Bricolo,
Feedback seems to make for a less than optimum resolution on the highest frequencys. Two examples, Gammut D200 and Aleph 5, both high biased mosfet amps with feedback. Both these amps have a high resolution with a slightly warm character. However listening to triangles, cymbals, steel-strings and cembalo I hear the upper harmonics in a slightly "muddy" way. It´s almost a little like pink noise.
I´m not sure this is all about feedback, it is possible that the rising distortion from about 1k on most mosfet amps, explains the thing. Or.... maybe the rising distortion IS the result of the feedback not being able to correct for the non-linearitys in the highest frequencys when it comes to the mosfet???
However, my amps (LC Audio Zapsolute) are complementary bipolar amps, DC-1MHz without feedback. These amps have a resolution of the above mentioned instruments, that are closer to reality than other amps I have listened to. The distortion does not rise in the high midrange, but are linear up to several hundreds kHz.
/Peter
Feedback seems to make for a less than optimum resolution on the highest frequencys. Two examples, Gammut D200 and Aleph 5, both high biased mosfet amps with feedback. Both these amps have a high resolution with a slightly warm character. However listening to triangles, cymbals, steel-strings and cembalo I hear the upper harmonics in a slightly "muddy" way. It´s almost a little like pink noise.
I´m not sure this is all about feedback, it is possible that the rising distortion from about 1k on most mosfet amps, explains the thing. Or.... maybe the rising distortion IS the result of the feedback not being able to correct for the non-linearitys in the highest frequencys when it comes to the mosfet???
However, my amps (LC Audio Zapsolute) are complementary bipolar amps, DC-1MHz without feedback. These amps have a resolution of the above mentioned instruments, that are closer to reality than other amps I have listened to. The distortion does not rise in the high midrange, but are linear up to several hundreds kHz.
/Peter
Nelson Pass said:
Nelson,
Of course, didn´t think of SOZ, silly me :-9
However, I guess I was thinking on a amp with more "normal" efficiency. Anyway... if the amp would serve as a mid and tweeter amp, do you feel the SOZ is up there with the XA amps?
Do you feel that in your amps, no feedback can result in something that can not be found in the feedback amps?
Hmm, maybe I should take a look at the SOZ with constant current sources. Or do the current sources have a price for raising the efficiency (sonically of course not $ 🙂.
What I am looking for is the ultimate resolution amp for a 90dB mid (and tweeter) in open baffle.
/Peter
I made an X out of the SOZ, remarkable improvement in all aspects. So I don´t think You can compare X-feedback with just ordinary NFB, it is so different.
Blitz said:
The schematic is here http://www.diyaudio.com/forums/showthread.php?s=&threadid=3924&highlight=schematic+explanation
Feedback...ah, now there's a thorny issue.
Think of negative feedback the way you would alcohol. A little is an okay thing, desirable even. But let the NFB demon seduce you and you will find yourself drawn into a never-ending spiral of increasing the gain so that you can increase the feedback so that you can get better specs, and back again.
It's a fools' game. You cannot win.
A reasonable (what's reasonable? that's another post entirely) amount of feedback will lower distortion and increase bandwidth. Add a little more and you'll find yourself with transient instabilities impossible to measure accurately, though they have sonic effects that are not consonant with the way real music sounds. Add a little more and your circuit can become an oscillator. That way madness lies.
Don't get me wrong; don't misquote me. Specifications are a useful thing...up to a point. But there is not a one-to-one correspondence between specs and sound quality. It's a very rapidly diminishing returns effect. Indeed, it doesn't take long before feedback begins to take away more than it gives.
Back in the seventies, there was a drive to see how many zeros you could add after the decimal point in various distortion categories. The results were astounding from a technical standpoint--not so amazing from a sonic point of view. How many '70s vintage solid state pieces can you name that are collectors' items? Got a number in mind? Good. Now subtract the ones that were cool simply because of cosmetics. Done? Okay...assuming that your mathematics hasn't lead you into negative numbers yet, take any survivors of this weeding process and look at the specs/amount of feedback. You will find no classic pieces (in terms of sound quality) with long strings of zeros in the distortion specs. Not one.
