grataku said:
I didn't use R11 and R33. V1 and V3 in my setup are 100k ea. I can adjust bias with those trimmers (47K resistance) and fine tune the differencial offset to 0V.
I didn't use R10 and R13 (as well as Q4) and corresponding parts on the other side. They don't adjust bias but act as protection in case too much current is flowing (when output shorted to ground).
Did you try not to use output to ground resistors at all? I didn't, but it would be interesting to see if the circuit can work without them. You also don't mention anything about 4.7k resistors from differential pair sources to output, but I'm assuming you are using them for further DC control.
Glad you have it working.😉
HI Peter, yes I am using the 4.7k resistors, I will try to eliminate the protection although I would like to make it so it kicks in a little later I think it maybe shaving off that extra 100mV that I need on Q2 and 11. I'll probably put a 500 ohms pot in place of R10 and see what that does.
As far as V1 and V3 I played with them a little with no apparent effect, I may need to remove the 47k in series to make the adjustment a little more drastic. With the different CS I really have some strange values come up. I am at the point where I really want to listen but I know I should do some more experiments. 😉 At the moment I am too lazy to remove the 4 200ohm output resistors, maybe later. It won't matter until I get those VM transformers anyway AND the cardas posts! PLUS I am thinking about where to get high quality AC receptacles, and those green brakers on that jeff rowland back panel picture you posted a while back.😉
As far as V1 and V3 I played with them a little with no apparent effect, I may need to remove the 47k in series to make the adjustment a little more drastic. With the different CS I really have some strange values come up. I am at the point where I really want to listen but I know I should do some more experiments. 😉 At the moment I am too lazy to remove the 4 200ohm output resistors, maybe later. It won't matter until I get those VM transformers anyway AND the cardas posts! PLUS I am thinking about where to get high quality AC receptacles, and those green brakers on that jeff rowland back panel picture you posted a while back.😉
R10/R13 and R39/R35 form voltage dividers that set the 'on' point for Q4 and Q9. Think of them as setting .65V for the Vbe. If the voltage drop across R6 or R41 gets too high, the voltage across R13 and R35 will approach .65V and the NPN will switch on, limiting the output MOSFET it is attached to. You can set this wherever you like or remove it entirely if it bothers you. To disable the current limiters, remove (or never install) R10, R13, Q4, R39, R35, and Q9. The circuit will hum along quite happily assuming that you don't have a fault condition at the outputs.
The sum of the resistance of V1/R11 and V3/R33 will alter the bias somewhat. In principle, you can remove them entirely (infinite resistance) and the Vbe of Q3/Q8 will take over entirely. Putting in resistance lowers that voltage, hence lowering the current in the MOSFET. By all means, experiment with different values there. This is going to be one of those issues that is influenced by the temperature of the MOSFETs, which in turn will be set by the bias current, rail voltage, and heatsinking.
I had the very devil of a time figuring out why my Alephs weren't biasing up properly. Eventually, I figured out that it was because they were stone cold (they're water cooled) compared to Nelson's production models which ran at supernova temperatures. Once I got that riddle solved, I was much happier. Everyone's thermal environment will be different so experimentation will be the order of the day.
The only way around this that I can see would be to specify heatsinks for specific variations on the circuit, but since DIY people are notorious for taking matters into their own hands, that solution wouldn't work well in the real world.
This is not a good beginner project...but then again, there are those budding DIYers who will scoff and try it anyway. Baptism by fire, you might say. A learning experience.
The value of the resistors from the output to the current source will be interrelated with the value of the resistors from the outputs to ground. The lower one is, the higher the other (there are limits to this, obviously--you don't want a dead short in either location).
What kind of voltage drop are you seeing across R5, R6, R40, and R41?
Grey
The sum of the resistance of V1/R11 and V3/R33 will alter the bias somewhat. In principle, you can remove them entirely (infinite resistance) and the Vbe of Q3/Q8 will take over entirely. Putting in resistance lowers that voltage, hence lowering the current in the MOSFET. By all means, experiment with different values there. This is going to be one of those issues that is influenced by the temperature of the MOSFETs, which in turn will be set by the bias current, rail voltage, and heatsinking.
I had the very devil of a time figuring out why my Alephs weren't biasing up properly. Eventually, I figured out that it was because they were stone cold (they're water cooled) compared to Nelson's production models which ran at supernova temperatures. Once I got that riddle solved, I was much happier. Everyone's thermal environment will be different so experimentation will be the order of the day.
The only way around this that I can see would be to specify heatsinks for specific variations on the circuit, but since DIY people are notorious for taking matters into their own hands, that solution wouldn't work well in the real world.
