Reading the posts. As you guys discuss how resistors affect the sound, I wonder if thermal distortion is the cause. The Vishays have lowest temperature coefficient.
says "iko" post 829
Perhaps in two years time I will hear the difference too
Now, I don't hear the difference between different resistors, but I'll take your word for it and try the fancier resistors.
I think I'll stay away from the expensive Vishay bulk foil resistors for now, but I got this idea, and you can call me crazy if you want. I'll wind my own resistors (as an experiment) from evanohm wire. They use evanohm in high precision resistors and fancy measuring equipment... or at least they did some years back.
says "iko" post 829
Perhaps in two years time I will hear the difference too
Now, I don't hear the difference between different resistors, but I'll take your word for it and try the fancier resistors.
I think I'll stay away from the expensive Vishay bulk foil resistors for now, but I got this idea, and you can call me crazy if you want. I'll wind my own resistors (as an experiment) from evanohm wire. They use evanohm in high precision resistors and fancy measuring equipment... or at least they did some years back.
Thermal coupling of the input pair is to attenuate drift of the output offset as internal temperatures vary.
An ordinary LTP with separated input pair is usually seen as sufficient for minimising signal related distortions without the need for "fast thermal-tracking".
Hi Andrew,
I think I'll disagree with you on that point. Thermally coupling the diff pair is always a benefit. Just watching the drifts on my matching jig (uses a diff pair circuit) between covered or not, or isolated (with foam around both) and covered shows dramatic changes. Assuming you have painstakingly matched pairs of the transistors, you may as well finish it and isolate & bond them together.
You use glue, I use thermal compound and heat shrink tubing. Isolating the pair from the ambient conditions will help stop drifts and air currents from affecting the balance (in high gain situations, like an RIAA amp).
The diff pair is where the distortion is subtracted from the signal (input), so anything that messes that up will have an effect. I probably wouldn't hear this on music, but I can measure the changes. For some who claim to hear that which cannot be measured, I'll assume this could be heard plain as day for those folks.
You matched them, finish the the job. Not to mention the fact that it tells the next technician that there is something special about that pair of transistors.
-Chris
I think I'll disagree with you on that point. Thermally coupling the diff pair is always a benefit. Just watching the drifts on my matching jig (uses a diff pair circuit) between covered or not, or isolated (with foam around both) and covered shows dramatic changes. Assuming you have painstakingly matched pairs of the transistors, you may as well finish it and isolate & bond them together.
You use glue, I use thermal compound and heat shrink tubing. Isolating the pair from the ambient conditions will help stop drifts and air currents from affecting the balance (in high gain situations, like an RIAA amp).
The diff pair is where the distortion is subtracted from the signal (input), so anything that messes that up will have an effect. I probably wouldn't hear this on music, but I can measure the changes. For some who claim to hear that which cannot be measured, I'll assume this could be heard plain as day for those folks.
You matched them, finish the the job. Not to mention the fact that it tells the next technician that there is something special about that pair of transistors.
-Chris
that sounds like we are agreeing, slow drifts due to different package temperatures are very noticeable.
again we are agreed.
It's the fast junction temperature change that I was commenting on, that is corrected by the circuit, particularly the feedback into the -IN. This cannot correct the drift, but it does very effectively correct for Tj. Any opamp shows us it works.
Most power amps that do not thermally couple the input pair show us it works. D.Self's "Blameless" does not show any Thermally Coupled input pair and yet he achieves low distortion over the whole audio band.
The thermally coupled input pair works for drift of output offset, but seems to be virtually unnecessary for the amplifier performance. Yes, I suspect that 7ppm distortion could probably be reduced by adopting this, but can we hear that? Many of us cannot even measure that.
Last edited:
The unwanted energy of distortion can be tightly located in time, thus very audible, or spread out in time and inaudible.
Consider a 10th harmonic of 1KHz *measured* to be 0.01%.
This distortion may be a tone-burst occurring once per second, thus powerful at 10% of fundamental amplitude (and a FFT distortion analyzer would detect this, if the observation time were 1 second, and the FFT distortion analyzer would report the 10KHz burst as being 0.01% strength of the 1KHz fundamental).
