Looking arround there are a few threads that talk about biasing the LM3886 A and most of them ended up with the conclusion that it is a bad idea because of power dissipation.
Taking a look at the numbers and assuming +/-20V supply and 1.5A CCS biasing we get a dissipation of 30W With 1W/ºC J-C (from the datasheet) and 1W/ºC case to heatsink made with a 0.5W/ºC Heatsink and 0.5W/ºC Case to heatsink the operating temperature at 25º should be 85º. This is hot, but it's almost a half from the spike temperature, so spike is not a problem and this is possible if well done.
I have just realized after a question from a forum member that with little extra investment a class A LM3886 can be made to allow the selection between emmiter follower and Sziklai output stages. This is of extraordinary interest because the sonic differences between these have been discussed a lot and the ability to switch would be of great use for those wanting to choose. I hope that the Baxandall diodes in the quasi-comp design won't change the Sziklai sound.
The idea is to load the LM3886 with either a source or a sink, these should be mounted on the same heatsink for economy as they will never work at the same time. They can also be turned off to allow class-B operation that would be handy in summer.
Seeing that thermal dissipation is a problem but it's not prohibitive and that there will be the option to choose between Sziklai and EF, have i missed something?
What would be the ideal CCS design for it, mosfet, diode, two-transistors?
Taking a look at the numbers and assuming +/-20V supply and 1.5A CCS biasing we get a dissipation of 30W With 1W/ºC J-C (from the datasheet) and 1W/ºC case to heatsink made with a 0.5W/ºC Heatsink and 0.5W/ºC Case to heatsink the operating temperature at 25º should be 85º. This is hot, but it's almost a half from the spike temperature, so spike is not a problem and this is possible if well done.
I have just realized after a question from a forum member that with little extra investment a class A LM3886 can be made to allow the selection between emmiter follower and Sziklai output stages. This is of extraordinary interest because the sonic differences between these have been discussed a lot and the ability to switch would be of great use for those wanting to choose. I hope that the Baxandall diodes in the quasi-comp design won't change the Sziklai sound.
The idea is to load the LM3886 with either a source or a sink, these should be mounted on the same heatsink for economy as they will never work at the same time. They can also be turned off to allow class-B operation that would be handy in summer.
Seeing that thermal dissipation is a problem but it's not prohibitive and that there will be the option to choose between Sziklai and EF, have i missed something?
What would be the ideal CCS design for it, mosfet, diode, two-transistors?
if you can get it to work, i.e. keep it cool enough, then you have a single ended ClassA amplifier.
The peak power output is Ipk^2*Rload.
The Maximum power output is Ipk^2*Rload/2
For 1.5Apk of CCS loading the maximum power output is about 9W into 8ohms.
If you use resistor loading for the bias current, the peak current in the worst polarity is halved, leading to quarter of the power, ~2W.
It might make a good headphone driver if the noise is low enough.
The peak power output is Ipk^2*Rload.
The Maximum power output is Ipk^2*Rload/2
For 1.5Apk of CCS loading the maximum power output is about 9W into 8ohms.
If you use resistor loading for the bias current, the peak current in the worst polarity is halved, leading to quarter of the power, ~2W.
It might make a good headphone driver if the noise is low enough.
Hi Ion,
your mind never seems to rest....
One could use another LM to provide the class-A bias (I've proposed that a few times and I once did an LM1875 amp with this trick). Offset one of the amps a little and combine outputs with load sharing resistor(s). Then you have many options, like adjusting the polarity/level of the offset (sourcing or sinking current from the main amp), choosing the tap-off for the speaker somewhere along the resistor (which lets you "move along" the sourcing/sinking config -- and adds some output resistance unless you feedback the master amp around them), and modulating the slave amp in various ways, like for constant current or constant power in the master output transistor and the like, even adding some slight nonlinearity to reduce max dissipation close to the rail. The nice thing with having several amps with basically the same gain and paralleled is that you can use class-B operation in all of them for occasional higher current demand. If you bridge the output then you have constant current draw on the supply as long all amps are in class-A and above that one can choose the A to B transitions with a little spreading, say with 4 amps only one at a time (and its distortion contribution from the switching will be damped down by the other three were the output current polarity doesn't change). There are many many options....
- Klaus
your mind never seems to rest....
One could use another LM to provide the class-A bias (I've proposed that a few times and I once did an LM1875 amp with this trick). Offset one of the amps a little and combine outputs with load sharing resistor(s). Then you have many options, like adjusting the polarity/level of the offset (sourcing or sinking current from the main amp), choosing the tap-off for the speaker somewhere along the resistor (which lets you "move along" the sourcing/sinking config -- and adds some output resistance unless you feedback the master amp around them), and modulating the slave amp in various ways, like for constant current or constant power in the master output transistor and the like, even adding some slight nonlinearity to reduce max dissipation close to the rail. The nice thing with having several amps with basically the same gain and paralleled is that you can use class-B operation in all of them for occasional higher current demand. If you bridge the output then you have constant current draw on the supply as long all amps are in class-A and above that one can choose the A to B transitions with a little spreading, say with 4 amps only one at a time (and its distortion contribution from the switching will be damped down by the other three were the output current polarity doesn't change). There are many many options....
- Klaus
AndrewT said:
My idea is to do a dorm room amplifier (my dorm room is small, close to the neighbours, and i'm there only at hours where playing loud is not a good idea) capable of delivering some 10W with very nice quality. This could also be a good project for people who want to try out class A without going too far from the standard gainclones.
KSTR said:Hi Ion,
your mind never seems to rest....
Well, i have to admit that i stole the idea of the class A chipamp from one of your posts!
KSTR said:
My idea was to do a simple and easy class A amp that did allow to choose between EF and CFP, but your ideas make me thing that it could be made an amp that was a combination of a class B powerful thing + a class A correcting amplifier that could at least push the crossover some decades lower. I belive that this is suggested in Self book but combining a class C amp + a class A correcting amplifier. Have you done any prototype?
Looks like for those 10 Watts a simple current sink/source could do the job, this also could be added to any existing amp. Though in terms of performance I guess that a decent discrete class-A amp (say, with that CFP output discussed in the other thread) will be ahead of any chipamp solution.ionomolo said:
While this is something completely different from simple class-A bias (paralleled amps or CCS/resistor-biased) it might be worth a try with chip amps. At the moment I'm not sure what exact concept you are referring to, is it the approach to look at the error voltage of the class-B amp and invert/amplify it to provide the other leg of a bridge?
Yes, I made a working stereo amp with 8 LM1875's (2x2 per ch., bridged) many years ago, when that chip was new. The first time I used this was again years before that with an OPA2544 when I studied the effects of offset/drift/gain errors with regular paralleling where normally offsets and class-A are not wanted (for maximum efficiency). But then I thought why not use it deliberatly for a class-A audio amp... I always thought about building a big chipamp with 8 LM3886 per ch. but then I feel all the effort for such a project would be better spent in a real amp... especially since I lurk around here at DIYAudio, learn so many things about discrete amplifier design and find many many promising amp ideas here, over and over again...
- Klaus
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.