This is not a good beginner project...but then again, there are those budding DIYers who will scoff and try it anyway. Baptism by fire, you might say. A learning experience.
The value of the resistors from the output to the current source will be interrelated with the value of the resistors from the outputs to ground. The lower one is, the higher the other (there are limits to this, obviously--you don't want a dead short in either location).
What kind of voltage drop are you seeing across R5, R6, R40, and R41?
Grey
Hey Grey,
I actually had a lot of fun with this project, I am completely new to MOSFET, if you don't consider the BOSOZ which was plug-and-play. I built a ton of transostor stuff and it was interesting to see conduction kickin in at ~4 volts. With transistor stuff happens much more gradually. Anyway, I listed the voltage you were asking about in my previous post. Q2 and q11 source drop is about 100-150mV lower than the Q1 Q10. I will set up the other beta board with a few mods among which there will be no current limiters.
I found it nice to work on this PCB I hope you will be pleased with them too. 😉
I actually had a lot of fun with this project, I am completely new to MOSFET, if you don't consider the BOSOZ which was plug-and-play. I built a ton of transostor stuff and it was interesting to see conduction kickin in at ~4 volts. With transistor stuff happens much more gradually. Anyway, I listed the voltage you were asking about in my previous post. Q2 and q11 source drop is about 100-150mV lower than the Q1 Q10. I will set up the other beta board with a few mods among which there will be no current limiters.
I found it nice to work on this PCB I hope you will be pleased with them too. 😉
Aleph-X High-Power Version: Draft Schematics are there to be reviewed
May I ask the members of the group to spend some minutes t ohave a look at a schematic which I oulled together for an Aleph-X 2 (100 Watt) based on Grey's input and the different discussions in this group ? I am looking especialy for any comments on flaws I put in.
Thanks & Best Regards
You find it here:
http://www.diyaudio.com/forums/showthread.php?s=&postid=104069#post104069
May I ask the members of the group to spend some minutes t ohave a look at a schematic which I oulled together for an Aleph-X 2 (100 Watt) based on Grey's input and the different discussions in this group ? I am looking especialy for any comments on flaws I put in.
Thanks & Best Regards
You find it here:
http://www.diyaudio.com/forums/showthread.php?s=&postid=104069#post104069
grataku,
You're a sneaky fellow to hide the voltage drops in plain sight like that.
(Shades of <i>The Purloined Letter</i>)
Grey
You're a sneaky fellow to hide the voltage drops in plain sight like that.
(Shades of <i>The Purloined Letter</i>)
Grey
The Beta AX lives...
Using Grey's original CCS.
From the beta schematic I changed the following:
R1/R4 and R44/R45 total 62-ohm instead of 31-ohm
C2 & C4 are 5pF instead of 10pF
R19 & R29 are 10K
The Results:
PS Voltage 14.3VDC w/ 220,000uF per rail
Avg DC offset is running 1.4mV
Avg Abs DC offset is 47mV / 55mV
R23 4.24V
R25 4.31V
R5 .499V
R40 .501V
R6 .504V
R41 .500V
My amp still has a slight hum with and without an input connected.
Is anyone else seeing the same?
Has anyone tried adding any of the optional caps?
Wessol
Using Grey's original CCS.
From the beta schematic I changed the following:
R1/R4 and R44/R45 total 62-ohm instead of 31-ohm
C2 & C4 are 5pF instead of 10pF
R19 & R29 are 10K
The Results:
PS Voltage 14.3VDC w/ 220,000uF per rail
Avg DC offset is running 1.4mV
Avg Abs DC offset is 47mV / 55mV
R23 4.24V
R25 4.31V
R5 .499V
R40 .501V
R6 .504V
R41 .500V
My amp still has a slight hum with and without an input connected.
Is anyone else seeing the same?
Has anyone tried adding any of the optional caps?
Wessol
Grataku,
The LM329 voltage reference is, as previously stated, a lower voltage than the 9.1V zener originally specified. With a ZVN3310 in place, I used R24=392, R26=150 and VR2=200 to get the same adjustment range as the 9.1V zener/IRF9610 current source. Tweaking VR2, I was easily able to get rid of any absolute DC offset. I do recommend 10-turn pots for this design, to give the fine adjustment ability. Or, you can do as Grey has, by altering the values of that triplet to reduce the range of adjustment of VR2.
VR1 and VR3 very definitely have a large impact, since they set the bias current through the output transistors. You can measure this across R6 / R41. I was a little surprised to find that very large variances in this current had little to no effect on the DC bias at the outputs. In any case, I'm running 2.5A bias per side for a total of 5A bias per channel, and I haven't had any difficulty with the standard component values. You may want to increase the value of R10 and R39 if you are running high bias currents.
