Output stage.......

[CBS240]

Hi

I have finally gotten some time to work on the first build of the new prototype of my amp circuit and was able to get the major bugs out of the system and able to do a listening test. It has the same dynamic precision as the original circuit I built before on vero-board. (which is awsome BTW ) There is a lot to the whole amplifier circuit, but for now I will share the output stage with all you audio gurus if there is any interest. A few very minor issues yet to contend with, but not really with the OPS.

This OPS circuit will change your mind about using complementary vertical Hex-Fet transistors as outputs, it certainly has for me. The amp is a balanced amp, so there are two separate output stages that are bridged by the speaker. On appearance it could be mistaken for a half bridge OPS with two pairs of paralleled outputs because of only one bias pot. The bias to each OPS is set by the conductance of a photo-diode from a IL300. The LEDs in the IL300s are servo by an LM358 duel op-amp. (Micro 8 package just behind the blue bias pot.) This way both OPS can be set with just a single pot and still remain completely electrically isolated from the other or the bias set circuit.

Here are the transistor datasheets for quick reference.

4092

5462

BC856

BC846

5179

H81

The C6026 and A2154 may be SOT-923 transistors but they are cascode and so Pd is only ~15mW. I chose these because, other than the fact that Toshiba makes great semis, they have such nice linear gain, for a transistor anyway. I’m not afraid to use small package components, you just have to design the circuit for them. Actually, the VAS comprises a balanced bridge using two SOT-563 duel device components, but that’s another chapter.

KSC2690 and KSA1220 are the drivers. Bias is 40mA, which comes from the low voltage, high current supply that powers the outputs and speaker. The extra drive voltage for the gates is provided by good ol’ bootstrap. This way the current needed for the drivers does not have to come from the higher voltage regulated supply and so each output stage only requires about 3mA from the high voltage supply.

Transistors labeled 'A' and 'B' is the FMB package of 2227A.

The output transistors are dirt cheap Hex-Fets from Fairchild, FQP47P06 ($1.60 ea) and FQP50N06 (cost me a whole 69 cents ea )
I realize this type of transistor is not the most linear device to use for audio outputs and most audio designers wouldn’t touch them with a 40m pole, but they really are more physically capable of handling a reactive load such as a speaker than any BJT of equal size, and they are much faster. But then no mosfet is really linear due to the inherent non-linear capacitances that exist in every mosfet. It is my opinion and there are others who would never use a mosfet as an output transistor without some sort of local error correction. Perhaps I've been spoiled.

One interesting thing I found with this circuit, other than using linear opto-couplers to bias the two stages simultaneously, is the method of thermal compensation. It is true that in order to thermally bias the HEC output stage the error amplifying transistors must be included in the thermal transfer function, thus in some physical way connected to the output devices. I picked up an idea that I saw first when Bonsai posted a photo of using a SOT 23 Vbe multiplier in contact with the collector pin of the output transistor. I figured this might just work with the drain pin of the mosfet in getting a better sample of the die temperature since that is where the heat is generated. Also I could reduce trace length and use very fast transistors in order to increase the BW of the error amplifier. In additoin there are only the four output devices to mount to the heatsink. I took a chance on whether it would even work or not by actually building it this way on the PCB I drew and ordered. It seems it was a good call, because I have it set now with a slight negative temperature co-efficient so that bias current decreases slightly with increase in temperature. Bias drops from initial setting of 200mA to about 140mA between initial turn on and when the transistor is too hot to touch without burning of the finger. The temperature co-efficient can be adjusted by moving the position of the error amp transistor with relation to the distance from the body of the output transistor package. I drew 'tracks' as pads in the PCB so the position could be adjusted, sort of like a thermal pot. Just figured all this might be interesting since I haven't seen anything similar and it seems to work quite well.


EDIT> The +/-10V is supplied by the VAS. Also each of the drivers will have a small heat sink eventually. They are operating a little too hot for comfort for me even though they are still within spec. I designed space for this for each one since they have to do Pd ~800mW.

Oh yeah, and thanks to Bob Cordell for his knowledge and contributions as this is basically his EC mosfet OPS with a few bells and whistles. (Figuratively of course)




Cheers

[kenpeter]

Yeah, I've had troubles with my own, leading to overcomplicated
fixes, leading to optos to try and get things simple again... Opto
can be slow if base is left open and not bled by a resistor, only
advice I can offer. You seem way further ahead already, so that
may be old news...

Your circuit shows diodes? CB of an opto transistor can be abused
that way. Fast too, by fully strapping emitter to base instead. But
no transistor current gain, and diode current swing is quite small.

--------

Oh, nevermind. You got them optos with two diodes, and one in
a send side local feedback loop. I don't know much how well the
loose diode might or might not track the trapped one? They are
not in the same circuit, they do not see the same voltage and
current challenge, how can one predict the right light needed to
linearly drive the other?

[CBS240]

Hi

Actually those diodes are matched so whatever conductance is set by the servo exists in both diodes. There are two IL300's, one for each OPS. There are two of the first circuit, both are controled by the second circuit servo-ing the optos. Each opto has one of the two matched diodes as servo sensing duty and the other for its respective OPS. They only set the DC bias so there are no transients or AC that could be frequency dependent. The temperature compensation is sensed by the error amplifiers. If you look closely at the outputs you will see the drain pin extends a bit further on the PCB. Just under where the pin exits the body of the output transistor is where the SOT 23 error amp transistors are mounted.

The original circuit has two separate pots. I just thought it would be neat to try this idea, even though those IL300 are over $3.50 each. Of course those THAT340 arrays are about $8 each. But then there is only $4.50 in all of the output transistors combined.

It's still under the 'testing phase' so not all the filter caps have been installed, some jumper wires need to be thicker, and some of the caps are just placed in and not soldered so I can take them out easily if needed.

[CBS240]

Here is an Eagle screen shot of that section......

[kenpeter]

Should 20ma drive bias cross to the other emitter?
I think this dumbs it down to a pair of single ended
drives with a 40mA CCS inbetween. instead of pair
of push-pull diamond buffers? Where else can the
total 40mA ( A + B ) current go?

Perhaps 20mA collectors should return to sources
of the opposing FQPs instead?

[CBS240]

The total 40mA goes from one driver to the other via the totem pole. The 'diamond buffer' of sorts, divides this current appropriately to accommodate the different drive currents needed for each of the output transistors. They are not complements, as no complementary parts can be had for this type of device, and require different drive currents for equal change in conductance. Ultimately the KSC2690 and KSA1220 drive the total current needed by both outputs, but this way the actual drive current per device at least for turn off action is relative to that specific device independent of the other. Is it truly necessary? Probably not, but it works like a charm. This circuit is a great way to use these cheap transistors that most here would declare 'unusable' for the functional use as output transistors in a linear amp. I put it up against any output stage using BJT's or Lateral fets in terms of distortion, slew rate, and SOAR. (component case size relative of course.)
BTW, replacing the photo diode with a pot in series with a resistor to set the bias is the way one would use this circuit if the amp is half bridge with only one output stage.


It doesn't have to take that much space either, in the photo I have outlined the first curcuit from post 1 in yellow. ~4cm X 4cm