Performance?
Pete,
I am suffering from home redecorat-itis so finding time to complete the project is hard. What I could do fairly quickly is get some distortion measurements for the input stage by itself, and I will attempt to get that info to you.
Still, the basic principle and design goal holds. Floating input --> floating output (tied to input, really). The temporary power supply I was using had significant noise which totally disappeared when I ran the unit differentially (according to plan)
Petter
Pete,
I am suffering from home redecorat-itis so finding time to complete the project is hard. What I could do fairly quickly is get some distortion measurements for the input stage by itself, and I will attempt to get that info to you.
Still, the basic principle and design goal holds. Floating input --> floating output (tied to input, really). The temporary power supply I was using had significant noise which totally disappeared when I ran the unit differentially (according to plan)
Petter
Alternative to Active Current Sources
Petter,
Warning! Long post.
I just came across this thread the other day. For the last few years, I have also been playing around with this circuit (along with some of the Zen designs), and I have to congratulate you on your circuit. It is pretty much what I pictured the X topology to look like, so I would agree that you are on the right track.
I have long been a fan of Nelson Pass' amplifiers, and over the years, their sound has continued to improve. One of the things that stands out in his recent designs is their simplicity. Given the choice between two different implementations of a circuit, he seems to choose the simplest. I light of that, I have a suggestion for a simpler current source implementation; it is quite possible that Mr. Pass used simpler, passive current sources (i.e. resistors) in some of the circuit locations.
Referring for a moment to your original schematic, current sources Is3 and Is5 should remain active to provide adequate drive for the followers in the second stage. Is2 and Is6 would be the easiest to change to passive, since the relatively large (and constant) voltage across a resistor would provide a decent current source. Is1 and Is4 could potentially be changed to resistors, as long as you plan on using the output of the amp in its intended differential mode (more on this later). Using resistors for current sources Is1, Is2, Is4, and Is6, in combination with careful matching between transistor pairs (including the transistors for current sources Is3 and Is5), it is possible to use a single potentiometer to null the DC offset. Here is the proposed design process:
1. Knowing the Vgs of the matched pair of transistors to be used for Is3 and Is5, select a fixed voltage reference for the gates (connected together) and a fixed resistor value for the two source resistors, which results in the desired current through each transistor (50mA in your case); no adjustment necessary.
2. Select a resistor value for Is2 and Is6 that will result in approximately the same current as Is3 and Is5. Referring to your original schematic, the resistors will have approximately 46V across them, so 910 Ohm resistors would provide slightly more than 50mA through each of the two input transistors (Q1 and Q20). Note that it is not necessary for Is2 and Is6 to equal Is3 and Is5, respectively. As long as Is1= Is2+Is3, and Is4=Is5+Is6, the DC offset will be nulled.
3. Knowing the Vgs of the matched pair of transistors to be used for Q3 and Q19, create an adjustable voltage reference (with a single trim pot) for the gates of Q3 and Q19 (connected together). Next select a resistor value for Is1 (and Is4) that will result in a current equal to Is2+Is3 (or Is5+Is6) when the gates of Q3 and Q19 are at the nominal value of the voltage reference.
4. Power it up, and adjust the trim pot to null the DC offset.
Now the disclaimer. I have not actually tried this out (and probably won't in the near future, the way things are going at work). However, I did do some simulations, and as much as I hate using simulations without validating the results on actual hardware, the results were very encouraging. Starting with active current sources in all locations, I first converted Is2 and Is6 to resistors. Measuring the distortion of a 1KHz sine wave, the single-ended distortion (either output to ground) dropped by a factor of five (from ~0.5% to ~0.1%), and the differential distortion (across the two outputs) dropped about 1/10 (from ~0.02% to ~0.018%); so far so good. I next converted Is1 and Is4 to resisters; the single-ended distortion went up (from ~0.1% to ~0.8%), but the differential distortion actually decreased slightly (~0.018% to ~0.017%). So it looks like the distortion increase in each half was canceled out by an equal and opposite distortion in the other half (just what a balanced circuit is supposed to do). Not bad performance for a substantial simplification of the circuit. My only other comment is that I haven't done a sensitivity analysis of variations in resistance values, but it would be prudent to match each pair of resistors anyway.