The LM329 voltage reference is, as previously stated, a lower voltage than the 9.1V zener originally specified. With a ZVN3310 in place, I used R24=392, R26=150 and VR2=200 to get the same adjustment range as the 9.1V zener/IRF9610 current source. Tweaking VR2, I was easily able to get rid of any absolute DC offset. I do recommend 10-turn pots for this design, to give the fine adjustment ability. Or, you can do as Grey has, by altering the values of that triplet to reduce the range of adjustment of VR2.
VR1 and VR3 very definitely have a large impact, since they set the bias current through the output transistors. You can measure this across R6 / R41. I was a little surprised to find that very large variances in this current had little to no effect on the DC bias at the outputs. In any case, I'm running 2.5A bias per side for a total of 5A bias per channel, and I haven't had any difficulty with the standard component values. You may want to increase the value of R10 and R39 if you are running high bias currents.
Pan,
I don't think I've received any email from you. Please click the "email" button at the bottom of this post to send me another message. I can add you on the list for spare boards.
I don't think I've received any email from you. Please click the "email" button at the bottom of this post to send me another message. I can add you on the list for spare boards.
Wessol,
My prototype board does have a very slight hum, but I am running directly from a C-only supply (62mF per rail) filter. My final version will use a CRC power supply filter, and from what I gather, this should eliminate the hum at the output. I don't think that fiddling with any of the optional C values in the circuit will have an effect on the hum, as their primary purpose is to ensure circuit stability only. I didn't need to use any of the caps, other than the 5pF caps in parallel with the 100K feedback resistors.
My prototype board does have a very slight hum, but I am running directly from a C-only supply (62mF per rail) filter. My final version will use a CRC power supply filter, and from what I gather, this should eliminate the hum at the output. I don't think that fiddling with any of the optional C values in the circuit will have an effect on the hum, as their primary purpose is to ensure circuit stability only. I didn't need to use any of the caps, other than the 5pF caps in parallel with the 100K feedback resistors.
My amp has a slight hum too. It is more when amps is cold, and goes away almost completely when warming up.
Oh yeah, forgot to mention... I'm planning to build a little test circuit for matching 9610 diff pairs. It should actually be quite simple... just a current source (or resistor from a high voltage) plus the 392 ohm drain resistors to ground and place on the breadboard to plug in the mosfets you're matching. Simply measure the differential voltage between the mosfet drains, and you can match that way... no need to have it all soldered into the circuit.
The reason the matching is so touchy is that both sides of the differential are being used. If you use only one side, the other can either accept surplus current or donate a bit to the other side, depending on what is necessary to cure a bit of DC offset. This circuit can't do that. If one side wants to get rid of a bit of current and tries to shove it off on the other side, it will drive the output MOSFET further into conduction. If that MOSFET was already in balance, it would now be out of whack. So it takes a well-matched set of gain devices to make it all come together.
The idea of matching MOSFETs out-of-circuit is a good one, but bear in mind that you're not only matching them against each other, you're matching them against the output devices. If you had a perfectly matched set of outputs, matching out of circuit would be easy. Since that's not going to be the case, you'll find that a well matched pair of inputs may not work just right in the circuit.
Hopefully you will feel that the effort was worth it once it's done.
Grey
The idea of matching MOSFETs out-of-circuit is a good one, but bear in mind that you're not only matching them against each other, you're matching them against the output devices. If you had a perfectly matched set of outputs, matching out of circuit would be easy. Since that's not going to be the case, you'll find that a well matched pair of inputs may not work just right in the circuit.
Hopefully you will feel that the effort was worth it once it's done.
Grey
I am probably going to open a can of worms but,
it is looking like the diff pair could use the
addition of a couple transistors to form a current mirror
to help balance everything, any one dare to give there thoughts on this???
wessol
it is looking like the diff pair could use the
addition of a couple transistors to form a current mirror
to help balance everything, any one dare to give there thoughts on this???
wessol
Or maybe use more output devices to balance everything better. With my 16 mosfets per channel I'm not experiencing any problems.😉
More 'experiments':
I found out why the V1 and 3 was unresponsive, bad trimpots, I guess what you find in the junkbox sometimes is just that. The outs I was using where stuck on 0. I now have experienced that too.
I now have a straight pair of 200k, with no resistors and finally things are working. Sorry for the confusion.
Another thing I have learned is that I need more voltmeters! 😉
Grey, I don't understand what you mean by the diff devices being matched to the outputs: are we looking at the Vgs being the same on ALL the mosfets?
