With Toshiba JFETs getting rare, a set of matched quads is now approaching 50€.
Similar amount would have to be fetched out for a quad of well matched 2SK1530/2SJ201.
And they will be sold out one day.
Then we are left with all the unmatched devices still to be had.
So the question is, can we tweak the circuit to reduce the even harmonics if we have unmatched devices.
Of course, the most well know solution is the famous P3 from Nelson.
This essentially changes the close-loop gain of the top and bottom half of the F5 (making them unequal deliberately) to cancel out the even harmonics.
The pre-requisite is access to a distortion analyser.
Quite a few years back, in the F5X project, we have investigated another route for the compensation, namely in the open loop.
Rather than making the feedback network asymmetrical, we manipulate the open loop gain of the two halves instead,
by changing the values of the 1st stage drain resistor and the 2nd stage source resistor values.
Attached are two spice files to demonstrate the effectiveness of such an approach.
The first only contains mismatched MOSFETs, while in the second both JFETs and MOSFETs have quite some mismatch.
One can see from the FFT analysis that in both cases, the even harmonics can still be largely compensated for, without causing a notable increase in 3rd harmonics.
Of course it is not incorrect to say that you still need a distortion analyser to make this compensation perfect.
And once you have one, the P3 solution is much easier.
But if you have good device models, you can actually get quite close by simulation.
And it is not too difficult to change the models to tailor to your particular JFETs (Idss) and MOSFETs (Vgs at bias).
The floor is open for discussions.
Patrick
.
Similar amount would have to be fetched out for a quad of well matched 2SK1530/2SJ201.
And they will be sold out one day.
Then we are left with all the unmatched devices still to be had.
So the question is, can we tweak the circuit to reduce the even harmonics if we have unmatched devices.
Of course, the most well know solution is the famous P3 from Nelson.
This essentially changes the close-loop gain of the top and bottom half of the F5 (making them unequal deliberately) to cancel out the even harmonics.
The pre-requisite is access to a distortion analyser.
Quite a few years back, in the F5X project, we have investigated another route for the compensation, namely in the open loop.
Rather than making the feedback network asymmetrical, we manipulate the open loop gain of the two halves instead,
by changing the values of the 1st stage drain resistor and the 2nd stage source resistor values.
Attached are two spice files to demonstrate the effectiveness of such an approach.
The first only contains mismatched MOSFETs, while in the second both JFETs and MOSFETs have quite some mismatch.
One can see from the FFT analysis that in both cases, the even harmonics can still be largely compensated for, without causing a notable increase in 3rd harmonics.
Of course it is not incorrect to say that you still need a distortion analyser to make this compensation perfect.
And once you have one, the P3 solution is much easier.
But if you have good device models, you can actually get quite close by simulation.
And it is not too difficult to change the models to tailor to your particular JFETs (Idss) and MOSFETs (Vgs at bias).
The floor is open for discussions.
Patrick
.
An interesting topic. Shame it has not generated more responses.
I'm afraid I don't have much useful experience in this area so I can't offer any real opinion. That said, I am also interested in how to control the gain of devices that don't match for a different reason, in my case N and P pucks as in the PL XA25. Given that in this application we don't want to employ degeneration, it seems that a different approach is required. So far, the only contender appears to be a the "gm trick" proposed by LHQuam in another thread which is not without its issues.
I'm afraid I don't have much useful experience in this area so I can't offer any real opinion. That said, I am also interested in how to control the gain of devices that don't match for a different reason, in my case N and P pucks as in the PL XA25. Given that in this application we don't want to employ degeneration, it seems that a different approach is required. So far, the only contender appears to be a the "gm trick" proposed by LHQuam in another thread which is not without its issues.
If one looks at the original First Watt F5, using Fairchild MOSFETs, they are not true complementary devices.
