srx revisited

[kevin gilmore]

James lin came up with this, I turned it into a circuit board.

cascaded current sources et all.

 

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

 

http://gilmore.chem.northwestern.edu/srx6.jpg

 

http://gilmore.chem.northwestern.edu/boards/srx6.zip

 

I did make a few changes including balanced input

Edited December 6, 2014 by kevin gilmore

[G600]

Looks like a very nice way to use some 6SN7 !

[PretentiousFood]

Very cool. I build something similar with an EL34 output stage, and no tail CCS on the output, based on Spritzer's ES-1 rebuild thread.

 

The DN2540 wasn't super happy in the input LTP. If the second 12AT7 is to be biased at ~1/2 B+ (170V) with a 300K load, each triode is running at just over 500uA. A LND150 would be perfect there (and is pin-compatible with the DN2540 in TO92). Looks like you plan on running it hotter though.

Edited December 6, 2014 by PretentiousFood

[nopants]

if possible make the pads a little larger so I can get rid of some mundorf supremes

[kevin gilmore]

6sn7gta or 6sn7gtb only, otherwise arcy sparcy

 

how big you want the holes on the caps?

[nopants]

I think supremes are 1mm lead diameter, but I was just thinking out loud.

 

It might be beneficial to include holes for multiple lengths when there's AC coupling in general. Nothing crazy like what diyaudio enables, maybe just 2 pr 3 mafia-approved caps.

Something that came to mind while build the megatron

[kevin gilmore]

1mm is .039, and the holes are already .040

[JimL]

Thanks for posting this, Kevin.  I've submitted a full article to AudioXpress which I hope will be coming out in a few months.  It contains a very detailed analysis of the basic SRX circuit and the modifications I have made.  In the meanwhile just a few comments.

 

1)  the use of current sources in the output stage is very important.  With anode resistors in the output stage, the resistors are much lower impedance than the headphones themselves, whereas with cascoded current sources, the headphones are significantly lower impedance than the current sources.  To put it another way, they convert an amp for driving anode resistors into an amp for driving headphones.  Because the 6SN7 tubes are not nearly as high power as a EL34, this makes a significant difference.

 

2) My crude measurements of a cascoded current load puts the impedance at > 160 megohms, which means at 20 Hz about 2/3rds of the signal current is going to the headphones, and by 100 Hz, 90% of the signal current is going to the headphones.

 

3) all the current sources are adjustable.  In the input stage the current source should be set at about 1.3 mA, so that the plate of the upper 12AT7 is about halfway between ground and B+.  The output current loads should be set for 7 mA, and the output current sink should be set for about 17 mA.  The resistor string between B+ and the output current sink gives a pathway for the excess 3 mA of current difference between the current loads and the current sink.  The output current sink should be adjusted to zero the output plate voltage - anything less than 10 volts from ground is acceptable.

 

4) I'm currently running my unit with B+ at about 325 volts, B- at about -325 volts using a TL431 based shunt regulated power supply.  The schematic of that power supply will be in the article.  I know there have been experiments posted here using the TL431 for a regulator that commented on turn-on instabilities and oscillation, however I found a Texas Instruments application note on how to correct that problem, and my version seems to be stable. 

 

5) For the input stage I recommend ONLY 12AT7 tubes for the lower cascode tube, as substituting a 12AX7, for example, will decrease the gain of the input stage by about 15%.  However, substituting a 12AX7 or 5751 for the upper cascode tube will have negligible effect on the input stage gain or output impedance, and may alter the sonic flavor of the circuit.

 

6) The measured frequency response with my Fluke 189 meter was flat between 20 Hz and 20 kHz and -1 dB at 50 kHz at 20 volts RMS output.  With an additional 100 pf load (equivalent to a Stax SR007 or 009, plus the load of the meter) the high frequency response was -3dB at 46 kHz at 100 volts RMS output.

 

7) I built this in the spirit of a Stax Mafia Volksamp.  I appreciate the fact that Kevin Gilmore thought well enough of it that he spent his time and effort to design a really nice board for it.

 

Finally, I want to thank Kevin Gilmore for correcting a beginner mistake on the amp schemata.  I built it correctly, I just drew it incorrectly. )

[JimL]

A couple more comments:

 

1)  Since the SRX circuit is inherently balanced, it's easy to have a balanced input - just disconnect the grounded half of the input circuit and use it for the negative input.  For unbalanced input, just connect the negative input to ground.

 

2)  Haven't tried the LND150 for the input stage current sink, but my unit the DN2540 worked fine. I tried a J113 and 2SK170 as the lower device in the cascode with the DN2540 as the upper device but the DN2540 for both devices worked best.  The input current sink runs about 1.3 mA - 650 uA/section - I used the TO92 version rather than the TO220 that Kevin has on the board.

