What is correct location of the temperature sensing Vbe multiplier transistor?
I see different approaches here.
On most PCBs it's located on the main heatsink, between drivers.
On some builds - attached on top of one of the output devices.
Douglas Self recommends placement on one of the drivers for CFP
output, and on output device for EF output.
Is this correct?
Are most of the PCBs I see here on DIY wrong?
Placement of the temp sensor on the heatsink, 3-4 cm away from output devices will definitely slow down the response.
What about quasi output? Upper device works as EF, bottom one as CFP, so what would be the best place? Driver or output device?
I see different approaches here.
On most PCBs it's located on the main heatsink, between drivers.
On some builds - attached on top of one of the output devices.
Douglas Self recommends placement on one of the drivers for CFP
output, and on output device for EF output.
Is this correct?
Are most of the PCBs I see here on DIY wrong?
Placement of the temp sensor on the heatsink, 3-4 cm away from output devices will definitely slow down the response.
What about quasi output? Upper device works as EF, bottom one as CFP, so what would be the best place? Driver or output device?
In a properly designed circuit, the quiescent current is managed by way of an accurate maximum amount. When the output transistors are cold, their gain is less than when hot, hence the bias amplifier. When correctly fitted, on the heat sink, it takes care of all output devices. If mounted on a single output transistor, it only looks after that transistor, primarily. If there was one per transistor, independently monitoring that transistors current, that would be utopia but this is not good practice as when one transistor heats up, the bias for all the transistors will be affected, because it is only a single driver stage and may cause cross over distortion.
An EF type class AB design with a single pair of output transistors has no problem using the most direct form of temperature sensing, right on one output transistor case or virtually on the collector lead, as shown and tested on Bonsai's Ovation e-amp and discussion paper which, incidentally, has multiple output pairs. (Hifisonix.com)
There, with close to direct sensing of output transistor die temperature, you have a very low temperature sensing error and hence very low distortion due to bias tracking error. This is what we should want, regardless of how much distortion we can tolerate with less effective i.e. slow responding methods where air currents, radiant heat losses or insulating materials and relatively long thermal paths are imposed. This is what you get in typical builds by manufacturers and DIYs just following the age-old low assembly cost, easy conventions. By contrast, Bob Cordell discussed the benefits of power transistors fitted with temperature sensing diodes (On-Semi Thermaltrak types) in his book, discussed here in the big book thread too. Sanken's latest series STD02 audio Darlingtons are also fitted with thermal tracking diodes.
Douglas Self explained his methods with standard components pretty thoroughly too, first for CFP designs in his early handbook editions and more recently for EF designs, since the 5th edition.
Manufacturers and DIYs probably see anything departing from traditional practice or simply "what the other guy did" as too hard or unusual to be any good when the approach is a proven worthwhile improvement for Hifi audio in either CFP or EF format. If you consider the quirks of some commercial products, where 30min+ warm ups are still required just to stabilise bias - not even attempting to get it to track, it becomes clear that simply putting the temperature sensor in the same box as the power amplifier or even close to the heatsink is literally a waste of time, for both the performance quality and user satisfaction. IMV, optimum thermal coupling and hence bias accuracy and speed, is at the heart of consistent and optimum performance, whatever your criteria for good sound quality may be.
Commercial products all take shortcuts to reduce assembly costs. Whilst simply gluing or bolting a sense transistor to a convenient location on the heatsink may work to prevent thermal runaway, it won't often be much use as a reference for accurate, optimal bias. CFP designs of course, are much easier to optimise as the thermal mass of just the drivers and separate small heatsink can be so much less.
There, with close to direct sensing of output transistor die temperature, you have a very low temperature sensing error and hence very low distortion due to bias tracking error. This is what we should want, regardless of how much distortion we can tolerate with less effective i.e. slow responding methods where air currents, radiant heat losses or insulating materials and relatively long thermal paths are imposed. This is what you get in typical builds by manufacturers and DIYs just following the age-old low assembly cost, easy conventions. By contrast, Bob Cordell discussed the benefits of power transistors fitted with temperature sensing diodes (On-Semi Thermaltrak types) in his book, discussed here in the big book thread too. Sanken's latest series STD02 audio Darlingtons are also fitted with thermal tracking diodes.
Douglas Self explained his methods with standard components pretty thoroughly too, first for CFP designs in his early handbook editions and more recently for EF designs, since the 5th edition.
