I think shunt, and Alexander comp as well for that matter, work better than Miller because they allow you to close the loop at higher ULGF. If you are using a TAS ( Transadnittance = current mirror), the input capacitance is very low, the the BW high, hence the phase shift is also lower. So, the phase margin budget is better in this case as a result. In the case of the Hawsford, the input capacitance is also low. In both cases, this can translate directly into more FB at HF and higher SR's (comp cap is smaller and ULGF higher).
However, ULGF's of 6 MHz is pushing it a bit 😉.
Miller is very sub-Optimal for CFA for reasons I outlined in my CFA vs VFA write-up. Come to think of it, it's probably not a good idea for VFA anymore either - things have progressed remarkably over the last 10 years or so wrt compensation design.
However, ULGF's of 6 MHz is pushing it a bit 😉.
Miller is very sub-Optimal for CFA for reasons I outlined in my CFA vs VFA write-up. Come to think of it, it's probably not a good idea for VFA anymore either - things have progressed remarkably over the last 10 years or so wrt compensation design.
Incorrect. You have shown no proof of this, not even a reference to the "Sassen paper".
Incorrect on all counts. (Large signal) current on demand has nothing to do with the (small signal) frequency compensation. For the same (first) order compensation, the higher ULGF means lower stability margins, there's no way around. Two pole compensation has nothing to do with shunt compensation.
Alexander comp is shunt compensation, it shunts indirectly to ground. The same effect is had if one shunts via a resistor directly to groud.
Please could you expand on this?
I see what looks like a shunt in the classic TPC (where the resistor between the two caps is grounded / taken to supply rail).
The shunt part of TPMC is often overlooked. Notice TPC = Two Pole Compensation, so it is a name of a special Bode plot curve, not a name of any specific circuit or topology. So the shunt part of TPMC (Two-Pole Miller Compensation, the most common form) could be seen as a "leakage" path, which may improve or worsen stability depending on other parts of the design. However if it plays a significant role in the compensation scheme, then what you have is a combination/hybrid of TPC and something else. If such a scheme has shown to be beneficial in an emitter/source-feedback amp, then recognizing and studying it as a special case may allow us to develop a better scheme.
See the CFA amp I posted, most overlooked the fact that it uses 2 pole compensation. It has two shunts on each vas polarity. Most thought its just a shunt compensated amp. Thats the reason for the excellent performance.
If you can get hold of Sassens paper its a little easier to understand but I warn you I hope your maths is still fresh in your mind, its daunting stuff in there.
The design is standard VFA... differential input etc... the output stage is same topology as shown in the www. site except that there are 3 OP transistors paralleled. I measure the same results as shown in the theory of that www site ref to in #5848, above. Power at 10W-40W at 1KHz into 8 Ohms. Best bias point gives = or <-100dB THD.
Do you have ideas about this discrepancy here for the higher bias used?
If the thermal tracking isnt perfect and fast enough, such amplifiers with a narrow bias null for THD could be dynamically shifting into higher distortion when listening to music. Bench testing is too slow and the bias is stable enough when testing continuous steady state signals. This is something I want to look into. It could be yet another situation where measured data doesnt always match up with listening data. It might be blameless on tests and SIM but rises to a high THD under dynamic conditions as the sides away from null are steep and THD rises quickly to higher levels with a small bias change.
THx-Richard
Last edited:
Have to admit, don't recall noticing this in your amp.The discovery was made, at least for me, from Edmond Stewart's SuperTis design.
So, why is a two pole shunt only compensated amp better than a two pole shunt + TMC compensated (my last prototype) amp?
Not fresh at all. Managed to get a degree in Electronics (although didn't have much of a grasp on analogue) so probably understood it at one time.
I don't think so, and will be away of this thread till the nice atmosphere of those last weeks will be back again.
Now could you once and forever post a reference to that mysterious "Sassen paper" on shunt compensation in CFAs? Perhaps it's [Willy] Sansen? And if so, which paper is that? His book "Analog Design Essentials" is a study requirement in our university, but then I don't recall anywhere preaching shunt compensation for CFAs.
Thank you.
Hi Guys
In the bias discussion, Vbe was erroneously mentioned earlier (post-5849_ where Vq was meant. For many, Vq is the total voltage across the complementary Re's, where Self defines Vq as being across only one Re (one side of the circuit). In any event, SDelf determined that vq is the critical parameter, not the specific idle current that results.