Like Prohibition, chasing low distortion figures was an experiment that didn't work. Let's learn from the mistake and go on.
Proponents of high feedback always tend to use a particular catchword--the word is 'Accuracy.' Like religious people who try to stake out the moral high ground by using the word Truth, Accuracy is a semantic attempt to fence in critics who point out that the amplifier no longer sounds like real music. If you let them box you in, you will lose the debate before it has even begun.
The word Accurate, in the hands of a feedback proponent, means an overly etched upper midrange and a treble that leans towards the sizzly side. They insist that this is the way music sounds. Why? Well, it's obvious! If the distortion figures are low, then clearly the signal must be an Accurate reproduction of what came before. Right? The only known antidote to this train of logic is a concert hall. Sorry, rock concerts will not do. You're seeking unamplified music. You must be able to hear the instruments themselves, not an industrial strength stereo version. Once you hear a real ride cymbal, you will no longer be fooled into thinking that the grotesque hissing sound you'd heard before is an Accurate reproduction of the real thing.
Bass reproduction can improve with more feedback. In fact, it can stand more feedback than the mids and highs, which is part of the seduction. If you add another few dB of NFB and the bass sounds better, then you tend to fool yourself into thinking that the high end is getting better, too. Not necessarily so. There's a way to back into this one. Note how many times you've heard of bi-amped systems with solid state on the bottom and tubes for the mids and highs. Now remember that tube amps generally use very, very little feedback, and that solid state amps almost always heap it on. Setting aside questions of the differences in distortion spectra between tubes and solid state, focus on the feedback question. It's a hint that all is not right with high feedback designs if they cannot be trusted with the entire musical signal.
Life is an ongoing process of compromise. Sometimes you have to use feedback; sometimes you end up using more than you might like. Don't let yourself be drawn into saying,"If some is good, then more is better, and too much is just enough!"
Grey
P.S.: Not all circuits that claim to be feedback free, are.
Think of negative feedback the way you would alcohol. A little is an okay thing, desirable even. But let the NFB demon seduce you and you will find yourself drawn into a never-ending spiral of increasing the gain so that you can increase the feedback so that you can get better specs, and back again.
It's a fools' game. You cannot win.
A reasonable (what's reasonable? that's another post entirely) amount of feedback will lower distortion and increase bandwidth. Add a little more and you'll find yourself with transient instabilities impossible to measure accurately, though they have sonic effects that are not consonant with the way real music sounds. Add a little more and your circuit can become an oscillator. That way madness lies.
Don't get me wrong; don't misquote me. Specifications are a useful thing...up to a point. But there is not a one-to-one correspondence between specs and sound quality. It's a very rapidly diminishing returns effect. Indeed, it doesn't take long before feedback begins to take away more than it gives.
Back in the seventies, there was a drive to see how many zeros you could add after the decimal point in various distortion categories. The results were astounding from a technical standpoint--not so amazing from a sonic point of view. How many '70s vintage solid state pieces can you name that are collectors' items? Got a number in mind? Good. Now subtract the ones that were cool simply because of cosmetics. Done? Okay...assuming that your mathematics hasn't lead you into negative numbers yet, take any survivors of this weeding process and look at the specs/amount of feedback. You will find no classic pieces (in terms of sound quality) with long strings of zeros in the distortion specs. Not one.
Like Prohibition, chasing low distortion figures was an experiment that didn't work. Let's learn from the mistake and go on.
Proponents of high feedback always tend to use a particular catchword--the word is 'Accuracy.' Like religious people who try to stake out the moral high ground by using the word Truth, Accuracy is a semantic attempt to fence in critics who point out that the amplifier no longer sounds like real music. If you let them box you in, you will lose the debate before it has even begun.
The word Accurate, in the hands of a feedback proponent, means an overly etched upper midrange and a treble that leans towards the sizzly side. They insist that this is the way music sounds. Why? Well, it's obvious! If the distortion figures are low, then clearly the signal must be an Accurate reproduction of what came before. Right? The only known antidote to this train of logic is a concert hall. Sorry, rock concerts will not do. You're seeking unamplified music. You must be able to hear the instruments themselves, not an industrial strength stereo version. Once you hear a real ride cymbal, you will no longer be fooled into thinking that the grotesque hissing sound you'd heard before is an Accurate reproduction of the real thing.