 

3) Let me re-emphasize that with B+ and B- > 300 volts, you CANNOT use 6SN7 or 5692 tubes.  Use 6SN7GTA or GTB tubes ONLY in the output unless you enjoy burning up expensive tubes.  The 6SN7 is specified at 300 volts max plate voltage and 5 watts combined plate dissipation, the 6SN7GTA/B are specified at 450 volts max plate voltage, 7.5 watts combined plate dissipation and 1500 peak positive pulse plate voltage.  Most of the new old stock (NOS) 6SN7s that are going for silly money are the original version.  NOS 6SN7GTA/B are still available at reasonable prices, and there are also new manufacture 6SN7GTAs.

 

4) I haven't tried this, but if you use 6BX7 tubes for outputs you could crank the power supply up to 400 volts, increase the output current loads to 10 mA and the output current sink to 23 mA - might have to upgrade the heatsinks though.  Frequency response should be about the same but more current drive for the headphones.

[nikongod]

Why not run the input tail off of B-? Maybe with a resistor to dissipate some heat. Saves the -20V power supply. 

 

Any thoughts on using the output CCS's in "combined CCS & mu-follower mode" (see link to Pimm Labs below - about half way down the page)? 

 

http://www.pimmlabs.com/web/Active_loads_and_signal_current_control.html

[nopants]

isn't using the hv rail like that similar to what the exstata does for its input stage? I thought there were drift problems or something

[kevin gilmore]

no current source in the input stage of the exstata

just resistors

[PretentiousFood]

Why bother with Mundorfs when you could be the proud owner of 11lb of teflon?

 

2)  Haven't tried the LND150 for the input stage current sink, but my unit the DN2540 worked fine. I tried a J113 and 2SK170 as the lower device in the cascode with the DN2540 as the upper device but the DN2540 for both devices worked best.  The input current sink runs about 1.3 mA - 650 uA/section - I used the TO92 version rather than the TO220 that Kevin has on the board.

 

When I simmed it, the LND150 showed marginally better linearity at high frequency, and a few 10s of dB better supply rejection-- probably nothing audible. They'd need to be selected though, since they only guarantee an Idss of 1mA. I'll see if I can't measure the two cascodes.

[JimL]

re: Nikongod:

 

"Why not run the input tail off of B-?" 

 

Kevin didn't show my power supply schematic, but that is exactly what I did - calculated the current draw for both input tail current sinks and used a resistor from B-. 

 

Re nopants: "isn't using the HV rail like that similar to what the exstata does...?"

 

The current source isolates the input tail from any variations in the power supply, which are minimal since the power supply is regulated.

 

"Any thoughts on using the the output CCS's in combined CCS and mu follower mode?"

 

Interesting idea, I hadn't noticed that before even though I've looked at Gary Pimm's CCS postings several times in the past.  Gary specifically mentions that it works well for high capacitance loads, and electrostatic headphones aren't really that high capacitance - around 100 pf give or take.

 

More importantly, though is that I wanted to maintain the "character" of the original SRX circuit.  As I mentioned in a previous post, a cascoded 10M90S/DN2540 has a very high measured impedance - I measured a DC impedance greater than 160 megohms, which is my measurement limit with the crude equipment I have.  By comparison, a single 10M90S current source measured the same way works out to about 170 kilohms DC impedance.  In comparison with all the other impedances in the circuit, the cascoded current sources are orders of magnitude higher impedance.  A stat headphone with a typical 100 pf impedance is j80 megohms at 20 Hz, and drops from there at higher frequencies.

 

What this means is that the current sources are almost completely out of the circuit as far as the music signal is concerned, all they do is optimize the function of the tubes as active devices.  So effectively all the output tube sees is the headphone, and all the headphone sees is the tube driver.  As I said, it converts an amp for driving anode resistors to an amp for driving headphones.

 

By the way, another advantage of having both current sources and sinks on the output stage is that they do a very good job of isolating the output tubes from the power supply and from the other channel.

 

Re: Pretentious fool:  My main concern with the DN2540 was whether it would work at currents around 1 mA, since most of the designs I seen using it run much higher currents.  Fortunately it seems to be stable at that current, the current didn't vary when I changed the voltage by 10-15 volts, and it's quiet.

[nopants]

thanks for the isolation info, I also came across something similar in pimms notes

[smitty1110]

Are there any plans to make some runs of these boards?

[Laowei]

Great project JimL. Lots to think about.

I was wondering how you arrived at the 100 ohm value for the grid stopper on the lower DN2545. I have used cascaded CCS's on tube plate loads in the past and most of the schematics used >1K.