However, the CFP can attain its lowest THD with idle currents nearly one-tenth those of EFs, presumably because of the local feedback loop within the CFP.
Since the low level signal lives within the class-A region of the output stage where both halves of the circuit are working and gm is constant, and then distortions appear once one side turns off even though total gm begins to rise, it would seem that a means to make gm constant over the large signal cycle might be something to investigate. Previously, a curve was shown illustrating how THD varies with idle current which would probably have a correlating gm vs THD curve. Error correction is the usual way to try to control gm without resorting to class-A for large signals.
It would also seem worth investigating to have a bias system that monitors Vq directly, rather than having to anticipate all the Vbe variations with temperature etc. Self's Trimodal amp uses such a scheme for the class-A mode. When it switches to class-AB THD is higher than for either A or B, and there is likely a way around that occurring for those who don't mind adding more BJTs.
Have fun
Kevin O'Connor
In the bias discussion, Vbe was erroneously mentioned earlier (post-5849_ where Vq was meant. For many, Vq is the total voltage across the complementary Re's, where Self defines Vq as being across only one Re (one side of the circuit). In any event, SDelf determined that vq is the critical parameter, not the specific idle current that results.
However, the CFP can attain its lowest THD with idle currents nearly one-tenth those of EFs, presumably because of the local feedback loop within the CFP.
Since the low level signal lives within the class-A region of the output stage where both halves of the circuit are working and gm is constant, and then distortions appear once one side turns off even though total gm begins to rise, it would seem that a means to make gm constant over the large signal cycle might be something to investigate. Previously, a curve was shown illustrating how THD varies with idle current which would probably have a correlating gm vs THD curve. Error correction is the usual way to try to control gm without resorting to class-A for large signals.
It would also seem worth investigating to have a bias system that monitors Vq directly, rather than having to anticipate all the Vbe variations with temperature etc. Self's Trimodal amp uses such a scheme for the class-A mode. When it switches to class-AB THD is higher than for either A or B, and there is likely a way around that occurring for those who don't mind adding more BJTs.
Have fun
Kevin O'Connor
There is no shunt in TPMC, there is an equivalent lead-lag effect. Big difference, as from a pole to a pole + a zero.
Incorrect. Alexander amp has "input inclusive" Miller compensation, from the VAS output to the inverting input node. The fact that the inverting input node is low impedance doesn't change this fact, it is still Miller compensation.
Shunt via a resistor directly to ground is not "shunt compensation", it is called lead-lag compensation. It introduces a zero in the forward path, which changes completely the picture of a single pole shunt.
Lead lag compensation is in general useful for compensating the Miller loop, so the lead - lag pole + zero are at frequencies much higher than the Miller dominant pole.
RNMarsh,
Why not use a real musical signal on the bench rather than a sine wave or such steady state signal if you are wanting to look at a change due to dynamic loading of the system? Could you develop a two or three tone test with varying output level that would more easily show the THD under load than trying to analyze a musical signal for distortion?
Why not use a real musical signal on the bench rather than a sine wave or such steady state signal if you are wanting to look at a change due to dynamic loading of the system? Could you develop a two or three tone test with varying output level that would more easily show the THD under load than trying to analyze a musical signal for distortion?
Hi, all good. I am using the idle current as a marker... not as an explanation of how the circuit actually operates or works in detail. I could have said the same occures for a narrow range of Vbe. Doesnt matter. I am still wondering about the descrepency of using high bias (60-80mA) for low distortion and low bias where a null in distortion can be found. .
The Sziklai transistor pair info can be found in the literature since first described and patented in 1963. "Some advantages are: more immune to thermal runaway. And, a Sziklai derived complimentary push-pull arrangement would be more immune to NPN and PNP mismatches. As an interesting aside, the Sziklai pair most resembles the IGBT equivalent circuit" Linear Audio Vol 3 pg 34.
Note: current gain is similar to that of a Darlington pair.... the product of the gains of the two transistors.
TH-RNMarsh
Last edited:
Thanks for educational info Waly, will benefit from lecture.
Can you please post the continuation of your input stage presented here? What's the output and PSU associated with it?
Can you please post the continuation of your input stage presented here? What's the output and PSU associated with it?