Bass reproduction can improve with more feedback. In fact, it can stand more feedback than the mids and highs, which is part of the seduction. If you add another few dB of NFB and the bass sounds better, then you tend to fool yourself into thinking that the high end is getting better, too. Not necessarily so. There's a way to back into this one. Note how many times you've heard of bi-amped systems with solid state on the bottom and tubes for the mids and highs. Now remember that tube amps generally use very, very little feedback, and that solid state amps almost always heap it on. Setting aside questions of the differences in distortion spectra between tubes and solid state, focus on the feedback question. It's a hint that all is not right with high feedback designs if they cannot be trusted with the entire musical signal.
Life is an ongoing process of compromise. Sometimes you have to use feedback; sometimes you end up using more than you might like. Don't let yourself be drawn into saying,"If some is good, then more is better, and too much is just enough!"
Grey
P.S.: Not all circuits that claim to be feedback free, are.
Very nice to have you back, Grey.
I would just like to add that the feedback is not the only
issue that has seen the waxing and waning of fashion;
things like slew rate, TIM, damping factor, Class A thru X,
high current, (fill in some more here) come and go.
After all, this is the entertainment industry.
😉
I would just like to add that the feedback is not the only
issue that has seen the waxing and waning of fashion;
things like slew rate, TIM, damping factor, Class A thru X,
high current, (fill in some more here) come and go.
After all, this is the entertainment industry.
😉
Pan said:
Well, they're just different, like Merlot and Cabernet. SOZ
might work for you since you have an open back baffle, but
keep in mind that is has almost no damping factor. If the
drivers have a relatively low Q to their resonance, it would
probably work well IMO.
After months of delay finally I hacked together the AX on the beta board. I guess I failed to do a job good enough with matching, but this is my first attempt.
The following mods where applied to Grey's circuit:
R19, 29 from 68k to 10k, 2 pair of panasonic 3.3uF where used on the hot and cold input, 200k and 5pF on the XFB loop, R23 and 25 are 300ohm, ~120 OHMS as the combination of r24, r26 and v2, 4kohm for r17, r17 to negative rail, lm329 as voltage reference but an irf9610 as the CS mosfet device, r to ground are 100 ohms. 4k7 from the output back to the source of the input differential. I believe that's it. I decided that I didn't want to deal with cooling the zvp3306 at this testing stage.
Initially I fired up the amp with r24 and r26 of 390 and 300 ohms, respectively. Nothing worked. The bias to the output mosfet is largely determined by the voltage across r23 and 25 and regulated by the mpsa18. Initally I had 2.9V across the 300 ohms resistors, nowhere close to the Vgs of 3.7 of the output mosfets. I lowered r24 and r26 to get 4.2V across r23-25 and my test transformer finally started to show signs of fatigue.
The differential irf9610 are not properly matched, the voltage across r23-25 is different by about 200mV. The Vgs was within 10mV so I guess I will have to try to match the transconductance. I bought 15 irfp9610 and I found only 4 pairs that were more than 50 mV apart from one another so I don't know exactly what I am going to do. Maybe I'll adjust the r23 and 25 to compensate.
At warmup I have about 70 mV DC offset which is not too bad, but I have a constant 8V absolute offset
which doesn't go away! The current is not equally shared across the source resistors at all. It looks like the Vgs on the mosfet is different. Note that I gain matched the MPSA18 and all the resistors. I not sure what's going on here. Adjusting the 100k pot in series with r11 appears to be completely ineffective.
My guess is that for this amp one either needs to use parallel devices to average out the effects of mosfet mismatch or use a much lower value of out to gnd resistor to better mask the poor matching or match the devices in the working circuit that sounds hard. 8V are a hell of a lot of volts, though I have no clue where this voltage is coming from.
Back to work. Will report more news later.
