srx revisited

[JimL]

I heard from AudioXpress, they plan to publish my article in two parts, the first part in October which analyzes the SRX circuit and modifications (which I've already discussed above in more detail) and the second part which covers the shunt regulated power supply.  In my build I used a surplus transformer, hybrid tube/SS rectification and choke input raw PS (mainly because I'm twisted), and a shunt regulator.  The shunt regulator is based on ideas from John Broskie's TubeCAD shunt regulator article of  July, 1999 and uses his TL431 circuit, but with stacked MOSFETs instead of the pass tube, an RC network for stability of the TL431, and connects the regulator between B+ and B- as suggested at the end of the article where he discusses bipolar supplies.  The latter helps keep signal currents out of the ground line.  I also used a current source instead of the series resistor.  The measured broadband noise between B+ and B- on my build is less than 4 mV and the voltage varies less than 0.2 volts between cold turn-on and warmed up. 

Edited July 1, 2015 by JimL

[mwl168]

Congrats JimL!

[Kerry]

Congrats

[JimL]

Thanks, guys!

[Remolon]

Thanks Jim for all the writing about the CCS.

I'm planning to use this CCS to feed a triode of a ECC99 tube.

B+ of 350VDC

Q1 10M45S (TO-220)

Q2 DN2540 (TO-92)

Ra 133R, to provide 10mA

Q1 is a TO-220 with a heatsink but DN2540 would be a TO-92 because I believe it will dissipate only a few mW and does not need a heatsink, what simplifies the PCB.

I will appreciate any comment or correction.

[JimL]

That looks OK. The only thing I would suggest is to make a test jig to check that the current source actually puts out 10 mA - there is some variation among DN2540s. Or you can use a fixed resistor plus trim pot to adjust the current to 10 mA in circuit.

[Remolon]

Thanks, your response allows me to order the PCB with more confidence.

[JimL]

If you use a 100 ohm fixed resistor plus trim pot, you can use the 100 ohm resistor as a sense resistor for the current.  Then 10 mA = 1 volt.

[Remolon]

I measure at  the cathode  resistor, to control the point of operation of the valve, but this is a good tip. Thanks.

[PretentiousFood]

I ended up opening mine to fix a ground loop, and well, one thing led to another until it morphed into this. The outputs are EL34 triodes. Mine never had the tail CCS in the output stage, but rather a shared cathode resistor as well as a C- supply, per this fellow's post over yonder.

The old arangement had a C- supply as well as the common cathode resistor that worked quite well. The SR-X Plus' tail CCS seemed like a smart improvement, but wasn't really as practical to implement in my old point to point build.

Another alternative was fixed bias. I don't think that gain mismatch between the EL34s would actually be an issue; due to the CCS loading, both plates end up in series with each other through the load. There's a single current loop like in the CCS-loaded LTP. The grid is biased by a LND150 CCS into the grid resistor. The grids sit around -440V, which conveniently means that the output will clip into the rails before drawing any grid current. The plate CCS makes this arrangement quite safe-- at worst, the plate voltage might drift, but the tube shouldn't burn up.

I was curious to try the amp without global NFB, so that went next. Considering that a gain of 40 or so is sufficient to drive the output to clipping, I reduced the cascode's plate loads to 100K, which had a few interesting consequences. The lower output impedance is more suitable for driving the EL34's Miller capacitance, which, as JimL said, is a pretty big deal when running without NFB. The 12AT7s are also now idling at 1.25mA per section, and have greater voltage across them. 275V still gives the cascode plenty of headroom.

It's still waiting for a -460V supply, which i should finish up tomorrow. Next step is to add some output transformers, and try it on speakers.

[JimL]

That's an interesting variation, will be interested in hearing how it turns out.  I would make one comment - without any global feedback you could have some issues with channel-to-channel balance as Dr. Gilmore has mentioned in his articles about the KGSS and BH which were on Headwize (not sure if they are still there or not).

[PretentiousFood]

It works!

I haven't put it on the bench yet but channel balance seems good. The difference in sound is much greater than I expected, going off of memory.