If you decide to try any of the above modifications, please let me know how it comes out.
Dale
Petter,
Warning! Long post.
I just came across this thread the other day. For the last few years, I have also been playing around with this circuit (along with some of the Zen designs), and I have to congratulate you on your circuit. It is pretty much what I pictured the X topology to look like, so I would agree that you are on the right track.
I have long been a fan of Nelson Pass' amplifiers, and over the years, their sound has continued to improve. One of the things that stands out in his recent designs is their simplicity. Given the choice between two different implementations of a circuit, he seems to choose the simplest. I light of that, I have a suggestion for a simpler current source implementation; it is quite possible that Mr. Pass used simpler, passive current sources (i.e. resistors) in some of the circuit locations.
Referring for a moment to your original schematic, current sources Is3 and Is5 should remain active to provide adequate drive for the followers in the second stage. Is2 and Is6 would be the easiest to change to passive, since the relatively large (and constant) voltage across a resistor would provide a decent current source. Is1 and Is4 could potentially be changed to resistors, as long as you plan on using the output of the amp in its intended differential mode (more on this later). Using resistors for current sources Is1, Is2, Is4, and Is6, in combination with careful matching between transistor pairs (including the transistors for current sources Is3 and Is5), it is possible to use a single potentiometer to null the DC offset. Here is the proposed design process:
1. Knowing the Vgs of the matched pair of transistors to be used for Is3 and Is5, select a fixed voltage reference for the gates (connected together) and a fixed resistor value for the two source resistors, which results in the desired current through each transistor (50mA in your case); no adjustment necessary.
2. Select a resistor value for Is2 and Is6 that will result in approximately the same current as Is3 and Is5. Referring to your original schematic, the resistors will have approximately 46V across them, so 910 Ohm resistors would provide slightly more than 50mA through each of the two input transistors (Q1 and Q20). Note that it is not necessary for Is2 and Is6 to equal Is3 and Is5, respectively. As long as Is1= Is2+Is3, and Is4=Is5+Is6, the DC offset will be nulled.
3. Knowing the Vgs of the matched pair of transistors to be used for Q3 and Q19, create an adjustable voltage reference (with a single trim pot) for the gates of Q3 and Q19 (connected together). Next select a resistor value for Is1 (and Is4) that will result in a current equal to Is2+Is3 (or Is5+Is6) when the gates of Q3 and Q19 are at the nominal value of the voltage reference.
4. Power it up, and adjust the trim pot to null the DC offset.
Now the disclaimer. I have not actually tried this out (and probably won't in the near future, the way things are going at work). However, I did do some simulations, and as much as I hate using simulations without validating the results on actual hardware, the results were very encouraging. Starting with active current sources in all locations, I first converted Is2 and Is6 to resistors. Measuring the distortion of a 1KHz sine wave, the single-ended distortion (either output to ground) dropped by a factor of five (from ~0.5% to ~0.1%), and the differential distortion (across the two outputs) dropped about 1/10 (from ~0.02% to ~0.018%); so far so good. I next converted Is1 and Is4 to resisters; the single-ended distortion went up (from ~0.1% to ~0.8%), but the differential distortion actually decreased slightly (~0.018% to ~0.017%). So it looks like the distortion increase in each half was canceled out by an equal and opposite distortion in the other half (just what a balanced circuit is supposed to do). Not bad performance for a substantial simplification of the circuit. My only other comment is that I haven't done a sensitivity analysis of variations in resistance values, but it would be prudent to match each pair of resistors anyway.
If you decide to try any of the above modifications, please let me know how it comes out.
Dale
Spice simulators
I have used both PSpice and Electronic Workbench on this circuit in the past. I don't trust the Electronic Workbench models as much, but that is what I used for this most recent experiment because I didn't have much time, and I find that tool faster to use for simple "what if" experiments.
As a personal preference, I generally use simulation only for investigating the behavior of subcircuits rather than entire designs, since the more complex the circuit, the less likely the simulation results will match reality. In this case I only simulated the first stage (everything except the followers).
Dale
I have used both PSpice and Electronic Workbench on this circuit in the past. I don't trust the Electronic Workbench models as much, but that is what I used for this most recent experiment because I didn't have much time, and I find that tool faster to use for simple "what if" experiments.
