Thanks Nick, thanks Marcel, "several times" - "in great detail" ...i know and understand perfectly.
;-)
My question is and remains the following: why does the universally valid postulate not seem to have been taken into account in the examples I have given? Why does the SUPRA work without disturbing shot noise???
But we don't have to argue and want to be right, that doesn't help.
HBt.
Regarding your second question: it's no big mystery, in fact I hinted at it in post #17. Noise optima are rather broad and there is simply quite a difference between the noise level that the average user would find disturbing and the noise level of a carefully optimized moving-magnet amplifier. As a result, you have to be very far from the optimum before it gets disturbing to most users.
Like I wrote, when the 47 kΩ termination resistance is made with just a resistor shunting the input, it injects 0.5869 pA/√Hz of noise current into the node that connects the cartridge, the resistor and the input of the rest of the amplifier. I have shown in my 2003 article that the thermal noise of the cartridge is of the same order of magnitude as that of a 47 kΩ termination resistor when you take into account things like iron losses. Hence, when you add an NPN stage/bunch of NPN stages with 1.075 µA total base current and a PNP stage/bunch of PNP stages with 1.075 µA total base current, the noise with no record playing gets 3 dB worse than when you only had the cartridge and resistor as noise sources and 6 dB worse than when the cartridge were the only noise source.
That's simply not bad enough to annoy most users, particularly when they only care about the noise when there is a record playing. Record noise is usually far stronger than cartridge thermal noise and phono amplifier noise. Nick is an exception to this rule: as the current noise gets converted into a voltage across the inductive impedance of the cartridge, it sounds sharper than normal record surface noise. He finds that quite annoying.
Still, I have seen a thread on this forum from someone, not Nick, who had built a phono preamplifier with an LT1028 and found the hiss annoying. He wasn't aware of the issue of noise current, so it rather surprised him that his expensive ultralow noise op-amp produced annoying hiss. Mostly due to their base current compensation circuit, the LT1028/LT1128/LT1115 family of op-amps have an equivalent input noise current of about 3.4 pA/√Hz when measured under realistic circumstances. (The LT1028 moving-magnet version of the Elektor Supra 2.0 is even worse, it uses four LT1028s to produce as much as 6.8 pA/√Hz.)
Regarding your first question: moving-magnet phono amplifiers are more often than not measured with a 600 Ω source impedance. Each pA/√Hz produces only 0.6 nV/√Hz across 600 Ω, but 12 nV/√Hz across the 12 kΩ effective RIAA- and A-weighted average cartridge impedance that I estimated in my 2003 article for a 500 mH cartridge. It is therefore very easy to largely overlook the effect of noise current. In fact, even Douglas Self did it wrong until I pointed it out to him in the mid 1990's - it was kind of funny, because he was already a famous writer back then and I was completely unknown and in my mid 20's.
Apparently Stereophile also uses 600 Ω rather than 600 Ω + 500 mH or so. That means that for an amplifier with a bipolar input stage, the designers can choose between as low a noise level as possible with a cartridge connected to the input, or as good numbers as possible during a Stereophile test - you can't have them both...
Besides, one of the circuits you showed looks like an amplifier for moving coil cartridges that also supports moving magnet. Because of the much lower impedance and much lower output voltages of moving coil cartridges, equivalent input noise current is not much of an issue, but equivalent input noise voltage is much more critical than for moving magnet. If you want to support both moving coil and moving magnet, you can either choose a compromise collector current, or make it switchable, or optimize for one of the two and accept suboptimal performance for the other type. Or use really high transconductance JFETs - not having any base current and practically no gate current, they produce almost no equivalent input noise current.