I'm more than 20% overcompensated on all my IPS's . If you raise the
unity gain point on a typical VFA you will increase usable NFB at a given
frequency.
In layman's terms ... reduce Cdom by 20%.
BUT ! ..... you might not have enough margin at unity gain by this point.
A small lead capacitance across the feedback R (2-5pF) extends the
phase "hump" to a slightly higher frequency.
The CFA's don't have this problem ... they have full gain available for
NFB right up to 10+ khz and usually have 12-15db "extra" gain at even
20-25khz !
PS ..002% is not really bad in the "real world" (better than most OEM's).
EDIT - the CFA's should be just as quiet as any of the VFA's.
Another EDIT - at 10K+ , just a slight OPS bias change will drop THD from 20+ to 5 ppm !
OS
Last edited:
Well I know that 0.002% on the HD2 and HD3 isn't anywhere close to being bad, I was just wondering why the simulation is so different from the real world. I am not really interested in altering the compensation as I'd rather have properly stable amplifier.
By the sound of things though, it would appear that I don't have the OPS biased correctly. I adjusted it with respect to what Self recommends, but setting the bias correctly for the EF3 is quite difficult because there's a decent amount of thermal lag involved as the main heatsink needs to equilibriate every time you make a bias adjustment, which of course then alters the point at which you set it to previously.
I'll have a play with the bias setting when viewing higher frequencies on the spectrum analyser. All I did before was take a peak at the 1kHz spectrum and watched as the switching products disappeared into the noise. No matter what I try however, I can never seem to get spikes to appear again as a result of GM doubling.
Maybe all I need to do is observe the THD at higher frequencies to set the bias properly. I did notice that when doing say 1/10th of watt power tests that the THD fell completely off the chart.
By the sound of things though, it would appear that I don't have the OPS biased correctly. I adjusted it with respect to what Self recommends, but setting the bias correctly for the EF3 is quite difficult because there's a decent amount of thermal lag involved as the main heatsink needs to equilibriate every time you make a bias adjustment, which of course then alters the point at which you set it to previously.
I'll have a play with the bias setting when viewing higher frequencies on the spectrum analyser. All I did before was take a peak at the 1kHz spectrum and watched as the switching products disappeared into the noise. No matter what I try however, I can never seem to get spikes to appear again as a result of GM doubling.
Maybe all I need to do is observe the THD at higher frequencies to set the bias properly. I did notice that when doing say 1/10th of watt power tests that the THD fell completely off the chart.
That was the point of my statement. The VAS (and OPS) is sub PPM in class A , 90+
% of THD is at the OPS (when in class B).
For some reason , the VFA's are very sensitive to a "window" of OPS
biases (50-75ma per device = single digit PPM).
The CFA's seem to tolerate a much greater range of bias settings. I
believe this has to do with the "extra" loop gain available for NFB ...
(especially at HF).
I also ran some old Sankens on one of my "pre-badger" blameless's ... 35ma was the
"sweet spot" (device dependent optimal bias points).
OS
Last edited:
Revelation !!
Yes ,
CFA-X is SO fast and has SO much extra FB gain that even with
gross X-over distortion , it can actually correct the class AB artifacts
with very little total THD gain.
25ppm @ 20k - 75ma bias
35ppm @ 0 OPS bias.
VFA's will fare far worse under this test.
Of course , we should bias correctly ... why "cripple" a good thing ?
PS - this also may be why the CFA "fanclub" exists ... with class AB at ANY
level - the speed/NFB may be able to actually "fake" class A performance/sound Q ?
OS
Yes ,
CFA-X is SO fast and has SO much extra FB gain that even with
gross X-over distortion , it can actually correct the class AB artifacts
with very little total THD gain.
25ppm @ 20k - 75ma bias
35ppm @ 0 OPS bias.
VFA's will fare far worse under this test.
Of course , we should bias correctly ... why "cripple" a good thing ?
