Bob, could I ask you as to why transistors with integrated temperature sensing never really became popular. They theoretically appear close to the ideal and yet such transistors namely the thermal track devices from ON and STD03x from sanken, are the only ones from major manufacturers and still remain fringe devices amongst amplifier builders. What went wrong?
The Sankens are different as they have the diode integrated on the same die as the darlington, while the ThermalTraks have a separate small chip within the case (of a regular diode, I think some MUR type). So the thermal characteristics of the two types probably are different.
Anyway, I am not impressed with this principle, I did (probably the only one here) some testing and the results were not what was expected (and not what was hoped).
See for yourself :
Thermal transient variation of power amp quiescent current | Linear Audio NL
A peek:
Conclusions from part 2
So called ThermalTrak devices, with an on-die diode that can be included in the bias control loop, offer no advantage regarding compensation for thermal transient bias current changes due to dynamically varying dissipation in the output devices. These diodes react to dissipation changes with a delay of 150 to 200 milliseconds. Furthermore, their reaction appears heavily low-pass filtered at fractional Hz frequencies, and is much slower than the transient thermal changes in the output devices themselves.
Jan
Anyway, I am not impressed with this principle, I did (probably the only one here) some testing and the results were not what was expected (and not what was hoped).
See for yourself :
Thermal transient variation of power amp quiescent current | Linear Audio NL
A peek:
Conclusions from part 2
So called ThermalTrak devices, with an on-die diode that can be included in the bias control loop, offer no advantage regarding compensation for thermal transient bias current changes due to dynamically varying dissipation in the output devices. These diodes react to dissipation changes with a delay of 150 to 200 milliseconds. Furthermore, their reaction appears heavily low-pass filtered at fractional Hz frequencies, and is much slower than the transient thermal changes in the output devices themselves.
Jan
Last edited:
Fashion and misunderstanding. Also, SIMs do not show dynamic distortions caused by change of virtual geometry of transistors by the signal itself, due to tiny dies dissipating huge power.
Thank you very much for this exhaustive and very scientific analysis. Your conclusion that there was no significant change in amplifier linearity due to transient thermal bias changes of upto 10% puts things in a practical perspective. Once again, thanks for such a exhaustive analysis, the design of your measuring circuitry itself is very logical and impressive.
Thanks, Jan, to remind me of this serious study. At the time you published this, I got very interested because I was busy simulating these bias variations. Whatever the time constants, within reasonable guess estimations and available data, I could plug in the thermal model of a temp sensor / transistor combo, I had similar conclusions as from your measurements. Then, when I saw your measured delays of 150 to 200 mS or so, I got convinced the so called ThermalTrack solution was a born dead solution. Then I switched to other interests.
All options are still on the table. I am opened-minded with no horse in this race. Yes, I did look at those Sanken Darlingtons with the built-in diodes. I am not inclined toward a Darlington or a triple configuration at this time (maybe later), as I have another driver scheme in mind that I want to test. Also those diodes are electrically connected to the first bases, which won’t suit my plan. The On-Semi ThermalTrak diodes are electrically isolated, so there are more connection options.
I am considering the Sanken 2SC6145A/2SA2223A which do not have integral diodes but offer fairly constant hfe vs Ic, like the 3281/1302 do. Ft values are spec’d higher than the On-Semi devices but in practice, I don’t know how real that is. Also the Sanken NPN and PNP do not appear to be very closely complementary to each other, especially with regard to Ft and Cob. Nevertheless I have ordered enough Sanken devices to build up a full pair of amps and compare them to the On-Semis.
It is not a delay, but a phase shift really. It is not completely useless. Nothing can solve the problem completely, except admitting that the problem is in the core of class AB solid state amplifiers biased with feedback through thermal phase shift. This design solution accepted as a standard today is wrong and flawed.
Still, Denon receivers sound better than other amps, however I would never call it "High End".
Its really delay, Anatolyi. See the article figure. If it was phase shift, it would react at the instant that the power was changing, just what you see when you change the current into a cap, the voltage reacts at the very same time, and it is phase shift.
In my figure you see clearly that there is no reaction when the power changes, the diode curve just continues and reacts only 150 to 200mS later. A pure delay.
But you are right, the root cause is the class (A)B concept.