Or in other words, they have an intrinsic "mismatch".
https://www.diyaudio.com/forums/att...r-amplifier-toshiba-complementary-mosfets-pdf
https://www.diyaudio.com/forums/att...6759318-f5-power-amplifier-fqa-comparison-pdf
There were already a number of posts in the original F5 thread about this, some 10 years ago.
F5 power amplifier
You can take a more detailed look at posts '1457, 1462~1465, 1470.
In particular, using the MOSFET source resistor to trim 2nd harmonics was already mentioned by NP himself :
"For trim, I simply add parallel resistance to the Source resistor on
either the P or N channel outputs while measuring the distortion at 1
watt, 1 KHz. At some value, the 2nd harmonic essentially disappears,
and there you are.
If your distortion is .01% or so, then trimming will get you down to
.001% to .003%."
In those days with plentiful of Toshiba audio devices, we never bothered.
We have always gone for perfectly matched parts. And we still do today.
But for those who are on a tighter budget or want to challenge themselves technically, trimming is a good way out.
Another good way is to build the F5X and let the two differential phases cancel out the even harmonics for you automatically.
What I want to show here is the principle of how that can be done.
And it is not necessarily limited to the F5 circuit alone.
But then maybe most people just want to solder and listen, and not care about understanding how things work.
Which is of course not too surprising.
Cheers,
Patrick
Or in other words, they have an intrinsic "mismatch".
https://www.diyaudio.com/forums/att...r-amplifier-toshiba-complementary-mosfets-pdf
https://www.diyaudio.com/forums/att...6759318-f5-power-amplifier-fqa-comparison-pdf
There were already a number of posts in the original F5 thread about this, some 10 years ago.
F5 power amplifier
You can take a more detailed look at posts '1457, 1462~1465, 1470.
In particular, using the MOSFET source resistor to trim 2nd harmonics was already mentioned by NP himself :
"For trim, I simply add parallel resistance to the Source resistor on
either the P or N channel outputs while measuring the distortion at 1
watt, 1 KHz. At some value, the 2nd harmonic essentially disappears,
and there you are.
If your distortion is .01% or so, then trimming will get you down to
.001% to .003%."
In those days with plentiful of Toshiba audio devices, we never bothered.
We have always gone for perfectly matched parts. And we still do today.
But for those who are on a tighter budget or want to challenge themselves technically, trimming is a good way out.
Another good way is to build the F5X and let the two differential phases cancel out the even harmonics for you automatically.
What I want to show here is the principle of how that can be done.
And it is not necessarily limited to the F5 circuit alone.
But then maybe most people just want to solder and listen, and not care about understanding how things work.
Which is of course not too surprising.
Cheers,
Patrick
There is some speculation that Pass is intentionally mismatching input jfets on some amps to adjust harmonics...rather than resort to trimming (and increasing parts count). I believe there is some discussion of this in the f7 thread.
But yes, any discussion that helps us work around the lack of toshiba input jfets is certainly worthwhile. Thanks evul.
But yes, any discussion that helps us work around the lack of toshiba input jfets is certainly worthwhile. Thanks evul.
> There is some speculation that Pass is intentionally mismatching input jfets on some amps to adjust harmonics...
> rather than resort to trimming (and increasing parts count).
You are suggesting that a serious manufacturer of high end audio would risk L/R channel mismatch by saving 2x 2W resistor ?
Or that he would use poorly matched devices in his products ?
Patrick
> rather than resort to trimming (and increasing parts count).
You are suggesting that a serious manufacturer of high end audio would risk L/R channel mismatch by saving 2x 2W resistor ?
Or that he would use poorly matched devices in his products ?
Patrick
Iam saying that on some amps like the f7 the jfets may not be perfectly matched but rather selected in a intentionally mismatched pair that gives the desired harmonics. It’s not a random set.
The f7 is an exercise in low parts count, makes sense in that application.
The f7 is an exercise in low parts count, makes sense in that application.
- Status
- Not open for further replies.