A few interesting side effects of opening the feedback loop: the amp is now a fair bit more microphonic than I recall, and the heater noise is much more noticeable. I ended up elevating the heaters about 30V above the cathode potential, and it's now very very quiet. The bass also seems a bit weaker, I think because the coupling caps aren't in the feedback loop. I'll probably replace them with these.

[PretentiousFood]

Just tried out a pair of push-pull output transformers on the output. These guys:

They're 8k CT : 16/8/4 rated for 15W/80mA on a core just a bit larger than the SRD-7 stepups, but only have a primary inductance of 11H or so. I was expecting the bass to suck, but the SR-X is happily (?) driving a pair of KEF LS50s now.  Might mount them to the box and add an output selector.

[nopants]

I thought of doing that at one point, seemed more logical than using transformer-output tube amplifiers specifically intended to drive speakers. How do they sound?

[PretentiousFood]

Sounds good enough that I'd consider putting nicer transformers in. After listening a bit longer, the bass isn't quite as nice as I'd like-- similar faults to that I've heard with the Stax step-up transformers, but still quite a bit better and with much more extension. The rest makes up for it, though, and the whole thing gets plenty loud despite the agressive step-down.

[JimL]

Cool!  I bet this is a first - of course Stax built transformers to use with speaker amps but I haven't heard of a headphone amp used with transformers to drive speakers.  I wonder how a T2 or BHSE with transformers would sound driving, say, original Quad ESLs, which don't need all that much power.

[MLA]

I have 4 spare amp boards (consequences of a a minimum order of 5 and a successful first build attempt :)).

Anybody need one, send me a PM and I'll be happy to ship it your way for cost of postage. 

Note: at least in my build, I had to lower source and sink resistors 20% to get to the intended currents so make sure you measure your dn2540s.

Update: all boards spoken for!

Edited September 2, 2015 by MLA

[JimL]

Great to hear that another SRX Plus is alive.  The DN2540s definitely vary in the resistance needed to achieve the proper current.  How do you like the sound?

[MLA]

Haven't had that much time to spend with it yet. Would say though that it reminds me of a beer commercial that ran a few years back "[insert brand here] - worth waiting for..."

You did a great job designing this! (and many thanks to Kevin Gilmore for drawing up the board!)

[JimL]

Most of the design credit belongs to the engineers at Stax who designed the original circuit.  I just did a logical update with technology that wasn't available when they did the original design.

Edited September 3, 2015 by JimL

[MLA]

True true, the original circuit really is a very clever design within the design goals and constraints of the day. Your analysis has been a most interesting read.

However, it's not for everyone to undertake that next logical update based on new technology so still: much kudos to your work on this!

[JimL]

Thanks!

[JimL]

The publication date for the SRX Plus in AudioXpress has been pushed back a month, to Nov and Dec.  The first part covers the amp circuit, which I’ve already discussed, and the second the power supply. 

 

In the interim, I thought I would start discussing the power supply.  My raw power supply design is a bit different, in some ways a throwback to pre-World War II days.  There isn’t really anything new about it, and in fact I used a similar design in a phono preamp project published a decade ago.  Nor do I claim it is better or worse than current orthodoxy, but it is different, with both advantages and drawbacks that are worth discussing.  Pretty much all the power supplies in Head-Case projects that I have seen are solid state rectifier capacitor input supplies.  If you look at the schematic above, you will see that mine is a hybrid rectifier choke input filter supply.  The bias supply is a direct steal from Dr. Gilmore - if you're going to steal, steal from the best.

 

Transformer

The first thing you may notice is the transformer has a higher secondary voltage than usual, although the ultimate output voltage is +/- 340V or so.  This is because a choke input filter outputs a voltage that is close to the RMS voltage of the secondary, as opposed to a capacitor input filter, which outputs a voltage close to the peak voltage of the secondary – the latter is about 1.4x higher than the former.  I used a surplus transformer with an electrostatic shield between primary and secondary to minimize the amount of noise and hash coming in through the AC line.  The secondaries were specified at 780VCT center tapped/50 mA with two 6.3 volt/1.2 amp filament windings and a 5volt winding, which I used with a voltage doubler to power the HV delay (not shown).  A separate 6.3V/1.6 amp transformer was used to power the 6BY5GA dual damper diode rectifier. 

