I think you didn't read B.Hofer article.
He showed that thermal distortion ( some) is important only under 200hz ( which is an important freq band in audio); your quick calculation is out of bounce. His negative comments on metal foil resistors although they have vanishing thermal coef should be revealing in that respect.
His interest is in decreasing as much as possible non linear distortion because he is designing lab instruments; it is clear in his mind and in the paper that it is perhaps not related to good sounding; it seems anyway that many peole are in a quest for the last ppm of distortion. For them, details are important as B. Hofer is writing.
Would people hear the difference between amp made of thick and thin films ?
As you, I think no but the idea is interesting; Why not build two and test them
Regards
JPV
When I get around to it, I'm tempted to try the following. I'll take my THD analyzer and run it back-to-back at 1kHz, 5V rms output, through a 11:1 resistor divider made up of a 10k series resistor and a 1k shunt. My analyzer has a THD residual on the order of <0.0003% at 1kHz. I'll take the distortion output and put it into a spectrum analyzer to get rid of the noise and see lower.
I'll use a good quality 1k metal film shunt resistor and a DUT 10k resistor. I'll try carbon composition, carbon film, and metal film. I'll try to see if there is any difference in observed distortion, either up or down, for each harmonic (allowing for the possibility of harmonic cancellation).
The question is whether this setup is sensitive enough to see the effects of resistor voltage distortion, since the voltage being applied is only 5V rms (max output of my analyzer).
I'm guessing the main place to look is at the 3rd harmonic.
If my analyzer has a 0.0003% residual 3rd (3ppm), and I see a 1dB change in the third (and it is fully constructive or destructive between the analyzer and the resistor nonlinearity), I might be able to discern 0.3ppm.
According to Bruce's equation for 3rd, a 1ppm/V resistor driven at 5V rms should produce something on the order of 1ppm, right?
Did I get this right? Does this make sense?
Any thoughts or suggestions?
Cheers,
Bob
I'll use a good quality 1k metal film shunt resistor and a DUT 10k resistor. I'll try carbon composition, carbon film, and metal film. I'll try to see if there is any difference in observed distortion, either up or down, for each harmonic (allowing for the possibility of harmonic cancellation).
The question is whether this setup is sensitive enough to see the effects of resistor voltage distortion, since the voltage being applied is only 5V rms (max output of my analyzer).
I'm guessing the main place to look is at the 3rd harmonic.
If my analyzer has a 0.0003% residual 3rd (3ppm), and I see a 1dB change in the third (and it is fully constructive or destructive between the analyzer and the resistor nonlinearity), I might be able to discern 0.3ppm.
According to Bruce's equation for 3rd, a 1ppm/V resistor driven at 5V rms should produce something on the order of 1ppm, right?
Did I get this right? Does this make sense?
Any thoughts or suggestions?
Cheers,
Bob
Hi Jean-Pierre,
>Would people hear the difference between amp made of thick and thin films ?
I don't think so, just as you can't hear the difference between 1 or 10ppm distortion. But on one occasion I've measured the difference quite easily (see post 4615) between SMD resistors of unknown brand and construction versus MELF(2211) metal film R's from Vishay 1% 50ppm/K, which gave a significant lower distortion. I've measured it at 1, 5 and 20kHz, so it had nothing to do with the thermal coefficient or self heating. Besides, all voltages were quite low (<1V). So I concluded that the voltage coefficient of these (bad!) SMD's were extreme high. I don't why, as I don't have the tools to look "under the hood" of these parts.
Cheers, E.
>Would people hear the difference between amp made of thick and thin films ?
I don't think so, just as you can't hear the difference between 1 or 10ppm distortion. But on one occasion I've measured the difference quite easily (see post 4615) between SMD resistors of unknown brand and construction versus MELF(2211) metal film R's from Vishay 1% 50ppm/K, which gave a significant lower distortion. I've measured it at 1, 5 and 20kHz, so it had nothing to do with the thermal coefficient or self heating. Besides, all voltages were quite low (<1V). So I concluded that the voltage coefficient of these (bad!) SMD's were extreme high. I don't why, as I don't have the tools to look "under the hood" of these parts.
Cheers, E.
I did not, but I guess you missed the actual numbers.
As Mr. Cordell mentioned, the thermal effect in resistors has a low pass effect; the resistor temperature will follow the signal amplitude only at low frequencies.
I don't know how to calculate the -3dB point of this low pass filter, since it depends on many non-electrical quantities (thermal mass, environment temperature, etc...) but my guts are telling me that a 0.15K (zero-point-fifteen degrees, to match the voltage coefficient effect) temperature variation is well up in the KHz range. It is 48dB below the nominal low frequency temperature coefficient effect of 250ppm/K.