Jan
In my figure you see clearly that there is no reaction when the power changes, the diode curve just continues and reacts only 150 to 200mS later. A pure delay.
But you are right, the root cause is the class (A)B concept.
Jan
Physically it is not a heat wave Johannes, it is gradual increase of an energy; it only looks as a delay because thermal conductivity of silicon is pretty low.
The thermal conductivity of silicon | Electronics Cooling
One of solutions that I found back in 1970'Th, as you know, is, what Peter Walker invented independently, patented and called "Current Dumping"
Another solution is class A, better if with augmented output devices, and even better if single-ended.
The thermal conductivity of silicon | Electronics Cooling
One of solutions that I found back in 1970'Th, as you know, is, what Peter Walker invented independently, patented and called "Current Dumping"
Another solution is class A, better if with augmented output devices, and even better if single-ended.
Nice article Jan.
I have used a small SOT23 placed very close to the collector lead of one of the output devices for all my amps since about 2008. It seems to work very well and the response is fast (though not as fast as an integrated diode).
On my e-Amp, I augmented the sense transistor with an SMD RTD to flatten the compensation curve by forcing it to intercept the ideal bias current at two temperatures - one at ambient and the other at about 60 degrees.
The other option is a micro (one of those 8 pin devices) to perform 2nd order feed forward compensation (the standard sensor would still do the first order compensation). You could make the whole thing self calibrating. But then I think to myself, why bother? The standard approach works remarkably well and you can still get to single digit ppm with it.
The upshot of all this is that there’s quite a few ways to solve the bias problem!
Here are some thoughts I put down on paper a few years ago
Some Ideas on Temperature Compensation for Audio Amplifier EF Triples
I have used a small SOT23 placed very close to the collector lead of one of the output devices for all my amps since about 2008. It seems to work very well and the response is fast (though not as fast as an integrated diode).
On my e-Amp, I augmented the sense transistor with an SMD RTD to flatten the compensation curve by forcing it to intercept the ideal bias current at two temperatures - one at ambient and the other at about 60 degrees.
The other option is a micro (one of those 8 pin devices) to perform 2nd order feed forward compensation (the standard sensor would still do the first order compensation). You could make the whole thing self calibrating. But then I think to myself, why bother? The standard approach works remarkably well and you can still get to single digit ppm with it.
The upshot of all this is that there’s quite a few ways to solve the bias problem!
Here are some thoughts I put down on paper a few years ago
Some Ideas on Temperature Compensation for Audio Amplifier EF Triples
Last edited:
Yes, many ways to skin a cat! Many of us messed around with transistors or diodes glued to output devices, with wiring to a pcb that always breaks when it shouldn't, that sort of thing. I believe Douglas Self advises to glue it on top of an output device case.
The nice thing about using (at least) one ThermalTrak device in an output stage is that it is all much more repeatable and no-hassle construction. But you still need to think up a clever way for a stable, flat control loop, and it doesn't do anything to counter the alleged 'thermal distortion'. Which, based on my research, I think is no issue at all anyway.
Edit: great article Andrew! Did you ever get around to the uC controlled version?
Jan
The nice thing about using (at least) one ThermalTrak device in an output stage is that it is all much more repeatable and no-hassle construction. But you still need to think up a clever way for a stable, flat control loop, and it doesn't do anything to counter the alleged 'thermal distortion'. Which, based on my research, I think is no issue at all anyway.
Edit: great article Andrew! Did you ever get around to the uC controlled version?
Jan
Last edited:
At ease, Colonel! In civilian life the rules are different, exponential. ;-)
And diodes on the die are much better than separate.
When I bought a Denon receiver for my home theater, I had an intention to use it with external power amps, but found that it is already good enough not only for movies, but also for occasional music listening.
And diodes on the die are much better than separate.
When I bought a Denon receiver for my home theater, I had an intention to use it with external power amps, but found that it is already good enough not only for movies, but also for occasional music listening.
Jan, Unfortunately I only got as far as a circuit and a very high level flow chart. I maybe should do it just out of interests sake - but as mentioned, I don’t feel modern amps that obey all the Cordell and Self rules need it - they are down at low double digit or single digit ppm levels at high powers. Purely something for the intellectual challenge!
