I just finished reading this post by Paul Joppa when I saw your question here
Posted by Paul Joppa at www.bottlehead.com forum on May 19, 2002 at 17:13:11
In Reply to: Re: Anticipation Pics (& long) posted by kphill on May 19, 2002 at 16:07:29:
"'Tube hiss' is usually thermal noise; the result of a large number of electrons hitting the plate at more or less random times. It is a characteristic of pretty much anything that has resistance, and is white noise, that is equal energy per Hz bandwidth at all frequencies. In the case of a triode, the equivalent noise resistor is approximately 2.5/transconductance. That's about 2500 ohms for a 12AU7 at 1mA. The cathode resistor, if unbypassed as in the Foreplay, adds to this equivalent noise resistor. So does the source resistance. There's not a lot you can do about it, but here are a few things:
1) Increasing the current will increase transconductance and decrease the noise resistance, but only by the cube root of current.
2) Bypassing the cathode resistor will reduce the resistance, but not by much (less than 3dB in the Foreplay), and will damage the sonics to some degree. LEDs and batteries in the cathode are sometimes used, and can sound better than caps - but that's another can of worms.
3) Reducing the impedance of the level controls will reduce noise, for instance a 10k pot will be 10dB less noisy than a 100k pot. But it may be too low a load for some source devices. A transformer volume control should be quieter than a resistor pot.
4) A higher transconductance tube will have a smaller equlvalent noise resistor.
Sometimes the term "tube rush" is used. That carries the suggestion of more low frequency content. The terms are not universally agreed to, but the main source of excess noise (above the basic physics of resistors described above) seems to be something in the cathode that results in "1/f noise", also called "flicker noise". It's called that because the spectral level is inversely proportional to frequency. Most devices have something with this behavior (tubes, transistors, FETs) but in highly varying degrees. Often this noise is dominant below about 10kHz. Fortunately, there is something you can do about it: select tubes, or buy those pre-selected, for a low noise figure.
Beyond those considerations, the best practice for noise is to operate every device at its maximum output level before distortion, reducing the input sensitivity of the following device as necessary. For instance, your Adcom takes 1.5v for 200 watt, which is 335mV for 10 watts. That means it is 15.5dB less sensitive than the 6V6 at 2v for 10 watts, and the input noise will be 15.5dB lower. You could achieve the same effect with a 15.5dB pad at the input of the Adcom."
I remain,