Your SMD resistors of unknown brand were probably low grade thick film resistors. These can be worse than carbon film devices, this is a well known fact.
Why do people like to compare apples and pears, is beyond my understanding.
Hi Bob,
>The question is whether this setup is sensitive enough to see the effects of resistor voltage distortion, since the voltage being applied is only 5V rms (max output of my analyzer).
I think your setup is sensitive enough, but maybe it is more difficult to find such a bad resistor that will show a significant distortion.
BTW, you also have a sound card from ESI. I hope it contains better R's than mine.
Cheers, E.
>The question is whether this setup is sensitive enough to see the effects of resistor voltage distortion, since the voltage being applied is only 5V rms (max output of my analyzer).
I think your setup is sensitive enough, but maybe it is more difficult to find such a bad resistor that will show a significant distortion.
BTW, you also have a sound card from ESI. I hope it contains better R's than mine.
Cheers, E.
Most likely. According to wikipedia most SMD R's are of this type (thick film).
Clearly not to ESI, otherwise they wouldn't equip their so called 'professional' sound cards with crappy resistors.
To find out whether they are apples or pears.
Last time I've checked, sound cards are not intended to be instruments for measuring ppm distortions, but are sound reproduction devices. As far as we agree that thick film resistors are not anywhere close to the audibility threshold, ESI's decision to use them (instead of much more expensive thin film devices) was perfectly justified.
You attempted to use a device outside the scope it was intended for, simply as that.
Hi Walter
Do you mean the actual quantum effect?
Or do you use the term descriptively, for classical conduction?
Either would be extraordinary here, but for different reasons.
What was your source?
Best wishes
David
There is no "classic conduction" tunneling, and there's nothing extraordinary in that. My source was school (yes, they teach such stuff here, if you have an interest in) but I guess Google would also provide some references. Took me 10 seconds to find one: http://www.barthelectronics.com/pdf_files/Application%20note%201%20Voltage%20Coefficient%20Products_Pulse%20Page.pdf
"Thin metal film can also be used to achieve reasonable resistance values, but these resistors have a high voltage coefficient. This may result from the extremely thin metal film deposited on a very rough ceramic substrate, that allows tunneling or current flow across the ceramic valleys."
Resistor voltage distortion measurements
Here are some results on resistor distortion measurements that I made. The setup included my Cordell THD analyzer and an HP3580A spectrum analyzer monitoring the distortion output of the THD analyzer. The THD analyzer was operating at its maximum output of just below 5.0V into a resistor divider comprising a 6.8k series resistor (the DUT) and a 681 ohm shunt to ground (a quality 2-watt metal film resistor). Attenuation of the divider was about 21dB. Resulting input to the analyzer was about 0.45V. THD+N of the analyzer reads about 0.0006%, which is mostly second harmonic and noise. Voltage across the DUT resistor was 4V.
The spectrum analyzer was set to a 3Hz bandwidth and a very slow sweep to minimize the noise floor.
With a 2W metal film DUT, H2 was down 111dB at about 0.0003%, or 3 ppm. More importantly for resistor voltage distortion, H3 was buried in the noise floor, at below –134dB, or 0.2 ppm.
With a 1/4W carbon film DUT, H3 was just above the noise floor at –130dB, or 0.3 ppm.
To see some better evidence of resistor voltage distortion, I then connected a 40-year-old half-watt 10% carbon composition DUT. It finally showed a clearly discernable H3 at –117dB, or 1.4 ppm.
I note for reference that according to Bruce’s equation, a 10 ppm/V resistor with 4V across it should show H3 of 10ppm X 4 / 5.9 = 40/5.9, or about 7ppm. Since the carbon composition DUT showed 1.4 ppm under these conditions, it looks like this crummy carbon composition resistor has a Kv of only 2 ppm.
Moreover, at 0.3 ppm H3, it looks like the humble carbon film resistor has a Kv of only about 0.5 ppm.
These results are surprising to me. I hope I have not goofed in some way. These were difficult measurements, but I checked them a few times. As a sanity check, I injected a known –80dB 6kHz pilot tone at the output of the analyzer’s oscillator so as to cause the THD analyzer to read 0.01% and indicate the –80dB point on the spectrum analyzer.
Over a small sample size, these results, if correct, suggest that resistor distortion may not be as bad as we thought.
There was one more interesting observation. With the carbon composition as the DUT, I dropped the operating voltage level by 3dB. The H3 dropped by 6dB. This runs contrary to Bruce’s equation, which suggests that H3 falls in proportion to level for resistor voltage distortion.
