I guess that nonswitching output stages are intended for the 3rd edition?
I already have plans to touch on it in the new added chapter on output stages, but that chapter is still a work in progress and I have never built a non-switching output stage except for one that uses the LT1166.
I always remind people that I do not consider many of the non-switching output stages to be a version of class A just because neither output transistor turns completely off. Such terminology can be misleading and often originates from the marketing department.
In my view, the proper description of a circuit that can be considered a version of class A is one where both transistors are on AND contributing some amount of transconductance to the output stage as a whole. A circuit that just in some way keeps the bias from going to zero when it otherwise would is not causing that transistor to contribute gm when it otherwise would not. For example, some kind of a diode arrangement that prevented Vbe from going to the point of no conduction is a type of such circuit. Indeed, that kind of arrangement does have a switching output stage, but the switching that occurs is in that "commutating" diode, which is much faster. Such circuits also do not eliminate the static crossover distortion that occurs at crossover as a result of the total output stage transconductance going through a wingspread change.
Traditional proper class A has a more strict definition in my view. In such a class A amplifier, both output transistors are contributing on the order of the same transconductance throughout the cycle (of course, in the real world there are some departures from this, so perhaps substitute "substantial").
That having been said, and semantics aside, more gentle switching of the slow output devices can sometimes reduce or eliminate switching glitches.
If I have time and space, and I find one that I am comfortable with, I might simulate it next to an otherwise-identical one without the non-switching feature. Suggestions for the non-switching stage are welcome.
Sliding bias is an interesting candidate for discussion.
Cheers,
Bob
So you wanna meet Bob in person? He's a great guy! as you know from this thread..
He'll be giving a talk at Burning Amp this year on Nov 13th thanks to Linear Integrated Systems, who are flying him out..
http://www.diyaudio.com/forums/clubs-events/289409-burning-amp-2016-a-10.html#post4859491
He'll be giving a talk at Burning Amp this year on Nov 13th thanks to Linear Integrated Systems, who are flying him out..
http://www.diyaudio.com/forums/clubs-events/289409-burning-amp-2016-a-10.html#post4859491
Last edited:
Thanks! I'm really looking forward to going out to Burning Amp, but I am actually doing it on my own dime.
I'll be describing a complete re-do of the venerable Hafler DH-220 lateral MOSFET amp, with emphasis on implementing a fully complementary JFET input stage using the Linear Systems LSK489 N-channel and LSJ689 P-channel dual monolithic JFETs.
Cheers,
Bob
I had that Hafler I think, but sold it to a guy and one channel stopped working. Hmmmm I think I might be able to get it back for half price!
Their power transformers are really a good size and the heatsinks are nice. Easy to work on as well. If you can, check to see if the output MOSFETs in the bad channel are still good; if not, you will need a replacement pair for the N or P channel pair that died (or both in the worst case).
Cheers,
Bob
Bob, I suggest you mention that NPN transistors usually have much higher Early voltage than their PNP counterparts, a fact which often makes NPNs a much better choice for second stages "VAS" in power amps -- at least, if the second stage is not cascoded.
I also suggest that you present Bob Cordell's Rule Of Thumb for the minimum acceptable gain of a second stage. Is it (Vposrail - Vnegrail)/0.5V ?? (that's my rule, but you're not me). If Posrail = +40V and Negrail = -40V then min 2nd stage gain is 160X (or your recommended number). You could then remind readers that 2nd stage gain is set by 2nd stage emitter resistance, the output impedance of the current source load, the input impedance of the OPS, and the output impedance of the VAS (if not cascoded). So to get 160X gain you need each of those parallel impedances to be at least 3*160*Re ohms. When Re=22 ohms as in your Figure 3.10, you need each of the impedances to be at least 10.5K. Aha! Notice that the input impedance of a double EF OPS, driving a 2 ohm loudspeaker, is too low. It takes the gain of the VAS below your Rule Of Thumb.
I also suggest that you present Bob Cordell's Rule Of Thumb for the minimum acceptable gain of a second stage. Is it (Vposrail - Vnegrail)/0.5V ?? (that's my rule, but you're not me). If Posrail = +40V and Negrail = -40V then min 2nd stage gain is 160X (or your recommended number). You could then remind readers that 2nd stage gain is set by 2nd stage emitter resistance, the output impedance of the current source load, the input impedance of the OPS, and the output impedance of the VAS (if not cascoded). So to get 160X gain you need each of those parallel impedances to be at least 3*160*Re ohms. When Re=22 ohms as in your Figure 3.10, you need each of the impedances to be at least 10.5K. Aha! Notice that the input impedance of a double EF OPS, driving a 2 ohm loudspeaker, is too low. It takes the gain of the VAS below your Rule Of Thumb.
