The ultimate DIY? A Stax SRM-T2!

[Pars]

^ Begin with a "P"?

[kevin gilmore]

yup

[Linear]

Hello everyone,

I'm new here, and I have spent the last few days reading this thread. It certainly was interesting!

My congratulations to all involved in this project, especially KG and Spritzer for starting it all, and making the first unit.

I think that Craig Sawyer also deserves special mention. His perseverance in solving the "defective (low hfe) 2sc3675" problem was inspiring!

Hopefully, I will be able to make some small contribution to all that has gone before, now that I am a member of Head Case.

Best regards to all,

Linear

[kevin gilmore]

My T2 is just about 1 year old now.

The hundreds of hours of joy it has given me

and will continue to give me in the future is

something i will remember forever.

I just don't see any way to improve on the

design. Maybe ways to make it a bit cheaper

and easier to build, but doing anything

to increase the slew rate, or further lower

the distortion does not seem to be possible.

[Linear]

Craig,

Your link, quoted below, seems to be broken. Perhaps it's because you moved to a new ISP?

Could you post a new link where I could look at your "block diagram" of the T2?

Thanks for your help.

Best regards,

Linear

Bored and twiddling my thumbs waiting for semiconductors to arrive to fix the zapped T2, I thought I'd try to get a grip on what the actual signal level circuit was doing.

I've replaced all the current sources and 740V battery with ideal symbols. I've also left out compensation components, and some of the details of one or two current sources which I think are error handling, to prevent the thing blowing up if a tube goes down badly. I think the result shows the rather neat symmetry of the design, and the clever way in which DC voltage shifts are generated using the FETs and batteries to move the ground referenced input signal to -500V.

File is here http://www.tech-ente...deconstruct.pdf

[Fitz]

I just don't see any way to improve on the

design.

What about water resistance?

[Craig Sawyers]

Craig,

Your link, quoted below, seems to be broken. Perhaps it's because you moved to a new ISP?

Could you post a new link where I could look at your "block diagram" of the T2?

Thanks for your help.

Best regards,

Linear

OK - just restored it at http://www.tech-enterprise.com/tekstuff/T2deconstruct.pdf

Anyone else need any files that I have supplied links for that are now dead, let me know and I restore them. I just don't have a record of what was stored, because my half-assed ex-ISP corrupted the damned folder when they had the fatal crash. Complete wasters that aren't worth the desk they sit at.

[Linear]

OK - just restored it at http://www.tech-ente...deconstruct.pdf

Thanks, Craig, for posting the T2 diagram.

After looking at this drawing, a question comes to mind:

Is the beautiful sound of the T2 determined almost completely by this topology, the 4 tubes, and the 6 MOSFETs? How critical are the current sources?

In other words, if we built the current sources and the 740v battery using "modern" (ie different) components, would it sound significantly different?

Comments anyone?

[jcx]

...I just don't see any way to improve on the

design. Maybe ways to make it a bit cheaper

and easier to build, but doing anything

to increase the slew rate, or further lower

the distortion does not seem to be possible.

ghosh, thank you Buddha Gilmore for choosing the path of a bodhisattva when as an Enlightened One you can just step off of the wheel...

as someone liking beer and barbeque too much to sit around staring at a single grain of rice all day long I would try some tricks from Hawksford's Slope Distortion Reduction paper on the ccs, along with the current modulation trick I've already shown

but what I'd really do is add lots of loop gain...

[Craig Sawyers]

but what I'd really do is add lots of loop gain...

Why not build the T2 (instead of endless spice analysis an what-if's) and then carry out mods using real components so that you can evaluate them by actual measurements and (shock, horror) what it actually sounds like?

I've got precisely this problem on my major project at the moment, where the subcontract engineering company insists on only using analysis tools, and pooh-poohing measurement. This is a system for lauch to Mercury in 2014 - and they are studiously missing the point that you don't lauch analysis - you launch actual hardware that must not fall to pieces when subjected to launch loads. They are in the last chance saloon at the moment on this topic.

So, jcx - lets see some building and measurements. And does it sound like audio nirvana or a bag of hammers?

[kevin gilmore]

Well i'm 30+ lbs LESS buddha than i was when i started this project 19 months ago.

