Yes, you need power supplies to power the active device that float on the bias spreader. that power supplies has to get power from outside, you are going to have to have long wire connections to noisy power supply circuits. Bias spreader is a high impedance node that can pick up noise, that last thing you want to do is to have long wires that easily form a loop and pick up magnetic field around.
Also Power supplies have capacitance coupling from the source (be it 60Hz AC or switching frequency) to the bias spreader section.
uP always have to have the clock running even though it's asleep. Or else, it would not wake up!!! they run on square wave clock that produce harmonics, you can have f1-f2 IM that can be down to audio frequency.
If you really sit down to design-in such a cirtcuit you will find that all this can be solved with some smart stuff. Just as an example, the uC can easily run off the voltage across the bias spreader, just an additional uA worth of current shunt.
Yes you can put it in 'deep sleep/static mode' with no clocks running. I have the feeling you are still in 'can't be good' mode but not worked it out. It's a whole different paradigma as I noted before, but that also means you wouldn't be in a position to judge it either.
Jan
AFAIK, the first mention of it in DIYAudio was by JPV here >http://www.diyaudio.com/forums/soli...n-thermal-trak-transistors-3.html#post1639411
Best wishes
David
Well, to be fair, I haven't seen too many "pure" Cherry amps here, other than your work.
Toni found TMC aka "partial Output Inclusive Compensation" to work better in practice.
And Dr Cherry's own amps had a reputation for stability issues when built by others, perhaps without his expertise in layout details.
Still a little unclear, so what's your answer to my earlier question? here >http://www.diyaudio.com/forums/soli...lls-power-amplifier-book-128.html#post4401051
Best wishes
David
Last edited:
Absolutely wrong, I'm afraid.
As I believe I said before, I have used 0R1 as standard for a couple of decades, in amplifiers in serious mass production. Thermal stability has never been an issue.
As it happens I have a four-output-pair amplifier under detailed measurement at the moment. Re tolerance is 5%. No thermal stability issues whatsoever. The bias generator is naturally mounted on the top of one of the output devices.
A better question is if 0R05 resistors are practical politics. Probably it would be necessary to take a hard look at PCB track resistances. Two-ounce copper would help there, but I always use it for power amp PCBs anyway.
Another good question is if 5% is close enough to prevent the generation of distortion because the gains are slightly different in the +ve and -ve directions. I have done some work on this...
Last edited:
I don't think this is established at all. Personally I don't believe it to be true.
If most of the circuitry is actually sitting on the output rail, rather than the VAS node, with the power supply bootstrapped, this is really not a problem. The voltages you want to look at are accessed with a differential amplifier. See the bias regulator of the Class-A amplifier in Chapter 17 of APAD6. Getting that working was completely straightforward.
I think this is true. I wouldn't dream of sitting a microcontroller on the VAS output node. I did a design for one client where all the housekeeping was done in resistor-transistor logic, so there would be no possibility of clock noise etc getting into the audio. It may have looked a bit old-fashioned, but by the time you factor in all the level shifting required to interface with a micro it is competitive. On the other hand, if you need to decode IR data...
It's not a question of horizons. It's a question of prioritising the use of resources. I've seen nothing so far to make me think that CFA amplifiers would get anywhere near the top of my to-do list.
And perhaps you could do with some refreshment on the rules of Polite Discourse.
Just for the record, I do have a copy of Bode's book, and I have read it.
Can you say the same?
wiring
Hi Douglas,
Do you mean the "laws of physics" of the wiring as shown below?
IF your test circuit was built this way, of course it shouldn't work.
Cheers, E.
[snip]
A simple way to achieve this is to adopt Pure Cherry Compensation.
Guru Self will tell us he is unable to make it work but that doesn't stop the rest of us from doing so
Hi Douglas,
Do you mean the "laws of physics" of the wiring as shown below?
IF your test circuit was built this way, of course it shouldn't work.
Cheers, E.
Attachments
This is music to my ears. I always want to use 0.12ohm( hey, at least it's 0.02ohm over 0.1
) but I was told it's dangerous. This is my first amp, I did put in the 0.22ohm to power it up first. I put the resistors on the bottom side, I ordered extra 0.22, 0.18 and 0.33ohm. I am going to experiment putting them in parallel to see.You never mention about Vbe and hfe matching of transistors. I think this is very important if using 0.1ohm. I matched and group transistors in sets of 5( I have 5 pairs of output transistors). I run 200mA current and matched Vbe to 1mV and hfe to 2% for each set of 5. I hope this will make it even safer.