An analog distortion analyzer may or may not detect the once-per-second tone burst at 10KHz; depends on how the frequency sweeping does or does not align with when the tone burst occurs. The analog DA might report the 10KHz as being 10% of the 1KHz fundamental strength. Or not.
Now lets consider 0.001% (-100dBc) energy events. This is only 10X weaker than my initial example. Can we hear them? Depends on how concentrated-in-time the energy is.
Regarding bipolar transistor thermal distortion, the critical region is the Emitter-Base junction. It has low-reverse-breakdown, thus is a thin/narrow region, and is located just the base-region-width away from the *heat* of the collector region. I've seen bipolars (operating on silicon, in cooperation with other transistors) run-away thermally at 15 volts; we added emitter-degen resistors (we had several emitter strips so added several such resistors) and the thermal-runaway no longer occurred. The bipolars operated at 4mA per strip (20micron by 150micron, from memory) and 15 volts, thus 60 milliWatts per strip. This led to self-destruction via positive feedback thermally. Took about 6 months to let the design-team come to that conclusion, and install the emitter-degen Rs.
Thermal distortion is real, it can be positive-feedback leading to a snappy overshoot on any transient event; this means thermal-positive-feedback (TPF) is a fast bandwidth narrow-in-time energy event; a 10 micron region has 1.14 microsecond thermal time constant; in the real world, the emitter/friends form an ugly distributed thermal structure.
This experience with bipolar self-destruction, and years of musing about distortion narrow-in-time or spread-in-time, is why we've included "thermal distortion" with adjustable thermal-masses and thermal-resistances, in both the operational amplifiers and the resistors of Signal Chain Explorer (robustcircuitdesign.com, free for download).
Consider a 10th harmonic of 1KHz *measured* to be 0.01%.
This distortion may be a tone-burst occurring once per second, thus powerful at 10% of fundamental amplitude (and a FFT distortion analyzer would detect this, if the observation time were 1 second, and the FFT distortion analyzer would report the 10KHz burst as being 0.01% strength of the 1KHz fundamental).
An analog distortion analyzer may or may not detect the once-per-second tone burst at 10KHz; depends on how the frequency sweeping does or does not align with when the tone burst occurs. The analog DA might report the 10KHz as being 10% of the 1KHz fundamental strength. Or not.
Now lets consider 0.001% (-100dBc) energy events. This is only 10X weaker than my initial example. Can we hear them? Depends on how concentrated-in-time the energy is.
Regarding bipolar transistor thermal distortion, the critical region is the Emitter-Base junction. It has low-reverse-breakdown, thus is a thin/narrow region, and is located just the base-region-width away from the *heat* of the collector region. I've seen bipolars (operating on silicon, in cooperation with other transistors) run-away thermally at 15 volts; we added emitter-degen resistors (we had several emitter strips so added several such resistors) and the thermal-runaway no longer occurred. The bipolars operated at 4mA per strip (20micron by 150micron, from memory) and 15 volts, thus 60 milliWatts per strip. This led to self-destruction via positive feedback thermally. Took about 6 months to let the design-team come to that conclusion, and install the emitter-degen Rs.
Thermal distortion is real, it can be positive-feedback leading to a snappy overshoot on any transient event; this means thermal-positive-feedback (TPF) is a fast bandwidth narrow-in-time energy event; a 10 micron region has 1.14 microsecond thermal time constant; in the real world, the emitter/friends form an ugly distributed thermal structure.
This experience with bipolar self-destruction, and years of musing about distortion narrow-in-time or spread-in-time, is why we've included "thermal distortion" with adjustable thermal-masses and thermal-resistances, in both the operational amplifiers and the resistors of Signal Chain Explorer (robustcircuitdesign.com, free for download).
Last edited:
Hi Andrew,
The problem with DC servos is that they work by driving an unmatched pair out of balance even more. Personally, I would probably apply correction from a DC servo after the voltage amp stage to balance the stuff that is really not balanced. That assumes a well matched diff pair to begin with. If they aren't, then all bets are off.
