Best way to reduce noise when reducing gain, triode pre-amp stage?

[Merlinb]

I took on board the grid leak idea, and lowered it's value to 22K, which is 8.5 times the tanks output impedance (~2.6K at 1KHz). This in theory should reduce resistor noise.

A smaller grid leak will increase resistor noise, or rather, it will reduce the signal more than the noise, so you still end up with a poorer signal-noise ratio. On the other hand, it does damp the transducer resonance, as you noted.

Might also need to increase the input grid stopper as tubeswell suggests.

That will increase resistor noise. Johnson noise is almost entirely due to large resistance being in series with the signal. The basic rule of thumb for minimising it is to maximise all resistances that shunt the signal, and minimise all those in series with the signal. I would go for a 100R stopper, 470k-or-less reverb pot, and metal film resistors throughout (and try another valve in case it is the culprit). If you have the option of using a 12AT7 instead of 12AX7-variant then go with it, as that too should reduce hiss.

But what is the best way to reduce the gain in order to give the greatest reduction in noise?????

Anything that reduces the resistance in series with the signal. This may include a smaller reverb pot, a smaller anode resistor, or split the anode resistor in two and take the signal from the junction.

[jimboyogi]

Thanks for your replies everybody , I had no idea that LEDs could generate current under light, I went back and tried covering each of the LEDs in my circuit and was still unable to hear any change, but my bench light source is not very bright and is an incandescent globe, not a flouro.

Merlinb, thanks for your help I have some questions that are in bold type below..

A smaller grid leak will increase resistor noise, or rather, it will reduce the signal more than the noise, so you still end up with a poorer signal-noise ratio. On the other hand, it does damp the transducer resonance, as you noted.

This impedance/inductance effect is working very well with the 22K grid leak, so well that I was able to remove the treble shunt cap and still have the eq that I wanted. I'll try a higher value and use a cap from plate to ground as JM Fahey suggested, and see if that is less noisy.

That will increase resistor noise. Johnson noise is almost entirely due to large resistance being in series with the signal. The basic rule of thumb for minimising it is to maximise all resistances that shunt the signal, and minimise all those in series with the signal. I would go for a 100R stopper, 470k-or-less reverb pot, and metal film resistors throughout (and try another valve in case it is the culprit). If you have the option of using a 12AT7 instead of 12AX7-variant then go with it, as that too should reduce hiss.

Is this because the 12AT7 has a lower plate resistance than a 12AX7? At present I am using a 12AU7 which is giving enough gain for the recovery, as well as having a low plate resistance. This also allows me to use a lower value anode resistor (47K) and a lower value reverb pot (at present 500K, but I should go lower!)

Thanks for the tip about maximising shunt resistances, and minimising series resistances. I don't exactly understand how the reverb pot fits into this though, from an AC perspective it is in parallel with the anode resistor, so minimising it's value lowers the effective series resistance for Johnson noise. But isn't the reverb pot also a signal shunt, between the output of the coupling cap and ground?

Anything that reduces the resistance in series with the signal. This may include a smaller reverb pot, a smaller anode resistor, or split the anode resistor in two and take the signal from the junction.

[pdf64]

' don't exactly understand how the reverb pot fits into this though, from an AC perspective it is in parallel with the anode resistor'
only when it's set to max. At settings other than max and zero, it will be adding series resistance to the source, which Merlin explained should be minimised to reduce noise.

[raiken]

Randall: I'd bet you (or Atmel...) had a race condition or metastability issue in that design, and the extra leakage current from ambient light was skewing the thresholds of your gates enough to resolve it.

I never found out the exact cause, and Atmel denied having a problem, until the new part came out, then I finally got one engineer to admit they'd seen it.

I didn't change my design, which was completely synchronous, with only one clock domain, so I doubt there was metastability in the top-level design, however, this particular PLD did have separate product-term clocks for each of the 48 buried registers, so anything could happen inside the chip to affect the clocks on their way through the sea of gates.

Since the exact same code worked fine on their fixed silicon rev and the problem never reared it's head again in production, I'm blaming it on them.


