NickKUK,
you may have disregarded a relevant snake oil aspect.
The ratio of magnetic energy and thermal energy is a rough thermodynamic measure of efficiency. Another measure of efficiency is the magnitude of excitation energy. A higher efficiency impairs the paramagnetic property which decisively determines signal transfer quality. Who does not like high efficiency and low distortion? Sadly, they cannot coexist in the same physical system.
Again, distortion is much more than just thermal noise.
you may have disregarded a relevant snake oil aspect.
The ratio of magnetic energy and thermal energy is a rough thermodynamic measure of efficiency. Another measure of efficiency is the magnitude of excitation energy. A higher efficiency impairs the paramagnetic property which decisively determines signal transfer quality. Who does not like high efficiency and low distortion? Sadly, they cannot coexist in the same physical system.
Again, distortion is much more than just thermal noise.
If this thesis is fact, then we may never get lower distortion than metal film.
The foil resistors introducing a casimir effect.
Electronic zero-point fluctuation forces inside circuit components
https://arxiv.org/pdf/1612.03250.pdf
The foil resistors introducing a casimir effect.
Electronic zero-point fluctuation forces inside circuit components
https://arxiv.org/pdf/1612.03250.pdf
Attachments
Hello All,
I am away from my tools for a few days at the beach.
Sitting here looking out the window, at past data and Bruce Hofer’s Audio Xpress article.
Bruce gave the impression that there was some lumping or local peaking at 5 to 200Hz, “Problems occur in a middle zone of approximately 5 to 200 Hz”. The sliding FFT data for the Vishay PTF65 resistor did not show any localized distortion for the PTF65 1000Ohm resistors in that 5 to 200Hz range. The data did show gradually, smoothly increasing distortion with decreasing frequency. The increasing distortion looked like the graph in Bob Cordell figure near page 370 in his new Power Amplifier book 2nd edition hardback.
Looking at the Vishay sliding FFT data for the Vishay Bulk Foil S102 1K Ohm resistor,
This resistor smoothly increased distortion with decreasing frequency. The slope of the increasing resistance with decreasing frequency was much steeper. The measured distortion was much higher for the Bulk Foil resistor compared to the thin film NiCrome metal film resistor.
The Bulk Foil resistor has a much lower Thermal Coefficient of Resistance than the thin film NiChrome resistor, yet it increases distortion much faster and higher with decreasing frequency.
I also tested some Vishay RN65 milspec metal film resistors. Lower TCR RN65’s had lower distortion than higher TCR RN65 resistors. Smaller size (smaller thermal mass) RN55 milspec metal film resistors had higher distortion than the RN65.
What is the mechanism of injury? What causes resistor distortion? Don’t know! The cause of distortion seems to be similar but different for NiChrome metal film and Bulk Foil.
Increasing TCR increases distortion, I am thinking the assumption that the mechanisms are different but both mechanisms are functions of heat and frequency.
Thanks DT
I am away from my tools for a few days at the beach.
Sitting here looking out the window, at past data and Bruce Hofer’s Audio Xpress article.
Bruce gave the impression that there was some lumping or local peaking at 5 to 200Hz, “Problems occur in a middle zone of approximately 5 to 200 Hz”. The sliding FFT data for the Vishay PTF65 resistor did not show any localized distortion for the PTF65 1000Ohm resistors in that 5 to 200Hz range. The data did show gradually, smoothly increasing distortion with decreasing frequency. The increasing distortion looked like the graph in Bob Cordell figure near page 370 in his new Power Amplifier book 2nd edition hardback.
Looking at the Vishay sliding FFT data for the Vishay Bulk Foil S102 1K Ohm resistor,
This resistor smoothly increased distortion with decreasing frequency. The slope of the increasing resistance with decreasing frequency was much steeper. The measured distortion was much higher for the Bulk Foil resistor compared to the thin film NiCrome metal film resistor.
The Bulk Foil resistor has a much lower Thermal Coefficient of Resistance than the thin film NiChrome resistor, yet it increases distortion much faster and higher with decreasing frequency.