The following mods where applied to Grey's circuit:
R19, 29 from 68k to 10k, 2 pair of panasonic 3.3uF where used on the hot and cold input, 200k and 5pF on the XFB loop, R23 and 25 are 300ohm, ~120 OHMS as the combination of r24, r26 and v2, 4kohm for r17, r17 to negative rail, lm329 as voltage reference but an irf9610 as the CS mosfet device, r to ground are 100 ohms. 4k7 from the output back to the source of the input differential. I believe that's it. I decided that I didn't want to deal with cooling the zvp3306 at this testing stage.
Initially I fired up the amp with r24 and r26 of 390 and 300 ohms, respectively. Nothing worked. The bias to the output mosfet is largely determined by the voltage across r23 and 25 and regulated by the mpsa18. Initally I had 2.9V across the 300 ohms resistors, nowhere close to the Vgs of 3.7 of the output mosfets. I lowered r24 and r26 to get 4.2V across r23-25 and my test transformer finally started to show signs of fatigue.
The differential irf9610 are not properly matched, the voltage across r23-25 is different by about 200mV. The Vgs was within 10mV so I guess I will have to try to match the transconductance. I bought 15 irfp9610 and I found only 4 pairs that were more than 50 mV apart from one another so I don't know exactly what I am going to do. Maybe I'll adjust the r23 and 25 to compensate.
At warmup I have about 70 mV DC offset which is not too bad, but I have a constant 8V absolute offset
which doesn't go away! The current is not equally shared across the source resistors at all. It looks like the Vgs on the mosfet is different. Note that I gain matched the MPSA18 and all the resistors. I not sure what's going on here. Adjusting the 100k pot in series with r11 appears to be completely ineffective.My guess is that for this amp one either needs to use parallel devices to average out the effects of mosfet mismatch or use a much lower value of out to gnd resistor to better mask the poor matching or match the devices in the working circuit that sounds hard. 8V are a hell of a lot of volts, though I have no clue where this voltage is coming from.
Back to work. Will report more news later.
Okay...DC offset problems are most likely to be found in the front end current source/load resistor combination.
The values you choose for the resistance to the Source of the current source MOSFET are critical. Since the current source is a direct derivation from the current source Nelson used in the Aleph/Volksamp circuits, think of the combination as a nominal 221 ohm resistor, with a little bit of play to one side or the other. But only a little bit. As was noted back at the beginning of the thread, there's a knife edge to be tread. I only designed in little bit of variation to the current source output. If you stray too far from that, you'll slam the output to one rail or the other. If you have 'only' 8V, you're getting close.
My original values were (for 392 ohm loads):
R24 475 ohms
R26 332 ohms
V2 200 ohms
I later posted a somewhat tighter combination:
R24 332 ohms
R26 562 ohms
V2 200 ohms
The second set of values gives you finer control, but over a narrower range.
In the great scheme of things, the exact values that you choose for those three resistors aren't important, but it is essential that they give you a control range centered on 221 ohms, plus or minus a fairly small percentage. The original group was 221 ohms +- roughly 12%. The second group was roughly +-5%.
Assuming that you're still using a 200 ohm pot for V2, 390 for R24 and 300 for R26 will give you a range between 169 and 219 ohms. 300 ohm resistors for the front end load can be done, but you'll have to be very careful matching everything else to that.
Yes, I found that simple matching of the front end MOSFETs was a good start, but that it takes a bit more fiddling to get the pair matched properly. Not only are you matching them to each other, but to the output MOSFETs. Matching the output MOSFETs across the amp, i.e. left and right halves matched together, will help with this.
If you want, you can make R23 and R25 both adjustable, but I'd recommend using the same triplet layout that I used for the current source, as you don't need much absolute adjustment. In fact, if you want to go that route, you can simply revert to a simple 221 ohm under the current source MOSFET and do all of your adjusting on R23 and R25. That's the way I originally did it, but I got tired of going back and forth between the two sides, so I went to the single adjustment under the current source.
If you're a real glutton for punishment, you can set it up with all three.
The bias/relative offset adjustment (V1 & V3) will only work when everything else is close. That will be the last thing on the to-do list.
100 ohms for the resistors to ground from the outputs should work only if everything else is close. As a temporary measure, go ahead and tag in some 30 ohm resistors until you get everything else settled.
I need to get some numbers posted over in the high-powered Aleph-X thread. I'll try to check back here when I get done with that.
