Negative rail - Decoupling needed?

[J M Fahey]

Yes, definitely !!
After all, the *first* tubes (remember: older=holier) were battery powered !!!!
Not forgetting the "bias batteries"
Using a cathode resistor for biasing, once was a newfangled, cutting edge technology.
It even had its own name: "automatic biasing".

[kg]

The problem with the coupling caps is that they have a static -6db/octave slope, which could be bad if I set the rolloff freq too high.

dude, don't get too hung up on numbers. this is guitar amp stuff! i purposely use ceramic coupling caps in a few key places because they have such lousy DA.

if the channel has too much fart, reduce coupling caps. you'd be surprised how small you can go before you really start to lose body and volume.

[exclamationmark]

Hahahaha. The problem was indeed my design. I had a couple of DC-Coupled cathode followeres and the cathode resistors on the cathode follower section were much too small, making the grid draw something like 0.7 mA at quiescent conditions! (Which is a HUUUUUGE current considering considering my B+ is only 100V - The grid of the CF was actually drawing more current than the gain stage itself!). Must of been some awesome square wave clippage. Increased the values to something reasonable and now the nasty intermodulation is gone... But I've somehow developed oscillation problems, probably due to the increased gain. Not a problem though, just gotta route some wires (like not having my output wires next to my input ones LOL).

[Gingertube]

Another thing to watch for eliminating mud in high gain preamps is to make sure no stage is going into grid rectification (high grid current).
High value grid stops are your friend - try 47K or even 100K. (Jim Kelly used up to 2M2 into some low gain stages where he had only low Miller Capacitance to worry about))
You can also adjust interstage attenuators and if you need to get back some high mid and tops you can bypass the upper arm of the attenuator. That is, you can implement shelving filter response in an interstage attenuator.
Attenuators are also your friend in another sense. As you look away from the grid you have a certain impedance to 0V. You want to keep that lowish (< 500K) for low noise. When an attenuator is used that impedance will nearly always be lowish.
I've always found zener and diode bias to be noisy. LED bias should be better but I haven't tried it.
An LM317 was mentioned above. Sorry, you "pressed one of my buttons". LM317 is one of the nastiest sounding voltage regs ever produced, don't use them.
Cheers,
Ian

[exclamationmark]

I'm actually using 12au7's for this build, which allows me to stick humongous grid stoppers on the tube without eating into the HF response. Also, the entire amplifier is fixed biased - hence the question about the negative supply rail , so my circuit is already chock full of shelving filters. Actually turns out cheaper without the poly cathode bypass caps! A really huge advantage of fixed bias, is that for a relative novice like myself, it allows you to change the bias on the fly, without screwing up the frequency response (which is what would happen to cathode biased stage, with a pot across the bottom, unless it's unbypassed) and without spending half your time desoldering things.

One thing interested that I noted, is that by biasing a stage cold enough to make a cold clipper stage, it somehow starts to ooze mids from every orifice (that isn't there when biased hot/middle). Probably due to the mass accumulation of shelving filters upstream. Sounds nice, but I have to switch my topology around to get more gain to the stage using only 12au7's, and drop some shelving filters. Might save that for a V2...