Manufacturers and DIYs probably see anything departing from traditional practice or simply "what the other guy did" as too hard or unusual to be any good when the approach is a proven worthwhile improvement for Hifi audio in either CFP or EF format. If you consider the quirks of some commercial products, where 30min+ warm ups are still required just to stabilise bias - not even attempting to get it to track, it becomes clear that simply putting the temperature sensor in the same box as the power amplifier or even close to the heatsink is literally a waste of time, for both the performance quality and user satisfaction. IMV, optimum thermal coupling and hence bias accuracy and speed, is at the heart of consistent and optimum performance, whatever your criteria for good sound quality may be.
Commercial products all take shortcuts to reduce assembly costs. Whilst simply gluing or bolting a sense transistor to a convenient location on the heatsink may work to prevent thermal runaway, it won't often be much use as a reference for accurate, optimal bias. CFP designs of course, are much easier to optimise as the thermal mass of just the drivers and separate small heatsink can be so much less.
Thanks Ian.
Any ideas about quasi config?
What should be tracked in your opinion? Driver or output device?
Assuming drivers have their own heatsinks.
I'm mostly into quasi amps...
Any ideas about quasi config?
What should be tracked in your opinion? Driver or output device?
Assuming drivers have their own heatsinks.
I'm mostly into quasi amps...
I once asked the question of D Self; how best to control the bias for a Quasi comp. design?
I got stony silence. The reason, I assumed as I hastily changed the subject, is that there are 2 different bias current conditions required that need to be controlled independently but are in series. The net result is that when you try to satisify both conditions you can't meet either. Typically, you might require only 13 mA for the CFP side and perhaps 90 mA for the EF. Both will also have different tempcos and so it becomes impossible to do with any conventional biasing control circuit.
The only simple practical way to bias them that I'm aware of is still the same compromise current of perhaps 30-40 mA that results in lowest THD on test with the particular transistor type, output stage design, voltages, gain/feedback etc. However, this isn't as bad as it may appear. Take a look at these interesting tests carried out on a little Nait2 integrated Quasi-complementary amplifier, where distortion was measured at ~ 0.01%. That's surprising when you consider the craptanium output transistors and dated Naim amplifier design, quirks and all: Naim, NAPs, Naits, distortion and bias-current settings
I tinker with quasi designs too, because I think they can sound quite good with little effort, use up the non-complementary transistors lying about and the results are just fine in many home and medium PA applications. There are always threads here with details and helpful comments on getting the best from them too
I got stony silence. The reason, I assumed as I hastily changed the subject, is that there are 2 different bias current conditions required that need to be controlled independently but are in series. The net result is that when you try to satisify both conditions you can't meet either. Typically, you might require only 13 mA for the CFP side and perhaps 90 mA for the EF. Both will also have different tempcos and so it becomes impossible to do with any conventional biasing control circuit.
The only simple practical way to bias them that I'm aware of is still the same compromise current of perhaps 30-40 mA that results in lowest THD on test with the particular transistor type, output stage design, voltages, gain/feedback etc. However, this isn't as bad as it may appear. Take a look at these interesting tests carried out on a little Nait2 integrated Quasi-complementary amplifier, where distortion was measured at ~ 0.01%. That's surprising when you consider the craptanium output transistors and dated Naim amplifier design, quirks and all: Naim, NAPs, Naits, distortion and bias-current settings
I tinker with quasi designs too, because I think they can sound quite good with little effort, use up the non-complementary transistors lying about and the results are just fine in many home and medium PA applications. There are always threads here with details and helpful comments on getting the best from them too
No. In a properly designed circuit, the bias is set to whatever is optimum for the output stage topology. This requires tracking a proxy for the junction temperature of either an output device (EF or Darlington) or a driver (CFP). Heatsink temperature is a poor proxy, but better than nothing.JonSnell Electronic said:
I agree completely.
I have recommended gluing a sot23 to the collector lead of the EF BJT output stage for fastest response to changing Tj and for closest approach to Tj.
Ts responds very slowly to changes in Tj and never gets close to Tj.
Heatsink location is poor by any standard. But we have made do with that for a long time.
On top of a metal packaged output device may be quicker and may get closer to Tj, D.Self certainly thinks so.
But does mounting on top of a plastic package get close to Tj? It certainly does not get there quickly. It is my opinion that mounting on top of a plastic packaged output device is poor as a tempco adjsutment.
I've checked this article, this is interesting thing I didn't know:
Naim amps don't have thermal compensation for the bias setting
They do have Bias spreader transistor, so is it just mounted on the board, without any thermal contact with devices?
Wouldn't biasing be easier (than for BJTs) for a quasi hexfet configuration?
This amp:
Project 54 | BuildAudioAmps
that I built one year ago, seems to have pretty good THD (for a quasi).
Temp tracking is attached to the output device, not a driver.
