cycle by cycle thus the gain was changing cycle by cycle (poor thermal mass). Changing the power and adding two resistors in series resolved the problem.
I had designed a mixer product @ Mackie and the production engineers made a change to the part type. The +48 volt decoupling 100 ohm 1/2 watt through whole resistor was replaced by 0805. The phantom power worked 1 or 2 times before the resistor opened from the inrush charging current (poor thermal mass).
At BGW we used a lots of low ohm wire wound’s, I used many tests to evaluate new sources. Measure the TC (what is the change in resistance vs temperature) and the overload power were the major tests. Overload test was to charge the 20 Joule cap and discharge the cap through the resistor and SCR. Many went open or had a large change in resistance.
Small wire, end caps, welding, low thermal mass and high TC’s do have undesirable traits that comprise our designs.
Small non linearity’s usually are not a problem if they are inside the feedback loop.
It is imperative that you know what all the parameters of the parts are, because all designs have some types of tradeoffs to meet the design goals.
Duke
Distortion in Resistors is real. I would like to share some of mine events.
When I was designing the ATS2 (first all SMD Audio analyzer @ AP) I started to test / debug the first PCB Assy. The generator DAC output test was 0.0003% @ 1 kHz but 0.01% @ 10Hz. The DAC current to voltage stage feedback resistor 0805 was running about 62% rated power. The parts were 25ppm type. The resistor was heating and cooling cycle by cycle thus the gain was changing cycle by cycle (poor thermal mass). Changing the power and adding two resistors in series resolved the problem.
I had designed a mixer product @ Mackie and the production engineers made a change to the part type. The +48 volt decoupling 100 ohm 1/2 watt through whole resistor was replaced by 0805. The phantom power worked 1 or 2 times before the resistor opened from the inrush charging current (poor thermal mass).
At BGW we used a lots of low ohm wire wound’s, I used many tests to evaluate new sources. Measure the TC (what is the change in resistance vs temperature) and the overload power were the major tests. Overload test was to charge the 20 Joule cap and discharge the cap through the resistor and SCR. Many went open or had a large change in resistance.
Small wire, end caps, welding, low thermal mass and high TC’s do have undesirable traits that comprise our designs.
Small non linearity’s usually are not a problem if they are inside the feedback loop.
It is imperative that you know what all the parameters of the parts are, because all designs have some types of tradeoffs to meet the design goals.
Duke
Distortion in Resistors is real. I would like to share some of mine events.
When I was designing the ATS2 (first all SMD Audio analyzer @ AP) I started to test / debug the first PCB Assy. The generator DAC output test was 0.0003% @ 1 kHz but 0.01% @ 10Hz. The DAC current to voltage stage feedback resistor 0805 was running about 62% rated power. The parts were 25ppm type. The resistor was heating and cooling cycle by cycle thus the gain was changing cycle by cycle (poor thermal mass). Changing the power and adding two resistors in series resolved the problem.
I had designed a mixer product @ Mackie and the production engineers made a change to the part type. The +48 volt decoupling 100 ohm 1/2 watt through whole resistor was replaced by 0805. The phantom power worked 1 or 2 times before the resistor opened from the inrush charging current (poor thermal mass).
At BGW we used a lots of low ohm wire wound’s, I used many tests to evaluate new sources. Measure the TC (what is the change in resistance vs temperature) and the overload power were the major tests. Overload test was to charge the 20 Joule cap and discharge the cap through the resistor and SCR. Many went open or had a large change in resistance.
Small wire, end caps, welding, low thermal mass and high TC’s do have undesirable traits that comprise our designs.
Small non linearity’s usually are not a problem if they are inside the feedback loop.
It is imperative that you know what all the parameters of the parts are, because all designs have some types of tradeoffs to meet the design goals.
Duke
Hi Duke, good points. Note that my comments were not to deny this; I'm on the same sheet you are. My comment was to 'resistor distortion depends on the voltage across it not the current through it' and my point was that the two are always connected.
I've been at a few presentations byAP's Bruce Hofer and he made similar points. Like the one that in a power amp feedback loop of say 20k and 1k, that 20k can be the limiting factor on high level distortion as it has to bear the full Vout.