PS - this also may be why the CFA "fanclub" exists ... with class AB at ANY
level - the speed/NFB may be able to actually "fake" class A performance/sound Q ?
OS
Last edited:
5th element
My 3 cents about EF3 biasing. I have predrivers biassed at about 5mA without heatsink above the pcb and drivers with o/p tranies at main heatsink. The bias of output tranies is approx 10-15% overcompensated.
The amp is not the same as in this topic but I may help.
What I found about CFA - the top octave is more delicate, sometimes bit hidden but very nice presented.
On the measurements (real prototypes) - there is allways some knee where thd is rising in VFA (2-5kHz), CFA has rather flat thd threw all audio freq spectrum.
My 3 cents about EF3 biasing. I have predrivers biassed at about 5mA without heatsink above the pcb and drivers with o/p tranies at main heatsink. The bias of output tranies is approx 10-15% overcompensated.
The amp is not the same as in this topic but I may help.
What I found about CFA - the top octave is more delicate, sometimes bit hidden but very nice presented.
On the measurements (real prototypes) - there is allways some knee where thd is rising in VFA (2-5kHz), CFA has rather flat thd threw all audio freq spectrum.
Put the thermal sensor ( TO-126 case or better TO-92, or small diode) right on the top of one output transistors (for EF output configuration), directly above transistor chip, here is thermal transfer delay shortest (thermal capacity needed to "charge" via thermal resistance chip to case), thermal tracking is the best, very near to thermal track transistors . If you are perfectionist, use two sensors connected in series, each sensor on one output transistor for each polarity.
This originated in (usually) higher OLG in VFA amps at lowest frequencies. As loop gain is decreasing (with frequency and needed compensation to hold it stable), so distortion is increasing. But it is no disadvantage, if distortion at higher frequencies is the same or even lower, than compared to CFA "rival", and at lower frequncies is VFA THD far better.
Need to hold constant THD across audio band is only one of many dogmas. It is no problem hold in VFA loop gain (and so resulting THD will be the same way flat as in CFA amp) constant even to 30-50kHz, but it is possible only for price worsening linearity at lower frequencies, simply to throw away usable loop gain in this region. It is very simple to load VAS with resistor to ground (very poor solution, loading VAS worsens basic linearity), or bypass Cdom (much, much better, local NFB in VAS ) in VAS with about from 330k to 1mega resistor .
Revelation!
P.S. "fakeing" is actally feedback's ability for correction with error signal in a real time. High harmonic distortions needs fast and accurate correction system
Imagine what CFA can do with the normal signal if it can do that with 0 OPS bias crossover distortions. I told you months ago about CFA's preserving original signal, maybe now you have more of a clue of what I thought.Yes ,
CFA-X is SO fast and has SO much extra FB gain that even with
gross X-over distortion , it can actually correct the class AB artifacts
with very little total THD gain.
25ppm @ 20k - 75ma bias
35ppm @ 0 OPS bias.
VFA's will fare far worse under this test.
Of course , we should bias correctly ... why "cripple" a good thing ?
PS - this also may be why the CFA "fanclub" exists ... with class AB at ANY
level - the speed/NFB may be able to actually "fake" class A performance/sound Q ?
OS
P.S. "fakeing" is actally feedback's ability for correction with error signal in a real time. High harmonic distortions needs fast and accurate correction system
How can one mimic Thermaltrack BJT?
Well by using SMD BJT as a temp sensor by glueing it to the central pin of a power transistor, to the point where it comes out from a plastic case. There's usually wider area of a copper pin just of the right size to accomodate SMD BJT. Copper pin actually serves as the best temperature transferring media from a die to the outside world.
Well by using SMD BJT as a temp sensor by glueing it to the central pin of a power transistor, to the point where it comes out from a plastic case. There's usually wider area of a copper pin just of the right size to accomodate SMD BJT. Copper pin actually serves as the best temperature transferring media from a die to the outside world.
i used it several years ago to make a opamp multiloop line stage to experiment on, just to see how their process worked.
it was fine, a little expensive. the tool was easy to learn/use.
mlloyd1
it was fine, a little expensive. the tool was easy to learn/use.
mlloyd1
Improving the bias transistor a little.
