No, he specifically talked about using 0.1ohm for multiple pairs. He even gave examples of 3 output pairs with 0.1ohm. And that's scary. He insists on not doing any transistor hfe and Vbe matching also. That's unthinkable.
This question has been going on for many posts here lately.
Cherry doesn't use the heretical words "current drive"
but 'pure Cherry' compensation essentially does that for the output 'emitter followers'.'Pure Cherry' IS just another different feedback loop but different feedback loops have different impedances at different points in the circuit. These change the 'inherent' distortions at those points.
Instead of joining the interminable debate, I adopted a different view and investigated whether undistorted voltage or current resulted in undistorted voltage or current for each stage in an attempt to find out what caused the remaining distortions in a simple Blameless topology. I think I posted the results in my 'pure Cherry' thread but that's now too huge to find stuff at short notice. Also the remaining distortions in a 'pure Cherry' amplifier are 2nd & 3rd .. mostly in the VAS .. so I don't address (cos I don't have to) ops xover Oliver bla bla
I remember the first time I tried 'pure Cherry' for real in an amp. The residual THD20k showed MUCH clearer xover which disappeared into the noise as you turned Iq up above 20mA
Previously, I'd been running 100mA & more with loadsa different compensation including TPC .. and never getting the xover spikes to go away until the amp was nearly in Class A.
The 2 Cherry papers that discuss this are
- Feedback, Sensitivity and Stability of Audio Power Amplifiers
- Nested Differentiating Feedback Loops in Simple Audio Amplifiers
S[Gm3] is of course sensitivity of the Closed Loop gain to changes in mutual conductance in the output stage ie Oliver criteria, Iq bla bla.
4) in Section 3 of 1. says "The only method for significantly reducing S[gm3] is to increase the closed-loop cutoff frequency GTiB/C .. " Hence all this obsession with Oliver bla bla with simple Miller comp. AND INDEED ANY TOPOLOGY & COMPENSATION THAT RESULTS IN 'VOLTAGE DRIVE' OF THE OUTPUT STAGE including Edmond's & Great Guru Baxandall's bastardized TMC
.The rest of that section expounds at length on the effect of hfe in both VAS & OPS.
The bit in 2. after Eqns (19) & (20) pontificate on the advantage of using 'pure Cherry'. He mentions 'deadband' which is a concept I've used much earlier in my novice amp days .. also Blomley which is an explicit example of current drive. Note the bit at the bottom of that page where he explains what happens when hfe of the 2 ops halves are equal.
My $0.02 is that even if they are not, the THD introduced is far lower order than xover even with perfect Oliver bla bla
Of course even with 'pure Cherry', you do want to have Iq in the right ballpark. The current drive breaks down eg in a simple Blameless cos the VAS has yucky Cob. Cascoding it helps a lot and indeed, if yucky Cob is the major THD source, this is worthwhile.
In the huge [url]http://www.diyaudio.com/forums/analog-line-level/218373-discrete-opamp-open-design.html[/URL] thread, I show a discrete OPA example where cascoding the VAS was preferable to 'enhancing' the VAS ala Blameless.
But Iq is FAR LESS CRITICAL in a 'pure Cherry' design and this has implications for stuff with VFETs & also how accurately you need to control Iq too.
Any free paper on Cherry Amp? I have no idea what it is.
Finally I power up my IPS/VAS. I just connected the collector of the top VAS to the collector of the bottom VAS and drive the 20K feedback resistor. Just cheesy way to test the IPS/VAS without the OPS. I got about 500KHz at large signal -3dB BW and gain of 21(20K and 1K resistor). I am a little disappointed because I up the current of the LTP to 2mA tail and push pull VAS at 14mA idle. That's a lot of current to overcome dV/dt. I can see it's slew rate limited as the sine turned into triangular wave. The SR is about 35 to 40V/uS. I was hoping to see 1MHz. I am using TMC compensation with 18pF,180pF and 1K resistor. I have no peaking as I run the frequency up, that's good news. But it's too slow. It's not going to get better with the OPS and with 4ohm load.
I need to check all the DC bias and also see what I can do to increase slew rate, I am surprised I've gone this far first time power up.
Thanks
Finally I power up my IPS/VAS. I just connected the collector of the top VAS to the collector of the bottom VAS and drive the 20K feedback resistor. Just cheesy way to test the IPS/VAS without the OPS. I got about 500KHz at large signal -3dB BW and gain of 21(20K and 1K resistor). I am a little disappointed because I up the current of the LTP to 2mA tail and push pull VAS at 14mA idle. That's a lot of current to overcome dV/dt. I can see it's slew rate limited as the sine turned into triangular wave. The SR is about 35 to 40V/uS. I was hoping to see 1MHz. I am using TMC compensation with 18pF,180pF and 1K resistor. I have no peaking as I run the frequency up, that's good news. But it's too slow. It's not going to get better with the OPS and with 4ohm load.