I also tested some Vishay RN65 milspec metal film resistors. Lower TCR RN65’s had lower distortion than higher TCR RN65 resistors. Smaller size (smaller thermal mass) RN55 milspec metal film resistors had higher distortion than the RN65.
What is the mechanism of injury? What causes resistor distortion? Don’t know! The cause of distortion seems to be similar but different for NiChrome metal film and Bulk Foil.
Increasing TCR increases distortion, I am thinking the assumption that the mechanisms are different but both mechanisms are functions of heat and frequency.
Thanks DT
Last edited:
Hello Jan, Yes measured by me. Some of the plots are posted in this thread. Some data was less formal, as in after the test/plot I wrote a note in my lab book.
Yes I believe that the distortion is all about heat input (watts) plus cyclic heating and cooling.
I will retest a few resistors and post some test plots here after the weekend if you would like.
Thanks DT
The answer is that the TCR specification for foil resistors is quoted at DC. The low TCR of Vishay foil is achieved by using a resistance foil that has a specific thermal expansion property, bonded to an engineered ceramic substrate that has an equivalent and opposite thermal/mechanical expansion property.
Stated more simply: the foil is designed to expand with temperature at a specific rate, and it is bonded to a ceramic substrate that contracts at the same rate with temperature. So, to a first order, the foil's mechanical expansion is countered by the ceramic substrate's contraction. At very low frequencies (or very high frequencies), the substrate and foil will be in mechanical equilibrium over all parts of the signal waveform, and the net effect is to reduce the effect of the foil's native TCR, especially as it is caused by signal induced heating.
Now, imagine a frequency somewhere between the extremes of DC and HF. At such a frequency, when a signal passes through the foil, it heats the foil, expands the foil, and then the heat travels to the ceramic substrate, heating it after a little delay as the heat diffuses from the foil to the ceramic. But, the frequency is low enough that the contraction of the substrate is 180 degrees out of phase with the expansion of the foil. When the foil is expanding, the substrate is also expanding, making the effects of signal induced heating twice as large. In this worst case, the native TCR of the foil is effectively enlarged by 2x, making the TCR induced 3rd harmonic distortion even worse than it would be with a simple substrate.
This is the reason why the "trick" reactive substrate of a foil resistor is not useful for audio frequency uses. Sure, distortion is harder to notice at LF, but still, the numbers bear this out, and if low distortion is the goal, as Bruce Hofer says, foil resistors should not be used. Traditional, high quality metal film resistors with inherently low temperature coefficients, achieved by engineering the thin film itself, will provide the best performance.
Hello,
I believe that it is easy to get lost.
Fist the heat modulated distortion shown in the posted FFT’s smoothly and continuously decreases with increasing frequency. There was no peaking at any particular band of low frequencies. There was no peaking at any band of frequencies up to over 90 khZ.
The opposing thermal expansion and contraction discussion is interesting. As discussed there would be band of low frequencies where distortion would peak. That did not happen in the measurements that I made.
Thanks DT
I believe that it is easy to get lost.
Fist the heat modulated distortion shown in the posted FFT’s smoothly and continuously decreases with increasing frequency. There was no peaking at any particular band of low frequencies. There was no peaking at any band of frequencies up to over 90 khZ.
The opposing thermal expansion and contraction discussion is interesting. As discussed there would be band of low frequencies where distortion would peak. That did not happen in the measurements that I made.
Thanks DT
Out of interest to try DT's experimental setup, but from a poor-man's equipment perspective, I enjoyed a few hours preparing a bridge and isolated excitation signal supply to get a feel for what I (or any mere mortal 
) could achieve on the bench.
I only have a EMU0404 USB soundcard, so set up a low distortion isolated sinewave supply by connecting the headphone output through a TDA7492 class D amp module with 12V battery supply, in to a 1:4 line transformer. I used REW's ability to suppress harmonic levels on the 21Vrms output going to the bridge to about 0.004% (about -100dB below fundamental). I could then use my standard scope probe connection across the bridge to have as a soundcard unbalanced input, such that one bridge terminal was effectively at local soundcard ground.