His trick: replace that 20k with 20 times 1k, of the same type and batch of the other 1k. Not only does the power level on each resistor drop, but also the remaining tempco and voltage coefficient factors exactly cancel! Problem gone!
Jan
To reduce the voltage I often use series connected resistors of half value.
That reduces the power dissipation by 50% for each resistor and it reduces the voltage coefficient effect as well.
The B.Hoffer technique is just a twenty fold version.
This is used when trying to measure distortion of a resistor.
One side of the bridge is 20*1k over 1k and the other side of the bridge is 20k (DUT) over 1k.
The tempco and voltage coeffs of all the 1k are near enough identical.
The error across the bridge is then down to what ever the 20k is creating.
That reduces the power dissipation by 50% for each resistor and it reduces the voltage coefficient effect as well.
The B.Hoffer technique is just a twenty fold version.
This is used when trying to measure distortion of a resistor.
One side of the bridge is 20*1k over 1k and the other side of the bridge is 20k (DUT) over 1k.
The tempco and voltage coeffs of all the 1k are near enough identical.
The error across the bridge is then down to what ever the 20k is creating.
No, it is fundamentally different. Because all the resistors are equal (to a first extend), the ratio of 1/21 (in this example) is exactly maintained while still each resistor on itself 'distorts' due to tempco and volt coefficient.
Jan
The resistance is not just 'R' (constant value) but something like 'R' = R(constant) + a*V + b*T where V and T are the voltage and temp coefficients.
So the resistance value changes during the signal cycle and also depending on instantaneous temp. Which means that for instance in a voltage divider, the divider ratio changes during the voltage cycle and that, if symmetrically, causes the peaks of the signal to flatten or to become peaky. And that means odd order harmonics.
Jan
What did I get wrong?
The error across the bridge is a measure of the distortion that the big resistor creates.
All right, I'll explain in more detail.
I have a power amplifier.
It uses 5 resistors in parallel as the upper NFB leg.
It feeds into a single resistor for the lower NFB leg.
Changing the 5 resistors in parallel to 5 in series reduces the effect of voltage coef to very much closer to that of the single resistor . Not there yet since I would need 27 resistors in series.
It is the same philosophy, try to get the voltage across any single resistor to match the voltage across the lower leg resistor.
If one can get the power dissipation in each resistor in the upper leg down to match that of the resistor in the lower leg, then the effect of tempco would also be reduced.
To get closer to that, I replaced each of the 5 single resistors with a parallel pair. I now have 10 resistors in series parallel to act as the upper leg of the voltage divider.
To me that is much better than 5 in parallel. The theoretical cancellation would require 27 resistors : 1 rresistor
That for the PCB would be impractical, but the implementation is much closer than the original.
I have a power amplifier.
It uses 5 resistors in parallel as the upper NFB leg.
It feeds into a single resistor for the lower NFB leg.
Changing the 5 resistors in parallel to 5 in series reduces the effect of voltage coef to very much closer to that of the single resistor . Not there yet since I would need 27 resistors in series.
It is the same philosophy, try to get the voltage across any single resistor to match the voltage across the lower leg resistor.
If one can get the power dissipation in each resistor in the upper leg down to match that of the resistor in the lower leg, then the effect of tempco would also be reduced.
To get closer to that, I replaced each of the 5 single resistors with a parallel pair. I now have 10 resistors in series parallel to act as the upper leg of the voltage divider.
To me that is much better than 5 in parallel. The theoretical cancellation would require 27 resistors : 1 rresistor
That for the PCB would be impractical, but the implementation is much closer than the original.
Last edited:
note that I said for the two resistor implementation
I did not claim complete cancelation of thermal or voltage coef generated distortion.
Hi Jan,
In your post No.38 you say the following;
“I don't think you can separate this - according to Mr. Ohm, the voltage across and current through a resistor are bound together through the resistance. So whenever you change the V across the R, it changes the I trough the R, and vice versa.”
My understanding of Ohms law comes from the following equation, R=V/I.
Where R is a linear resistance, but you have ignored the fact I was referring to a non-linear resistance in the sentence you quoted me on. This is made more confusing since in your post No.49 you actually describe an equation for non-linear resistance as “'R' = R(constant) + a*V + b*T).”
To my knowledge, Ohms law has never been defined as R(constant) + a*V + b*T) = V/I.
