Hello
From the backengineering thread I got interested in building an X series amplifier myself. Someone seems like doing it already, and Mr. Pass said couple of times he has no objections.
I have some practical questions concerning the details of the construction process, the X circuit, it's performance and inner workings. This is why I founded this thread. While planning this and the amplifiers in sauna some time ago I found lots more to say, but below is what I still remember.
1. I wonder how far do the X amplifiers work in class A, and how is it related to quiescent dissipation. In A40 and ordinary push-pull amplifiers I know that for every 40W of class A output one must dissipate 100W heat. As both sides of the end stage conduct current, the amplifier is in class A, and if one side ceases conducting, then the amplifier is no more in class A. So one must use bias that is half of the maximum output current.
I get this intentionally wrong here, but don't know exactly where, so please tell me. So... the X circuit is internally bridged. X150 dissipates 100W per channel. 50V output swing is 25V to one direction, that is like operating from 70V or +/-35V supplies (just a guess). That gives me about 1.4A of current per channel, divided between 2 bridge halves, 0.7A in each. That would mean class A operation up to 1.4A output current, which is about 15W peak into 8 Ohms. Not even close to stated 50W in class A.
2. How close must the end-stage FETs be matched? If I buy a batch of 50 IRF9240 and use the 12 of them that are closest to each other, do you think that'll do it? What kind of measuring equipment will I need? Is a homemade circuit with a cheap digital voltmeter good enough?
3. If I happen to win a million in lottery, then how much do you think would it cost to get one X350 to Finland? Of course all X series models deserve my advocacy, in case someone else here gets wealthy.
4. I hope it will not be a problem if I use much smaller transformers than NP uses in the originals. If I did a copy of X150, I'd use one 540VA 2x24V transformer for end stages and one with higher voltage output (maybe 2x38V) for the drive stages. The drive stage would also get good regulation and/or a capacitance multiplier.
5. How about replacing all the current sources with resistors? Multiple (maybe nine) 1/2W 1% metal film resistors made to look like one, to get less thermal noise. I have some reservations about the effect a constant current source has to the sound.
6. Power supply capacitors. One 100V 10 000uF from positive rail to negative rail equals two a) 5000uF, b) 10 000uF or c) 20 000uF caps from the rails to ground?
7. The virtual grounds next to the inputs are connected to each other and to nowhere else? Maybe to the input cable ground?
More later. If you want to tell me something, but at the same time keep it out of public domain, I welcome emails to musher@mbnet.fi
-Kimmo Sundqvist
From the backengineering thread I got interested in building an X series amplifier myself. Someone seems like doing it already, and Mr. Pass said couple of times he has no objections.
I have some practical questions concerning the details of the construction process, the X circuit, it's performance and inner workings. This is why I founded this thread. While planning this and the amplifiers in sauna some time ago I found lots more to say, but below is what I still remember.
1. I wonder how far do the X amplifiers work in class A, and how is it related to quiescent dissipation. In A40 and ordinary push-pull amplifiers I know that for every 40W of class A output one must dissipate 100W heat. As both sides of the end stage conduct current, the amplifier is in class A, and if one side ceases conducting, then the amplifier is no more in class A. So one must use bias that is half of the maximum output current.
I get this intentionally wrong here, but don't know exactly where, so please tell me. So... the X circuit is internally bridged. X150 dissipates 100W per channel. 50V output swing is 25V to one direction, that is like operating from 70V or +/-35V supplies (just a guess). That gives me about 1.4A of current per channel, divided between 2 bridge halves, 0.7A in each. That would mean class A operation up to 1.4A output current, which is about 15W peak into 8 Ohms. Not even close to stated 50W in class A.
2. How close must the end-stage FETs be matched? If I buy a batch of 50 IRF9240 and use the 12 of them that are closest to each other, do you think that'll do it? What kind of measuring equipment will I need? Is a homemade circuit with a cheap digital voltmeter good enough?
3. If I happen to win a million in lottery, then how much do you think would it cost to get one X350 to Finland? Of course all X series models deserve my advocacy, in case someone else here gets wealthy.