Hi Jan,
A couple of comments. The Sanken devices are not referred to as ThermalTrak transistors, as the term ThermalTrak is a registered trademark of ON Semiconductor. The ON Semiconductor ThermalTrak devices are based on an ON Semiconductor patent, referenced in my book, but I don't have that in front of me. As you point out, the Sanken and ThermalTrak devices are rather different beasts, even though they are based on the same clever concept.
I have never used the Sanken devices, so I cannot comment on their diode response time. The diode response time of the ON Semi ThermalTraks is quite fast, with a time constant on the order of 20 ms. This appears to be very different from the 200 ms you quote for the Sankens, but I don't know why. Nevertheless, even 200 ms is quite a bit better than a sensing diode or transistor mounted on the plastic package of an output transistor. BTW, a great many good amplifiers are built with the sensing junctions mounted on the package of one of the power transistors and do work quite well and reliably. This approach is much better than mounting the sensing junction on the heat sink close to the power transistor. It is just that the ON Semi ThermalTrak devices are far better.
I have a bias vs. time graph in my book for a class AB amplifier with 3 or 4 scenarios of bias junction placement. The first is use of a ThermalTrak transistor, the second is mounting of the sensing junction on the power transistor, the third is mounting the junction on the heat sink very close to the output transistor, and I believe there is a fourth where the sensing junction is mounted maybe 2 inches from the power transistor. The differences in bias as a function of time are obvious.
One useful way to measure bias stability is to run the amplifier at some high-dissipation power level, perhaps 1/8 or 1/3 power, until a stable high heat sink temperature is reached. Then one monitors the quiescent bias as a function of time beginning immediately after the load is disconnected and the amplifier enters the quiescent state.
One other thing about the Sankens that makes me less inclined to use them is that they are Darlingtons with an internal driver pull-down resistor from base to emitter of the power device. This means that the user has no control of the driver pull-down current, which likes to be moderately high (perhaps 30 mA or more) for adequate timely turn-off of the output device. I don't recall what the value of this internal pull-down resistor is.
Cheers,
Bob
Bob, yes, the Sankens are quite different from the ON devices, apart from being darlingtons. I've build an amp with the Sankens which was well received, but that in itself doesn't say a lot, I know ;-)
Anyway, I put quite some effort in measuring the thermal characteristics of those ON devices as described in the article I linked to earlier. Of course for medium/long-term thermal stability they work as advertised and are a very efficient way to build the stage.
My investigation was inspired by some articles from way back when from a French guy who wrote by the nom de plume 'Hephaïstos' in Jean Hiraga's L'Audiophile, which was more about cycle-by-cycle modulation of the device operation by the cycle-by-cycle changes in dissipation. He called it Memory Distortion, assuming that the device would be starved of bias current right after a strong signal dropped away. I wanted to know if the thermal diodes would help against that.
But my measurements showed that there's quite a delay between the device thermal excitation and the reaction of the diode voltage, and the thermal variation of device bias current from relatively fast variations in dissipation was clearly visible, and could not be cancelled by the diode circuit because the diode didn't react fast enough.
But distortion measurements also showed that the 3-4% variation in bias current from thermal cycling didn't make a difference in linearity, so the whole thing turned out to be moot anyway.
Jan
Anyway, I put quite some effort in measuring the thermal characteristics of those ON devices as described in the article I linked to earlier. Of course for medium/long-term thermal stability they work as advertised and are a very efficient way to build the stage.
My investigation was inspired by some articles from way back when from a French guy who wrote by the nom de plume 'Hephaïstos' in Jean Hiraga's L'Audiophile, which was more about cycle-by-cycle modulation of the device operation by the cycle-by-cycle changes in dissipation. He called it Memory Distortion, assuming that the device would be starved of bias current right after a strong signal dropped away. I wanted to know if the thermal diodes would help against that.
But my measurements showed that there's quite a delay between the device thermal excitation and the reaction of the diode voltage, and the thermal variation of device bias current from relatively fast variations in dissipation was clearly visible, and could not be cancelled by the diode circuit because the diode didn't react fast enough.
But distortion measurements also showed that the 3-4% variation in bias current from thermal cycling didn't make a difference in linearity, so the whole thing turned out to be moot anyway.
Jan
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.