Bear in mind that the cascomp developed by Pat Quinn at Tek deals with this issue in a rather elegant way. The cascomp is covered in my book. I also show how the cascomp principle can be applied to JFETs. I seem to recall that the cascomp has a 3-dB noise penalty associated with it, however.
Pat's mentor at Tek was Barrie Gilbert.
Cheers,
Bob
Hi Samuel,
Thanks for posting these patent numbers. I finally got around to looking them up and checking them out. They are definitely pretty much the same idea, with the common mode signal for control of the two current source loads taken from the tail of the second diff pair instead of from merging resistors off of the emitter follower buffers as I did. Basically the same idea, and with the same advantages.
BTW, one can tell from the topology I used that I was involved in linear IC design, sometimes using more transistors than resistors
. Small-signal transistors are cheap, and I think if they result in better performance their liberal use in discrete circuit design is just as valuable as in IC design.I think that concerns that some have expressed about high-impedance nodes can be tempered by the fact that the usual feedback compensation circuits tend to make these nodes have a much lower impedance at high frequencies, where it counts. Of course, those who prefer amplifiers with wide open-loop bandwidth would justifiably disagree.
Cheers,
Bob
I expressed concern a few posts back, partly because I don't plan to use the usual Miller compensation. For instance Miller Input Compensation would not lower the VAS input node impedance so much I think, but I haven't worked it out yet.
Any comments on how this worked out on your JAES amp?
Best wishes
David
Good point, David. It turns out that the compensation loop in Miller Input Compensation itself needs to be compensated. In my MOSFET power amplifier with error correction, this is accomplished by the series R-C network across the differential collectors of the input stage. So this tends the lower the differential impedance at these nodes at high frequencies.
One other thing about using resistors instead of current sources, as Marshall Leach did, is that sometimes the presence of the load resistance makes the input stage work a bit harder at audio frequencies, increasing its distortion a bit.
In some no-NFB amplifiers, the gain and pole frequency of the VAS is established by a shunt resistance to ground from the high-impedance VAS output node. This also makes the VAS work harder and generate more distortion. An alternative to set the gain of the no-NFB to a defined value is to instead incorporate a local, flat feedback loop from the VAS to its input or back to an earlier stage, just like Miller compensation. That relives the VAS of the heavy burden of a load resistor. However, no-NFB purists would objevt to this, as it than constitutes a feedback amplifier where the output stage only is excluded from the NFB.
Cheers,
Bob
When I wrote that I wanted to lower the impedance I did not mean to imply the use of resistors rather than current sources, or the use of shunt loads.
That is precisely why I am interested in your differential current mirror to lower the common mode impedance without any sacrifice of differential impedance.
I want to take the rationale of your JAES amp one step further and compensate the MIC loop itself with nested Miller rather than shunt an R-C load across the IPS. This should offer similar benefits to the use of Miller rather than shunt in the overall compensation. But I suspect it lowers the impedance of the amp somewhat differently.
I am surprised that I can't find more theory for this, so thank you for the response.
Best wishes
David
Hi Mike,
Thanks for these suggestions. I had already planned to try to go deeper on the TPC and TMC matter, and with more results. This is obviously an area where more discussion is needed and where more proof of the pudding is needed. More guidance on choosing values for the networks is also needed. Finally, I expect to include results on an actual amplifier where ordinary Miller compensation, TPC and TMC are used, and compare the results.
I will also try harder to explain HEC. I already thought I did as best as I could explaining HEC, but I'll certainly try to improve it.
Cheers,
Bob
DIY Ideas!!
Hi Bob & others,
As we all know, writing something in a book/design spec. and getting it to work in practice, is not always the same. Lots of variables.
I see this in Doug Self's books too, okay I realize that you can buy the stuff from Signal Transfer company. Is there a way to make it more available/affordable to the masses?
How about saying that it is made in North America for a change Too bad lots of my parts are made/assembled in Asia, but that I have little control over and learn to live with.
My suggestion is to actually take one or more ckts as described in the book and make them work. I know Bob has done his own implementations, but they are not generally available to people who just want to buy the pcb(s), parts & solder them up. Most people on this site can figure it out. I know chassis are available. If not I'll do it for you, since I am looking for work
Not sure if this a proper thing to discuss on this site, since they offer there own stuff?
What do you think Bob, I can offer to do a layout or two for you for $0, maybe some things for your web site. Maybe like what you did with VinylTrack, still gotta buy that copy of LinAudioSanta did get me that one.
I have basically done this of late, just have to fab/test:
1) Input AC assy with auxillary power(+5Vunreg/3v3 lin reg, for logic, soft-start relay, remote on/off.
2) AC rectifier assy, big ecaps, fusing.
3) Linear regulators (+/- 65 -75V),+/-15V, +5/3V3, PA protection ckts
4) LME49830 (WireAmp) clone. (Could be convinced to make this as a super gain clone)
I always like to have others review my stuff since I have no trade secrets with this design. Since the AC stuff is very important, this is where I would like to have a second eye.