I only had a box of 1.2k PRO2 to play with, so sorted out two sets of arms with 0.01% resistance matching (using a HP3497A and kelvin clips). One set of arms used 2 series/2 parallel for each 1.2k net resistor arm. The other set of arms just used a single resistor for each arm. That gave a nominal -90dB reduced H1 across the bridge, along with 0.06% H2 (-70dB below H1) and 0.1% H3 (-60dB below H1) signals sufficiently (at least 10-20dB) showing above the noise floor.
On first glance, the bridge output HD levels have degraded by about 30-40dB. I didn't try to see if those harmonic levels 'improved' with lower excitation signal level, as this setup doesn't have low enough noise floor to expect to track any change much further, and the base resisitor I used in the bridge probably has way too much tempco.
Given the PRO2 batch of tested resistors all had a noticeable positive tempco (circa 50-100ppm), and a spec of up to 150ppm, it doesn't seem appropriate to use these bridge arms for testing exotic resistors (which was not my intent anyway as I don't have any), as they aren't anywhere near being 'blameless'. But for fun I will try to check out comparative harmonic levels with any vintage resistor or two I can find.
) could achieve on the bench.I only have a EMU0404 USB soundcard, so set up a low distortion isolated sinewave supply by connecting the headphone output through a TDA7492 class D amp module with 12V battery supply, in to a 1:4 line transformer. I used REW's ability to suppress harmonic levels on the 21Vrms output going to the bridge to about 0.004% (about -100dB below fundamental). I could then use my standard scope probe connection across the bridge to have as a soundcard unbalanced input, such that one bridge terminal was effectively at local soundcard ground.
I only had a box of 1.2k PRO2 to play with, so sorted out two sets of arms with 0.01% resistance matching (using a HP3497A and kelvin clips). One set of arms used 2 series/2 parallel for each 1.2k net resistor arm. The other set of arms just used a single resistor for each arm. That gave a nominal -90dB reduced H1 across the bridge, along with 0.06% H2 (-70dB below H1) and 0.1% H3 (-60dB below H1) signals sufficiently (at least 10-20dB) showing above the noise floor.
On first glance, the bridge output HD levels have degraded by about 30-40dB. I didn't try to see if those harmonic levels 'improved' with lower excitation signal level, as this setup doesn't have low enough noise floor to expect to track any change much further, and the base resisitor I used in the bridge probably has way too much tempco.
Given the PRO2 batch of tested resistors all had a noticeable positive tempco (circa 50-100ppm), and a spec of up to 150ppm, it doesn't seem appropriate to use these bridge arms for testing exotic resistors (which was not my intent anyway as I don't have any), as they aren't anywhere near being 'blameless'. But for fun I will try to check out comparative harmonic levels with any vintage resistor or two I can find.
Unless we discover a percular that magically works at room temps vs absolute zero. A quantum distill fridge thermal output may cause tube owners mouths to drop..
I just re-did the bridge test using a full bridge of single 1k 0.4W MF (unknown make and spec) that were resistance matched as pairs (taken from a tape reel). With 20Vrms across the bridge and up to H5 nulled for < 100dB below H1, the basic bridge output gave H1 about 90dB below excitation level, and H2 about 63dB below H1, and H3 about 53dB below. NF was only about 5dB below H2.
So outcome is nearly similar to the bridge using PRO2 resistors, so not much to critically pull out except for consistent results.
I then replaced one arm with a vintage CC 0.5W'ish 1k resistor - compressed brown body, 10% tolerance as typically used in 1950's amps - it measured 1.019k so must have had an easy life. I tweaked the other arm to get within an ohm or so using the 1k MF plus a few low value MF's in series. H1 null wasn't as good, but harmonic levels were high, with H2 about 18dB below H1, and H3 about 14dB below H1.
At least I have another way now to generate low distortion, high signal level, isolated test tones if needed.
So outcome is nearly similar to the bridge using PRO2 resistors, so not much to critically pull out except for consistent results.