If you read Self, forget which page, he does show that a matched pair has a dramatic effect on how much distortion can be reduced when the pair is matched and in balance. Remember that this is the area where distortion is subtracted from the signal, and that diff pair is the detector for this process. I think we are talking about more than a few ppm difference here.
It would be an interesting experiment to see if the DC balance can be isolated from the diff pair completely and handled by a servo alone. That would allow a diff pair to remain in balance to handle the AC portion of the signal without interference from the servo.
In the BGW 750C and similar amplifiers, they use an op amp as the gain stage and some transistors to attain the swing they need. Interestingly, the DC offset adjustment is carried out by adjusting the current in the predrivers directly while you watch the DC offset of the output of the LM318 op amp. They want to see less than 100 mV by adjusting that current. It would appear that they are onto something that has been pretty much ignored by the rest of the industry. I really believe that they did DC offset correction properly in this amp. It is also very probably why some designers don't like the sound of a DC servo. The implementation of the servo is maybe more important than whether one exists or not. The take-away from this? Don't mess with the balance of a differential pair.
There's a direction to experiment for you.
-Chris
The problem with DC servos is that they work by driving an unmatched pair out of balance even more. Personally, I would probably apply correction from a DC servo after the voltage amp stage to balance the stuff that is really not balanced. That assumes a well matched diff pair to begin with. If they aren't, then all bets are off.
If you read Self, forget which page, he does show that a matched pair has a dramatic effect on how much distortion can be reduced when the pair is matched and in balance. Remember that this is the area where distortion is subtracted from the signal, and that diff pair is the detector for this process. I think we are talking about more than a few ppm difference here.
It would be an interesting experiment to see if the DC balance can be isolated from the diff pair completely and handled by a servo alone. That would allow a diff pair to remain in balance to handle the AC portion of the signal without interference from the servo.
In the BGW 750C and similar amplifiers, they use an op amp as the gain stage and some transistors to attain the swing they need. Interestingly, the DC offset adjustment is carried out by adjusting the current in the predrivers directly while you watch the DC offset of the output of the LM318 op amp. They want to see less than 100 mV by adjusting that current. It would appear that they are onto something that has been pretty much ignored by the rest of the industry. I really believe that they did DC offset correction properly in this amp. It is also very probably why some designers don't like the sound of a DC servo. The implementation of the servo is maybe more important than whether one exists or not. The take-away from this? Don't mess with the balance of a differential pair.
There's a direction to experiment for you.
-Chris
Hello anatech
Would you please show us a schematic including your way to connect the servo after the voltage amp so we can visualize your idea ?
Resistor and Capacitor Properties for RIAA success
hello "Simplistic NJFET RIAA" forum
I've been reading, from post #1 forward, to learn about necessary properties of resistors and capacitors. Now at page 162 (post 1,620)
and still reading, I want to offer the attached attempt, intended to summarize what properties of resistors and capacitors are important for RIAA success.
Please be very critical of my ideas, because its a learning experience for me.
thanks for sharing your DIY efforts in the past
tankcircuitnoise
hello "Simplistic NJFET RIAA" forum
I've been reading, from post #1 forward, to learn about necessary properties of resistors and capacitors. Now at page 162 (post 1,620)
and still reading, I want to offer the attached attempt, intended to summarize what properties of resistors and capacitors are important for RIAA success.
Please be very critical of my ideas, because its a learning experience for me.
thanks for sharing your DIY efforts in the past
tankcircuitnoise
Attachments
Hi tankcircuitnoise,
Sometimes too much detail will only stop you dead in your tracks. Some of the issues you are concerned about are completely swamped by other issues. There is always a certain amount of inductance in leads and wires. But those frequencies where you would be worried about are so far above the audio range that you can normally ignore them.
Dielectric absorption is normally a problem with high "K" dielectrics. So polystyrene, polypropylene and NP0/C0G ceramic capacitors are the preferred types of capacitors to work with. The worst would be solid tantalum and aluminum electrolytic capacitors.
