Ok THD20. 
This amplifier starts to clip at +/- 17V.
THD20K @ +/-16V output into 8R = 0.000579%
THD20K @ +/- 2.4V output into 8R = 0.000146%
This amplifier has never really been designed for low THD. It has been the goal to maximize loop gain while keeping good stability margins while hoping THD took care of itself.
Paul
This amplifier starts to clip at +/- 17V.
THD20K @ +/-16V output into 8R = 0.000579%
THD20K @ +/- 2.4V output into 8R = 0.000146%
This amplifier has never really been designed for low THD. It has been the goal to maximize loop gain while keeping good stability margins while hoping THD took care of itself.
Paul
Last edited:
Yes, there is an error in Cordell's book and he acknowledged that. Search in the Cordell book thread. I had a problem with this, made tens PCBs with that error on them.
Redesign...
Next update...
The major change here is the removal of diode strings with them being replaced by vbe multipliers. This gives another variable to play with. Should be able to get the VAS current to perfectly match the IPS tail current. And may be able to use the vbe multiplier to temperature compensate the VAS? Need to think about that one...
Paul
Next update...
The major change here is the removal of diode strings with them being replaced by vbe multipliers. This gives another variable to play with. Should be able to get the VAS current to perfectly match the IPS tail current. And may be able to use the vbe multiplier to temperature compensate the VAS? Need to think about that one...
Paul
Attachments
Redesign 2
The evolution continues...
This is the next experiment. Have tried to remove complexity where possible and to simplify things within the amplifier itself. Have removed some complexity from the front end and moved it to the front end PSU. As long as the front end rails are stable, minimal ripple and at defined voltages then I don't see any penalty.
A minimalist DC servo has been added. It's not strictly necessary as the first version of this amplifier (DC coupled) has been playing at work for weeks and the offset problems have not been an issue.
The most interesting part is the stability plots. Through experimentation with worst case loads have found that GM is far more important than PM for this amplifier. The plots look sub optimal at 0V output into a resistive load. And that the VAS loop is the critical one that requires massive GM. Maybe its to do with the bootstrapped shunts....?
Attached is the schematic and some loop gain plots. The first two plots are of the output stage loop and the VAS loop at 0V output into an 8R load. looks sub optimal in that the it looks like more PM can be had through optimization. This is the wrong way to go. If you optimize this way when the amp is presented with a pure reactive load the VAS loop goes unstable. Optimize for GM and things improve into pure reactive load. See the next two loop gain plots for OPS loop and VAS loop. These are into pure 100nF (no resistive element series or parallel).
Loop Vx (VAS) into 8R with 0v amp output: PM = 88 degrees Gm = 41dB
Loop Vz0 (OPS) into 8R with 0v amp output: PM = 60 degrees Gm = 23dB
Loop Vx into 100nF with 0v amp output: PM = 54 degrees Gm = 0.5dB
Loop Vz0 into 100nF with 0v amp output: PM = 50 degrees Gm = 0.4dB
Need to try and understand why GM is so important to this amp.
And for the THD20: 0.4V from clipping into 8R = 8ppm
Thank you for reading the ramblings of a madman...
Paul
The evolution continues...
This is the next experiment. Have tried to remove complexity where possible and to simplify things within the amplifier itself. Have removed some complexity from the front end and moved it to the front end PSU. As long as the front end rails are stable, minimal ripple and at defined voltages then I don't see any penalty.
A minimalist DC servo has been added. It's not strictly necessary as the first version of this amplifier (DC coupled) has been playing at work for weeks and the offset problems have not been an issue.
The most interesting part is the stability plots. Through experimentation with worst case loads have found that GM is far more important than PM for this amplifier. The plots look sub optimal at 0V output into a resistive load. And that the VAS loop is the critical one that requires massive GM. Maybe its to do with the bootstrapped shunts....?
Attached is the schematic and some loop gain plots. The first two plots are of the output stage loop and the VAS loop at 0V output into an 8R load. looks sub optimal in that the it looks like more PM can be had through optimization. This is the wrong way to go. If you optimize this way when the amp is presented with a pure reactive load the VAS loop goes unstable. Optimize for GM and things improve into pure reactive load. See the next two loop gain plots for OPS loop and VAS loop. These are into pure 100nF (no resistive element series or parallel).