4. I hope it will not be a problem if I use much smaller transformers than NP uses in the originals. If I did a copy of X150, I'd use one 540VA 2x24V transformer for end stages and one with higher voltage output (maybe 2x38V) for the drive stages. The drive stage would also get good regulation and/or a capacitance multiplier.
5. How about replacing all the current sources with resistors? Multiple (maybe nine) 1/2W 1% metal film resistors made to look like one, to get less thermal noise. I have some reservations about the effect a constant current source has to the sound.
6. Power supply capacitors. One 100V 10 000uF from positive rail to negative rail equals two a) 5000uF, b) 10 000uF or c) 20 000uF caps from the rails to ground?
7. The virtual grounds next to the inputs are connected to each other and to nowhere else? Maybe to the input cable ground?
More later. If you want to tell me something, but at the same time keep it out of public domain, I welcome emails to musher@mbnet.fi
-Kimmo Sundqvist
OK here goes,
X-series is push-pull and is inherently bridged so ...
X-150 idle dissipation is 200W which is 100W per channel which is 50W per terminal which is 25W per mosfet bank.
Ok now the output swing is 50V which is about +/- 28V
25/28~0.9Amps bias per mosfet bank.
now lets calculate the class A point into 8ohms.
1.8x1.8x4=12.96W peak / terminal which is 6.48Wrms
12.96x2=25.92W/channel which is 12.96Wrms / channel.
25.92x2=51.84W total.
Therefore the 50W Class A must be 25W peak Class A per channel. Either that or Nelson's done something I dont know about.... (wouldnt surprise me if he has
)
You'll need big transformers as Nelson Pass uses because it has a high bias point.
X-series is push-pull and is inherently bridged so ...
X-150 idle dissipation is 200W which is 100W per channel which is 50W per terminal which is 25W per mosfet bank.
Ok now the output swing is 50V which is about +/- 28V
25/28~0.9Amps bias per mosfet bank.
now lets calculate the class A point into 8ohms.
1.8x1.8x4=12.96W peak / terminal which is 6.48Wrms
12.96x2=25.92W/channel which is 12.96Wrms / channel.
25.92x2=51.84W total.
Therefore the 50W Class A must be 25W peak Class A per channel. Either that or Nelson's done something I dont know about.... (wouldnt surprise me if he has
)You'll need big transformers as Nelson Pass uses because it has a high bias point.
5. Yes constant current sources can be replaced by resistors but the idle dissipation of the circuit will increase and the performance can actually be worse depending on the situation.
6. 10000uF between +ve and -ve will look like 20000uF between +ve and ground and 20000uF between -ve and ground
7. the virtual grounds are definately not connected to the input cable ground because the other end of the cable will be @ true ground which will be different.
6. 10000uF between +ve and -ve will look like 20000uF between +ve and ground and 20000uF between -ve and ground
7. the virtual grounds are definately not connected to the input cable ground because the other end of the cable will be @ true ground which will be different.
1) There's nothing inherent in the X/Supersymmetry circuit that means that you have to hit the same bias point that Nelson uses. All the "special" stuff that happens is in the front end. You can use any back end or operating class you like. Feel free to bias it for class A, AB, or B if it makes you happy. I wouldn't recommend going so far as class C.
2) The Aleph devices were specified as being matched to within .1V. Although closer is better, that would be a good goal for output devices for an X.
3) There are Pass dealers listed in several northern European countries according to the Pass Labs website.
4) A 24-0-24VAC transformer would give you roughly +-30V rails. That should get the job done. The VA rating will depend on how hard you intend to bias the output (see #1). There are several options for the front end rails, including voltage doublers, separate transformers, tagging smaller transformers onto each end of the main transformer, etc.
5) Although I have expressed reservations about current sources from time to time, they do have their uses. Depending on how you assemble the front end, and whether you intend to use unbalanced or balanced inputs, you may need to use them.
