The C4S stands for "Camille Cascode Constant Current Source." It was designed by Buddha (John Camille). This current source is simple, robust and in my opinion functions very well. Bottlehead sells the basic C4S as a kit.

I had a friend ask how to cascade two C4S in a power supply so that he could get more power supply rejection. He said that Budda had mentioned that this could be done. You would start with the raw B+ and have a C4S feed a capacitor. This capacitor feeds the C4S that runs the output stage. This can get you a lot of attenuation in a very small space as opposed to cascading RCRCRC stages or LCLC stages.

The problem with implementing this with a standard C4S is mostly with the variation in current in the C4S. The C4S changes in current because of  value/gain changes in the parts, changes in current over temperature and changes over time (mostly the LED).

The problem:

If the first (top) C4S drifts up in current, it will pull the capacitor it feeds up in voltage until the first C4S saturates and now all the benefits of the top C4S are lost.
If the first (top) C4S drifts down in current, the voltage on the feed capacitor will start to drop and the output stage will clip on loud passages or stop to work entirely.

This is basically how the C4S gets set up:
"OUT1" is the high volts that feeds the final C4S.
"OUT2" is a reference RCRCRC filter stage with the same voltage drop as the modified C4S.
"PLATE" what the voltage on a 6K plate of a tube will look like for a given B+ ripple.

Notice that I have my normal protection resistor in the collector of Q2 and Q6. A good value for this resistor is to use the same value as you are using for R1 or to use R = B+/ 0.5 A. I used a 1N4148 for the base emitter protection resistor. A UF1007 is a better choice, but I don't have a model for that (yet.)

Here is what the different voltages look like with 1V of ripple on B+. Notice that the RCRCRC stage actually works better.
I'd guess the RCRCRC version would occupy 3 times the volume as the C4S version.

The OUT1 peak at 0.2 Hz worries me. This peak could mean that there is a sensitivity to low frequency noise. I don't fully understand the OUT1 peak, but I have only spent enough time to throw together a simple model.

I am also showing a Hawksford Cascode connected C4S in the schematic!

The Hawksford Cascode has a capacitor or a zener between the emitter of the current set transistor to the base of the cascoding transistor. This way of making a cascode has more feedback than a standard cascode. The feedback is from the base of the cascode transistor to the emitter of the transistor that attaches to the current set resistor. This feedback  feeds the current from the collector to base capacitance of the cascode back to the local feedback point (the emitter of the current set resistor.) Because of this feedback, the output impedance is  greatly increased and the output distortion is reduced.

Unfortunately, this extra feedback can make the Hawksford Cascode ring at RF frequencies. I added a resistor and a capacitor (C3) to turn the Hawksford Cascode back into a standard cascode at RF frequencies so we get our stability back. The added capacitor (C3) should be a mica or ceramic with very short leads, do not use a metalized film.

If we remove C3, we see a nasty notch in the output impedance of the modified C4S. It is very easy for this notch to get out of control if we leave out C3 and R4
Notice look at the improved output impedance in the audio band and out to 100 kHz. A full 25 dB of improvement.

In the plot below, I kept C3 and R4 the same (220 pF and 200 ohms) and swept the load current and associated resistor values in the modified C4S.

The Hawksford version of the C4S works much better.

With all said and done, I would personally use the RCRCRC filtering if I could afford the space needed for the extra resistors and capacitors. The last C (the one that attaches to the tube) is the one that needs to be the best quality. The first two Cs (220 uF) just need to have long life and low ESR.

With RSET = about 50 ohms for 3V drop at 60 mA throughput and using the Hawksford Cascode mod, the graph above says we should see about 1 megohm at 100 Hz. To get 1 megohm at 100 Hz, we would need 1561 henries of inductance. That's a big choke. This big choke will probably have more capacitance at high frequencies than the C4S.

60 mA * 30V drop = 1.8 W, that is a do able heatsink size for a C4S. Make sure you watch the SOA during power up. I did not draw the SOA curves here. I learned about SOA the hard way using a current limited transistor to charge up a capacitor. The transistor died about every 15 power ups.

The C4S approach should be smaller than the RCRCRC approach and does show promise. There still are a few questions left to resolve about power up and the low frequency peak.

A few ideas:
How about replacing the 270 ohm resistor in the Paramour with a C4S that works like this!
I think I have a new mod to try after VSAC!
I may just make a version of the active filter already on may Paramour page and just leave the delay out.
Adding a choke is much easier to do. Since when is the easy way the fun way!

The problem with using the C4S in a power supply near 300V is the turn on surge. When the C4S sees more than 300V, it will be toast. When it fails, it will probably cause a few more parts to toast.

This example is for a shunt regulated supply:

D1 and D2 adsorb the turn-on surge between B+ and OUTPUT through D3. Remember, C1 looks like a short at turn on.
With R1 to bias D1 and D2 on in normal operation, D3 should be reverse biased isolating the output from the high capacitance of D1 and D2.