I am sure that there are many ways to implements this stuff, but the idea is to make it small, not overly complicated/expensive, use easy to get parts, Mouser, Digi-key, Newark. Make it of quality construction
As it is, I have already bought all of the expensive parts for this PA project. As well I have got the radio/media player/pre-amp designed, working/tested, so I am going through with my stuff.
Listening to a Stones LP on the FM (Q107/Toronto) radio, "Going Home", first album "After Math", the record crackle was a welcome change.
Nothing like this PA
http://www.diyaudio.com/forums/solid-state/221979-prototype-power-amplifier.html
I'd like to see his invoices!
Cheers
Rick
Hi Bob & others,
As we all know, writing something in a book/design spec. and getting it to work in practice, is not always the same. Lots of variables.
I see this in Doug Self's books too, okay I realize that you can buy the stuff from Signal Transfer company. Is there a way to make it more available/affordable to the masses?
How about saying that it is made in North America for a change
Too bad lots of my parts are made/assembled in Asia, but that I have little control over and learn to live with.My suggestion is to actually take one or more ckts as described in the book and make them work. I know Bob has done his own implementations, but they are not generally available to people who just want to buy the pcb(s), parts & solder them up. Most people on this site can figure it out. I know chassis are available. If not I'll do it for you, since I am looking for work
Not sure if this a proper thing to discuss on this site, since they offer there own stuff?
What do you think Bob, I can offer to do a layout or two for you for $0, maybe some things for your web site. Maybe like what you did with VinylTrack, still gotta buy that copy of LinAudio
Santa did get me that one.I have basically done this of late, just have to fab/test:
1) Input AC assy with auxillary power(+5Vunreg/3v3 lin reg, for logic, soft-start relay, remote on/off.
2) AC rectifier assy, big ecaps, fusing.
3) Linear regulators (+/- 65 -75V),+/-15V, +5/3V3, PA protection ckts
4) LME49830 (WireAmp) clone. (Could be convinced to make this as a super gain clone)
I always like to have others review my stuff since I have no trade secrets with this design. Since the AC stuff is very important, this is where I would like to have a second eye.
I am sure that there are many ways to implements this stuff, but the idea is to make it small, not overly complicated/expensive, use easy to get parts, Mouser, Digi-key, Newark. Make it of quality construction
As it is, I have already bought all of the expensive parts for this PA project. As well I have got the radio/media player/pre-amp designed, working/tested, so I am going through with my stuff.
Listening to a Stones LP on the FM (Q107/Toronto) radio, "Going Home", first album "After Math", the record crackle was a welcome change.
Nothing like this PA
http://www.diyaudio.com/forums/solid-state/221979-prototype-power-amplifier.html
I'd like to see his invoices!
Cheers
Rick
Last edited:
Thermal Trak revisited.
The recent discussions over in the Self thread have made me revisit the Thermal Trak Vbe issue.
Bob posted thisThermal Trak Circuit before the book. It looks attractive but it never seems to have been followed up, I am curious to learn why.
Best wishes
David
The recent discussions over in the Self thread have made me revisit the Thermal Trak Vbe issue.
Bob posted thisThermal Trak Circuit before the book. It looks attractive but it never seems to have been followed up, I am curious to learn why.
Best wishes
David
That is basically an Allison driver circuit, with emphasis on thermal matching. These require extreme care to stabilize, and it can be very difficult. Perhaps not as difficult when the outputs aren't included in the feedback loop. Furthermore aberrations in the circuit's response may limit the global stability. Refer to Ed Stuart's Super TIS page for the best way to stabilize Allison-related bias circuits.
I used this part in my TT amp:
"The two “extra” ThermalTrak diodes, D3-4, are used between the emitters of the driver transistors to establish the idle current of the drivers while keeping the impedance between the emitters very small, so that the drivers can operate in push-pull to provide turn-on and turn-off current to the output transistors. Keeping the impedance between the driver emitters very low at high frequencies is essentially what is often done with the speedup capacitor. The ThermalTrak diodes make it possible to do this in a d.c. fashion with the necessary precision because they track the output transistor Vbe’s."
BR Damir
IIRC the need for low impedance here was to prevent charge suckout from twanging the output bias.
In this case the diodes serve no purpose but to keep the driver idle more constant, because the Allison feedback loop lowers the driver output impedance greatly. This does not necessarily eliminate the need for the cap though, because regardless of any feedback loop, the drivers can't sink current in reverse. By the time the diodes are conducting the suckout current, the odd driver will already be in reverse saturation.
So a suckout cap here may still be needed. Unfortunately because the driver Zout has been reduced so much, you'll need a much larger cap in order to get the same level of drive sharing between the drivers.
So, actual dynamic ability of this output stage at high powers and high frequencies may be considerably less than that of an ordinary driver stage.