Wessol,
I have had no problems finding a matched 9610 with ordinary matching technique, I used a 680 Ohms resistor. I watched the mosfet for about a minute. So the differential is the section of the circuit I feel most comfy about. The voltage drop on the 390 ohms is dead even.
After doing all measurements several times I can now state my findings with more certainty, I don't claim to understand the results but I claim this to be the truth to the best of my knowledge.
I basically I can get to two conditions :
1) Voltage drop on the sources R of the CS mosfets 0.5V
Voltage drop on the sources R of the Gain mosfets 0.43V
This gives 65 mV DC at the load offset with the absolute DC pretty much at 0.0 V DC. Note that the drop are symmetric on either half.
2) One side has 0.57V and 0.5V across the CS and gain sources, respectively and 0.38V and 0.35V across the CS and gain sources on THE OTHER side.
Under this condition both DC measurements are virtually 0.
Nedless to say I don't particulary like state 2 and I don't know how can I have no DC if two sides are not equally sinking current. I also don't understand the different drop on the sources from top to bottom where the extra 300-400 mA go? It is possible that my rail voltage is not perfectly symm. I haven't checked for that in a while. I have noticed a bit of change in the rail voltage as the bias settles, the trafo is ~350 VA.
Other than an understanding I would also like to fix it, probably I need better matching.
I wonder If I should get two pairs one matched with higher VGS for the CS and one with lower VGS for the gain devices in order to equalize the drop. I ripped out the protection on the gain mosfets. Without it the circuit seems to be getting an extra 100 mV at the gate. However the source drop still limits the Vgs that is seen by the mosfet to about 4.17 V.
The 65 mV is really very stable over hours, I will not hesitate to attach real speakers to this version amp as soon as it's complete.
My version is dead quiet. I have a 93 dB dual concentric Tannoy driver that I use for testing, with input connected or without input connected.
The test PS I am using is a 22 VAC trafo and rectifier followed by a 1ohm-50mF-1ohm-50mF I end up with about +/- 14 VDC. The PS grounds are all symmetrically laid out with all equal lengths etc.
I hear no turn on-off transients. The only thing I noticed is that when the mosfet stop conducting a +/- 8 Vdc are left hanging on the filter caps but that doesn't seem to cause problem If I switch the amp back on.
ok enough BS! 😉
I found out why the V1 and 3 was unresponsive, bad trimpots, I guess what you find in the junkbox sometimes is just that. The outs I was using where stuck on 0. I now have experienced that too.
I now have a straight pair of 200k, with no resistors and finally things are working. Sorry for the confusion.
Another thing I have learned is that I need more voltmeters! 😉
Grey, I don't understand what you mean by the diff devices being matched to the outputs: are we looking at the Vgs being the same on ALL the mosfets?
Wessol,
I have had no problems finding a matched 9610 with ordinary matching technique, I used a 680 Ohms resistor. I watched the mosfet for about a minute. So the differential is the section of the circuit I feel most comfy about. The voltage drop on the 390 ohms is dead even.
After doing all measurements several times I can now state my findings with more certainty, I don't claim to understand the results but I claim this to be the truth to the best of my knowledge.
I basically I can get to two conditions :
1) Voltage drop on the sources R of the CS mosfets 0.5V
Voltage drop on the sources R of the Gain mosfets 0.43V
This gives 65 mV DC at the load offset with the absolute DC pretty much at 0.0 V DC. Note that the drop are symmetric on either half.
2) One side has 0.57V and 0.5V across the CS and gain sources, respectively and 0.38V and 0.35V across the CS and gain sources on THE OTHER side.
Under this condition both DC measurements are virtually 0.
Nedless to say I don't particulary like state 2 and I don't know how can I have no DC if two sides are not equally sinking current. I also don't understand the different drop on the sources from top to bottom where the extra 300-400 mA go? It is possible that my rail voltage is not perfectly symm. I haven't checked for that in a while. I have noticed a bit of change in the rail voltage as the bias settles, the trafo is ~350 VA.
Other than an understanding I would also like to fix it, probably I need better matching.
I wonder If I should get two pairs one matched with higher VGS for the CS and one with lower VGS for the gain devices in order to equalize the drop. I ripped out the protection on the gain mosfets. Without it the circuit seems to be getting an extra 100 mV at the gate. However the source drop still limits the Vgs that is seen by the mosfet to about 4.17 V.
The 65 mV is really very stable over hours, I will not hesitate to attach real speakers to this version amp as soon as it's complete.
My version is dead quiet. I have a 93 dB dual concentric Tannoy driver that I use for testing, with input connected or without input connected.