6) J=.5*C*V**2, where
J= Joules of storage
C= capacitance (in Farads)
V= volts
The V squared part is where the kick comes in. Note that if you use one cap across both rails, you'll need to create a ground one way or another. Never met a circuit yet that didn't ground <i>somewhere</i>.
7) Ah...the mystery of the grounds. Actually, you can do all sorts of things with the grounds. It's not important to the X front end.
Grey
2) The Aleph devices were specified as being matched to within .1V. Although closer is better, that would be a good goal for output devices for an X.
3) There are Pass dealers listed in several northern European countries according to the Pass Labs website.
4) A 24-0-24VAC transformer would give you roughly +-30V rails. That should get the job done. The VA rating will depend on how hard you intend to bias the output (see #1). There are several options for the front end rails, including voltage doublers, separate transformers, tagging smaller transformers onto each end of the main transformer, etc.
5) Although I have expressed reservations about current sources from time to time, they do have their uses. Depending on how you assemble the front end, and whether you intend to use unbalanced or balanced inputs, you may need to use them.
6) J=.5*C*V**2, where
J= Joules of storage
C= capacitance (in Farads)
V= volts
The V squared part is where the kick comes in. Note that if you use one cap across both rails, you'll need to create a ground one way or another. Never met a circuit yet that didn't ground <i>somewhere</i>.
7) Ah...the mystery of the grounds. Actually, you can do all sorts of things with the grounds. It's not important to the X front end.
Grey
1. This is not important, you can configure anything you want and even have a low/high switch if you like.
2. This depends on a number of factors, one of which is the value of your source resistor. Naturally, matched units work better together -- if you want perfect matching, try using single devices ... It is most important to match each 4 sections (top left, TR, BR, BL) internally. Bear in mind that the whole point of the X circuitry is to make it take advantage of cancellations in very similar components. You can find information on what matching results I got out of a population of 60 in the reverse engineering thread (X100 ...)
4. You can use any transformer you like, it is not a very important question. However large transformers seem to have a very positive effect on power amps in general. NP has info on power supplies on his sites (commentary for consumers?).
5. Sorry, this will not work well unless you increase the voltage and value of resistors radically. I have no idea why you would want to do this as there are very few components to make a decent current source. A resistor is only a constant current source if the voltage drop is very large or the change of voltage is low (which it is actually is here, but hey!). The folded cascode still needs to be there, but it is not a cc anyway. If you build a preamp, you might want to experiment with constant current diodes, but the circuit still needs to be trimmed to work well. An alternative to trimming is matching current mirrors. (top to bottom)
6. If the caps are identical, 4 times the capacitance at 1/4 (the impedance when eyeballed from a circuit diagram (this ties well with the stored energy calculation as well). Tip: use 2 caps and see what happens when you put them in parallel instead of series.
However, you would typically use larger caps with lower voltage ratings and so it is impossible to say for a practical case without choosing components first.
7. One place to go looking is at the original patent found at www.passlabs.com. As long as you reference something to some ground, you are OK. I use my own grounding scheme where I reference center tap of transformers to source component ground. Your mileage may vary.
When you make boards, don't make them too small like I did with my first set! Focus on the important stuff (getting the circuit right) and make it as a preamp first. That way you are experimenting at low cost and you can learn from your results with a simpler circuit. It is quicker to build a prototype than to make everything perfect before you start (look who is talking!)
Good luck!
2. This depends on a number of factors, one of which is the value of your source resistor. Naturally, matched units work better together -- if you want perfect matching, try using single devices ... It is most important to match each 4 sections (top left, TR, BR, BL) internally. Bear in mind that the whole point of the X circuitry is to make it take advantage of cancellations in very similar components. You can find information on what matching results I got out of a population of 60 in the reverse engineering thread (X100 ...)
4. You can use any transformer you like, it is not a very important question. However large transformers seem to have a very positive effect on power amps in general. NP has info on power supplies on his sites (commentary for consumers?).