Good luck.
Grey
The values you choose for the resistance to the Source of the current source MOSFET are critical. Since the current source is a direct derivation from the current source Nelson used in the Aleph/Volksamp circuits, think of the combination as a nominal 221 ohm resistor, with a little bit of play to one side or the other. But only a little bit. As was noted back at the beginning of the thread, there's a knife edge to be tread. I only designed in little bit of variation to the current source output. If you stray too far from that, you'll slam the output to one rail or the other. If you have 'only' 8V, you're getting close.
My original values were (for 392 ohm loads):
R24 475 ohms
R26 332 ohms
V2 200 ohms
I later posted a somewhat tighter combination:
R24 332 ohms
R26 562 ohms
V2 200 ohms
The second set of values gives you finer control, but over a narrower range.
In the great scheme of things, the exact values that you choose for those three resistors aren't important, but it is essential that they give you a control range centered on 221 ohms, plus or minus a fairly small percentage. The original group was 221 ohms +- roughly 12%. The second group was roughly +-5%.
Assuming that you're still using a 200 ohm pot for V2, 390 for R24 and 300 for R26 will give you a range between 169 and 219 ohms. 300 ohm resistors for the front end load can be done, but you'll have to be very careful matching everything else to that.
Yes, I found that simple matching of the front end MOSFETs was a good start, but that it takes a bit more fiddling to get the pair matched properly. Not only are you matching them to each other, but to the output MOSFETs. Matching the output MOSFETs across the amp, i.e. left and right halves matched together, will help with this.
If you want, you can make R23 and R25 both adjustable, but I'd recommend using the same triplet layout that I used for the current source, as you don't need much absolute adjustment. In fact, if you want to go that route, you can simply revert to a simple 221 ohm under the current source MOSFET and do all of your adjusting on R23 and R25. That's the way I originally did it, but I got tired of going back and forth between the two sides, so I went to the single adjustment under the current source.
If you're a real glutton for punishment, you can set it up with all three.
The bias/relative offset adjustment (V1 & V3) will only work when everything else is close. That will be the last thing on the to-do list.
100 ohms for the resistors to ground from the outputs should work only if everything else is close. As a temporary measure, go ahead and tag in some 30 ohm resistors until you get everything else settled.
I need to get some numbers posted over in the high-powered Aleph-X thread. I'll try to check back here when I get done with that.
Good luck.
Grey
When everything is matched properly, the absolute offset is adjusted by changing current to input differential. I don't know why but it works very well (with HH CCS) and my absolute offset is set to within 40mV when amp is warm. The differential offset between outputs can be adjusted by fine tuning the bias and in my case is below 3mV steady.
Peter, is the HH's CCS you're talking about the one with a LM329?
How good is it? Have you tested with other ones?
How good is it? Have you tested with other ones?
Ok, I´ll try again..
In one of my earlier posts in this thread I asked what to do if I want to buy a couple of Aleph-X boards, anyone???
/Peter
In one of my earlier posts in this thread I asked what to do if I want to buy a couple of Aleph-X boards, anyone???
/Peter
Hello Peter,
The Rev 1 boards are being fabricated and the current order has been closed. The organizers may have some extras when all is said and done.
I ordered eight, but will not use all of them. I would be happy to send you a set. I think that the final price was around USD$5.00 ea. Shipping would be USD$6.00
Best Regards,
Dale
The Rev 1 boards are being fabricated and the current order has been closed. The organizers may have some extras when all is said and done.
I ordered eight, but will not use all of them. I would be happy to send you a set. I think that the final price was around USD$5.00 ea. Shipping would be USD$6.00
Best Regards,
Dale
Bricolo said:
Yes. We called it HH CCS, because he was the one who post it on a forum (actually he post it, because I asked him about it, after reading the discription of his amp). I don't really care if he designed it or not. You can find it here: http://www.diyaudio.com/forums/showthread.php?s=&postid=70140#post70140
I didn't compare it against the other configurations.
I used 200ohm trimmer for R2, and with it I can adjust the current very nicely. Very slow variations in a current, change the absolute offset by a big margin. Initially, at start up, Absolute offset is around 2-3V and when the amp warms up it goes down to almost 0V.