Thinking of adding a lower DN2545 cascade to the 10M90 CCS to my KGST. Wondering what value GS to use. Do you recommend the 100 ohm?

[luvdunhill]

121-332 is what I have used. I think it doesn't really matter as long as the result is stable.

[Laowei]

Thanks!

[JimL]

Re:  grid stopper for the cascode CCS (BTW, the spell checker seems to substitute cascAde for cascOde unless you manually correct it).  The original circuit was taken from Walt Jung and Gary Pimm's work, I used 100 ohms for the DN2540 because that's what Jung used and I figured he knew what he was doing, and 1 kilohm for the 10M90S because that's what KG used.  

[Laowei]

I like your reasoning, JimL.

[PretentiousFood]

I just "traced" a DN2540 with a tube tracer with various degeneration resistor values. I'm not sure how valid the method is, but when you have a hammer everything looks like a nail. It looks like the published curves are very optimistic (or perhaps compressed by scale) in the region of interest for plate load design. The output impedance is the dVd/dId curve. All measurements taken from a single device (no averaging) with a 100R gate stopper. The TO92 sample measured differently, but It might just be sample variation.

 

 

The output impedance is much closer to constant as a function of bias when some degeneration is applied. For these curves, Vgs is actually Vgs + I*Rs. It looks like a BJT current source might beat the single DN2540 for high currents, where you won't have much degeneration.

 

Link here for all the measurements. I'll try measuring a cascode and some other devices, if the methodology is sound.

[nikongod]

Still mucho better than a resistor

 

I would be curious to see what results you get by setting the current purely by varying the value of a resistor between the source and ground, instead of applying a voltage to the gate. 

I know it should not make any difference, but I think it might. 

 

My reasoning is as follows: 

In the full graphs you can see that the slopes of current vs voltage are different with different Rs. As Rs increases the slope of the current VS voltage line decreases at similar currents*. My thinking is that if Rs was not important, the slopes of the curves closest to 10mA would not change.

 

*Its kind of hard to see whats going on in Rs=0ohm, but comparing the lines closest to 10mA from Rs=0 & 10ohm - 11 to like 19 (8mA) vs 10 to like 13 (3mA) 

Rs=10ohm VS 24.9ohm @10mA - 3mA(10 ohm as above) VS either 8 to 9mA or 11 to 12.5mA (about 1.25mA) for the 24.9ohm curve.

The 49ohm curves dont go to 10mA, but at lower currents they are flatter still than the 24.9ohm curves if you compare at ~5mA.

[JimL]

PretentiousFood:

I'm not quite sure how you have hooked up the FET to the tube tracer.  If by degeneration resistor you mean a resistor going from source to ground, but you connected the gate stopper resistor to grid voltage rather than the bottom end of the source resistor, then this is not going to give accurate results for the impedance of a current source.  The reason is that in a FET current source, the gate has to be connected to the bottom end of the source resistor. - see the current sources in the SRX schematic.  This results in a negative feedback loop that increases the impedance, and also sets the static current.

 

Here is how it works:  suppose we try to increase the current running through the FET.  The increase in the current increases the voltage across the source resistor, which in turn means that the gate voltage becomes relatively more negative which tends to turn off the FET, making the current decrease.

 

The correct way to see its behavior as a current source is to treat it as a two-terminal device rather than a three-terminal device, which is what a tube tracer is usually doing.  Instead of plugging in the FET like a tube, you have to use both gate and source resistors, and connect the gate resistor to the bottom end of the source resistor, then run the I vs V curves - you should get a dV/dI in the hundreds of kilohms, which is what both Pimm and Jung measured.  The gate should NOT be connected to a constant grid voltage, which appears to be how you did the measurement.  Jung specifically stated his measurements of the DN2540 were made at around 30 mA static current, where he measured about 300 kilohms, which is about 10x what you measured.  If what I suggest is in fact the way you did the measurement then I apologize for the misunderstanding, however your results are so far off from his and Pimm's I suspect that that is not how you did it.

 

To go one step further, the reason a cascode current source works so well is that a large variation in the drain-to-source voltage of the upper device causes relatively little change in the gate-to-source voltage of that device.  Almost all of the variation occurs in the drain-to-gate voltage.  The drain and source of the lower device in the cascode is connected between the gate and source of the upper device, therefore its drain-to-source voltage is nearly constant even while there is a wide variation in the voltage across the two device combination.  Since the lower device controls the current, and since it is running at close to a constant voltage regardless of voltage variations across the cascode, and the lower device is also getting the negative feedback from the source resistor as described above, it should output a constant current - or very very close to it.  And that is the definition of a current source.  With the cascode, you should get a dV/dI on the order of a hundred megohms or more.

[luvdunhill]

JimL: have you tried the bias multiplier version of the cascode current source?