You can be pretty sure that if you do a spice analysis, and the results are poor, then

if you actually build and test the device, it will be poor or worse. But a good spice result

does not necessarily translate to a wonderful sounding amplifier.

The ixys10m90s devices sound just fine, but blow up at about 925 volts. Rather consistently.

Anything else at 1000 volts does not sound as good. Replacing the 2sc3675's with 2sc4686's

does in fact work fine, and sound the same but you have to turn up the bias currents a bit.

Same thing with the active batteries.

The open loop gain is plenty high already. Anything over 15db of headroom seems to make

lots of different amplifiers sound bad. Many of the portable opamp based toys with 50db+

of extra gain are a good example. The slew rate is more than sufficient with a rise time

of under a microsecond large signal.

Replacing the input tubes is still open to discussion.

Edited March 6, 2011 by kevin gilmore

[Kerry]

Well i'm 30+ lbs LESS buddha than i was when i started this project 19 months ago.

...

Congratulations

[Craig Sawyers]

Well i'm 30+ lbs LESS buddha than i was when i started this project 19 months ago.

Wow - well done sir!

You can be pretty sure that if you do a spice analysis, and the results are poor, then

if you actually build and test the device, it will be poor or worse. But a good spice result

does not necessarily translate to a wonderful sounding amplifier.

The only problem is tracking down spice models for the typical transistors we use in audio - particularly (but not uniquely) of Japanese origin. You can get quite a long way by writing a simple spice description file for the major transistor parameters from the spec sheet. Good enough for DC and low frequencies where the various capacitances can be ignored. There is enough on the web describing how to do this. But a complete Gummel-Poon model needs quite a bit of measurement/guesstimation to put a full description in place if the manufacturer has not made a full Spice model available. Plus the fact that there is a large spread in parameters typically, so to do the full nine yards you need to do a sensitivity analysis over the parameter spread. Not counting layout parasitics......

I did the simple description route to work out a foolproof method for how to set the battery voltage. The graphs (Excel spreadsheet) used to be on my webspace, but they probably got trashed in the ISP debacle. If anyone has a burning desire to get these, let me know and I'll upload them again.

[Linear]

Well, I may be late to the party, but I've decided to take the plunge and build a T2! I've just ordered some 2sc4686a parts from Mouser for what seemed to be a good price ($0.82 ea). Can any of you give me a lead on where I can get 2sa1486, 2sj79, 2sk216, and 2sc3381 at a better price than B+D? They seem to be expensive ($2.82 for a 4896a!).

I guess that I'm too late for "group buy" PCBs, heat sinks, etc, so I'll do the best I can on my own.

T2 fever is very contagious!

Best regards,

Linear

[deepak]

Well, I may be late to the party, but I've decided to take the plunge and build a T2! I've just ordered some 2sc4686a parts from Mouser for what seemed to be a good price ($0.82 ea). Can any of you give me a lead on where I can get 2sa1486, 2sj79, 2sk216, and 2sc3381 at a better price than B+D? They seem to be expensive ($2.82 for a 4896a!).

I guess that I'm too late for "group buy" PCBs, heat sinks, etc, so I'll do the best I can on my own.

T2 fever is very contagious!

Best regards,

Linear

Just curious if you have the T2 pictured in your avatar? If so looking forward to hearing what you think on the DIY-T2 vs Stax T2.

[Linear]

..........

change out the fet and led's in the batteries for the precision regulator and a resistor, remove one adjustment.

might have to fiddle a bit with the range on RV1.

Neither should make any change in the sound whatsoever. But its not better.

..........

No practical way to replace the active batteries with a tube. Gas tubes

have too high an impedance.

Since I'm going to be using 2sc4686a devices in my T2, the active battery circuit will have to be modified. So, I thought that I'd take a look at re-designing it with a "modern" reference. The original JFET reference seemed cumbersome and "out of date". Well, it looks like Dr. Hayashi had a very good reason for using the JFET. Here's the deal:

Obviously, minimizing noise in the T2 is very important, and since the active batteries are in the signal path, their noise contribution can be critical. The reference voltage on the base of Q16 (in the battery circuit) is about 6.6 V, so it gets multiplied up by over 100X to generate 740 V. Therefore, the noise component in this 6.6 V also is amplified by 100X. So, let's compare the various reference types in nV/(root Hz) @ 10Hz:

Bandgap (eg TL431): 220

Burried Zener (eg LM329): 100

JFET Idss current: 16

22K resistor: 20

It looks like the JFET and resistor combination (25 nV/rt Hz when summed, I think) is 4X to 10X better noise-wise, versus "modern" technology!