I then replaced one arm with a vintage CC 0.5W'ish 1k resistor - compressed brown body, 10% tolerance as typically used in 1950's amps - it measured 1.019k so must have had an easy life. I tweaked the other arm to get within an ohm or so using the 1k MF plus a few low value MF's in series. H1 null wasn't as good, but harmonic levels were high, with H2 about 18dB below H1, and H3 about 14dB below H1.
At least I have another way now to generate low distortion, high signal level, isolated test tones if needed.
Last edited:
Hello All,
Here on my bench I have rounded up several bags of 1000 Ohm Vishay RN series axial metal film resistors; RN55D, RN55E, RN60D, RD60E, RN65D and RN65E. The RN55 is the smallest, the RN60 is mid-size and the RH65 is the largest and heaviest. I am taking a look at the distortion of these resistors when they are operated at 10volt AC input.
D is 100ppm TCR. E is 25ppm TCR
At 10volt input all the resistors will produce the same amount of heat. Because of their smaller size the RN55 resistors will heat faster and reach a higher temperature. The RN65’s will heat slower and not get as hot. The mid-size RN60’s will be down the middle in terms of heating rate and temperature.
Turn off the input voltage and I am thinking that the RN55’s will cool the fastest and that the largest RN65’s will cool the slowest. Sweep the input frequency and things get more complicated in terms of rates and limits of temperature change. Heating rate (change in temperature/change in time), cooling rate (change in temperature/change in time) and degrees of temperature swing. I am thinking that slower heating and slower cooling means smaller temperature changes. Smaller temperature change means less distortion.
Now we toss in the idea of Temperature coefficient of Resistance TCR, the change of resistance per degree C temperature change, things get more complicated. The changing temperature of the resistor causes primarily H3 distortion. A lower TCR means lower resistance change due to temperature change and lower distortion.
To minimize metal film resistor distortion, we need to minimize TCR and minimize the range of temperature swing. I am thinking that metal foil resistors heat and cool faster leading to wider temperature swings than metal film resistors.
More to be said later.
Now to testing the resistors. To follow with the previous test methods, I will use the test bridge. To help calibrate the bridge method I will use brute force and the analyzer to test the resistor with a large enough input voltage to cause resulting distortion levels to rise above the analyzer floor.
Thanks DT
Here on my bench I have rounded up several bags of 1000 Ohm Vishay RN series axial metal film resistors; RN55D, RN55E, RN60D, RD60E, RN65D and RN65E. The RN55 is the smallest, the RN60 is mid-size and the RH65 is the largest and heaviest. I am taking a look at the distortion of these resistors when they are operated at 10volt AC input.
D is 100ppm TCR. E is 25ppm TCR
At 10volt input all the resistors will produce the same amount of heat. Because of their smaller size the RN55 resistors will heat faster and reach a higher temperature. The RN65’s will heat slower and not get as hot. The mid-size RN60’s will be down the middle in terms of heating rate and temperature.
Turn off the input voltage and I am thinking that the RN55’s will cool the fastest and that the largest RN65’s will cool the slowest. Sweep the input frequency and things get more complicated in terms of rates and limits of temperature change. Heating rate (change in temperature/change in time), cooling rate (change in temperature/change in time) and degrees of temperature swing. I am thinking that slower heating and slower cooling means smaller temperature changes. Smaller temperature change means less distortion.
Now we toss in the idea of Temperature coefficient of Resistance TCR, the change of resistance per degree C temperature change, things get more complicated. The changing temperature of the resistor causes primarily H3 distortion. A lower TCR means lower resistance change due to temperature change and lower distortion.
To minimize metal film resistor distortion, we need to minimize TCR and minimize the range of temperature swing. I am thinking that metal foil resistors heat and cool faster leading to wider temperature swings than metal film resistors.
More to be said later.
Now to testing the resistors. To follow with the previous test methods, I will use the test bridge. To help calibrate the bridge method I will use brute force and the analyzer to test the resistor with a large enough input voltage to cause resulting distortion levels to rise above the analyzer floor.