Cheers,
Bob
Here are some results on resistor distortion measurements that I made. The setup included my Cordell THD analyzer and an HP3580A spectrum analyzer monitoring the distortion output of the THD analyzer. The THD analyzer was operating at its maximum output of just below 5.0V into a resistor divider comprising a 6.8k series resistor (the DUT) and a 681 ohm shunt to ground (a quality 2-watt metal film resistor). Attenuation of the divider was about 21dB. Resulting input to the analyzer was about 0.45V. THD+N of the analyzer reads about 0.0006%, which is mostly second harmonic and noise. Voltage across the DUT resistor was 4V.
The spectrum analyzer was set to a 3Hz bandwidth and a very slow sweep to minimize the noise floor.
With a 2W metal film DUT, H2 was down 111dB at about 0.0003%, or 3 ppm. More importantly for resistor voltage distortion, H3 was buried in the noise floor, at below –134dB, or 0.2 ppm.
With a 1/4W carbon film DUT, H3 was just above the noise floor at –130dB, or 0.3 ppm.
To see some better evidence of resistor voltage distortion, I then connected a 40-year-old half-watt 10% carbon composition DUT. It finally showed a clearly discernable H3 at –117dB, or 1.4 ppm.
I note for reference that according to Bruce’s equation, a 10 ppm/V resistor with 4V across it should show H3 of 10ppm X 4 / 5.9 = 40/5.9, or about 7ppm. Since the carbon composition DUT showed 1.4 ppm under these conditions, it looks like this crummy carbon composition resistor has a Kv of only 2 ppm.
Moreover, at 0.3 ppm H3, it looks like the humble carbon film resistor has a Kv of only about 0.5 ppm.
These results are surprising to me. I hope I have not goofed in some way. These were difficult measurements, but I checked them a few times. As a sanity check, I injected a known –80dB 6kHz pilot tone at the output of the analyzer’s oscillator so as to cause the THD analyzer to read 0.01% and indicate the –80dB point on the spectrum analyzer.
Over a small sample size, these results, if correct, suggest that resistor distortion may not be as bad as we thought.
There was one more interesting observation. With the carbon composition as the DUT, I dropped the operating voltage level by 3dB. The H3 dropped by 6dB. This runs contrary to Bruce’s equation, which suggests that H3 falls in proportion to level for resistor voltage distortion.
Cheers,
Bob
Interesting idea. I think one would need to take a differential output from the bridge into a diffamp with very low common mode distortion, then pass the signal to either a THD analyzer or spectrum analyzer, or both. I don't think a bridge by itself would give a deep enough null to make such measurements. Did I understand correctly what you had in mind?
Cheers,
Bob
Hi David,
Not sure what to do here, or maybe I don't understand. When I dropped the level by 3dB, it was with the carbon composition DUT. Are you suggesting doing the level drop with the high-quality "reference" resistor in the DUT position?
I completely understand the need to make sure that the analyzer residual is not changing as a result of the level change. However, the 3H analyzer residual is buried in the noise in both cases. The only way I was able to do the level drop test and get useful results was to use a DUT that made enough 3H to be at least 6dB above the noise floor to begin with. Its actually a bit worse than that, since when I drop the level by 3dB the noise floor comes up 3dB.
I hope I have answered your question and understood what you had in mind.
Cheers,
Bob
You can put a trimmer across the reference resistors to trim out the residual. If the resistors are all 1% or so, then the trimmer will only contribute no more than 1% or so of the result. So if the trimmer THD is say 0.0005%, the contribution would be something like <0.000005%.
With this method you don't necessarily need a low-distortion input signal either, since when the residual is isolated, it can only be harmonics of the input signal and it's harmonics. So if you want your result to be within say 20%, then 20%THD is in fact acceptable for an input signal as long as the harmonics don't disrupt the measurement equipment.
This means you could couple your SG to the input through a signal transformer to have an isolated measurement so you wouldn't need a differential amplifier to measure the result.
I'm not sure about this, but at any rate, Conrad Hoffman shows what can be done with a good bridge setup:
Measured Differences Between Capacitors for Audio Applications
With this method you don't necessarily need a low-distortion input signal either, since when the residual is isolated, it can only be harmonics of the input signal and it's harmonics. So if you want your result to be within say 20%, then 20%THD is in fact acceptable for an input signal as long as the harmonics don't disrupt the measurement equipment.
This means you could couple your SG to the input through a signal transformer to have an isolated measurement so you wouldn't need a differential amplifier to measure the result.
I'm not sure about this, but at any rate, Conrad Hoffman shows what can be done with a good bridge setup:
Measured Differences Between Capacitors for Audio Applications
I suppose if you can't see the distortion because it's buried in the noise then a baseline check isn't possible. Trying to take measurements so close to the residual is difficult and always taken with a grain of salt.
The possibility of analyzer / generator additive, subtractive or partial influence on the DUT distortion is a concern and leaves me scratching my head. I wish there was an easy way of separating the two. It would help if we new the phase relation of both.