I need to check all the DC bias and also see what I can do to increase slew rate, I am surprised I've gone this far first time power up.
Thanks
Last edited:
Instead of waiting somebody to baby feed you, why don't you get a good book from your BFF Google and start reading? You should be able to ask intelligent questions before requesting peer help. Alternatively, since this goes above and beyond the realm of DIY, I can send you my hourly rate for distance education.
Otherwise, I am not going into another debate, which would be once again an irreconcilable conflict between EE foundations and practices and your infinite pool of free time.
Let me suggest you to bone up Negative Feedback and how it lower distortion, lower output impedance. How loop gain affect the accuracy. This is well covered in both Mr. Cordell and Self's book.......and any other book on negative feedback.
Really? I never knew that, thanks for bringing the feedback properties to my attention.
Attachments
Yes, for you to make that comment, that shows you really don't understand open loop gain, loop gain and how negative feed back works. It is important for you to know these very very basic things to talk.
Waly, Alan made a statement, you said it was wrong in multiple ways, Alan asks why, you tell him to go study. I don't know either of you, but I can't see why he has any reason to go look up things, other than your statement, one that you didn't support. And after all, even with you wanting to show everyone how trivial it all seems to you by saying "blah blah" over and over... you're still only Some Guy On The Internet.
Attachments
Load Invariant is a 2pair output stage using 0r1 for 8ohms duty that performs well into 4r0 dummy resistor load. And it has no emitter resistors in the CFP output devices.
Load Invariant is detailed in my 2nd edition copy dated 2000
Last edited:
Be careful here. If you go for a high loop gain CFA, you have the same issues to deal with that you have in a VFA. The solution which is de riguer now days is multi- pole comp of which two pole is the most commonly used. You can apply that to CFA just as well as VFA.
If you look at the article on my website that discusses this, you will see that the real benefits of CFA (fast rise fall times, high slew rate, wide bandwidth and no gain-bandwidth dependence can only be met under specific conditions - and one of those is that the open loop gain is low and the open loop - 3dB bandwidth is high. This translates into higher ULGF and in general 6-8 dB of greater loop gain at HF in classic CFA topologies.
If you look at the article on my website that discusses this, you will see that the real benefits of CFA (fast rise fall times, high slew rate, wide bandwidth and no gain-bandwidth dependence can only be met under specific conditions - and one of those is that the open loop gain is low and the open loop - 3dB bandwidth is high. This translates into higher ULGF and in general 6-8 dB of greater loop gain at HF in classic CFA topologies.
Last edited:
Hi dadod,
Thanks for presenting this plot. It does have remarkable loop gain at 20kHz, but it is not defying any laws of physics. If you drop off at 40dB/decade for 2+ decades, with a 5MHz ULGF, it all adds up and there is no magic.
The real question is why we can't do the same thing with a VFA? We can certainly shape the LG rolloff of a VFA pretty much the same way using TPC.
Put differently, is there less phase lag inherent in the CFA topology than in a VFA topology that has been designed with great care? We know they both share the same output stage, which is the big elephant in the room.
Where might the possible reduced phase lag come from in the CFA?
Looking at the feedback loop around the whole path, we see pretty much the same thing in the VAS and output stage. BUT the input stage is different. In a CFA the input stage looks more like a cascode to the feedback path, while in a VFA the input stage looks like a common emitter stage to the feedback loop. We normally think of a common base stage as being faster and with less phase lag than a common emitter stage.
I think the key question is whether, with careful design, we can make a VFA IPS as wideband (or low phase delay) as the IPS in a CFA as seen by the feedback path.
How close can we get? I think the first part of this is to drive the VFA LTP with very low impedance on both sides at high frequencies. This means a low-impedance feedback network on the feedback side and a capacitance to ground on the input side which will be part of the input LPF anyway (with the possible exception of a base resistor of small value if needed for stability).
That having been done, the next thing to do is to probably cascode the IPS and use fast, lower-voltage transistors in the IPS (breakdowns maybe <30V). But we also need to be aware that even the cascode transistors will add a bit of additional phase lag.
Cheers,
Bob
I am scratching my head right now. I just tested my IPS/VAS board without OPS and only load driving is the 20K feedback resistor and the 10X scope probe. I use VFA with TMC of 18pF/1K/180pF. I have 2mA complementary IPS with active load, 14mA push pull VAS. I only get 500KHz large signal BW of 40Vpp, 40V/uS. Even small signal of less than 4Vpp is 900KHz.
I am not that brave to use TPC as I afraid of over shoot and ringing if not careful. With the safer TMC, I think I am pushing very hard already.
Only other thing is the 20K feedback resistor that can create a pole with the input capacitance. But I already have cascode in the LTP to eliminate the miller capacitance already. Seems like I am reaching the end of the road with the VFA. Only thing left is lower the feedback resistor to 10K and use 500ohm to get back gain of 21. But then I will have problem on the low cutoff frequency as I am using a 220uF cap for AC ground for the feed back.