I got laughed here that this is not engineering. As an engineer that worked with designs for production, guaranty by design and all, I feel this is the best way even if it is for production. I matched at least 50 each of the MJW1302/3281 in less than 2 hours. This kind of high end amplifiers, you don't sell large quantity. I can match enough for over 100 amps in one day even with my stupid set up. I am sure with real tester, people can do a lot better. To me, this is engineering. Compare to all the uP, individual servo and all, I consider this is better engineering.
For the extra pcb area, cost overhead for programming and the extra component, I much prefer to add more output pairs in parallel to improve thermal stability. This have the benefit of lower crossover distortion as so well put p238 Fig.9.6 in your book. More pairs make the apparent load resistance to each output pair higher and lower the crossover distortion, move the harmonics to lower frequency and easier to be cancelled by feedback. So just by matching and more output pairs, not only I avoid current hogging, I improve the sound. I don't think the uP route can claim it improve the sound like more pairs.
Thanks
I really think I should get out of this debate, I don't believe in this approach at all as stated in my last response to you. I don't even want to spend any more time thinking about how to do this..
This is music to my ears. I always want to use 0.12ohm( hey, at least it's 0.02ohm over 0.1
I think I've said all I want to on the eminent usability of 0R1. As a side-issue, the route to safety in amplifiers is to make sure it will work (in the sense of giving a visibly OK sinewave) off rails of only a couple of volts. You can then cautiously variac it up from nothing, and even if you've accidentally swopped over the NPN and PNP devices, you shouldn't blow any parts. I speak as one who has done this...
That's because I never do it.
I'm not laughing. The prospect is too frightening. I know of one amplifier company (you would recognise the name) which foundered on this very issue, because their output stage depended on the close matching of 16 (IIRC) power FETs. I talked to the biker charged with this task, while around us stretched dozens of trays each holding 200-odd devices. He was losing the will to live...
The trouble is that if you depend on device matching it may be do-able at first, but if the manufacturer then relaxes his specs or changes his process it may become effectively impossible. It's not just an issue of the time it takes. Danger, Will Robinson!
Hello Edmond
Do you mean that it won't work as shown because the device wiring is too long? I assure it also doesn't work if the power devices are mounted straight on the PCB, as intended for this design.
I don't think I've ever said Cherry compensation can never work; perhaps there are amplifier configurations where it does. Maybe. But I have never been able to make it work with the Blameless configuration/concept, and that is the most stable and predictable configuration I know.
With CFA I can easily get the loop gain of more than 80 dB flat from zero up to more than 20 kHz, and this will lower THD20k below 2 to 4 ppm for the HEXFET OPS and close to 1 ppm for BJT OPS.
I built the HEXFET variant and Richard Marsh measured it with the result quite close to the simulated ones.
I don't say the CFA is better then VFA, I like both.
BR Damir
Last edited:
I am not saying I only depend on device matching. In fact I don't depend on device matching.
1) You said it is safe to use 0.1ohm. I am adding extra level of protection by matching. I am not pushing 0.05ohm or eliminate the Re by matching.
2) I am also using more pairs to reduce peak current of each transistor to reduce junction temperature even with a given heat sink.
This is very conservative approach, to start out with Re=0.12ohm that you deem safe regardless.
I don't use MOSFET, it is much harder to have close match of Vgs as they vary. It's a different ball game, apply and orange.
Last edited:
Once and forever, since this is coming again, and again, and then again... Given a phase margin, an ULGF, and a compensation method order, a CFA circuit topology does NOT provide more loop gain than what you could get with a standard VFA. To understand this, before going to the Lurie stability or the Bode integrals, there is an almost trivial justification: whatever Mr. Zan is going to mumble and moan, audio circuit topologies are still of the minimum phase breed. As such the gain and phase are NOT independent, therefore pushing up the loop gain, pushes down the phase margin, no exceptions (or free lunch, if you prefer). Changing the compensation order is the only way to break this rule, by adding N poles and N-1 zeroes in the unity circle (N=1 Miller compensation, N=2 TPC or TMC, N=3 Cherry NDFL compensation). I've heard of amplifiers (both linear and class D) compensated up to the 5th order, but I have never seen a schematic. The CFA circuit topology has other interesting properties, none of which really matters for audio.
Your example with 80dB loop gain @ 20KHz is either a typo, or you are talking a different and strange Bode language. Even if a two pole type compensation is involved, such an amplifier would have a ULGF higher (I would estimate some 8-10 MHz) that what could be of any practical importance for a power amplifiers (with the power stage limiting the bandwidth).