In my last prototype (very much like the classic design by D.Self but added cascode for long tail pair and hybrid CFP in power stage) did not use miller compensation at all but 50pF from VAS (collector) point directly to long tail pair negative input (like the basic NFB).
Now I wonder what disadvanteges this design may have?
My own first assumption was that there must exist TIM distortion in the VAS stage input. The simulation of the input and VAS stages together shows that without NFB nor compensation the open loop bandwith is about 20-30kHz (and amplification >100db). So may be TIM distortion is not probable, or is it?
Now I wonder what disadvanteges this design may have?
My own first assumption was that there must exist TIM distortion in the VAS stage input. The simulation of the input and VAS stages together shows that without NFB nor compensation the open loop bandwith is about 20-30kHz (and amplification >100db). So may be TIM distortion is not probable, or is it?
Eva, my stability observations are indeed from the real amp not from simulation, however I have not tested it with complex load.
Nelson Pass, how should I understand your comment "good old days"
You are right it is not lag compensation but I don't figure out if the miller capacitor is needed at all. The role of that compensation is to shape the phase and amplitude behaviour so that when the real NFB is added the amp is still stable.
The differential stage and VAS stage together closed (by capacitor) is actually very much like closing the pnp driver and npn power device into CFP kind of loop. Only difference is that the both tails of pair are used to drive second stage (VAS).
Without the miller capacitance the VAS act more like voltage driven current source even on higher fregs. but when miller is added VAS acts like current drive current source. I beleave this is good but it leads to the higher impedance in the VAS output which is not good when driving nonlinear driver BJTs of the output stage. On the other hand there is much amplification to be used. This sounds like the old BJT amps with TIM and other problems but with exception that the diff stage and VAS stage together seems to be fast.
A bit similar approach is also used by A. Holton in his amp. The difference is that he used also miller compensation.
So, what is wrong in my thinking? There must be a reason why this way is not used.
Nelson Pass, how should I understand your comment "good old days"
You are right it is not lag compensation but I don't figure out if the miller capacitor is needed at all. The role of that compensation is to shape the phase and amplitude behaviour so that when the real NFB is added the amp is still stable.
The differential stage and VAS stage together closed (by capacitor) is actually very much like closing the pnp driver and npn power device into CFP kind of loop. Only difference is that the both tails of pair are used to drive second stage (VAS).
Without the miller capacitance the VAS act more like voltage driven current source even on higher fregs. but when miller is added VAS acts like current drive current source. I beleave this is good but it leads to the higher impedance in the VAS output which is not good when driving nonlinear driver BJTs of the output stage. On the other hand there is much amplification to be used. This sounds like the old BJT amps with TIM and other problems but with exception that the diff stage and VAS stage together seems to be fast.
A bit similar approach is also used by A. Holton in his amp. The difference is that he used also miller compensation.
So, what is wrong in my thinking? There must be a reason why this way is not used.
Most of the time if you need lag compensation you will
need it just driving an 8 ohm resistor, or even no load at all,
and the symptom will be oscillation, which is easy to see if you
have a scope.
You can also look for ringing on square waves.
Still, feeding the high frequency signal back to the
(-) input is usually a better approach - assuming it works.
need it just driving an 8 ohm resistor, or even no load at all,
and the symptom will be oscillation, which is easy to see if you
have a scope.
You can also look for ringing on square waves.
Still, feeding the high frequency signal back to the
(-) input is usually a better approach - assuming it works.
The late J.L.L. Hood used that form of compensation in his mosfet amp. He says (in his book, "valve and transistor audio amplifiers"), that it is better than miller comp at avoiding TIM, and is better with reactive loads. Square waves are reproduced better too. He had a resistor in series with the cap though, and chose the values appropriately for the phase shifts of the amp.
I've also used AC feedback from the VAS stage several times with good results. As a starter, I set the capacitor to break against the global feedback resistor connected at the output at about 100kHz or so, with a resistor to introduce a zero at 10x this frequency. I then verify performance with square wave response (keeping the amp out of saturation) and/or doing a gain/phase plot with the analyzer at my work place. I haven't had an amp oscillate in a long time, and even then it was inner loop oscillation in the ouput compound pairs (don't use them any more).