I'll check this out on the bench, but all I can say for now is "hats off" to Dr. Hayashi!

Linear

[Linear]

Just curious if you have the T2 pictured in your avatar? If so looking forward to hearing what you think on the DIY-T2 vs Stax T2.

Yes, that is a real Stax SRM-T2 in my avatar, but I don't have one! I got the photo from a Japanese web site back in 2002. At that time I was lusting after a T2 or a schematic ( so I could build one). I even wrote to Stax in Japan, requesting a schematic, but they very politely declined my request. So, imagine my joy when I discovered this thread, where KG and Spritzer accomplished what I could not! Thanks again guys!

I did, however, get a copy a a review of the unit from the May 1995 issue of the Japanese magazine "MJ". That article has a very detailed block diagram of the T2 with the tubes, the 2sk216, and 2sj79 MOSFETs shown. The text, of course, is in Japanese, but I'll try to post it on the group here. Perhaps one of our members could translate it.

Linear

[jcx]

the Pass A40 might be considered an ancestor of SS modulated push-pull Class A bias, Pass still uses a variant of the modulated current source in his recent Class A amps

I previously showed splitting the "sense" R to get push-pull output with a fet ccs

this can easly be done with the T2 Vref+bjt ccs too

obviously the sims are just circuit sketches - and the V and Cloads are scaled down/upto keep currents ~ the same as T2 output ccs - the Q models are just what LTspice include

common to the T2 output ccs circuit: triple pass Q with 2n2222 models are really bad approximations of the 2sc3675, cascaded pnp ccs is approximated with the spice ccs

in the 1st sim I show a circuit with no more parts than already exist in the T2 output ccs but theoretically gives 2x output current swing - which could double slew rate for the same quiescent bias current setting (I assummed ~10 mA output bias)

the 1st circuit uses a higher Vref 12 V Zener to get some added "loop gain" (relax, its all local degeneration) - since the circuit modulates the bjt current we can expect distortions from Vbe/Ic and Vref bias varying when the pass bjt base currents change

does the 2x gain, doubled slew rate compensate for the now varying ccs Q Vbe, Ib caused errors? - someone would have to build and measure

the next two circuits use the same triple pass Q but use local loop feedback to control the current and may be seen as derived from the "ring of two" bjt feedback ccs

the feedback Q Vbe is the Vref for the circuit in the 2nd push-pull current source

the third uses Hawksford/Baxandall "Super Pair" to give cascode like cancellation of Zcb, and the extra current gain reduces the feedback Q/Vref bias variation gain - the feedforward increases the "Q" and the circuit will have RF problems - but so does the Cascode that is AC referred to the "control" Q emitter - I believe it can be tamed with Zobel like RF damping that has little effect on the performance improvements at audio frequency

I also added Zener Vref in the feedback Q emitter to allow higher current sensing R values so the feedforward is more effective (although the improvement may just be from the added current gain)

I probe the load C, Vsource currents and look at fft plot to get an idea of the circuit topology effects - all have falling harmonic level with increasing order - the current control feedback versions I(Cload)distortions start ~ 40 dB lower than the 1st circuit, Vsource current are what an "ideally" controlled output tube would have to do to make the V distortion =0 at this point, I(Vsource)is lowest with 3rd circuit

the third "kitchen sink" circuit's higher frequency harmonics fall faster with increasing order so some of Hawksford's "slope impedance distortion reduction" appears to be taking place in the sim

T2_pp_cs.asc.txt

Edited March 7, 2011 by jcx

[kevin gilmore]

In all 3 cases above, its the current source that is the problem.

Has to have 1000 volts compliance. This requires stacked pnp's

similar to the original circuit. And the capacitor (c1,c2,c3)

necessary to stabilize the circuit is going to affect the

quality of the current source at higher frequencies.

the devil is in the details. Build it and test at appropriate

voltages, then let me know.