As a personal preference, I generally use simulation only for investigating the behavior of subcircuits rather than entire designs, since the more complex the circuit, the less likely the simulation results will match reality. In this case I only simulated the first stage (everything except the followers).
Dale
I have chosen not to post full schematics etc. on this page out of respect to NP. Nevertheless, there is enough information on these pages to do the required reverse engineering. The conceptual schematic on page 1 is what you need to emulate -- of that I am quite confident.
Another reason for not posting is that I have been to busy to complete but the input stage -- and therefore not proven even to myself that the sound is good and the circuit stable. Now, this will be changing sooner rather than later -- which probably means Xmas
Petter
Another reason for not posting is that I have been to busy to complete but the input stage -- and therefore not proven even to myself that the sound is good and the circuit stable. Now, this will be changing sooner rather than later -- which probably means Xmas
Petter
Mail has gone out to DIY Audio
http://www.diyaudio.com/projects/petter
Seems like Nelson Pass has put an OK stamp on going public with the schematic in the X-100 "reverse engineering" thread. Very cool of him. However, posting before testing the complete unit is not really optimal. However, feedback would be great, and I hope you will understand that I am slow in completing the project because I am very busy at work. ETA for completion is presently this calendar year.
Trim such as Zener gate protection is not shown (and indeed not used ... but for output stage). There are hardly critical component values, but watch out for current sources (which are set up for 30mA bottom and 60mA top – again not critical by themselves, but need to work together). There is, however a delicate balance.
Input stage built, and it works. Modify Rfb/2Rinp for gain. Set up voltages to suit required power, particularly at output stage. Play with Re to find optimal value. Change output stage source resistors to suit your taste and output FET matching level. Set up bias in output stage by twiddling Rbias. Balance circuit by adjusting triplets which should ideally be set up for the actual current source FET’s used (do the balance before you connect output stage …).
I am using radically high drop voltages for current sources with the aim of improving performance, electrical stability with thermal variations, improve PSRR etc. I suspect some will find this tradeoff somewhat over the top. My usage of triplets is also unusual, but using the Excel spreadsheet I sent you last time should make it easy for those who wish to optimize usage of trimmers in this fashion (which I believe is useful given the sensitivity of the circuit to balance), but other methods are also possible (dare I say series connection ...)
Thanks again for your support
Petter
[Edited by Petter on 09-10-2001 at 10:48 AM]
http://www.diyaudio.com/projects/petter
Seems like Nelson Pass has put an OK stamp on going public with the schematic in the X-100 "reverse engineering" thread. Very cool of him. However, posting before testing the complete unit is not really optimal. However, feedback would be great, and I hope you will understand that I am slow in completing the project because I am very busy at work. ETA for completion is presently this calendar year.
Trim such as Zener gate protection is not shown (and indeed not used ... but for output stage). There are hardly critical component values, but watch out for current sources (which are set up for 30mA bottom and 60mA top – again not critical by themselves, but need to work together). There is, however a delicate balance.
Input stage built, and it works. Modify Rfb/2Rinp for gain. Set up voltages to suit required power, particularly at output stage. Play with Re to find optimal value. Change output stage source resistors to suit your taste and output FET matching level. Set up bias in output stage by twiddling Rbias. Balance circuit by adjusting triplets which should ideally be set up for the actual current source FET’s used (do the balance before you connect output stage …).
I am using radically high drop voltages for current sources with the aim of improving performance, electrical stability with thermal variations, improve PSRR etc. I suspect some will find this tradeoff somewhat over the top. My usage of triplets is also unusual, but using the Excel spreadsheet I sent you last time should make it easy for those who wish to optimize usage of trimmers in this fashion (which I believe is useful given the sensitivity of the circuit to balance), but other methods are also possible (dare I say series connection ...)
Thanks again for your support
Petter
[Edited by Petter on 09-10-2001 at 10:48 AM]
About the sound of the Pass X series:
Yesterday I visited the Top Audio Show in Milan Italy. IMHO the best sounding two channel system in the show was powered by an X0 and a X250.