5. Sorry, this will not work well unless you increase the voltage and value of resistors radically. I have no idea why you would want to do this as there are very few components to make a decent current source. A resistor is only a constant current source if the voltage drop is very large or the change of voltage is low (which it is actually is here, but hey!). The folded cascode still needs to be there, but it is not a cc anyway. If you build a preamp, you might want to experiment with constant current diodes, but the circuit still needs to be trimmed to work well. An alternative to trimming is matching current mirrors. (top to bottom)
6. If the caps are identical, 4 times the capacitance at 1/4 (the impedance when eyeballed from a circuit diagram (this ties well with the stored energy calculation as well). Tip: use 2 caps and see what happens when you put them in parallel instead of series.
However, you would typically use larger caps with lower voltage ratings and so it is impossible to say for a practical case without choosing components first.
7. One place to go looking is at the original patent found at www.passlabs.com. As long as you reference something to some ground, you are OK. I use my own grounding scheme where I reference center tap of transformers to source component ground. Your mileage may vary.
When you make boards, don't make them too small like I did with my first set! Focus on the important stuff (getting the circuit right) and make it as a preamp first. That way you are experimenting at low cost and you can learn from your results with a simpler circuit. It is quicker to build a prototype than to make everything perfect before you start (look who is talking!)
Good luck!
http://www.passlabs.com/products.htm, the part about X150 has 50W quoted, and I can't figure out how could I have misunderstood it.
About the actual topic of the thread... I have bought about a year ago a 1000mm piece of Fischer SK110 heatsink, assuming it is the H-shaped 12x12cm big big one and I got the number right. It's 0.05K/W in that size, and I have plans to put it into four 10" parts, 0.2K/W each.
Now, into 25cm of the heatsink I only know how to fit 8 TO3s. An X250 has 32 TO3s, 16 for one channel. X350 1.5 times that, 24 per channel. X150 has 20, 10 per channel, though I am a bit lost on why that 10 is not a multiple of 4.
I don't want to use more than 25cm for a channel, since a piece that big can dissipate 150W quite well. And since I can easily get chassis made of 0.75mm galvanized steel sheet but not of aluminium, I more like the idea of having the heatsinks towering up from the amplifier (like in that A40 project in passdiy project gallery) than covering the sides of the chassis.
Also, if you know any MOS-FETs rated for more than 125W (of IRF9240 and 240) then please tell me. 150V of voltage rating would suffice, maybe even as little as 100V, but for that low it must otherwise be _really_ good.
-Kimmo S.
About the actual topic of the thread... I have bought about a year ago a 1000mm piece of Fischer SK110 heatsink, assuming it is the H-shaped 12x12cm big big one and I got the number right. It's 0.05K/W in that size, and I have plans to put it into four 10" parts, 0.2K/W each.
Now, into 25cm of the heatsink I only know how to fit 8 TO3s. An X250 has 32 TO3s, 16 for one channel. X350 1.5 times that, 24 per channel. X150 has 20, 10 per channel, though I am a bit lost on why that 10 is not a multiple of 4.
I don't want to use more than 25cm for a channel, since a piece that big can dissipate 150W quite well. And since I can easily get chassis made of 0.75mm galvanized steel sheet but not of aluminium, I more like the idea of having the heatsinks towering up from the amplifier (like in that A40 project in passdiy project gallery) than covering the sides of the chassis.
Also, if you know any MOS-FETs rated for more than 125W (of IRF9240 and 240) then please tell me. 150V of voltage rating would suffice, maybe even as little as 100V, but for that low it must otherwise be _really_ good.
-Kimmo S.
generally speaking the lower voltage mosfets have less distortion so you should try to match them fairly close to the rails allowing for a safety margin ....... given that the rails for a X150 are less than +/-30V 100V devices would get plenty of safety margin so long as the wattage is large enough .... as Nelson has stated that the X150 has approximately 15W Class A per channel as we calculated and that the 50W stated is the total peak class A and that he will have this rectified on the Pass Labs site soon .... each channel should have a multiple of 4 mosfets in the output stage because there is 4 banks per channel so i dont know how it uses 10 devices per channel but you could also use 12 devices per channel giving 4 or mosfets per bank and a little more safety margin.
