CCS Tube Line Stage

As my knowledge and skills in tube gear progresses, I have become confident enough to develop my own designs from scratch. Make no mistake: I am not an expert. I know just enough to be dangerous. I have a resolving system to determine what sounds good through trial and error, and I have enough test equipment to fine tune with measurements.

That being said I’m confident enough in the results to release my designs here, as “open source” free to use designs – including gerber files to print your own PCBs if so desired.

Both designs use constant current source plate loading, diode cathode biasing, and mu-follower outputs for the most linear response, the greatest fidelity, and the lowest possible output impedance – in the case of my line stage, measured at 30 ohms.

10M45S depletion mode MOSFETs are used for the CCSs as these are cheap, readily available, and don’t require any additional power or reference to ground – just stick em in where the plate resistor normally lives, set the current with the appropriate resistor, and you are done. Gate stopper resistors are included to prevent oscillation. To calculate the current set resistor: 3000/Ia = resistance in ohms (so 9ma is 3000/9 = 330R, 1ma is 3000/1 = 3K).

Before the purists start scoffing at the silicon bits I’d like to point out that Conrad Johnson has used diode bias for decades and Audio Research uses elaborate constant current source schemes on nearly all their products.

These designs lean towards the neutral side of tube sound – think Audio Research but a touch smoother with a bit more midrange and you’ll have an idea of what to expect. Bass is tight and well defined. Highs are extended but never harsh. Midrange is slightly forward. I voice my equipment to maximize the breadth and depth of soundstage while maintaining balanced tonality and a smooth top end – I hate harsh highs. It isn’t the last word in microdetail but you’d need to spend a lot more to get better. If you are looking for that old fashioned warm, syrupy sound like grandpa’s Dynaco don’t bother with these.

Recommended B+ is 180-280V; with CCS plate loads the precise voltage isn’t as important as having a dead quiet well regulated supply with sufficient current and as close to zero ripple as possible. Solid state or tube, your choice. I prefer remote external supplies to keep noise to a minimum. I only use 6.3VDC heaters. Heater ground is elevated with a voltage divider connected to B+ voltage: thus DO NOT CONNECT HEATER GROUND TO CHASSIS OR SIGNAL GROUND. I have provided my design for an all-tube series regulated power supply here.

I do not include provisions for local filtering caps on my boards because I use large film capacitors mounted remotely in my designs – I recommend you do the same. Solen makes affordable large value film caps that are ideal for tube power supplies.

---

First up is my 6922 zero-feedback single ended parallel triode line stage. This can use 6DJ8 or 6922 tubes: my personal recommendation is Amperex JAN 7308s, with British Mullard E88CC as a close second.

Gain is 29db at 0.06% distortion into a 100K load. Yes, really. Yes, with a single 6922 per channel. CCS, diode bias and vanishingly low output impedance makes for huge gain.

CCS plate currents are set to 9ma. Cathode bias is set to 2.5V via a standard red 5mm LED in series with a 1N4148 diode.

For coupling caps I recommend Vcap ODAM or TRT Stealth for best results, Jantzen Superior or Obbligato as runner ups if you aren’t feeling so spendy - you do you and choose whatever you prefer as long as they are rated for at least 300V. All resistors are 1/2W metal film Vishay Dale unless otherwise noted. A 100K stepped attenuator or quality log potentiometer should be placed ahead of the line stage for a nominal 100K input impedance.

Part II - Enraging the Purists, or how to stop worrying and love the feedback.

An optional but recommended tweak to this circuit is to add negative shunt feedback from the output to the grid. Why would you want to do this, when every tube head will scream bloody murder about adding feedback to anything powered by hot glass?

Well: 

1. The open loop gain without feedback is high (29db) and you'll probably want to tune it to suit the input sensitivity of your amp. Otherwise you'll be using attenuation at the output, which is fine but it's nice to not need it.

2. Adding feedback/reducing gain reduces the noise floor. A lot.

3. Adding feedback increased the bandwidth a HUGE amount. Zero feedback -3db point is 40khz. With feedback it's BEYOND 100KHZ (my analyzer maxes out at 100KHZ so I can't tell you where the precise point is), with -1db at 65khz. It also tightens everything up, sharpening the detail and dynamics across the board.

4. Distortion is reduced, but this is purely academic as the open loop distortion is already low. It goes from 0.06% to 0.02% with NFB.

To do it you need to run a resistor between the output (after the coupling cap to block the plate DC) to the junction between the grid stopper resistor and the grid of the tube. The resistor values need to be tweaked to form the voltage divider you need to add the feedback. For example, I use 22K from cap to grid, 2.2K grid stopper shared between both grids in parallel. This nets me 13db of gain, so the NFB is -16db. I didn't even bother modifying the boards to accomplish this: the 22K is just run on the underside of the PCB and a bodge wire is put between pin 2 and 7 of the sockets to parallel the grids, then remove one grid stopper and replace the remaining with the value you come up with for your network (if you left 2 in place, they'd be paralleled).

"But what about the input impedance, I read on a certain Wiz site you need to run really high value grid/FB resistance to not decrease the impedance!?"

Yes, you did. That site you found is for guitar amps where bandwidth doesn't matter. And we are using a CCS plate load that doesn't care much about parallel resistance. 

You know what happens when you use a large grid stopper ahead of a stage, say 100K? You create a really effective low pass filter with the Miller capacitance of the tube. I tried it out to see what happens. The -3db point was 20KHZ. Real bad idea. 

Keep the grid stopper value below 5K to not screw up the frequency response.