Noise in resistors is minimized in good metal film resistors. Bulk foil and the like really aren't any better than metal film parts. I would go for the Dale series of resistors. Temperature coefficients aren't really a problem unless you allow the ambient temperatures or the dissipation in the resistors to climb to high values. It would be nice to stay in the 100 ppm per degree celsius.
When choosing component types, keep an eye on which properties are the most important, and the properties that don't affect the circuit strongly. Optimization for one issue normally means that the overall performance is compromised in one or more ways. So don't go for perfect in each part - within reason. Avoid the bigger problems and ignore the small ones unless they become performance limiting.
Design of an RIAA stage could be perfect on paper, yet building (execution) of the design is often where things break down and deliver lower performance than you anticipated. A good RIAA stage has a balance of the design, and the execution (building the actual circuit) of the design.
Best, Chris
Sometimes too much detail will only stop you dead in your tracks. Some of the issues you are concerned about are completely swamped by other issues. There is always a certain amount of inductance in leads and wires. But those frequencies where you would be worried about are so far above the audio range that you can normally ignore them.
Dielectric absorption is normally a problem with high "K" dielectrics. So polystyrene, polypropylene and NP0/C0G ceramic capacitors are the preferred types of capacitors to work with. The worst would be solid tantalum and aluminum electrolytic capacitors.
Noise in resistors is minimized in good metal film resistors. Bulk foil and the like really aren't any better than metal film parts. I would go for the Dale series of resistors. Temperature coefficients aren't really a problem unless you allow the ambient temperatures or the dissipation in the resistors to climb to high values. It would be nice to stay in the 100 ppm per degree celsius.
When choosing component types, keep an eye on which properties are the most important, and the properties that don't affect the circuit strongly. Optimization for one issue normally means that the overall performance is compromised in one or more ways. So don't go for perfect in each part - within reason. Avoid the bigger problems and ignore the small ones unless they become performance limiting.
Design of an RIAA stage could be perfect on paper, yet building (execution) of the design is often where things break down and deliver lower performance than you anticipated. A good RIAA stage has a balance of the design, and the execution (building the actual circuit) of the design.
Best, Chris
Dale/Vishay RN70E
Anatech....just thinking here. The Dale (now Vishay) resistors of metal-film, large in size and thus slow to respond to heating, with the 25PPM/degC temperature coefficient, costs $1+ each Mouser, compared to capacitors for RIAA or DC-blocking in the signal path. RN70E is physically big, 17 millimeters from left-to-right. To bend leads, you need 3+3 mm for the needle-nose-pliers, thus 23millimeters plus bend radius, perhaps 3+3mm more, plus 0.5+0.5mm lead radius, totally 23+6+1 = 30millimeters in center-to-center PCB hole spacing.
In exchange for allocating this space ---- for critical resistors exposed to changes in current and thus generating change in power and in temperature, what might we gain? The diameter of 4mm, or radius of 2mm, provides (2mm/1mm)^2 ) 11,400 microseconds thermal tau, or 4*11,400 =45,600 microseconds tau. This is about 22 radians/second or about 4 Hertz.
Which resistors are vulnerable? Those with high voltage & high current, thus the 2 Drain resistors, that set gain.
The "E" rating promises 25 ppm/degreeC thermal coefficient.
Given this is a transient condition with the music dynamics, the resistor never reaches thermal stability as the music changes. To model the distortion, to compute the temperature_rise, I'd just use the core_mass as the storage volume. The larger core, the larger resistor size, the RN70, seems the better choice even given 30mm center-center requirement.
Anatech....just thinking here. The Dale (now Vishay) resistors of metal-film, large in size and thus slow to respond to heating, with the 25PPM/degC temperature coefficient, costs $1+ each Mouser, compared to capacitors for RIAA or DC-blocking in the signal path. RN70E is physically big, 17 millimeters from left-to-right. To bend leads, you need 3+3 mm for the needle-nose-pliers, thus 23millimeters plus bend radius, perhaps 3+3mm more, plus 0.5+0.5mm lead radius, totally 23+6+1 = 30millimeters in center-to-center PCB hole spacing.