Quasi250, 350 and 500 amps from this forum, designed/built by Quasi several years ago (which I also built) had thermal sensor transistor installed on the heatsink, but very close to the drivers. Not sure if someone ever measured THD of these amps...
Mosfets are a different biasing matter to BJTs. BJTs have a well defined voltage at which they turn on and begin to conduct in a fairly linear relationship to the input signal. The bias spreader or Vbe multiplier circuit is therefore designed to generate a precise, thermally compensated voltage, approximately equal to and nulling out the sum of Vbe's (or simply the big crossover glitch) that appears as a discontinuity in the circuit's output. Sure, designers often specify higher resulting bias currents than necessary for minimal THD but there are other, often subjective matters that creep into design work too.
There is no such bias threshold voltage with Mosfets. In BJT quasi designs, it is confused by having an asymmetric number of junctions with an uncertain overall thermal coefficient. Many designs seem to treat the bias of BJT quasis much like Mosfets - just enough current to minimize distortion. It seems rare though that DIYs ever venture this far in refining their designs, even with the option of SPICE.
Mosfets of any type can use as much bias as the power supply and heatsinks allow - right up to class A operation for the lowest possible THD. There is no specific amount of bias required for optimal low distortion - just have enough of it for your efficiency and performance demands. However, the various mosfet types and topologies all have different tempcos. Lateral mosfets even have a useful negative coefficient, rendering them bulletproof in regard to thermal runaway.
The advantage of bias being non-critical with Mosfets is that you can use a bias spreader transistor conveniently attached to the heatsink or any other point that relates strongly to mean output device temperature. As long as the bias spreader circuit's tempco is worked out appropriate to the device types and topology with reasonable accuracy, current will remain as stable as necessary to limit distortion. That's all you really need there.
Typical Hexfets have a positive thermal biasing coefficient after the output rises more than a few volts. This needs to be accounted for in the bias spreader design, whatever the particular output stage topology, if you want adequate protection from thermal runaway - still likely if the heatsinks are sized for only relatively low bias current. Bob Cordell's book is a must-have if you want to stay focused on Mosfets and getting the best from them. It's not expensive in the US and worth the $$ many times over.
BTW, there is thermal bias compensation with Naim's universal quasi design, or their amplifier products would be toast in short order. I think the author is confusing thermal compensation per se, with the use of additional tempco (slope) adjustment components. The problem there, already mentioned, is that being a long way from the output stage, the Vbe multiplier transistor is mighty slow responding - like 30 mins in some low power versions of their common basic design. Heat transfer depends on the air temperature rise in the case and with low bias current, you can see the problem. It never seemed to phase Naim though - all part of their mystique that still fills the minds of their "flat-earth" following.
There is no such bias threshold voltage with Mosfets. In BJT quasi designs, it is confused by having an asymmetric number of junctions with an uncertain overall thermal coefficient. Many designs seem to treat the bias of BJT quasis much like Mosfets - just enough current to minimize distortion. It seems rare though that DIYs ever venture this far in refining their designs, even with the option of SPICE.
Mosfets of any type can use as much bias as the power supply and heatsinks allow - right up to class A operation for the lowest possible THD. There is no specific amount of bias required for optimal low distortion - just have enough of it for your efficiency and performance demands. However, the various mosfet types and topologies all have different tempcos. Lateral mosfets even have a useful negative coefficient, rendering them bulletproof in regard to thermal runaway.
The advantage of bias being non-critical with Mosfets is that you can use a bias spreader transistor conveniently attached to the heatsink or any other point that relates strongly to mean output device temperature. As long as the bias spreader circuit's tempco is worked out appropriate to the device types and topology with reasonable accuracy, current will remain as stable as necessary to limit distortion. That's all you really need there.
Typical Hexfets have a positive thermal biasing coefficient after the output rises more than a few volts. This needs to be accounted for in the bias spreader design, whatever the particular output stage topology, if you want adequate protection from thermal runaway - still likely if the heatsinks are sized for only relatively low bias current. Bob Cordell's book is a must-have if you want to stay focused on Mosfets and getting the best from them. It's not expensive in the US and worth the $$ many times over.
BTW, there is thermal bias compensation with Naim's universal quasi design, or their amplifier products would be toast in short order. I think the author is confusing thermal compensation per se, with the use of additional tempco (slope) adjustment components. The problem there, already mentioned, is that being a long way from the output stage, the Vbe multiplier transistor is mighty slow responding - like 30 mins in some low power versions of their common basic design. Heat transfer depends on the air temperature rise in the case and with low bias current, you can see the problem. It never seemed to phase Naim though - all part of their mystique that still fills the minds of their "flat-earth" following.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.