The SiC devices work great on paper. In practice, there are

all sorts of other issues because they don't like to work

at low currents for example.

[jcx]

um, I said the 1 mA spice ccs ~= the T2 cascaded pnp ccs - I'm trying to show the circuit mods fit in the existing T2 output ccs topology

the push-pull operation of my 1st circuit seems to be one way to increase slew rate using the same number of parts as the existing T2 ccs

Edited March 7, 2011 by jcx

[Linear]

I did the simple description route to work out a foolproof method for how to set the battery voltage. The graphs (Excel spreadsheet) used to be on my webspace, but they probably got trashed in the ISP debacle. If anyone has a burning desire to get these, let me know and I'll upload them again.

Could you upload any active battery stuff that you have? I'm working on the battery right now and your thoughts would be helpfull. I'll report my findings in detail a little later, but here are a few preliminary observations:

As mentioned a few posts back, the JFET reference was used because it has very low noise.

Indeed, RV2 (in series with the JFET) acts as a coarse voltage adjust, while RV1 is a fine adjust, BUT....

RV1 also adjusts(and minimizes) the temperature coefficient of the JFET reference!

So, what is the Stax recommended procedure for minimizing this Tc ? Has anyone ever seen an owners (or service) manual with this info? Perhaps the reason that Stax built the original active batteries as "modules" (little daughter PCBs in a heat sink) was so they could put each in an oven, minimize the Tc with RV1, and then install them in the T2?

More to come on all this.

Linear

[Craig Sawyers]

Could you upload any active battery stuff that you have? I'm working on the battery right now and your thoughts would be helpfull. I'll report my findings in detail a little later, but here are a few preliminary observations:

I think that I put the models (simple text) on the thread a while back - check and see what is there

So, what is the Stax recommended procedure for minimizing this Tc ? Has anyone ever seen an owners (or service) manual with this info? Perhaps the reason that Stax built the original active batteries as "modules" (little daughter PCBs in a heat sink) was so they could put each in an oven, minimize the Tc with RV1, and then install them in the T2?

Nice thought - but I suspect not. The innards of the original T2 get incendiarly hot - nothing like as much heatsinking and air circulation as KG's clone and *that* gets quite hot enough! I think they just used a method similar to the one I worked out - again on this thread a way back.

[kevin gilmore]

if you look at the insides of the original, the 22k resistor was

mounted on the backside about .5 inch from the board so that

you could put clipleads on it to adjust.

[kevin gilmore]

posted on behalf of linear

http://gilmore.chem.northwestern.edu/t2schemip.pdf

http://gilmore.chem.northwestern.edu/t2schemop.pdf

http://gilmore.chem.northwestern.edu/t2schems&b.pdf

http://gilmore.chem.northwestern.edu/mjt2a.jpg

http://gilmore.chem.northwestern.edu/mjt2b.jpg

http://gilmore.chem.northwestern.edu/mjt2c.jpg

http://gilmore.chem.northwestern.edu/mjt2d.jpg

[Linear]

posted on behalf of linear

http://gilmore.chem....u/t2schemip.pdf

http://gilmore.chem....u/t2schemop.pdf

http://gilmore.chem..../t2schems&b.pdf

http://gilmore.chem....n.edu/mjt2a.jpg

http://gilmore.chem....n.edu/mjt2b.jpg

http://gilmore.chem....n.edu/mjt2c.jpg

http://gilmore.chem....n.edu/mjt2d.jpg

KG has kindly agreed to host some files that I thought would be of interest .

The PDF files are redrawn from KG's large "t2schem" schematic, arranged on 3 pages: IP, OP, and Servo & Battery. This effort was inspired partly as an aid to understanding the circuit, and partly by Craig Sawyers "simplified schematic" diagram. Like Craig's drawing, it seems easier to follow the signal path with this arrangement.

The JPG files are an article on the t2 from the May 1995 Japanese magazine "MJ". The only thing in it that I can "read" is the block diagram! Note that the +/- 600 V rails are not accurate, but the components and topology seem OK. It reminds me of Craig's "simplified schematic", which he reposted a few messages back.

So, can anyone out there HELP?!! It would be great to have a translation of the text. I know there are several "Head-Case" members who live in Japan - Please HELP us out if you can.

Hope that the above files are useful.

Linear