I super glued a tiny SMD device into one of the 'hollowed' out parts of the transistor case.
This works far far better than the original method, but then again I don't know if in practice that it really matters. It's certainly helpful when setting the amplifier up, tweaking and measuring though.
I had a bit more of a play around and it became obvious that I was suffering from some induction based distortion and moving some of the internal flying leads around helped reduce this to the extent that the amplifier is now less load dependent and the distortion is lower overall.
Optimising the wiring is clearly needed on a 'per amplifier' basis to maximise performance. Although distortion levels, even with the wiring neglected somewhat, are still very low by any standard.
Here's the 2.83V into 9.4 ohm load again. And like last time ignore the spikes as they are a measurement issue.
Pretty much noise really. Distortionless one could say?
Now up is 12.5V into the 9.4 ohm resistor, which represents what one could call half the max RMS output swing.
Still very low and clearly the output stage is now NOT in class A, which is was previously I think.
Here is the same but into 4.7 ohms instead.
Is pretty much identical.
Now 25V RMS into 9.4 ohms, which is just a hair before clipping occurs.
Again distortion increases a little, which is to be expected.
And into 4.7 ohms.
As you can see, the amplifier seems pretty load invariant. It wasn't before. If any of the distortion mechanisms that rely on the injection of half sine-wave pulses into the sensitive areas of the amplifier were being contaminated in any way then I would expect the distortion to jump up when going from 9.4 to 4.7 ohms, here there's basically no change at all.
I don't think there's really much more to be squeezed out of the amplifier, unless one wanted to risk stability in some way (or go for higher power). The design now seems pretty free from layout and implementation issues and is certainly, I think, what one could call state of the art.
I super glued a tiny SMD device into one of the 'hollowed' out parts of the transistor case.
This works far far better than the original method, but then again I don't know if in practice that it really matters. It's certainly helpful when setting the amplifier up, tweaking and measuring though.
I had a bit more of a play around and it became obvious that I was suffering from some induction based distortion and moving some of the internal flying leads around helped reduce this to the extent that the amplifier is now less load dependent and the distortion is lower overall.
Optimising the wiring is clearly needed on a 'per amplifier' basis to maximise performance. Although distortion levels, even with the wiring neglected somewhat, are still very low by any standard.
Here's the 2.83V into 9.4 ohm load again. And like last time ignore the spikes as they are a measurement issue.
Pretty much noise really. Distortionless one could say?
Now up is 12.5V into the 9.4 ohm resistor, which represents what one could call half the max RMS output swing.
Still very low and clearly the output stage is now NOT in class A, which is was previously I think.
Here is the same but into 4.7 ohms instead.
Is pretty much identical.
Now 25V RMS into 9.4 ohms, which is just a hair before clipping occurs.
Again distortion increases a little, which is to be expected.
And into 4.7 ohms.
As you can see, the amplifier seems pretty load invariant. It wasn't before. If any of the distortion mechanisms that rely on the injection of half sine-wave pulses into the sensitive areas of the amplifier were being contaminated in any way then I would expect the distortion to jump up when going from 9.4 to 4.7 ohms, here there's basically no change at all.
I don't think there's really much more to be squeezed out of the amplifier, unless one wanted to risk stability in some way (or go for higher power). The design now seems pretty free from layout and implementation issues and is certainly, I think, what one could call state of the art.
Attachments
Okay so so much for calling this state of the art, there is clearly something *else* going on here that is reducing linearity a little.
Here is a distortion sweep of the amplifier doing 12.5VRMS into no load whatsoever.
That clearly isn't right as output stage distortion should be pretty much zero without any load attached, ie the amp should be running in class A with basically no extra current flowing through the OPS.
From memory input stage distortion doesn't depend on their being a load connected as it is loaded by the caps in the compensation network. Then there's the feedback network and compensation network itself. These don't depend on the load and if any component within is acting up it could cause something like this to show up.