Are you going to publish the schematic of the re-do?
Thanks for teaching me so much.
hopefully, he will or at least include it in the upcoming next edition of his book.
in the meantime, for those that are able, it might be fun to guess before he does.
with the parts bob mentions, you could almost do a one-to-one replacement referring to the XL-280 schematic.
the problem is ready (and affordable) access to the lateral mosfets used for the output stage. you could go with a bjt triple (and get a little extra output power for the same rails), but that probably differs from the "spirit" of the hafler. or you use some vertical mosfets, but you have to rethink and redo the biasing (and maybe step up the driver rails a bit to compensate for increased VGS needed).
i never looked that hard, but it seems to me it might be interesting to sniff around for performance fruit to grab by changing from the compensation scheme used in the hafler designs.
anyway, it sounds like fun and bob's the right guy for the task.
mlloyd1
in the meantime, for those that are able, it might be fun to guess before he does.
with the parts bob mentions, you could almost do a one-to-one replacement referring to the XL-280 schematic.
the problem is ready (and affordable) access to the lateral mosfets used for the output stage. you could go with a bjt triple (and get a little extra output power for the same rails), but that probably differs from the "spirit" of the hafler. or you use some vertical mosfets, but you have to rethink and redo the biasing (and maybe step up the driver rails a bit to compensate for increased VGS needed).
i never looked that hard, but it seems to me it might be interesting to sniff around for performance fruit to grab by changing from the compensation scheme used in the hafler designs.
anyway, it sounds like fun and bob's the right guy for the task.
mlloyd1
hopefully, he will or at least include it in the upcoming next edition of his book.
in the meantime, for those that are able, it might be fun to guess before he does.
with the parts bob mentions, you could almost do a one-to-one replacement referring to the XL-280 schematic.
the problem is ready (and affordable) access to the lateral mosfets used for the output stage. you could go with a bjt triple (and get a little extra output power for the same rails), but that probably differs from the "spirit" of the hafler. or you use some vertical mosfets, but you have to rethink and redo the biasing (and maybe step up the driver rails a bit to compensate for increased VGS needed).
i never looked that hard, but it seems to me it might be interesting to sniff around for performance fruit to grab by changing from the compensation scheme used in the hafler designs.
anyway, it sounds like fun and bob's the right guy for the task.
mlloyd1
Do you have a link to the XL-280 schematic?
Thanks
Hi Mark,
These are good points. Early effect is important in the VAS transistor. Using significant emitter degeneration in the VAS helps raise the output impedance of the VAS transistor, and I like at least 10:1 degeneration of the VAS. However, note that the degenration, all else remaining equal, decreases VAS voltage gain.
Overall, it is important to keep the output impedance of the current source high (using a feedback current source helps) and to keep high current gain in the output stage. For this reason, I do not consider an output stage double to be acceptable for a really high quality bipolar output amplifier. My favorite is usually the Locanthi T circuit (Triple), but other triples, like the Diamond Buffer Triple are also very attractive.
Note also that the use of a 2EF (Darlington) VAS also helps to increase the output impedance of the VAS. Transistors like the 2SC3503 and 2SA1301 have a high figure of merit in regard to the product of Beta and Early voltage. High Early voltage alone is not that great if Beta is not also high.
Cheers,
Bob
Bob, if they are good points, why not put them into the 2nd edition of your book?
I suspect that VAnpn >> VApnp is not well known among your readers, and I know that your first edition does not contain tables of either VA or (VA * Beta) for a dozen PNPs and a dozen NPNs. Similarly I suspect your readers may not have thought to ask themselves "how much signal swing from the first stage, does my second stage require?" (answer: (2*Vrail / VASstageGain)). Why not give them Bob Cordell's Rule Of Thumb? It will help answer the question: EFdouble or EFtriple? It will also answer the question: Why oh why does John Curl use complementary MOSFET source followers (rather than complementary BJT emitter followers) as the 2nd-to-last stage in his power amps, driving the bases of the final output BJTs?
edit- I've attached the schematic of John Curl's "Parasound HCA-1200" power amp. Take a gander at Q115 and Q121.