Loop Vx (VAS) into 8R with 0v amp output: PM = 88 degrees Gm = 41dB
Loop Vz0 (OPS) into 8R with 0v amp output: PM = 60 degrees Gm = 23dB
Loop Vx into 100nF with 0v amp output: PM = 54 degrees Gm = 0.5dB
Loop Vz0 into 100nF with 0v amp output: PM = 50 degrees Gm = 0.4dB
Need to try and understand why GM is so important to this amp.
And for the THD20: 0.4V from clipping into 8R = 8ppm
Thank you for reading the ramblings of a madman...
Paul
Attachments
Hi David,
The first prototype has impressed those at work. It does make a naff speaker sound good. Development continues...
Normally, would have an output inductor and second zobel. This "testing" in LTSpice was supposed to be unrealistically harsh. With a 1u inductor in place stability margins are better but that's cheating isn't it?
Stability is paramount in this design so the output inductor is more of a safety belt than a necessity. Not chasing THD. It is already low enough that other external (parasitic) factors are going to be more of an issue. For example, PCB layout, component matching etc.
Feel your pain with computer issues. I refused to fully rebuild mine after 2 major crashes in the space of a month. But being a computer it's fine now...
Look forward to your comments.
Paul
Hi Damir,
Is this any better?
Paul
P.S. Don't blame you for losing interest.
edit: Image quality isn't the best. Think I need to make the schematic a little smaller in LTSpice.
Hi Paul,
And I lose interest when seeing those horrible LTSpice schematics, haha
(by the way, those from Damir are the worst I've ever seen, sorry Damir). The first thing you could do is customizing the appearance, for example changing the default setting from a gray background to a white one and turning off those distracting grid pixels.
Apart from these minor issues, please go on with your good and interesting work.
Cheers, E.
Hi Paul,
And I lose interest when seeing those horrible LTSpice schematics, haha
The first thing you could do is customizing the appearance, for example changing the default setting from a gray background to a white one and turning off those distracting grid pixels.
Apart from these minor issues, please go on with your good and interesting work.
Cheers, E.
Hi Edmond,
Yes I agree, but I made progress. What's wrong with this LTspice schematic? http://www.diyaudio.com/forums/solid-state/243481-200w-mosfet-cfa-amp-74.html#post4278243 .Maybe a bit to much information on it.
BR Damir
Hi Damir,
Indeed, that looks much better, though still some lines and/or symbols are rather cramped together. Also there are several overlaps between text components and lines.
On the other hand, I do know it's a lot of work (just as any other art) to make it "ready for the press".
Cheers, E.
Hi Edmond,
For final schematic I use ExpressSCH not so good schematic software but free and result you can see here http://www.diyaudio.com/forums/solid-state/243481-200w-mosfet-cfa-amp-77.html#post4307021.
But for this one the same complain is valid I did to Paul's schematic, to big.
My apology Paul for stilling your thread
BR Damir
Hi Edmond,
Glad you are enjoying the thread. It's all based around your idea. Hopefully, I'll get the schematic presentation sorted.
Hi Damir,
You haven't stolen my thread at all. It's all relevant stuff. And if two people get going on their own theme, that's fine by me
Here I present two more attempts. Tell me what you think.
Many thanks,
Paul
Glad you are enjoying the thread. It's all based around your idea.
Hopefully, I'll get the schematic presentation sorted.Hi Damir,
You haven't stolen my thread at all. It's all relevant stuff. And if two people get going on their own theme, that's fine by me
Here I present two more attempts. Tell me what you think.
Many thanks,
Paul
Attachments
I don't think an output inductor is a cheat.
It only ensures a predictable load far above the audio spectrum, in reality there's speaker and cable inductance, so the Spice 8R plus capacitance load is a wildly inaccurate idealization anyway.
As usual, I find your compensation scheme difficult to analyse
In my own amp I have found that I can use the inevitable load of the TPC components (your C9,C8,R466) to eliminate the need for a shunt network (C16,R45).
