Each driver only provides drive in one direction. So, you need to combine them to provide both + and - drive.
The gate stoppers dampen local instabilities in the 100MHz range, so they have to be very close to the MOSFETS with no other stray capacitance.
The gate stoppers dampen local instabilities in the 100MHz range, so they have to be very close to the MOSFETS with no other stray capacitance.
Not sure what you mean by “assymetry” and “multitude of signal paths”. The VAS is asymmetric by design. It’s a single ended VAS with a constant current generator on top and the actual amplifying transistors (along with a SOA protection transistor) at the bottom. It’s about as tried and tested it can get. The rest is symmetrical as far as I know.
As far as I know, there’s only one signal path: Through the LTP and VAS and then the inevitable split between positive and negative drivers and output devices. What would the other signal paths be? And as others have pointed out, moving C8 and R23 would do much more harm than good.
Like I already mentioned, the ~0.7Vbe drop will be modulated independently for each driver: as one is heating up at its own pace due to V × I, the other one cools down at a different pace. So the V-drop between the gates is doing its own thing, independent of Q7. Never mind the slight lag introduced by the upper 68p capacitor. The simulator won't take temperature into account, so the estimated distortion will be based on cold devices with 'magic' heat-sinking.
The original version on page 1 actually looks more compelling to me. The VAS already had an improved 2-transistor CCS, where the "hot" transistor is corrected with local feedback, and then if distortion is too high just try to increase the current so the gates charge and discharge faster. Other versions of gate drivers that I've seen had current-limiting resistors in series, too.
However, you'd have to just build both versions and trust your ears to figure out what is real and what is not.
The original version on page 1 actually looks more compelling to me. The VAS already had an improved 2-transistor CCS, where the "hot" transistor is corrected with local feedback, and then if distortion is too high just try to increase the current so the gates charge and discharge faster. Other versions of gate drivers that I've seen had current-limiting resistors in series, too.
However, you'd have to just build both versions and trust your ears to figure out what is real and what is not.
Wait what? You’re saying that the output devices have time to heat up and cool down during a cycle of the signal and that it creates a noticeable difference in the parameters of the MOSFETs? I don’t think it works that way…
Unless something is severely wrong, they should dissipate about the same power and since they’re also tightly thermally coupled, the difference should be negligible.
http://www.douglas-self.com/ampins/thermald/thermald.htm
"Thermal non-linearities would presumably appear as second or third harmonic distortion rising at low frequencies, and the largest effects should be in Class-B output stages where dissipation varies greatly over a cycle. There is absolutely no such effect to be seen in discrete-component power amplifiers."
I think the problem with such ideas is that the authors never "do the math" which would show that the possible effect is "below the noise floor". In any case, it's not a new idea and experience has shown it's "not an issue".
"Thermal non-linearities would presumably appear as second or third harmonic distortion rising at low frequencies, and the largest effects should be in Class-B output stages where dissipation varies greatly over a cycle. There is absolutely no such effect to be seen in discrete-component power amplifiers."
I think the problem with such ideas is that the authors never "do the math" which would show that the possible effect is "below the noise floor". In any case, it's not a new idea and experience has shown it's "not an issue".
I read the post from @abstract again, and I guess he’s referring to thermal signal frequency coupling between the output devices and the drivers/VBE multipliers. I think the thermal capacitivity of the heatsink and the distance between the devices are no to WAY too large for that to be an issue. It’s a different story in an IC where the distances could be in the fractions of a millimeter.
I just did the math. Heat transfer is difficult to calculate (Fourier had to invent all new math to do it), but if we just model it as the transistors sitting on a uniform piece of aluminum weighing 50g, it would take 0.9s to increase the temperature by 1 degree C. Kind of a crude calculation, but it should be in the right order of magnitude.
Conclusion: It shouldn't matter one iota.
The above is assuming 50W of power. I think it would take a lot longer in reality, since the heat is dissipating to the surrounding material and air.