I am running a power amp prototype which likewise uses no miller caps.
The general topology is a cascoded JFET input differential into a summing current mirror, a cascoded VAS with current-sink load, and a compound output stage (with an additional distortion-cancelling mechanism built-in), biased to about 200mA per device. Current limiting and shut-down safety circuitry are also included.
The power supply is based on a new low-noise switching circuit (patent applied for). Size (for a 500W power supply) is perhaps 8x16cmx8cm.
The compensation consists of an RC network at the input of the VAS, a shunt cap from the VAS output to the negative input node, and another RC network at the input of the Sziklai output stage (negative side only). There is an output RC (Zobel) network, but no output series inductor/resistor network.
It is completely stable, and despite that this is a Class AB design and runs on a switching power supply, it sounds very good. We've compared this against various power amps (including a Halcro and Accuphase A-60), and the sound was judged to be at least comparable overall, and superior in areas such as smoothness and low-level timbral and dynamic resolution. OTOH, when it comes to a total sense of ease and unlimited headroom into demanding speaker loads, the Halcro still has the edge.
hth, jonathan carr
The general topology is a cascoded JFET input differential into a summing current mirror, a cascoded VAS with current-sink load, and a compound output stage (with an additional distortion-cancelling mechanism built-in), biased to about 200mA per device. Current limiting and shut-down safety circuitry are also included.
The power supply is based on a new low-noise switching circuit (patent applied for). Size (for a 500W power supply) is perhaps 8x16cmx8cm.
The compensation consists of an RC network at the input of the VAS, a shunt cap from the VAS output to the negative input node, and another RC network at the input of the Sziklai output stage (negative side only). There is an output RC (Zobel) network, but no output series inductor/resistor network.
It is completely stable, and despite that this is a Class AB design and runs on a switching power supply, it sounds very good. We've compared this against various power amps (including a Halcro and Accuphase A-60), and the sound was judged to be at least comparable overall, and superior in areas such as smoothness and low-level timbral and dynamic resolution. OTOH, when it comes to a total sense of ease and unlimited headroom into demanding speaker loads, the Halcro still has the edge.
hth, jonathan carr
This type of compensation does work well, but you are taking feedback around the amp before the output stages - so distortion at high frequencies goes up because of crossover. Agree that simulation at high frequencies is probably not accurate with the generic models available for most spice work, but it still provides an indication. You will hardly notice the difference at 1KHz, but at 20KHz there is a significant increase in distortion (cross over). I only use VAS to -ve input feedback if I need to gain a little phase margin - but not as a main compensation scheme.
rgds
rgds
ACR said:
With the compensation I explaned and without main NFB loop there is as much (or actually more) gain and phase margin left for NFB than there is with miller compensation.
The raw gain at 20kHz is >100db and after this compensation there is still 50db left. The other 50db what was used for compensation, linearises the output of VAS and lowers the output impedance of VAS too. Unfortunately the phase margin is coming a bit worse however the phase response if flat -90deg.This is similar to miller compensation except by miller the phase margin runs out too soon.
Please explain why there is a significant increase in distortion (cross over).
A Thought Exercise
I'd like to offer another question as a thought exercise:
Take an an amplifier such as the Citation or Tiger as a simple example and open the loop, no feedback. Front end DC offset will be amplified and not corrected by feedback so let's assume that there's no imbalance and the output has a 0V quiescient point.
Remove all feedback, compensation, all caps except for the power supply caps and assume they're ideal.
If we sweep frequency and measure distortion with an 8 ohm load, let's say at 10W avoiding the crossover issue, there's a significant increase at 20 kHz as compared to 1 kHz.
Why? What is the most significant cause?
What would be required but perhaps not physically realizable to have no rise in distortion from 1 kHz on up?
I'd like to offer another question as a thought exercise:
Take an an amplifier such as the Citation or Tiger as a simple example and open the loop, no feedback. Front end DC offset will be amplified and not corrected by feedback so let's assume that there's no imbalance and the output has a 0V quiescient point.
Remove all feedback, compensation, all caps except for the power supply caps and assume they're ideal.