I can only get transformers with 18V and 24V secondaries easily. I think 18V turns to +/-26V unloaded rails with a 545VA Finnish transformer that has 6 percent of voltage drop. That maybe isn't enough for me, so 24V gives me 35V unloaded rails, and about +/-34V loaded.
This way I'd have output a bit more than X150 has, but a bit less than X250 has. What figure should I use for MOS-FET voltage drop at maximum output? I don't believe I can saturate them and still retain full quality.
Now everyone go and see http://www.fischerelektronik.de/fischer/download_PDF/A-Seiten/A67.pdf
I have one metre of that lower one, SK110. I will put it into 4 pieces, each 10". If anyone has suggestions how to mount the maximum amount of transistors on one 10" piece, please share your ideas. It should be a beautiful, yet efficient way, because the heatsink will be left visible when the thing is completed. One channel will have one 10" piece, and I am planning for two stereo units total.
I can have 150W of quiescent per channel, if the transistors can take it, since the heatsink can. A 10" piece will be 0.2K/W, if I haven't said it earlier. Maybe, just maybe, I should settle for 26V rails, since I don't really need the power anyway.
Tell me what you think, so I can go on obtaining many many IRF9240/240 and matching them. Or any better suggestions than IRF? Then I will have the problem of determining how the front-end is exactly made. With parallel JFETs? I will use current sources, but how about led as a reference?
-Kimmo S.
This way I'd have output a bit more than X150 has, but a bit less than X250 has. What figure should I use for MOS-FET voltage drop at maximum output? I don't believe I can saturate them and still retain full quality.
Now everyone go and see http://www.fischerelektronik.de/fischer/download_PDF/A-Seiten/A67.pdf
I have one metre of that lower one, SK110. I will put it into 4 pieces, each 10". If anyone has suggestions how to mount the maximum amount of transistors on one 10" piece, please share your ideas. It should be a beautiful, yet efficient way, because the heatsink will be left visible when the thing is completed. One channel will have one 10" piece, and I am planning for two stereo units total.
I can have 150W of quiescent per channel, if the transistors can take it, since the heatsink can. A 10" piece will be 0.2K/W, if I haven't said it earlier. Maybe, just maybe, I should settle for 26V rails, since I don't really need the power anyway.
Tell me what you think, so I can go on obtaining many many IRF9240/240 and matching them. Or any better suggestions than IRF? Then I will have the problem of determining how the front-end is exactly made. With parallel JFETs? I will use current sources, but how about led as a reference?
-Kimmo S.
Kimmo,
If your speakers are 8 ohms or less, I'd use the +/-25V supply and power the front end from an elevated supply derived from a voltage doubler as in the Pass/Thagard A75 project. Using a doubler will give you a raw +/-50V, which you can knock down to a smoother +/-40V or so by adding a series R and shunt C in each leg of the supply. You can also regulate the elevated supply or add a capacitance multiplier as suggested above, but if you use current sources in the front end, I suspect that the amp will deliver very clean performance without it.
With this setup, you'd be able to bias the output stage up to 2.5A per channel, for 25W class A into 8 ohms, as well as drive the output stage to over 100W in class AB into 8 ohms. This strikes me as a pretty good compromise between having a pure class A amp and one that can drive moderately efficient speakers to realisitic levels without clipping.
OTOH, If your speakers are higher impedance, you might be happier overall with the higher voltage supply and a lower bias. Either way, using a separate front-end supply should let you drive the output to within 2-3V of the main supply rails. You can do this safely with MOSFETs as they don't saturate and "stick" the way bipolars do.
If your speakers are 8 ohms or less, I'd use the +/-25V supply and power the front end from an elevated supply derived from a voltage doubler as in the Pass/Thagard A75 project. Using a doubler will give you a raw +/-50V, which you can knock down to a smoother +/-40V or so by adding a series R and shunt C in each leg of the supply. You can also regulate the elevated supply or add a capacitance multiplier as suggested above, but if you use current sources in the front end, I suspect that the amp will deliver very clean performance without it.
