X100 backengineered here

X100 backengineered


I am very interested in feedback on this design (which is based on patents held by http://www.passlabs.com and thus not available for anything but research). To scale power, simply change rail voltages (and increase value of feedback resistor to get enough front-end voltage gain to drive outputs to saturation). To change level of Class A drive, increase bias resistors. To improve "supersymmetry" and reduce gain, increase value of resistor between bottom of input FET's.

The 1M resistors are there for simulation purposes and would ordinarily be smaller. And yes, I simulated this design using a Spice CAD package.

We are probably looking at something very close to a commercial X-series amp (without embellishment). If you like feedback, place the feedback loop around the output transistors.

[Edited by Petter on 02-21-2001 at 07:52 AM]
Hello all,

I too had problems seeing the picture. I signed in using my ordinary hotmail-account but still got up a blank page. When I joined the community (note: "Stax OTL DC coupled amp design (fun with fibrillation)", created a new member) it all started working, et voilà, there it was.

Petter: I'll look at it later, but I guess there are people on this forum that can give you more intelligent input than I can since I'm not an EE.


[Edited by Matti on 02-21-2001 at 02:48 AM]
Got the image up, and some answers

Finally managed to get the image to show (i.e got the file hosted on this server).

I simulated using Circuitmaker which is a great piece of software -- so easy to use (and they do have a demo version available at http://www.microcode.com). The frequency response, I eyeballed to 30KHz or so -- it would be higher with fewer output transistors (number of output devices is somewhat arbitrarily chosen -- that is I did some quick calculations but no deep analysis, and to allow for future expansion if I need more power), or feedback placed at speaker terminals. A particularly cool feature about this design is that it scales so well -- hardly any changes other than PSU voltages are required for different power output level. Also note that output transistors need only be mathed in quad's when you cut to the bone.

The key to this design is accuracy of current sources. I strongly believe they should be FET type sources since any gate current will mess up the balance between halves. I am probably being a bit paranoid about this "base current" and the accuracy of current sources, but I really don't want an amp that is going to need adjustment other than for quiescent curren in output stage. I have two basic designs ready for this, one relying on significant voltage drop (degradation), the other on active device physical parameters. Op-amp based versions would work, but I don't like them very much.

What is missing is ajustment for quiescent current (now shown as fixed resistors), gate protection for transistors (zeners) and schematics for current sources. Note that the design requires balanced input signal like the original. It is very easy to make an unbalanced to balanced input stage and improve performance (frequency response etc.). What is also missing is compensation to control phase margin at very high frequencies.

I am indeed planning to build this design, but will tinker with voltage levels, number of transistors and resistor values, particularly the one between input transistors, as well as feedback resistors (gain selection). As I will be needing a single-ended to balanced front-end anyway, I will probably change these values to reflect the gain I can get out of this input stage as well.

As a concequence, I am very interested in feedback whch you can post here or send to me by mail.
As a non-EE who briefly thought of doing something very similar, but soon realised it was beyond me, I'm very impressed!

Have you tried the values for the 'virtual ground' resistors cited in the patent, and was there a particular reason for using 1M/5k for the purposes of the simulation?

Don't know whether NP will be giving his usual support to this one!
Well, anyone who was going to exploit this commercially would have the skills and tools to do what Petter has done in any case. So I can't see any problem, I just meant that NP couldn't be expected to help Petter along in his usual generous manner.

I'm sure your Karma is fine, and you aren't risking coming back as an ant or something.


I am sure your Karma is fine. There really is nothing new in this image other than what is available on the patent (linked to from the Pass site), and schematics freely available on the Pass site.

Reason for 1M resistors is that you need a ground to run a simulation. I wanted to make sure that there was noe effect of the ground, and put 1M down for starters. Should probably be 100K instead.

The other resistor you refer to is the combination of the gate input resistor and the one connected as feedback resistor. The ratio of those determines gain (R40/R39 for each half).

The other resistor in question R43. This resistor if zero implies unimpeded gain. The higher this value, the lower the gain. I cannot remember what the virtual ground resistors you refer to (from patent) were, but I am totally convinced that ground resistors in combination R43 are useless and unnecessary and that the Patent is not totally correct with respect to real life use. You want ground where I have placed it, near input MOSFET gates.
Hi again Petter,

Looked again at the patent last night, and I see where my confusion stemmed from, as pointed out it is the shift of the ground from between the source pins to between the gates. Would you mind expanding on the reason for so doing, in what way doesn't the patent reflect real-life (as I believe you said)?

I'm not disputing this by the way, just interested (and I might not understand the explanation for a few years, but I'm still intrigued!).

The values for those resistors in the patent are 300 Ohms (to ground), and 47 Ohm (coupling resistor), by the way.

The rest of the schematic looked (perhaps unsurprisingly) right on the mark - but then, of course my opinion is limited by ignorance. It is actually a remarkably simple circuit, no? One of my mistakes when I looked at this was to presume that the circuit used had a complementary symmetry input as outlined later in the patent, but I see from re-reading the literature that it ain't that involved at all.

Anyway, cheers Petter, there are lots of us rooting for you I suspect.
Thanks for the kind words. If you separate out the input section from the output, you will see that the amp is actually not that complicated at all. Think about it like this: Assume voltage at gate is zero. Assume gate draws no current. Place 1V at one gate. Current flow is 1V/R39 towards gate, right? This current must flow through R40 and makes the voltage at the other end of R40 equal to 0V - R40*this current.

Now to input impedance. i have said that R40/R39 is the limit to how much voltage gain you can have for each half. If the voltage at the gate is zero (reads ground). What then is the input impedance? R39 or 5K per half -- that is 10K in balanced mode. Rings any bells from X specifications???