I am talking about something I would want to own and listen to all day long. Since I didn't listen to the top of the line like the big Dynaudio, Madrigal, etc, etc, etc, I don't exclude there may be something better out there.
The speakers were the new ProAc Future Point Five with a ribbon tweeter of unknown origin and what appeared to be a 3.5 in Seas Excel with the copper plug for the mids and a Scan Speak 7 inch driver for the bass (the one with the cone made of paper treated with carbon model W18-something).
The source was analog, the usual 300 pounds of machined aluminum that I know nothing about.
I touched the HUGE heatsinks on the amp and they must have been at no more than 45-50 C.
Good job NP!!
Yesterday I visited the Top Audio Show in Milan Italy. IMHO the best sounding two channel system in the show was powered by an X0 and a X250.
I am talking about something I would want to own and listen to all day long. Since I didn't listen to the top of the line like the big Dynaudio, Madrigal, etc, etc, etc, I don't exclude there may be something better out there.
The speakers were the new ProAc Future Point Five with a ribbon tweeter of unknown origin and what appeared to be a 3.5 in Seas Excel with the copper plug for the mids and a Scan Speak 7 inch driver for the bass (the one with the cone made of paper treated with carbon model W18-something).
The source was analog, the usual 300 pounds of machined aluminum that I know nothing about.
I touched the HUGE heatsinks on the amp and they must have been at no more than 45-50 C.
Good job NP!!
PSU and box completed
I built the PSU capacitor section yesterday. I think some of you will find the information on the capacitor bank section of interest.
In short, it consists of 9 pieces of 10.000uF/100V/105degC high quality caps. They are connected through holes and spacers to two .5mm copper parallell plates. Seen from one side it looks like this, and there are 3 rows behind. Should be a fairly good low impedance connection Periods used to force spaces, sorry.
___________________________+.CU plate
___________________________-.CU plate
+...-.....+...-......+...-
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
Seen from the top it would be somehing like this:
_______________________
|.......................| |
|+...-....+...-...+....-|
|.......................|
|.......................|
|+...-....+...-...+....-|
|.......................|
|.......................|
|+...-....+...-...+....-|
|.......................|
Now, where is the ground you may ask. Well, there isn't one, at least not yet. I am planning to use a derived phantom ground and more on this will follow.
So, I have 90.000uF -- not a great deal, really but not totally shabby. That is until you consider that the stored energy given that I am running rail to rail instead of rails to ground is 4 times that and at a quarter of the impedance as well. So it looks like 360.000uF when compared to a regular setup. In effect the whole thing is not 4 or 16 times better by any means since I am using higher voltage caps than would otherwise be necessary, and lower voltage caps have "better numbers" in everything but series inductance and resonant frequncy.
Bear in mind, this trick only works for "inherently bridged" amps such as the X-series. This in non-tech terms means that the ground is not used as one of the outputs of the amp as is traditional.
Petter
[Edited by Petter on 10-02-2001 at 09:39 AM]
I built the PSU capacitor section yesterday. I think some of you will find the information on the capacitor bank section of interest.
In short, it consists of 9 pieces of 10.000uF/100V/105degC high quality caps. They are connected through holes and spacers to two .5mm copper parallell plates. Seen from one side it looks like this, and there are 3 rows behind. Should be a fairly good low impedance connection
Periods used to force spaces, sorry.___________________________+.CU plate
___________________________-.CU plate
+...-.....+...-......+...-
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
C A P.....C A P......C A P
Seen from the top it would be somehing like this:
_______________________
|.......................| |
|+...-....+...-...+....-|
|.......................|
|.......................|
|+...-....+...-...+....-|
|.......................|
|.......................|
|+...-....+...-...+....-|
|.......................|
Now, where is the ground you may ask. Well, there isn't one, at least not yet. I am planning to use a derived phantom ground and more on this will follow.
So, I have 90.000uF -- not a great deal, really but not totally shabby. That is until you consider that the stored energy given that I am running rail to rail instead of rails to ground is 4 times that and at a quarter of the impedance as well. So it looks like 360.000uF when compared to a regular setup. In effect the whole thing is not 4 or 16 times better by any means since I am using higher voltage caps than would otherwise be necessary, and lower voltage caps have "better numbers" in everything but series inductance and resonant frequncy.