Also, you have included two terms in your equation for non-linear resistance defined as a*V and b*T, which I agree with since both of these terms were implicit in my post No.31.
Jan, I think you really should re-read my post No.31 more carefully and please do not take a sentence like this out of context.
Peter
In your post No.38 you say the following;
“I don't think you can separate this - according to Mr. Ohm, the voltage across and current through a resistor are bound together through the resistance. So whenever you change the V across the R, it changes the I trough the R, and vice versa.”
My understanding of Ohms law comes from the following equation, R=V/I.
Where R is a linear resistance, but you have ignored the fact I was referring to a non-linear resistance in the sentence you quoted me on. This is made more confusing since in your post No.49 you actually describe an equation for non-linear resistance as “'R' = R(constant) + a*V + b*T).”
To my knowledge, Ohms law has never been defined as R(constant) + a*V + b*T) = V/I.
Also, you have included two terms in your equation for non-linear resistance defined as a*V and b*T, which I agree with since both of these terms were implicit in my post No.31.
Jan, I think you really should re-read my post No.31 more carefully and please do not take a sentence like this out of context.
Peter
resistor distortion
Hello All,
Several months ago I studied resistor distortion in the audio domain.
Largely Mr. Ohm is correct, E=IR. At edge of measurability nonlinearity begins to creep in.
At high frequencies impedance is more complex than Mr. Ohm’s simple, E=IR, model. As frequency increases impedance increases to a peek and then falls. On a log scale this plots looks like a bell shape curve. The inductive and capacitive parasitics typically only play a part well above the audio frequencies. The higher the resistor value the lower the frequency the complex parasitic impedance begins to increase the measured impedance of the resistor. A Rhode and Schwarz LCR bridge was used to measure this parasitic effect with Vishay RND60 resistors. The measured results were confirmed using a Keysight frequency generator and 6-1/2 digit volt meter.
In Linear Audio, several issues ago, Simon7000 posted an interesting article regarding resistor distortion. Using a Wheatstone bridge and an Audio Precision audio analyzer Simon7000 measured resistor distortion to the ppm level. Remember noise is random and distortion is not. My guess is that there was a high level of averaging to reduce the noise floor in the frequency domain plots. In the time domain noise will be orders of magnitude greater than the measured ppm’s of distortion. Temperature Coefficient was the identified distortion culprit. The higher the Temperature Coefficient the higher the distortion.
Where TC gets interesting is not with a single resistor but with a voltage divider. Delta T will affect each of the resistors differently. The TC ratio of the resistors will change. Place this voltage divider in the feedback loop of an op-amp and we have resistor nonlinearity on steroids.
Thank you Audio1Man for pushing the limits of measurability.
DT
Hello All,
Several months ago I studied resistor distortion in the audio domain.
Largely Mr. Ohm is correct, E=IR. At edge of measurability nonlinearity begins to creep in.
At high frequencies impedance is more complex than Mr. Ohm’s simple, E=IR, model. As frequency increases impedance increases to a peek and then falls. On a log scale this plots looks like a bell shape curve. The inductive and capacitive parasitics typically only play a part well above the audio frequencies. The higher the resistor value the lower the frequency the complex parasitic impedance begins to increase the measured impedance of the resistor. A Rhode and Schwarz LCR bridge was used to measure this parasitic effect with Vishay RND60 resistors. The measured results were confirmed using a Keysight frequency generator and 6-1/2 digit volt meter.
In Linear Audio, several issues ago, Simon7000 posted an interesting article regarding resistor distortion. Using a Wheatstone bridge and an Audio Precision audio analyzer Simon7000 measured resistor distortion to the ppm level. Remember noise is random and distortion is not. My guess is that there was a high level of averaging to reduce the noise floor in the frequency domain plots. In the time domain noise will be orders of magnitude greater than the measured ppm’s of distortion. Temperature Coefficient was the identified distortion culprit. The higher the Temperature Coefficient the higher the distortion.
Where TC gets interesting is not with a single resistor but with a voltage divider. Delta T will affect each of the resistors differently. The TC ratio of the resistors will change. Place this voltage divider in the feedback loop of an op-amp and we have resistor nonlinearity on steroids.
Thank you Audio1Man for pushing the limits of measurability.
DT
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.