 

Unfortunately the AC power line is often contaminated with noise and spikes from appliances turning on and off, light dimmers, etc.  The electrostatic shield helps minimize the amount of power line crud that gets into the secondary of the power supply.  It is found in transformers from Antek, for example, which a number of Head-Case members have used for their builds, and is routinely found in professional, industrial and medical electronics, but is generally missing in “high end” audio equipment.

 

Rectification

Every time a rectifier turns on or off it generates voltage spikes.  These spikes produce broad band noise that has to be filtered out. The problem is that the high frequency noise generated is often not well shunted to ground by electrolytic capacitors and so gets passed down the line.  The common 1N4007 diodes are among the worst offenders.  Soft recovery diodes such as HEXFREDs or Schottky diodes, and tube rectifiers generate less noise, with tube damper diodes as a class generating the least noise.  Tube rectifiers also turn on slowly, allowing the filaments of the signal tubes to warm up before the high voltage turns on.  Of course tube rectifiers have their downsides: they need a filament supply, they wear out, and they cannot tolerate large input capacitors.  Together with their relatively high impedance, this results in significantly less “stiff” power supplies, with DC voltage varying with changing demand.  However, since most electrostatic headphone amplifiers are class A, differential and balanced, their current demand is pretty constant so a “soft” power supply isn’t too much of an issue.

 

Since the SRX Plus needs a bipolar high voltage supply, a bridge rectifier is required.  Implementing this with tube damper diodes would require 4 diodes and at least 3 filament supplies (the two positive legs of the supply can share a filament supply), which is a significant complication.

 

As I was in an experimental mood when I built the SRX Plus, I used a hybrid rectifier with a 6BY5 dual damper diode tube for the positive legs and soft recovery diodes for the negative legs, and a HV delay and ramp up circuit from K&K Audio with an 11 second delay and a couple seconds ramp-up on the B- side to roughly match the turn on characteristics of the damper diode tube.  This was an attempt to get some of the noise and turn on benefits of a tube rectifier while only needing one filament supply.  This experiment was partially successful, in that there was a nice delay in HV, however, since the turn-on of the positive and negative halves were not perfectly matched, the amplifier output voltages initially went negative then positive, before drifting down to zero over 30 minutes or so (the drift being due to amp warm-up).  While it can be a bit disconcerting watching the voltages moving up and down by a couple hundred volts or so within a few seconds, since the amp is fully differential, the positive and negative outputs tracked each other well, limiting turn on and turn off noises.  For those who don’t want to bother with a tube rectifier, a regular solid state bridge rectifier would work. High voltage HEXFRED soft recovery or Shottky diodes would generate less noise than other solid state diodes.

 

Filtering

There are two ways of filtering the raw rectified AC, capacitor input filters and choke input filters.  Pre WWII, when iron and copper were cheap and capacitors were expensive, choke input filters were the norm. Nowadays when iron is expensive and capacitors are cheap, capacitor input filters are nearly universal. However, there is one significant disadvantage to this topology. 

 

With a capacitor input filter, there are large current spikes, because the capacitor only fills up at the peak of the voltage waveform when the input voltage exceeds the voltage already on the capacitor.  So during that brief period of time, there is a large current surge to top up the cap. These brief current spikes may peak at many amps even though the average current may be only a fraction of an amp.  The late Allen Wright likened these spikes to whacking the power supply with a ball peen hammer 100-120 times a second.  Those spikes produce lots of broad-band noise.  Soft recovery diodes do little to reduce this source of noise as the current spike is largely independent of the type of diode it is running through.  Tube rectifiers produce less of this, because their relatively their high impedance limits the amount of peak current during the spike.  Thus, the strength of solid state rectifiers – their low forward impedance and ability to pass large currents to top up large capacitors, is also their weakness from the viewpoint of noise generation.