This is of course assuming that charge suckout is actually a problem. Is it really? The ONSemi devices are really fast, with very low Cbe. What does it take to saturate the drivers into these devices?
In this case the diodes serve no purpose but to keep the driver idle more constant, because the Allison feedback loop lowers the driver output impedance greatly. This does not necessarily eliminate the need for the cap though, because regardless of any feedback loop, the drivers can't sink current in reverse. By the time the diodes are conducting the suckout current, the odd driver will already be in reverse saturation.
So a suckout cap here may still be needed. Unfortunately because the driver Zout has been reduced so much, you'll need a much larger cap in order to get the same level of drive sharing between the drivers.
So, actual dynamic ability of this output stage at high powers and high frequencies may be considerably less than that of an ordinary driver stage.
This is of course assuming that charge suckout is actually a problem. Is it really? The ONSemi devices are really fast, with very low Cbe. What does it take to saturate the drivers into these devices?
Yes, I noticed that, Ovidiu Popa also used it, but few others.
It makes sense to me and I wondered why Bob did not follow up with it in his book.
Actually it was your plan to use this with 4xxx transistors that made me re-check to see if it would still work with the different diode Vf. Then I became curious!
Yes, I understood Bob's explanation. It seems an excellent idea.
Unfortunately I see no obvious way to use this in a small amp with only one output pair. Damn.
Best wishes
David
You always can use capacitor parallel to the drivers emitter resistor, but then the bias spreader should be a bit different(different resistor values).
BR Damir
Hi Bob & others,
As we all know, writing something in a book/design spec. and getting it to work in practice, is not always the same. Lots of variables.
I see this in Doug Self's books too, okay I realize that you can buy the stuff from Signal Transfer company. Is there a way to make it more available/affordable to the masses?
How about saying that it is made in North America for a change
My suggestion is to actually take one or more ckts as described in the book and make them work. I know Bob has done his own implementations, but they are not generally available to people who just want to buy the pcb(s), parts & solder them up. Most people on this site can figure it out. I know chassis are available. If not I'll do it for you, since I am looking for work
Not sure if this a proper thing to discuss on this site, since they offer there own stuff?
What do you think Bob, I can offer to do a layout or two for you for $0, maybe some things for your web site. Maybe like what you did with VinylTrack, still gotta buy that copy of LinAudio
I have basically done this of late, just have to fab/test:
1) Input AC assy with auxillary power(+5Vunreg/3v3 lin reg, for logic, soft-start relay, remote on/off.
2) AC rectifier assy, big ecaps, fusing.
3) Linear regulators (+/- 65 -75V),+/-15V, +5/3V3, PA protection ckts
4) LME49830 (WireAmp) clone. (Could be convinced to make this as a super gain clone)
I always like to have others review my stuff since I have no trade secrets with this design. Since the AC stuff is very important, this is where I would like to have a second eye.
I am sure that there are many ways to implements this stuff, but the idea is to make it small, not overly complicated/expensive, use easy to get parts, Mouser, Digi-key, Newark. Make it of quality construction
As it is, I have already bought all of the expensive parts for this PA project. As well I have got the radio/media player/pre-amp designed, working/tested, so I am going through with my stuff.
Listening to a Stones LP on the FM (Q107/Toronto) radio, "Going Home", first album "After Math", the record crackle was a welcome change.
Nothing like this PA
http://www.diyaudio.com/forums/solid-state/221979-prototype-power-amplifier.html
I'd like to see his invoices!
Cheers
Rick
Hi Rick,
Sorry to be so slow getting back to you. These are great DIY ideas, and I have given thought to something like this, but just have not had the time. You're right - its one thing to breadboard a prototype and measure it, and yet another to create it in a way that others can readily duplicate. I generally do my prototyping on perf-board where the components are placed and routed in a way that is close to how the corresponding PCB would be laid out.
Send me a PM.
Cheers,
Bob
Hi David,
I am guilty as charged for not following up on it. I was writing my book at the time and was wrestling with the need to pull charge out of the power transistors to turn them off fast enough and mitigate dynamic crossover distortion. There is a section in my book where I discuss the problem, but did not follow up on this scheme. The usual capacitor from emitter to emitter of the drivers appeared to need to be quite large to be effective, and the charge being pulled out of the power transistors could charge up this capacitor and upset the bias shortly thereafter. I thought that the availability of the ThermalTrak diodes could provide an opportunity to achieve the function in a DC way that would not involve a charge storage device (i.e., the speedup capacitor).
The effective base-emitter capacitance of a power transistor that is conducting significant current is surprisingly large, even for a fairly fast power transistor. The hybrid-pi capacitance (gm/wt) gives a hint of this, although the effective capacitance changes with current.
I should have followed up on this idea. Indeed, there may be real-world practicalities that make it difficult to implement reliably.
Cheers,
Bob