Dale
Very nice of you, I would be very glad to accept your offer 🙂
Shure we will chat before, otherwise we contact eachother
at the board as soon as the boards begins to ship (board..boards.. haha).
Have a great Sunday!
/Peter
Very nice of you, I would be very glad to accept your offer 🙂
Shure we will chat before, otherwise we contact eachother
at the board as soon as the boards begins to ship (board..boards.. haha).
Have a great Sunday!
/Peter
Grey,
thanks for your reply, what I found most annoying was the total lack of responsiveness of the 100K pot.
To answer your question I now have a 300, and in parallel a 200ohm pot and a 205 ohm in series and the pot is turned to 0 to get up to around 2A of bias per side.
I have no idea where the 8 volts are coming from, although they are not 'Peter-matched' my mosfets are matched within 50mV for sure. Even so I measured a different voltage drop across the .22 ohm resistor the top values are about .38 and the bottom values .48V so my guess is that, as of right now the 8 volt come from the top mosfets not being fully turned on at 4. something volts or by some other weird "X effect", the bottom mosfet are now fully on and dissipating ((15-8)+15)*2=50W. These mosfets are at about 76 C on the case while the top mosfets are at 45 C on the case. If I turn back the current of the differetial the current will drop across ALL the resistors but the difference will remain the same. With 300 ohm loads I have around 15mA/ device on the differential with no ill effects, the heating on the TO-220 CS mosfet seems completely reasonable, I can put my finger on and leave it no problem.
I have setup a proper rig to match the mosfets so I'll try it again today. I would also like to figure the best way to adjust the VGS on the CS output mosfets otherwise my guess is that I won't be able to get around 0 volts on the output even with matched mosfets. In the zen4 Nelson doesn't use any adjustments on the CS mosftets, just fixed 1.5k resistors, maybe I'll try to play with these.
I agree that with 100 ohm to ground everything is more critical BUT it's also a better way to go in the end since it wastes less power. I haven't injected any signal into it yet, but I am wondering if having more voltage drop across the amplifying mosfet allows for larger votage swing.
thanks for your reply, what I found most annoying was the total lack of responsiveness of the 100K pot.
To answer your question I now have a 300, and in parallel a 200ohm pot and a 205 ohm in series and the pot is turned to 0 to get up to around 2A of bias per side.
I have no idea where the 8 volts are coming from, although they are not 'Peter-matched' my mosfets are matched within 50mV for sure. Even so I measured a different voltage drop across the .22 ohm resistor the top values are about .38 and the bottom values .48V so my guess is that, as of right now the 8 volt come from the top mosfets not being fully turned on at 4. something volts or by some other weird "X effect", the bottom mosfet are now fully on and dissipating ((15-8)+15)*2=50W. These mosfets are at about 76 C on the case while the top mosfets are at 45 C on the case. If I turn back the current of the differetial the current will drop across ALL the resistors but the difference will remain the same. With 300 ohm loads I have around 15mA/ device on the differential with no ill effects, the heating on the TO-220 CS mosfet seems completely reasonable, I can put my finger on and leave it no problem.
I have setup a proper rig to match the mosfets so I'll try it again today. I would also like to figure the best way to adjust the VGS on the CS output mosfets otherwise my guess is that I won't be able to get around 0 volts on the output even with matched mosfets. In the zen4 Nelson doesn't use any adjustments on the CS mosftets, just fixed 1.5k resistors, maybe I'll try to play with these.
I agree that with 100 ohm to ground everything is more critical BUT it's also a better way to go in the end since it wastes less power. I haven't injected any signal into it yet, but I am wondering if having more voltage drop across the amplifying mosfet allows for larger votage swing.
Pan,
write to HiFiZen, he's printing out more boards than we need to accomodate late requests. However, I don't know how many more. There is a waiting list though, so be advised.
HiFizen will be back on jan 13, I think. He's checking this forum from time to time.
write to HiFiZen, he's printing out more boards than we need to accomodate late requests. However, I don't know how many more. There is a waiting list though, so be advised.
HiFizen will be back on jan 13, I think. He's checking this forum from time to time.
grataku
Thank you for the info.