 

Choke input filters avoid this source of noise because the current flows throughout the whole cycle, assuming a proper value choke, and so are inherently quieter.  I feel it is preferably to generate less noise to begin with rather than trying to filter it out afterwards – an ounce of prevention, etc.  Choke input filters also match well with tube rectifiers, which cannot tolerate a large capacitor input filter – most tube rectifiers are specified to work with input capacitors of 20-50 µf at most.  The inductance of a choke acts like an electrical form of inertia to keep the current flowing during the downside of the voltage cycle to smooth the current flow.  Finally, choke/capacitor filters do a better job of removing residual AC line crud (12 dB/octave) than RC filters (6 dB/octave).  The output of a choke looks more like a simple sinusoid - easy for the filter caps to deal with.  This also makes the job of the power supply regulator much easier.  Even the simplest regulators have relatively high rejection ratios for hum and other low frequency garbage.  Where they tend to have trouble is at high and ultrasonic frequencies where the rejection ratio diminishes.

 

Note that these benefits only occur if the input section is a choke.  If you use a capacitor input filter, you still have the large current spikes.  A choke section down the line will help with noise filtering, but real world chokes have shunt capacitance that can still allow some high frequency noise to pass through.

 

With choke input filters, there is no derating of the transformer needed.  A transformer rated at 50 mA AC will easily supply 40-some mA DC current without strain with a choke input filter.   With a full wave bridge cap input filter, the minimum AC rated current is 1.8 x the desired DC current.

 

I used two LC sections each on the B+ and B- supplies – that’s four chokes and four capacitors, which is probably overkill.  But the result is a nice, quiet raw power supply.  In fact I used a single ended version of this with damper diodes, 20H primary choke and electrolytic caps on both sections as the raw power supply for my tube phono preamplifier, which has a passive power supply (if you’re interested, it was published in AudioXpress 1/04 and 2/04). 

 

If choke input filters are so great, why doesn’t everyone use them?  Well, chokes are expensive, heavy, and bulky.  My tube phono and line preamps and SRX headphone amp, which all have choke filtered supplies, weigh around 40 lbs each, and the power supplies occupy at least 2-3 times as much space as the active electronics – you cannot build a mini-SRX, or mini anything else, for that matter, using a choke filter power supply.  Forget it if you want a small bedside table amp.  Also, you need a higher output voltage from the transformer – the RMS voltage has to be higher than the final raw DC voltage.  For a high voltage supply, sourcing transformers can be a problem.  Antek does have 400 to 500V RMS secondary transformers if you’re willing to go to 400VA, which would be total overkill - not that there’s anything wrong with that.  But it is additional buik, weight and expense.  The 500V RMS would work for building a choke filter supply for a Blue Hawaii or +/-400 volt KGSSHV or Carbon, with 5-6H chokes, if you’re so inclined to experiment with choke filter supplies – just make sure you’ve got a strong back or a working fork lift.

 

So, a different methodology for your consideration.

[JimL]

One other thing which I forgot to mention above.  I run AC filament supplies, but I also add small value resistors in series (i.e. 0.2-0.5 ohms) to lower the voltage across the filaments to 5.7-6.0V RMS rather than the 6.3V design center.  This results in a negligible decrease in tube emission but a significant increase in filament life.

 

Those of you who are old enough to remember tube TVs, which were ubiquitous in the 50s to early 60s, will remember that they would run day in and day out for years before needing any tubes replaced, and this is one of the ways they used to accomplish that. The other way is by designing conservatively to use the tubes well below their voltage, current and power limits, which is what I've done in the SRX Plus.  I estimate that tube life should be in the 10-20,000 hour range with NOS tubes.

[mwl168]

Hi JimL:

I came across this discussion at diyAudio while searching for information on tube heater power supply wiring. http://www.diyaudio.com/forums/tubes-valves/211731-heater-wiring-good-bad-ugly.html. It offers some observations about AC vs DC heater supply among other things. It also discussed the effect of reducing heater voltage and quoted a study done by RCA in post #27.

I am very interested in hearing your thoughts and comments.