With this setup, you'd be able to bias the output stage up to 2.5A per channel, for 25W class A into 8 ohms, as well as drive the output stage to over 100W in class AB into 8 ohms. This strikes me as a pretty good compromise between having a pure class A amp and one that can drive moderately efficient speakers to realisitic levels without clipping.
OTOH, If your speakers are higher impedance, you might be happier overall with the higher voltage supply and a lower bias. Either way, using a separate front-end supply should let you drive the output to within 2-3V of the main supply rails. You can do this safely with MOSFETs as they don't saturate and "stick" the way bipolars do.
It seems that IRF640/9640 are much better available here than 240/9240. Though I found 240 but not 9240, it was almost 6 euro apiece. 640 is 1.5 euro and 9640 is 2.7 euro. Last time I checked, $1 was 1.15 euro (and it's been a while).
640/9640 are TO220, but 125W, just like TO3 240/9240. How big difference that is practically? I think I should use double the amount of 640/9640. Or maybe 1.5 times. Anyway, is there anything or something I haven't noticed that would make 640 less suitable or worse for this project than 240?
-Kimmo S.
640/9640 are TO220, but 125W, just like TO3 240/9240. How big difference that is practically? I think I should use double the amount of 640/9640. Or maybe 1.5 times. Anyway, is there anything or something I haven't noticed that would make 640 less suitable or worse for this project than 240?
-Kimmo S.
I used IRF644s (also a TO-220 case) for my Aleph 2s. In addition, I'm running the bias current about 10% over stock. That works out to about Pd=25W per device. Nary a problem. Keep in mind that I'm running a water-cooled system, so I can dissipate heat easily; if you use plenty of heatsinking (a good idea, anyway) you'll be fine.
Grey
Grey
Then to the problems with front-end
As a starting point I will use the schematic posted by Petter. Since I am not an electronics professional, I am left with gathering pieces of information from nice people, and none of these things are my own invention.
1. I think I know how to make current sources. And if I don't, I know how to see a book or two for assistance. Nobody has said much about using a led as a voltage reference, but I will try that out unless something better appears.
2. All current sources could have the same heatsink, or at least heatsinks of equal size.
3. How can I parallel JFETs and is there an improvement in using them somewhere in the front end? Really odd that this thing seems like "plagued" by DC drift problems... wonder how Mr. Pass did it. Maybe with trying and testing, since I think he had to find the right combination only once.
4. I will use either 26V rails with 120W idle dissipation (10W for each device), giving me maybe an X100 =) or 35V rails, because 545VA transformers are not available here with 18V secondaries, only 24, 30 and 38. For 18V secondaries I'd have to use two 210VA or even 330VA.
5. If I use a capacitance multiplier with the front-end rails and end up with +/-60V in case of 35V endstage rails, do you think that is enough?
-Kimmo Sundqvist
As a starting point I will use the schematic posted by Petter. Since I am not an electronics professional, I am left with gathering pieces of information from nice people, and none of these things are my own invention.
1. I think I know how to make current sources. And if I don't, I know how to see a book or two for assistance. Nobody has said much about using a led as a voltage reference, but I will try that out unless something better appears.
2. All current sources could have the same heatsink, or at least heatsinks of equal size.
3. How can I parallel JFETs and is there an improvement in using them somewhere in the front end? Really odd that this thing seems like "plagued" by DC drift problems... wonder how Mr. Pass did it. Maybe with trying and testing, since I think he had to find the right combination only once.
4. I will use either 26V rails with 120W idle dissipation (10W for each device), giving me maybe an X100 =) or 35V rails, because 545VA transformers are not available here with 18V secondaries, only 24, 30 and 38. For 18V secondaries I'd have to use two 210VA or even 330VA.
5. If I use a capacitance multiplier with the front-end rails and end up with +/-60V in case of 35V endstage rails, do you think that is enough?
-Kimmo Sundqvist
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.