That refers to IC amplifiers, where high dissipation sections couple thermally to the INPUT. Yes, it can affect other stages - the VAS, drivers, etc. But those stages are inside the global NFB, so those distortions are reduced by the amount of feedback. This pushes the effect below the noise floor. When such audio-rate Vbe shifts occur in the input stage, it is indistinguishable from input SIGNAL and no feedback corrections can be done. Such couplings are strongest on a single die. With discretes there is distance and lots of thermal mass between the heat sources and the base-emitter junctions of the input transistor(s). Die layout solutions do exist to minimize this, but it simply takes more silicon to do it. That means the IC costs more.
On my PC, the CPU die temperature shows the effects of two time constants (roughly one second and one minute). When a load is applied, the CPU die temperature increases (by half) nearly instantly, and the other half comes over a minute or two. This implies that the thermal resistance between the die and heatsink is similar to the thermal resistance between the heatsink and air.
Ed
Ed
Thats a BIG die, with a lot of heat sources spread out over hell and gone. Even then the sections (which are comprised of thousands of small heat sources themselves) do not see heat load evenly. On chip thermal management will respond to overall average temperature, not the instantaneous temperature of a single TIP41-size transistor on it. One would expect the thermal time constants to be radically different. For single power devices, the first thermal time constant is on the order of a millisecond, in “series” with other slower ones.
I know a lot about CPUs. 🙂 My point was that the die, package, and heatsink all have thermal time constants and resistances.
Ed
Ed
But now you’re talking about the time constant for heating up the die itself, right? The discussion here is about the transfer of heat between devices that are some 10mm apart with lots of aluminum between them. That constant should be on the order of a second per degree.
Right. And seconds per degree can’t make distortion due to thermal feedback. It can just cause bias current lag. Which is the same thing really - just well below the audio pass band. Still can be annoying, because crossover distortion characteristics change dynamically based on average dissipation. You can’t track the temperature rise on the die, just to it’s environment. ThermalTrak transistors get you closer but there is a limit to how close. Too bad there isn’t a way to use the body diode in a mosfet for that because it is REALLY close to the channel. Unfortunately, not well isolated from it electrically.
You can get rid of that altogether with a class-(A)B bias loop, but that is hardly bog standard.
Hell, these mosfet amplifiers work well enough and sound good just following best practices. No real need to split hairs. It’s not a single-digit ppm design. Car radio chiplifiers that have real thermal feedback issues are orders of magnitude worse (and you can hear it).
Exactly. The previous iteration of this (the one without the drivers) has a pretty pronounced negative temperature coefficient on the bias current. I compensated for that by running it at a little higher bias current than strictly needed. I couldn’t measure any difference between a hot and a cold amplifier. Now, this version will obviously behave differently. It should be good enough. I strongly doubt you can hear a few tens of PPMs anyway.
Actually, you were the one bringing up crossover distortion changes due to temperature variations, see post #194 😉 I only mentioned a potential solution, if you should want to solve it.
I designed and built myself an amplifier with a class-(A)B bias loop back in 1994 for various reasons:
I find it conceptually more elegant than the more usual circuits.
It seemed more fun than designing the nth variant of the circuit everyone has designed since the 1950's.
It allowed me to use dirt-cheap BUZ10 power MOSFETs.
I wouldn't have to worry about whether the distortion on music would be higher than on sine waves.
None of this means that a bog standard amplifier can't be good enough for all practical purposes, of course.
For what it's worth, my amplifier distorts about 0.006 % at 10 kHz, 80 % of maximum power and about 0.0025 % at 10 kHz, 50 % of maximum power. I did a subtractive test that attenuates the music without attenuating the distortion. With the music attenuated some 60 dB, it still sounded like music rather than distortion to me (back in 1994, when I was in my mid-20's and still had good ears).
You have to pick and choose your battles, when trying to make a bog standard amplifier out of parts that you can get in every corner dime store in the country.
Gerber files sent to JLCPCB. Hopefully I have the boards next weekend!