There is still work to be done it seems.
Here is a distortion sweep of the amplifier doing 12.5VRMS into no load whatsoever.
That clearly isn't right as output stage distortion should be pretty much zero without any load attached, ie the amp should be running in class A with basically no extra current flowing through the OPS.
From memory input stage distortion doesn't depend on their being a load connected as it is loaded by the caps in the compensation network. Then there's the feedback network and compensation network itself. These don't depend on the load and if any component within is acting up it could cause something like this to show up.
There is still work to be done it seems.
Attachments
5th,
What had you choosing that position on the transistor to mount your smd device? I don't know the layout of the die inside the transistor case but would have suspected that the center recess would be closer to the actual junction point. Does anybody know where the ideal mounting position would be to actually use this type of temp feedback controlling the bias voltage?
What had you choosing that position on the transistor to mount your smd device? I don't know the layout of the die inside the transistor case but would have suspected that the center recess would be closer to the actual junction point. Does anybody know where the ideal mounting position would be to actually use this type of temp feedback controlling the bias voltage?
Mainly because the transistor fitted there
No I kid you, that was part of it, but if you look at the power transistor itself, that little recess actually has the back side of the part of the metal 'slug' in it that would normally press against the heatsink. I figured that'd give a pretty direct connection to the die. The bias is now extremely stable.
Before when turned on at cold, the amp would start out at about 20mV, then increase to around 50mV, after which it would decrease again and stabilise around 23mV. This took a while to happen as the main heat sink needed to constantly equilibrate about a changing bias value.
Okay so I'm getting somewhere with the distortion measurements. It appears that the issue is with the resistor divider I am using to cut down the voltage prior to the signal reaching the sound card. I'll have to have a play with it to totally figure it out, but with a small change I've managed to change the 12.5VRMS into the 9.4 ohm load into this.
Okay so the noise floor is increasing as I'm adding in extra outboard attenuation to my box of tricks, but the distortion is way down at the high frequencies.
This needs to be figured out though because I've built some measurement hardware into my microphone pre amp to make measuring amplifiers and loudspeaker impedance etc, a lot easier. This is the first time I've used the amplifier side of it. It's quite possible that some of the surface mount resistors I am using and the DACT stepped attenuator I had left over (it provides variable attenuation) are becoming non linear with the highish voltage they are seeing.
No I kid you, that was part of it, but if you look at the power transistor itself, that little recess actually has the back side of the part of the metal 'slug' in it that would normally press against the heatsink. I figured that'd give a pretty direct connection to the die. The bias is now extremely stable.
Before when turned on at cold, the amp would start out at about 20mV, then increase to around 50mV, after which it would decrease again and stabilise around 23mV. This took a while to happen as the main heat sink needed to constantly equilibrate about a changing bias value.
Okay so I'm getting somewhere with the distortion measurements. It appears that the issue is with the resistor divider I am using to cut down the voltage prior to the signal reaching the sound card. I'll have to have a play with it to totally figure it out, but with a small change I've managed to change the 12.5VRMS into the 9.4 ohm load into this.
Okay so the noise floor is increasing as I'm adding in extra outboard attenuation to my box of tricks, but the distortion is way down at the high frequencies.
This needs to be figured out though because I've built some measurement hardware into my microphone pre amp to make measuring amplifiers and loudspeaker impedance etc, a lot easier. This is the first time I've used the amplifier side of it. It's quite possible that some of the surface mount resistors I am using and the DACT stepped attenuator I had left over (it provides variable attenuation) are becoming non linear with the highish voltage they are seeing.
Attachments
Here is what is able an amplifier with schematic very similar to the one I posted in this thread, 2 pairs SANKEN output transistors, 3EF output configuration (predriver, driver, end stage), Ub+-60V, SR 90V/us. SOAR protection implemented. Very low bias, only 30mA for one pair with Re=0,1R.
Attachments