Come to think of it, maybe I´ll start a new thread try to organize the shipping to Sweden, if that is not taken care of allready.
It would be a lower cost to shipp the stuff as well as cheaper fees to transfer the money to US.
(edited) grataku, do you happen to have a e-adress to HifiZen?
I tried to send him a PM at this board but don´t know if he received it. Thanks!
/Peter
Thank you for the info.
Come to think of it, maybe I´ll start a new thread try to organize the shipping to Sweden, if that is not taken care of allready.
It would be a lower cost to shipp the stuff as well as cheaper fees to transfer the money to US.
(edited) grataku, do you happen to have a e-adress to HifiZen?
I tried to send him a PM at this board but don´t know if he received it. Thanks!
/Peter
For the uninitiated to the X configuration, like yours truly here is how I got my single channel AX to work.
It took me a couple of days of dicking around and here is what I would have liked to know but nobody told me about. 😉
I will try to be as specific as I can be, and I will be referring to Grey schematic for nomenclature:
First of all I setup a proper rig to test match the mosfet, and matched both the output and differential mosfets. Contrary to Peter findings I got a pair of 9610 with Vgs within 2mV over a 1 minute monitoring period and that was enough to get a corresponding voltage drop on r23 and r25 when I soldered them in place back to back with a silpad in the middle. The Output mosfet could not be matched to better than 10 mV on my sample of 20 I could only find 4 between 3.85-3.86 V Vgs. This translates in about 40mV DC offset. That IMHO is the result of different transconductance. Maybe this large effect would average out if several mosfets would be paralleled resulting in a much lower relative DC offset.
TO FIRST ORDER things that DON'T MATTER AT ALL:
-Value of out to gnd resistance.
I am assuming that some type of value is needed to control the DC offset between + and - out this to work HOWEVER, short of eliminating that completely, playing with values between 20 and 100 ohms made no difference at all in the absolute or relative DC offset. THEREFORE, this resistor might as well be 100 ohm to save some power. It is POSSIBLE that while music is playing the value of the DC will be kept to a more constant value with a lower resistance but as I said this is all to first order.
-V1 and V3 don't appear to affect anything at all. Basically a nominal fixed value can be picked could be 100K for the R1-V1 and R33-V3. Again I am assuming that something here is needed.
Things that matter ALOT:
-Values of R24 to R26 and V2.
R24, 26 and V2 decide how much current is going through the output differential.
R23 and R25 control the threshold at which Q2 and Q11 will start conducting.
-Values of R15 and R32. The voltage drop across these resistor control the conduction of the Q1 and Q10 mosfets. Not as effective as r24, 26 and V2 but somewhat effective to control absolute DC value at the outputs. The reality all these resistances need to be adjusted simultaneously I have a 2k trimpot in right now which is turned all the way up to 2k.
I used a strange CS a hybrid of HH and the standard that used an LM329 instead on the 9.1 zener and a 5k to –15V for r17. The current R24 and 26 are 205 ohms and V2 is 200 ohm. I initially used 301 ohm for r23 and 25 but reverted back to 390 ohms. It should make no difference at all. At one point I was listening to the sound of the input differential biased to 15mA just for kicks. On my test crappy speaker it sounded ok.
I went back to the 100K and 10pf loop (r16 c2) from the 200k and 5 (more XFB?).
Here are some voltages:
PS voltage +/- 13.5 VDC on my test setup. As soon as the VM guy replies to my email I will order the real transformer.
+/- DC ~50-70 mV when playing music with balanced in and 6.6 uF input caps on both hot and cold.
The absolute DC hangs around 0 and +/-35 mV.
Source R drop top 490 mV, bottom 460mV
Vgs top 4.2 V, bottom 4.07V
Differential load drop (r23,25) 4.510 and 4.512V.
At this point it would be nice to figure out how to mess with R9,10, and 13 to increase the bias on Q2 and Q11 thereby evening out the source resistor drops. Since adjusting the CS resistor affects both the lower and upper output mosfets AND r15 and r32 take care of the top bias only a third way of adjusting the bias of only the bottom mosfets could prove somewhat useful. Maybe at that point the resistor to ground would just be there for show.