I am not sure if it's theoretically better but it shouldn't be worse and it saves a few components, simplifies layout, more compact board etc.
But my main difficulty is C3,C4,R39 and the interaction with R21,R20.
This kind of delta network between amp sections has the potential to create multi-pole/zero compensation, a la TMC, but also to decrease performance from unintended feed-forward and RHP zeros.
Personally I don't like to use compensation I don't fully understand.
I will think some more, I have made some advances in nested loop theory.
In the mean time I would recommend you check it carefully.
Best wishes
David
Off topic.
The computer problem is that AVG antivirus has wrecked my system.
One of the AVG drivers stops the system boot so I can't even run the system to remove AVG.
I recall that Stuart disliked AVG and I was in accord with that, now I find "dislike" inadequate.
It'd say it's "shyte", my new word for bad software.
Also in accord with Stuart on removal of Grid for posts, it's just clutter.
PS
I wrote the previous comments before I read to the end of the thread.
Looks like you have already discovered my concerns.
It only ensures a predictable load far above the audio spectrum, in reality there's speaker and cable inductance, so the Spice 8R plus capacitance load is a wildly inaccurate idealization anyway.
As usual, I find your compensation scheme difficult to analyse
In my own amp I have found that I can use the inevitable load of the TPC components (your C9,C8,R466) to eliminate the need for a shunt network (C16,R45).
I am not sure if it's theoretically better but it shouldn't be worse and it saves a few components, simplifies layout, more compact board etc.
But my main difficulty is C3,C4,R39 and the interaction with R21,R20.
This kind of delta network between amp sections has the potential to create multi-pole/zero compensation, a la TMC, but also to decrease performance from unintended feed-forward and RHP zeros.
Personally I don't like to use compensation I don't fully understand.
I will think some more, I have made some advances in nested loop theory.
In the mean time I would recommend you check it carefully.
Best wishes
David
Off topic.
The computer problem is that AVG antivirus has wrecked my system.
One of the AVG drivers stops the system boot so I can't even run the system to remove AVG.
I recall that Stuart disliked AVG and I was in accord with that, now I find "dislike" inadequate.
It'd say it's "shyte", my new word for bad software.
Also in accord with Stuart on removal of Grid for posts, it's just clutter.
PS
I wrote the previous comments before I read to the end of the thread.
Looks like you have already discovered my concerns.
Last edited:
Another hasty post on a computer that's not mine.
Paul, I suspect that the shunt point you have probed is not indicative of a real stability problem. It isn't a "controlled source", to use Bode's nomenclature.
I had a bit of a discussion about this with Ric Lee and others on one of Damir's threads where he had a similar issue.
It is non minimum phase (due to multiple paths) and that usually means reduced stability but it doesn't seem to matter for nodes that are not sources.
So I would not be too concerned but I still don't understand it fully, ideas appreciated from anyone, as always.
Edmond, sorry that I typed "Stuart" rather than your proper name, I was a bit rushed.
Best wishes
David
Paul, I suspect that the shunt point you have probed is not indicative of a real stability problem. It isn't a "controlled source", to use Bode's nomenclature.
I had a bit of a discussion about this with Ric Lee and others on one of Damir's threads where he had a similar issue.
It is non minimum phase (due to multiple paths) and that usually means reduced stability but it doesn't seem to matter for nodes that are not sources.
So I would not be too concerned but I still don't understand it fully, ideas appreciated from anyone, as always.
Edmond, sorry that I typed "Stuart" rather than your proper name, I was a bit rushed.
Best wishes
David
AVG
/Off topic
>dislike ? Not just that, I'm through wit AVG!
It seems that this piece of "shyte" has erroneously removed one of your vital system files. I hope you have a rescue/backup CD or memory stick to repair your PC.
Now I'm using NOD32 (very affordable) and 'Macrium Reflect Free' for backups.
Cheers, E.
/Off topic
Hi David,
>dislike ? Not just that, I'm through wit AVG!
It seems that this piece of "shyte" has erroneously removed one of your vital system files. I hope you have a rescue/backup CD or memory stick to repair your PC.
Now I'm using NOD32 (very affordable) and 'Macrium Reflect Free' for backups.
No problem.
Cheers, E.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.