The test PS I am using is a 22 VAC trafo and rectifier followed by a 1ohm-50mF-1ohm-50mF I end up with about +/- 14 VDC. The PS grounds are all symmetrically laid out with all equal lengths etc.
I hear no turn on-off transients. The only thing I noticed is that when the mosfet stop conducting a +/- 8 Vdc are left hanging on the filter caps but that doesn't seem to cause problem If I switch the amp back on.
ok enough BS! 😉
Sorry...didn't mean to confuse you about matching the front end to the outputs. Think of it this way. Suppose you had a <i>perfectly</i> matched front end, say a 2SK389 manufactured on a day when there were no clouds in the sky. The voltage drop across the load resistors (R23 & R25) would be identical. Let's say for example that the DC drop across both of those resistors is exactly 4V. Now, suppose we have a tiny imperfection in the matching of the output MOSFETs. One side wants to go live at 4 volts, but the other (for shame!) wants 4.1V.
The 4v side is happy as a clam, but the other side isn't. So you adjust the current source a hair, such that there's 4.1V across the load resistors. Now the 4.1V side is happy, but the 4V side is starting to pull on its leash a bit.
The secret here is to slightly mis-match the front end MOSFETs so that one conducts slightly in advance of the other so as to shake hands with the outputs more readily.
A current mirror won't help. In fact, to use one you'd have to recalculate the feedback and Bog knows what else. Ugh. If you want to, go ahead. I'm putting my efforts into preamps at the moment.
Ideally, each side should be biased the same or you could find yourself with the two halves pumping unevenly under a heavy load. This would cause an asymmetry in the waveform and distortion would rise. (At least that's my thought experiment--even a best case hypothesis leads to a loss of some fraction of power.) It's probably not serious in the real world, but I wanted to try to lock down that particular variable if I could. It wasn't all that difficult to add the pots, so I did. Put 'em in, take 'em out. I don't care. The better everything else matches up, the less you'll need them unless you just want to experiment with cranking up the bias. Or you can lower it to match your dissipation to your heatsinks. Whatever makes you happy.
Yes, increasing the bias on an Aleph makes an audible difference.
Have you measured your Source resistors? If they're off (5% parts can be 10% apart just by being at the opposite ends of their range although it's rare), you could have identical current, but be reading different voltages.
Grey
The 4v side is happy as a clam, but the other side isn't. So you adjust the current source a hair, such that there's 4.1V across the load resistors. Now the 4.1V side is happy, but the 4V side is starting to pull on its leash a bit.
The secret here is to slightly mis-match the front end MOSFETs so that one conducts slightly in advance of the other so as to shake hands with the outputs more readily.
A current mirror won't help. In fact, to use one you'd have to recalculate the feedback and Bog knows what else. Ugh. If you want to, go ahead. I'm putting my efforts into preamps at the moment.
Ideally, each side should be biased the same or you could find yourself with the two halves pumping unevenly under a heavy load. This would cause an asymmetry in the waveform and distortion would rise. (At least that's my thought experiment--even a best case hypothesis leads to a loss of some fraction of power.) It's probably not serious in the real world, but I wanted to try to lock down that particular variable if I could. It wasn't all that difficult to add the pots, so I did. Put 'em in, take 'em out. I don't care. The better everything else matches up, the less you'll need them unless you just want to experiment with cranking up the bias. Or you can lower it to match your dissipation to your heatsinks. Whatever makes you happy.
Yes, increasing the bias on an Aleph makes an audible difference.
Have you measured your Source resistors? If they're off (5% parts can be 10% apart just by being at the opposite ends of their range although it's rare), you could have identical current, but be reading different voltages.
Grey
Maybe it's time to change to dual current sources for the diff pair? Only trouble is, you'll end up adding some source degeneration in order to allow for balance adjustment, though I suppose this could be compensated for by increasing the current, which in turn would be at least partially counteracted by the required reducition in drain resistance... hmm. Well, it's a thought, anyway. Certainly not preferable to fet matching.
The idea is tempting, but I chose not to follow that route because it would interfere with the ability of the front end to serve as a phase splitter; if you got the two sides sufficiently separated, you'd be forced to use balanced input signals.
Also, keep in mind that the connection between the front end Sources is part of the "X" topology. If you separate the Siamese twins too completely, you'll lose the X and have only a bridged Aleph. Not that that would be a bad amplifier topology, but...
Grey
Also, keep in mind that the connection between the front end Sources is part of the "X" topology. If you separate the Siamese twins too completely, you'll lose the X and have only a bridged Aleph. Not that that would be a bad amplifier topology, but...
Grey