At this point the only comment about the sound is that it was really audibly distorting when the absolute DC offset was 8V and now it sounds a heck of a lot better. Very promising actually.
Heating is under control, or so it seems it looks like I might have picked the right heatsinks for once! 😉 And yes the rectifier need it's own heatsink.
I will report later on the noises, so far I don't hear any turn on-off noises or hums. I have a high efficiency test speaker that I want to try before I really say.
It took me a couple of days of dicking around and here is what I would have liked to know but nobody told me about. 😉
I will try to be as specific as I can be, and I will be referring to Grey schematic for nomenclature:
First of all I setup a proper rig to test match the mosfet, and matched both the output and differential mosfets. Contrary to Peter findings I got a pair of 9610 with Vgs within 2mV over a 1 minute monitoring period and that was enough to get a corresponding voltage drop on r23 and r25 when I soldered them in place back to back with a silpad in the middle. The Output mosfet could not be matched to better than 10 mV on my sample of 20 I could only find 4 between 3.85-3.86 V Vgs. This translates in about 40mV DC offset. That IMHO is the result of different transconductance. Maybe this large effect would average out if several mosfets would be paralleled resulting in a much lower relative DC offset.
TO FIRST ORDER things that DON'T MATTER AT ALL:
-Value of out to gnd resistance.
I am assuming that some type of value is needed to control the DC offset between + and - out this to work HOWEVER, short of eliminating that completely, playing with values between 20 and 100 ohms made no difference at all in the absolute or relative DC offset. THEREFORE, this resistor might as well be 100 ohm to save some power. It is POSSIBLE that while music is playing the value of the DC will be kept to a more constant value with a lower resistance but as I said this is all to first order.
-V1 and V3 don't appear to affect anything at all. Basically a nominal fixed value can be picked could be 100K for the R1-V1 and R33-V3. Again I am assuming that something here is needed.
Things that matter ALOT:
-Values of R24 to R26 and V2.
R24, 26 and V2 decide how much current is going through the output differential.
R23 and R25 control the threshold at which Q2 and Q11 will start conducting.
-Values of R15 and R32. The voltage drop across these resistor control the conduction of the Q1 and Q10 mosfets. Not as effective as r24, 26 and V2 but somewhat effective to control absolute DC value at the outputs. The reality all these resistances need to be adjusted simultaneously I have a 2k trimpot in right now which is turned all the way up to 2k.
I used a strange CS a hybrid of HH and the standard that used an LM329 instead on the 9.1 zener and a 5k to –15V for r17. The current R24 and 26 are 205 ohms and V2 is 200 ohm. I initially used 301 ohm for r23 and 25 but reverted back to 390 ohms. It should make no difference at all. At one point I was listening to the sound of the input differential biased to 15mA just for kicks. On my test crappy speaker it sounded ok.
I went back to the 100K and 10pf loop (r16 c2) from the 200k and 5 (more XFB?).
Here are some voltages:
PS voltage +/- 13.5 VDC on my test setup. As soon as the VM guy replies to my email I will order the real transformer.
+/- DC ~50-70 mV when playing music with balanced in and 6.6 uF input caps on both hot and cold.
The absolute DC hangs around 0 and +/-35 mV.
Source R drop top 490 mV, bottom 460mV
Vgs top 4.2 V, bottom 4.07V
Differential load drop (r23,25) 4.510 and 4.512V.
At this point it would be nice to figure out how to mess with R9,10, and 13 to increase the bias on Q2 and Q11 thereby evening out the source resistor drops. Since adjusting the CS resistor affects both the lower and upper output mosfets AND r15 and r32 take care of the top bias only a third way of adjusting the bias of only the bottom mosfets could prove somewhat useful. Maybe at that point the resistor to ground would just be there for show.
At this point the only comment about the sound is that it was really audibly distorting when the absolute DC offset was 8V and now it sounds a heck of a lot better. Very promising actually.
Heating is under control, or so it seems it looks like I might have picked the right heatsinks for once! 😉 And yes the rectifier need it's own heatsink.
I will report later on the noises, so far I don't hear any turn on-off noises or hums. I have a high efficiency test speaker that I want to try before I really say.