If we sweep frequency and measure distortion with an 8 ohm load, let's say at 10W avoiding the crossover issue, there's a significant increase at 20 kHz as compared to 1 kHz.
Why? What is the most significant cause?
What would be required but perhaps not physically realizable to have no rise in distortion from 1 kHz on up?
Hi,
JLH went further, stating in his various articles and book that the VAS to inverting input feedback loop would sound better even though he accepted that distortion was not as low as Miller Comp. Basically his major claim was improved sound quality.
He also showed the location of about 4 other feedback mechanisms that seemed to be required to optimise the stability and response characteristics. I think this indicates that it is more difficult to set up than Miller comp. I have seen it stated by a number of authors that Miller comp is a simple way to ensure stability and reading between the lines I think they are suggesting that for DIY this gives a satisfactory amp that swaps qualtiy for ease of reproducability.
Finally a large improvement in slew rate and avoidance of transient distortion results from avoiding Miller comp. The Miller method is limited by the current ability of the LTP to supply or source the VAS cap. Again this is supported by a number of authors.
regards Andrew T.
JLH went further, stating in his various articles and book that the VAS to inverting input feedback loop would sound better even though he accepted that distortion was not as low as Miller Comp. Basically his major claim was improved sound quality.
He also showed the location of about 4 other feedback mechanisms that seemed to be required to optimise the stability and response characteristics. I think this indicates that it is more difficult to set up than Miller comp. I have seen it stated by a number of authors that Miller comp is a simple way to ensure stability and reading between the lines I think they are suggesting that for DIY this gives a satisfactory amp that swaps qualtiy for ease of reproducability.
Finally a large improvement in slew rate and avoidance of transient distortion results from avoiding Miller comp. The Miller method is limited by the current ability of the LTP to supply or source the VAS cap. Again this is supported by a number of authors.
regards Andrew T.
Electron, what I meant is that the crossover distortion in the output is not corrected as well as if you had NOT taken the feedback from the VAS collecter to the -ve input.
I have to add that in the designs I have been working on I only have 15 to 20dB feedback at 20Khz - so not the 50dB you are working with around the total loop at 20KHz.
rgds
I have to add that in the designs I have been working on I only have 15 to 20dB feedback at 20Khz - so not the 50dB you are working with around the total loop at 20KHz.
rgds
I think that this compensation scheme is better for the Hitachi/Maplin MOSFET amp discussed in this thread, than the Miller type. With a differential VAS, the two halves are loaded by different impedences, so Maplin's version doesn't seem right. Hitachi use one miller cap, but this method would be better, I think.
Food for thought
Backing up with this question, how about we start with why the frequency response rolls off as we go up in frequency, what is the major cause of loss of voltage gain?
Same conditions as below:
I'd like to offer another question as a thought exercise:
Take an an amplifier such as the Citation or Tiger as a simple example and open the loop, no feedback. Front end DC offset will be amplified and not corrected by feedback so let's assume that there's no imbalance and the output has a 0V quiescient point.
Remove all feedback, compensation, all caps except for the power supply caps and assume they're ideal.
If we sweep frequency and measure distortion with an 8 ohm load, let's say at 10W avoiding the crossover issue, there's a significant increase at 20 kHz as compared to 1 kHz.
Why? What is the most significant cause?
What would be required but perhaps not physically realizable to have no rise in distortion from 1 kHz on up?
Backing up with this question, how about we start with why the frequency response rolls off as we go up in frequency, what is the major cause of loss of voltage gain?
Same conditions as below:
I'd like to offer another question as a thought exercise:
Take an an amplifier such as the Citation or Tiger as a simple example and open the loop, no feedback. Front end DC offset will be amplified and not corrected by feedback so let's assume that there's no imbalance and the output has a 0V quiescient point.
Remove all feedback, compensation, all caps except for the power supply caps and assume they're ideal.
If we sweep frequency and measure distortion with an 8 ohm load, let's say at 10W avoiding the crossover issue, there's a significant increase at 20 kHz as compared to 1 kHz.
Why? What is the most significant cause?
What would be required but perhaps not physically realizable to have no rise in distortion from 1 kHz on up?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.