_
I suspect that VAnpn >> VApnp is not well known among your readers, and I know that your first edition does not contain tables of either VA or (VA * Beta) for a dozen PNPs and a dozen NPNs. Similarly I suspect your readers may not have thought to ask themselves "how much signal swing from the first stage, does my second stage require?" (answer: (2*Vrail / VASstageGain)). Why not give them Bob Cordell's Rule Of Thumb? It will help answer the question: EFdouble or EFtriple? It will also answer the question: Why oh why does John Curl use complementary MOSFET source followers (rather than complementary BJT emitter followers) as the 2nd-to-last stage in his power amps, driving the bases of the final output BJTs?
edit- I've attached the schematic of John Curl's "Parasound HCA-1200" power amp. Take a gander at Q115 and Q121.
_
Last edited:
hopefully, he will or at least include it in the upcoming next edition of his book.
in the meantime, for those that are able, it might be fun to guess before he does.
with the parts bob mentions, you could almost do a one-to-one replacement referring to the XL-280 schematic.
the problem is ready (and affordable) access to the lateral mosfets used for the output stage. you could go with a bjt triple (and get a little extra output power for the same rails), but that probably differs from the "spirit" of the hafler. or you use some vertical mosfets, but you have to rethink and redo the biasing (and maybe step up the driver rails a bit to compensate for increased VGS needed).
i never looked that hard, but it seems to me it might be interesting to sniff around for performance fruit to grab by changing from the compensation scheme used in the hafler designs.
anyway, it sounds like fun and bob's the right guy for the task.
mlloyd1
There are indeed similarities to the XL280 in that I use a full complementary JFET input with a floating tail (but with available LSK489 nd LSJ689). These devices provide lower transconductance but in the circuit provide quite low noise (the JFETs are but one source of noise in a usual IPS). As a bonus, they provide low capacitance. They are cascoded as in the XL280 for reasons of rail voltage. I use a loaded helpered current mirror load for the IPS. I use straight Miller compensation. Same lateral output MOSFETs, but one can get sufficiently close ones these days from Renasas. I also use a dc servo.
Cheers,
Bob
Hi Bob:
Thanks for your informative comments.
With a helpered current mirror load IPS, i would guess that you are not using the pseudo folded (sorry, i never knew if there was proper name for it) approach?
Sounds interesting ...
I must admit I was somewhat hoping you would also toss in your error correction OPS (but lateral MOSFET style), but I know how busy you are!
mlloyd1
Thanks for your informative comments.
With a helpered current mirror load IPS, i would guess that you are not using the pseudo folded (sorry, i never knew if there was proper name for it) approach?
Sounds interesting ...
I must admit I was somewhat hoping you would also toss in your error correction OPS (but lateral MOSFET style), but I know how busy you are!
mlloyd1
Last edited:
Hi Bob:
Thanks for your informative comments.
With a helpered current mirror load IPS, i would guess that you are not using the pseudo folded (sorry, i never knew if there was proper name for it) approach?
Sounds interesting ...
I must admit I was somewhat hoping you would also toss in your error correction OPS (but lateral MOSFET style), but I know how busy you are!
mlloyd1
By pseudo folded, I assume you are referring to the IPS/VAS topology where the main collector of the IPS is connected to the base of the VAS and thence to the rail through a load resistor, and the other collector is connected to the emitter of the VAS transistor. This is done with either a 1T VAS or a 2T VAS. The "other" collector is also often just connected right to the rail.
This connection does 2 things. First, it puts approximately the same signal voltage on both collectors of the IPS. This may mitigate some minor Early effect in the input stage by putting the signal voltage into the common mode. I think the effect of this is quite small, especially if the IPS is cascoded.
Secondly, some of the IPS signal current from the "other" output is added to the primary VAS signal current that results from the signal applied to the base. This will forward IPS current to the collector of the VAS by the cascode operation in the secondary path. I have not looked closely at this, but it appears that this secondary IPS current will appear in the collector of the VAS even at high frequencies where a Miller compensation capacitor is present. This may create a zero in the forward response.
I should probably take a closer look at this connection and touch on it in the second edition. I am not under the impression that it has any significant effect. If anyone is aware of beneficial effects of this connection that I have not mentioned, please let me know.
It would be interesting to try out the EC circuit with the amplifier, but time is a bit limited.
Cheers,
Bob