Thanks DT
DT, as I understand your comments, a bridge test will cause a 'static' resistance change in the DUT due to dissipation and ambient air and lead cooling conditions, as well as a signal induced dynamic resistance change around the static value due to the construction and performance of the part.
The static change will have a time-constant, and given time should reach an equilibrium state.
Two or three arms of the bridge would be assumed to be 'blameless', assuming they have also reached equilibrium temperature conditions, and do not exhibit a significant signal induced dynamic resistance change. The practical technique of using multiple series-parallel parts to make a blameless bridge arm is based on each resistor in the arm being exposed to 50%, 33% 20% .... lower voltage, and hence 25%, 11% ... lower dissipation.
As far as result confidence, I could suggest the following measurement setups may be worthwhile pursuing:
(a) the bridge output showing twice the harmonic levels when the bridge is configured with two blameless arms (ie. two arms with DUTs), as compared to the bridge configured with 3 blameless arms (ie. just one arm with a DUT).
(b) the bridge output showing negligible harmonic levels when the bridge is configured with four blameless arms (ie. negligible as compared to results from (a)).
I can appreciate that the 2 DUT arm configuration bridge may well cancel some signals such as excitation voltage harmonics, and double some signals such as DUT non-linear response, but have some concern about how well that plays out in a world that is well below -120dB.
The static change will have a time-constant, and given time should reach an equilibrium state.
Two or three arms of the bridge would be assumed to be 'blameless', assuming they have also reached equilibrium temperature conditions, and do not exhibit a significant signal induced dynamic resistance change. The practical technique of using multiple series-parallel parts to make a blameless bridge arm is based on each resistor in the arm being exposed to 50%, 33% 20% .... lower voltage, and hence 25%, 11% ... lower dissipation.
As far as result confidence, I could suggest the following measurement setups may be worthwhile pursuing:
(a) the bridge output showing twice the harmonic levels when the bridge is configured with two blameless arms (ie. two arms with DUTs), as compared to the bridge configured with 3 blameless arms (ie. just one arm with a DUT).
(b) the bridge output showing negligible harmonic levels when the bridge is configured with four blameless arms (ie. negligible as compared to results from (a)).
I can appreciate that the 2 DUT arm configuration bridge may well cancel some signals such as excitation voltage harmonics, and double some signals such as DUT non-linear response, but have some concern about how well that plays out in a world that is well below -120dB.
Why would one expect a narrow peak? If the thermal time constants were identical and 'optimally bad', then there is a maximum of 2x magnification of the TCR induced distortion. A factor of two away from those time constants, the summation would only be sqrt(2) extra distortion, not the 2x seen at the maximum. That's a pretty wide "peak" right?
I like to think of it as a phase shift between the heating/cooling cycle of the resistive body, and the substrate. How would that phase shift change with frequency?
I think we agree that at high frequencies, the phase shift is (near) zero and the design works as intended, countering the expansion of the resistive body with a contraction of the substrate, implying 180deg phase difference or opposite phase.
How does that evolve if we go down in frequency?
Jan
I think we agree that at high frequencies, the phase shift is (near) zero and the design works as intended, countering the expansion of the resistive body with a contraction of the substrate, implying 180deg phase difference or opposite phase.
How does that evolve if we go down in frequency?
Jan
So if everything were lumped, the transfer from power to resistance variations would be some small constant times
1/(s tau1 + 1) - 1/(s tau2 + 1) = s (tau2 - tau1)/((s tau1 + 1) (s tau2 + 1))
That's indeed a very wide bandpass. For s = j omega, it first increases with 20 dB/decade, then has a very wide peak, and then drops with 20 dB/decade. I don't know what effect the distributed rather than lumped nature of the heat transfer has.
1/(s tau1 + 1) - 1/(s tau2 + 1) = s (tau2 - tau1)/((s tau1 + 1) (s tau2 + 1))
That's indeed a very wide bandpass. For s = j omega, it first increases with 20 dB/decade, then has a very wide peak, and then drops with 20 dB/decade. I don't know what effect the distributed rather than lumped nature of the heat transfer has.