In exchange for allocating this space ---- for critical resistors exposed to changes in current and thus generating change in power and in temperature, what might we gain? The diameter of 4mm, or radius of 2mm, provides (2mm/1mm)^2 ) 11,400 microseconds thermal tau, or 4*11,400 =45,600 microseconds tau. This is about 22 radians/second or about 4 Hertz.
Which resistors are vulnerable? Those with high voltage & high current, thus the 2 Drain resistors, that set gain.
The "E" rating promises 25 ppm/degreeC thermal coefficient.
Given this is a transient condition with the music dynamics, the resistor never reaches thermal stability as the music changes. To model the distortion, to compute the temperature_rise, I'd just use the core_mass as the storage volume. The larger core, the larger resistor size, the RN70, seems the better choice even given 30mm center-center requirement.
Last edited by a moderator:
The thermal time constant of the metal film comes first.
That heat has to pass to the ceramic core before it's thermal time constant has any chance to contribute.
Then the heat escapes down the resistor leads, or goes in the opposite direction into the insulation layer, to be radiated/conducted into the air.
Your simplified thermal model is just too simple.
R.Cordell and D.Self look at multiple series connected thermal paths if you need ideas for creating a better model.
Just reducing the power through the components gets your amplifier into a better performing region.
I have recommended using <10% of the maximum power rating after reading much from proper Designers, i.e. the first 600mW 27k in the feedback path should not be exposed to more than 41Vpk, when a transient from a 100W into 8ohms amplifier nearly clips.
If this metal film never has to dissipate much power then it never gets hot and thus the 50ppm/C or 100ppm/C never gets excercised much. One could use 25ppm/C or better for these NFB resistors, but what gain will that bring?
I would need some mighty good equipment and knowledge to be able to compare 25ppm/C @ <10% to 100ppm/C @ <10% to be able to reach any experimental conclusions.
I am also fairly sure that using a 300ppm/C @ ~30% would give some good ammunition for rejecting Metal Oxide in the NFB path.
That heat has to pass to the ceramic core before it's thermal time constant has any chance to contribute.
Then the heat escapes down the resistor leads, or goes in the opposite direction into the insulation layer, to be radiated/conducted into the air.
Your simplified thermal model is just too simple.
R.Cordell and D.Self look at multiple series connected thermal paths if you need ideas for creating a better model.
Just reducing the power through the components gets your amplifier into a better performing region.
I have recommended using <10% of the maximum power rating after reading much from proper Designers, i.e. the first 600mW 27k in the feedback path should not be exposed to more than 41Vpk, when a transient from a 100W into 8ohms amplifier nearly clips.
If this metal film never has to dissipate much power then it never gets hot and thus the 50ppm/C or 100ppm/C never gets excercised much. One could use 25ppm/C or better for these NFB resistors, but what gain will that bring?
I would need some mighty good equipment and knowledge to be able to compare 25ppm/C @ <10% to 100ppm/C @ <10% to be able to reach any experimental conclusions.
I am also fairly sure that using a 300ppm/C @ ~30% would give some good ammunition for rejecting Metal Oxide in the NFB path.
Last edited:
Hi tankcircuitnoise,
Andrew and I do agree on most things. Not in this case as far as resistor heating is concerned.
You should be using a low tempco type of resistor, but you don't have to get crazy low in that area. For your feedback path you absolutely need a metal film (or better) type part. But there are lower limits where the extra performance does not benefit you. I wouldn't use any better than 50 ppm, and 100 ppm is a lot better than you will ever see in a carbon composition or metal oxide resistor. Don't forget too, some resistors have a voltage coefficient too. Consider any feedback path critical in this regard, and also tempco. Metal film resistors are fine for this location. If you really want to push the limits, they do make resistor networks where the two resistors you could use for the feedback network are together and share the same tempco. This generally makes the absolute temperature coefficient drop out of the equation. Pretty neat eh? Again, audio doesn't need this type of extreme performance. The thermal performance of your speakers will not be nearly as good as your amplifier. The sensitivity of the speaker drops as the voice coil temperature is increased. The types of circuits that use resistor networks and very low tempcos will be Military (guidance systems mostly) and Electronic test and measurement equipment. That's why they exist. With the variables that exist in home music systems you would be just throwing your money away by using the very low tempco resistors (anything below 50 ppm). Even the 50 ppm resistors are somewhat overdoing things.
