LCR phono stages we know about

Yes you can have a LTP differential stage with a current source tail and current source plate loads. I have one. The trick is the shunt resistor. It bleeds excess current from the plate load CCS to ground and establishes the plate voltage at the same time. This does present a problem if you are using an all vacuum tube cascode in that imbalance currents between the tubes will change as the tubes age and this will lead to differences in the two plate voltages of the pair. If you are direct connected to the second stage as in my phono stage, this is bad and will eat into the dynamic voltage "headroom" of the preamp. However if a SS/tube hybrid cascode is used, the bias is mostly determined by the SS devices used (matching is a must) which generally doesn't change over time. If the plate load current sources are mounted to the same heat sink so they track thermally the problem is solved. It's a shame that the Sonus Veritas phono pre which used this topology never got any traction in the marketplace. One TT manufacturer who will remain nameless compared in to the Ypsilon and he preferred the Sonas.

With the single ended version the cathode CCS is bypassed with a capacitor.

"With the single ended version the cathode CCS is bypassed with a capacitor." This is a sidebar to the topic, really, but are you imagining a single-ended phono input with CCSs between both the cathode and ground AND between plate and B+? Using a bypass capacitor around the lower CCS does not fit in with my understanding of the need for a bypass capacitor, the value of which needs to be inversely proportional to that of the more typical cathode resistor, in order to pass audio frequencies of interest. In this case, the CCS impedance is so high that a bypass cap would seem to be superfluous. Never seen that done.

I am guilty of going off topic in a thread that I myself started up.
I found a nice thread on Wigwam, among some English guys, one of whom built a LCR phono for another member, using inductors built by Dave Slagle. The builder explains that for an LCR, it either has to be driven by a source with an output impedance equal to its input impedance (e.g., 600R), in which case it can be terminated by an impedance that is ideally 10X higher, or the other way around. In other words, when I wrote that there needs to be 600R on either side of 600R LCR network, I was wrong. But 600R impedance is still very difficult to attain with tubes, without resorting to transformers, etc. This guy used 7K inductors made by Slagle.

You need a cathode bypass cap for the same reason you need one in an ordinary single ended triode circuit. The tube for the sake of analysis can be viewed as an AC signal generator whose output is Eg*u were Eg is the signal voltage presented to the grid times u, the gain factor of the tube. The AC signal loop with a cathode bypass cap consists of two resistors, Rp and Rl in series where Rp is the dynamic plate resistance of the tube and Rl is the load. This acts as a voltage divider on the output of our theoretical signal generator. The greater the load resistance in relation to the plate resistance, the greater the percentage of the signal is dropped on the load. With a CCS load the gain of this triode stage thus approaches u.

Now remove the bypass cap. The signal still has to return from ground to the cathode. It no longer has nice near zero impedance cap. It must pass through the cathode resistor. The AC voltage drop is -i*Rk where i is the AC signal current and Rk is the cathode resistor. The output of our tube (signal generator) which was Eg*u is now (Eg-iRk)*u. If Rk is very large this gain stage will have very little to no output.

I designed and built a handful of inductor-based phono stages about fifteen years ago, both tube and solid-state. I used pre-made LCR networks from Tango and S&B (both now out of production), and a couple of LCR and LR networks of my own design with inductors custom-wound by Sowter. And while I certainly have no inherent prejudice against inductive components in the signal path, I ultimately felt that it wasn't the best way to go . . . but I'll share a few of the conclusions (at least that I can remember).

As others have pointed out, designing the network to a lower impedance reduces the inductor values and makes them easier to wind, but it at the same time pushes the inductors' ultrasonic self-resonance frequency higher, and further from the audioband. The Tango and S&B units I had were both 600 ohms, and one of my Sowter-based networks was significantly higher (about 2K IIRC), and the latter's transient response was poorer in an immediately obvious way, even though the RIAA response through the audioband was at least as good.

The other impedance idiosyncrasy (not surprisingly) about all the networks was that distortion was markedly lower when driven by a zero-impedance source, and terminated by its characteristic impedance, rather with "equal" impedances at each end. The noise figure is obviously lower as well. It also seems that the better the winding techniques and core materials are, the more sensitive the inductor is to any amount of DC current leakage. So I'll admit that I'm scratching my head a bit as to why LCR equalization and (especially low-feedback) tube electronics seem to be frequently associated, as the "impedance comfort zones" of LCRs and tubes don't much overlap, and some traditional tube biasing techniques that put DC through the inductor are definitely a no-no with a precision LCR EQ network.

If you load the input stage cascode with an active current source you can effectively achieve an infinite output impedance. This allows you switch the RIAA series resistor from a series connection to a shunt connection. No issue with RIAA response changing as the tube ages. The excess current from the plate load CCS is shunted across this resistor to ground and sets the plate voltage. Neat trick. To maintain the set plate voltage the cascode pair needs to be biased with a cathode/source (tube or tube/SS hybrid cascode) CCS.

As I read it, the operating conditions you describe are identical to a conventional transconductance amplifier with a plate-load resistor to B+. To the audio signal, the load resistor goes to ground either way, it's just that in the traditional topology the current is capacitively coupled to ground through the B+ supply capacitor. The source impedance is still the tube's output impedance in parallel with the load resistor, and since this resistor loads the signal, its value is still the limiting factor for the open-loop gain. Just because there's a constant-current-source yanking the DC parameters into submission doesn't change any of this; rather, is simply adds one more uncorrelated noise source to the equation. Maybe I'm not seeing something fundamental about your circuit description, but (no offense) I don't get how this is any kind of improvement over the basic tube circuits of seventy years ago.

But this does highlight how many designers make broad topology decisions in conjunction with their approach to RIAA EQ . . . so in evaluating various designs, it's hard to know what are the characteristics of the LCR approach in general, and what's simply the sonic signature of the circuit design as a whole. Incidentally, my favorite of the tube designs from the aforementioned exercise was five tube stages from three tubes per channel: WE417 followed by 12AT7 (both plate-loaded) and a 12AX7 cathode-follower, with global NFB around these three for a very low output impedance. I then used the Tango iron loaded at 600 ohms, and then the other sections of the 12AT7 and 12AX7 (also plate-loaded and cathode-follower, respectively, and with global NFB around the two) as additional gain and buffer. Sounded great especially with an MM cartridge.

My favorite solid-state approach was two 990 opamps, the first one giving active 318uS and 3180uS EQ and driving a LR network (IIRC 200 ohms or so) for 75uS EQ, then the second for additional gain and an output buffer. This was better for MC cartridges, especially when I added a Jensen 346-AX input transformer for a bit more gain, and so could reduce the electronic gain of the first stage.

At some point I removed the LR network from the SS preamp and substituted an RC . . . and found I liked it better. I recall the RIAA tolerances for the LCR-based units to be better than +/- 1/4dB or so, the RC/LR design was significantly tighter as there are fewer interactions, that is the resistor values can be easily twiddled to match the inductors and caps. The S&B units seemed to have a bit of an ultrasonic peak that I tamed down with a small-value Zobel network - the same approach wasn't completely successful with the higher-impedance LCR network.

John, I think there is a misunderstanding between us that comes from trying to describe a circuit in words rather than by schematic. Can you reference a schematic that illustrates your point? For my part, I can only say that I've taken a particular interest in balanced differential input stages, and I have never ever seen a schematic wherein the CCS ideally used between ground and the cathodes of the two tube sections that constitute the balanced input is bypassed by a capacitor. Plus, my understanding of theory tells me that it's not necessary, would even be detrimental. However, I stand ready to be educated; if you can point to a schematic it would help.