Bear in mind, this trick only works for "inherently bridged" amps such as the X-series. This in non-tech terms means that the ground is not used as one of the outputs of the amp as is traditional.
Petter
[Edited by Petter on 10-02-2001 at 09:39 AM]
Active current sources
Petter,
do you still have fun with your X-Amp design? What is the current status?
I followed the thread for a while because I'm playing around with the idea to build a small version of a x-amp to drive a
wide-band speaker (100Hz-30KHz). Based on some experiences I made designing my balanced line stage I will try to use
current sources based on LED's and bjt transistors. The reason for this is, that both devices have a very close temperature
coefficient and such a current source is very stable with changing temperature. The other nice feature is that LED's produce
very less noise compared to typical zener diodes. The bad thing is that I have to use a lot of them in series to get enough voltage drop.
Right now I'm playing around in Pspice and Hspice to find the best current values and to decide where to use active current sources.
I agree with Dale, that it could be a good alternative to use resistors in some places instead of active current sources.
I will also use a constant voltage source for adjusting the bias of the output follower stage. It will have some temperature
compensation in it to keep the bias current independent from the MOS-temperature.
If the simulations are finished I will post the schematic here.
Good luck,
Maik
Petter,
do you still have fun with your X-Amp design? What is the current status?
I followed the thread for a while because I'm playing around with the idea to build a small version of a x-amp to drive a
wide-band speaker (100Hz-30KHz). Based on some experiences I made designing my balanced line stage I will try to use
current sources based on LED's and bjt transistors. The reason for this is, that both devices have a very close temperature
coefficient and such a current source is very stable with changing temperature. The other nice feature is that LED's produce
very less noise compared to typical zener diodes. The bad thing is that I have to use a lot of them in series to get enough voltage drop.
Right now I'm playing around in Pspice and Hspice to find the best current values and to decide where to use active current sources.
I agree with Dale, that it could be a good alternative to use resistors in some places instead of active current sources.
I will also use a constant voltage source for adjusting the bias of the output follower stage. It will have some temperature
compensation in it to keep the bias current independent from the MOS-temperature.
If the simulations are finished I will post the schematic here.
Good luck,
Maik
Status + ideas
Thanks for your input.
I have not finished yet as I have spent a lot of time on redecorating. That will soon end, though.
Current status: I have redesigned the entire board into a universal board (Eurocard size 100*160mm). This board is intended to be used with any input devices (say 4 JFET's in paralell per leg, or 6H30P tube [inherently matched ...]) and using large components such as Holco resistors + much more heatsinking. It is also intended to be used to build low power X-Zen's and preamps. The board is easier to build than the previous one which was just way too cramped -- you can have too much of something good!
I have given some thought to a Vbe multiplier in the voltage source to bias the output stage to have a negative tempco, and I might implement this at a later date. For now, I will try the old method. Other methods of implementing a negative tempco output stage bias would be appreciated as I am not happy with Vbe/Vgs multipliers per se ...
I have also made a layout for a PassDIY Pearl with 8 input devices, 4 output devices and more PSU filtering that fits on a Eurocard. I expect I will finish this project first, and plan on using Teflon boards and end up with 4 times the voltage gain of the original.
Petter
Thanks for your input.
I have not finished yet as I have spent a lot of time on redecorating. That will soon end, though.
Current status: I have redesigned the entire board into a universal board (Eurocard size 100*160mm). This board is intended to be used with any input devices (say 4 JFET's in paralell per leg, or 6H30P tube [inherently matched ...]) and using large components such as Holco resistors + much more heatsinking. It is also intended to be used to build low power X-Zen's and preamps. The board is easier to build than the previous one which was just way too cramped -- you can have too much of something good!
I have given some thought to a Vbe multiplier in the voltage source to bias the output stage to have a negative tempco, and I might implement this at a later date. For now, I will try the old method. Other methods of implementing a negative tempco output stage bias would be appreciated as I am not happy with Vbe/Vgs multipliers per se ...
I have also made a layout for a PassDIY Pearl with 8 input devices, 4 output devices and more PSU filtering that fits on a Eurocard. I expect I will finish this project first, and plan on using Teflon boards and end up with 4 times the voltage gain of the original.
Petter