-Chris
Andrew and I do agree on most things. Not in this case as far as resistor heating is concerned.
You should be using a low tempco type of resistor, but you don't have to get crazy low in that area. For your feedback path you absolutely need a metal film (or better) type part. But there are lower limits where the extra performance does not benefit you. I wouldn't use any better than 50 ppm, and 100 ppm is a lot better than you will ever see in a carbon composition or metal oxide resistor. Don't forget too, some resistors have a voltage coefficient too. Consider any feedback path critical in this regard, and also tempco. Metal film resistors are fine for this location. If you really want to push the limits, they do make resistor networks where the two resistors you could use for the feedback network are together and share the same tempco. This generally makes the absolute temperature coefficient drop out of the equation. Pretty neat eh? Again, audio doesn't need this type of extreme performance. The thermal performance of your speakers will not be nearly as good as your amplifier. The sensitivity of the speaker drops as the voice coil temperature is increased. The types of circuits that use resistor networks and very low tempcos will be Military (guidance systems mostly) and Electronic test and measurement equipment. That's why they exist. With the variables that exist in home music systems you would be just throwing your money away by using the very low tempco resistors (anything below 50 ppm). Even the 50 ppm resistors are somewhat overdoing things.
-Chris
OK, I read your next paragraph.
so far I don't see any disagreement
Again I agree.
Again, agreed on both points.
Not sure what to make of that section, since I have never used matched integrated resistors.
I think we are both agreeing that metal film 50ppm/C, or 100ppm/C, are good enough and provided the voltage coefficient does not ruin the ratio of upper to lower then all should be OK.
What part did we not agree on?
Something to do with the recommendation to not overheat the upper resistor by keeping dissipation below 10% of Pmax? where I said
Last edited:
Hi Andrew,
Feedback resistors are often called upon to dissipate large portions of their rated power and sometimes do exercise their thermal temperature coefficients. I've actually seen some that have burned a hole in the circuit board as they expired, amplifier following moments later. The exact amplifier that actually burned through the PCB? Carver Lightstar series 1 amplifier. I suspect it was allowed to run well into clipping without a load for an extended period of time. I forget which reviewer had it at the time, but he was completely irresponsible.
Anyway, the point being is that you cannot assume the feedback resistor going to the output will not be called upon to dissipate high amounts of power. This is especially true in a fault condition or operator abuse. Why do I assume that in the example above there was no load? Well, if there was one, there probably wasn't a load by the time the amplifier expired. I still have that circuit board around here somewhere. It reminds me to look past what would normally be considered a performance limit.
As I said Andrew, we normally do agree, but I wasn't certain that agreement held to everything I was about to post.
Best, Chris
Feedback resistors are often called upon to dissipate large portions of their rated power and sometimes do exercise their thermal temperature coefficients. I've actually seen some that have burned a hole in the circuit board as they expired, amplifier following moments later. The exact amplifier that actually burned through the PCB? Carver Lightstar series 1 amplifier. I suspect it was allowed to run well into clipping without a load for an extended period of time. I forget which reviewer had it at the time, but he was completely irresponsible.
Anyway, the point being is that you cannot assume the feedback resistor going to the output will not be called upon to dissipate high amounts of power. This is especially true in a fault condition or operator abuse. Why do I assume that in the example above there was no load? Well, if there was one, there probably wasn't a load by the time the amplifier expired. I still have that circuit board around here somewhere. It reminds me to look past what would normally be considered a performance limit.
As I said Andrew, we normally do agree, but I wasn't certain that agreement held to everything I was about to post.
Best, Chris