The patent with 2*300 Ohms to ground + 47 across is equivalent to 47 in paralell with 600 Ohms. between sources and about 175 or so Ohms to ground. The advantage of putting something to reference sources instead of gates is that you don't draw current to ground from signal directly -- all current flows towards gate -- the same is essentially taken care of if you use 100K resistors, and you can accept a common mode error voltage without loading down the input too much. Very nice.

The problem I have with the patent grounding scheme is that all this grounding business kind of messes with the current sources operation -- which I would like to be pure and never managed to simulate successfully anyway. You don't need current to flow to ground (which you would get because the gates are probably 3+ volts higher in potential (and be the same at the output unless you mess with where you take the feedback from). What I believe you want is current flowing in the source resistor.

This section of the patent had me scratching my head ever since I saw it. When I looked at the preamp schematics available on the Passlabs site, it became very clear -- it is not done the patent way at all.

So what I have done is to set the mean voltage at the gate. Thus, the voltage at the sources and output will sort themselves out, and the voltage at the output will be zero which is what we want.

The circuit is simple. It is in fact very similar to certain op-amp circuits (look in patent) and there are many application notes available from op-amp manufacturers on such op-amp connections. The only thing you need to worry about is the balance of current sources -- IS1=IS2+IS3, then everything will fix itself -- if it is precise which is easy to do. In fact, if you were to place the input at +10V average, it would still work without a hitch (I like current sources!). R43 will handle some of the imbalance.

It is interesting that Nelson Pass has progressed from relatively complex circuits to rather simple ones. I own a Stasis 3 which I quite like, but it is a little complex albeit logical.

I will be buying components next week, but don't hold your breath for me to come up with pictures etc. as I am really busy at work + have some hectic travel plans. At some point I guess I should post pictures of how the project is progressing.
There should be no major problem with the amp you designed. Since I've just finished building my own headphone amp with similar topology to yours (without the folded cascoding) and I used single-ended follower rather than push-pull), I have a few point to be careful about. Vgs difference between IRF610 can be quite significant ranging from 3.8 - 4.1 Volt at 50 mA. This causes DC offset ebwteen the drain pins of the two half (which will appear across the speaker finally). Not only that, the rising temperature will causes drift of Vgs which is unlikely to be the same rate between the two half. I chose to couple the two stages with capacitor because even 50mV DC offset is quite serious for headphones. In the worst case, the offset used to drift between +/- 200mV. The less-than-exact current source can also cause this offset variation and the current can also be drifted with temperature as well. Just a note.
I see from the various owners manuals that the later (and lesser powered) X amps use dual JFET inputs to ease the matching problems. On other hand this project was always going to call for very close matching of the transistors, and presumably that shopuld be carried out at something close to their eventual expected operating temperature.

Re getting the current sources balanced, would it possible to use current mirrors to force equalisation, Petter?
Thanks Namui


I am not sure that mismatching of the input FET's will cause DC at output per se. The current sources and feedback resistors should still handle the DC aspects because they will "automagically" set voltages to what they need to be in order to have correct current flowing (or adjust to the required gate voltage if you will). The gate voltage is tied to input voltage since no current flows through gate.

However, in order to have good AC performance (and achieve some level of noise cancellation which is the point) I was always going to match very tightly. At about 50 cents a piece, and only requiring matching in pairs, this should be no problem.

However, unless current sources are designed properly, mismatch of Vgs can cause them to be inaccurate and this shold have a profound effect on output DC voltage.

If I am wrong, don't sue me.
Thanks Jakeh

The later and lowered powered X models use JFET's but they also accept unbalanced inputs. The way I read this is that they have an extra gain stage or SE to balanced converter, probably with some gain at input (reads classical long-tailed pair). If you check out my "open source electrostatic speaker paper in my file cabinet page you will see what I am talking about. Basically classical long-tailed pairs.


In order to get high current going through the stage (to allow for high current drive at output), JFET's really don't cut it. If they do, you need to do a lot of cascoding etc. to get them to work without blowing up since their power-rating sucks. However, it is possible by cascoding -- but either way you are putting extra semiconductors in the signal path.

So base on all this, I would say that the JFET's are not running in Super Symmetric mode -- because they would then not be able to handle single ended inputs. Because there is some gain at this extra input stage, you are not quite so sensitive to matching, but still need to match up pairs.

As I recall, in the commercial units, the input devices are connected to the same heatsinks to maintain the same temperature for both.

Curren mirrors: Good idea. The critical matching is for top source to equal sum of bottom sources for each half. That they also match between halves is not critcal, but probably smart. Now the problem of having current sources matched at different voltage levels can be solved with mirror's I suppose but I would have to look at it and probably look up Horowitz & Hill's Art of Electronics. I think I have seen other methods of creating tracking sources as well in there.

Now, I had planned to use very high levels of degradation, possibly even in addition to physical parameter methods(reads BJT's on the control side) and then trim out any errors. Since there is not much power being lost in this segment, I have significantly increased the supply voltages of the input stage to allow for the simple method. Op-amps and precision resistors should be able to do the same, and the SuSy resistor helps a little too. One final method to do it is to use precision current sources from companies like Burr Brown (now TI) or Analog Devices to set up currents. You will find white papers on doing this kind of thing on their pages http://www.analog.com and http://www.burr-brown.com

I am beginning to think we are getting a little paranoid about this. However, I am thinking about putting a resistor between output terminals of say 50-500 Ohms to set up another current path, at least for testing purposes.