RA

[Merlinb]

Is this because the 12AT7 has a lower plate resistance than a 12AX7?

Exactly- it has almost as much gain as a 12AX7, but lower internal resistance. However, your 12AU7-variant probably(?) isn't the problem.

Thanks for the tip about maximising shunt resistances, and minimising series resistances. I don't exactly understand how the reverb pot fits into this though,

When you turn down the pot you are basically adding resistance in series with the signal!
As a quick test, do you get maximum hiss when the reverb pot is at maximum, or when the pot is somewhat less than maximum? (I'm just curious really, as we already know most of the noise is coming from the transducer or grid stopper, so the pot is presumably not a big issue).

[raiken]

:|


Test setup :

...

A. Measurement one. Multimeter on A/C setting.

...

Uh...we're measuring photovoltaics though a diode - you may want to put your meter on DC mV... I doubt your meter can read the high frequency switching of a compact fluorescent bulb (around 40-50kHz), although it should be able to read the lower-level 60Hz AC superimposed on it, or possibly emission from an incandescent bulb. At close range, the high-frequencies will dominate, so you may want to use a scope instead of a meter to look at the AC noise. This isn't rocket science, so you're forgiven.

The vast majority of the change you can read will be a DC shift that varies with the amount of light. If you bypass it with a small capacitor, you will see a very clean, flat DC that varies with the ambient light (the capacitor makes a peak-detector circuit that stores the peak level). The time constant will depend on the value of the capacitor you use. Something in the range of 0.01uF to 0.1uF when driving a high-impedance will give a nice result when viewing on a scope. A larger capacitor, such as 10uF, will take a long time to charge up at the low current capability of the LED, but it will eventually build up a voltage and stay there, because it can't quickly discharge back through the diode.

This shouldn't cause any problems in a cathode circuit, because of the low impedance of the circuit, which effectively shunt things to ground, although, at higher frequencies, the impedance will rise, so it wouldn't hurt to bypass the LED with a capacitor, just in case.

Here's another video (made with a different .3g2 to .avi converter, so hopefully a bit better). It shows the high-frequency (50kHz) modulation caused by the compact-fluorescent lamp, along with the superimposed 60Hz mains noise, and in the later part of the video, I put a bypass capacitor across the LED so you can see the smooth DC shift up and down with changes in light intensity.

http://www.aikenamps.com/led_photosensitivity_2.avi

As you can see, I have the cursors set at 0.87V, and you can see it exceed that voltage and go up to about 1.2V at the end of the video when I move the lamp closer to the LED.


This is great fun, by the way...I've never taken the time to actually see how much voltage an LED could generate.

RA

[raiken]

Exactly- it has almost as much gain as a 12AX7, but lower internal resistance. However, your 12AU7-variant probably(?) isn't the problem.

I used a 12AT7 for the input stage of my Tomcat amps. Incredibly nice, open, chimey tone, but unfortunately, non-microphonic, non-noisy 12AT7's are as scarce as hen's teeth. JJ ECC81's were the best. All of the NOS I tried were horrible, including the Mullard CV4024 (or whatever it was, I forget the number at the moment). I literally would have to screen 50 or 100 of them to get 10 that were usuable.

RA

[tubeswell]

Thanks Merlin. (I guess I was confused about what you had written in your first book in relation to higher grid leak resistance and increasing noise. I thought there was an optimum point for grid leak resistance which was related to the source impedance - as far as noise was concerned, but obviously there is also an optimum point below which the signal attenuation is such that the noise to signal ratio is increased. I should have thought about that some more).

I have used 39k grid stopper on a recovery stage grid and found it good for reducing surplus noise but (being the humble hobbyist I am) I obviously have a lot more yet to learn. Thank you for taking the time to explain it.

[Enzo]

And gary, I brought up the LED thing as just one possibility in the situation the OP reported. That has no bearing on the efficacy of your bias applications, nor was that intended. You get good results, fine, enjoy the sucess of the design. The OP was having noise, and that one, easily confirmed or dismissed potential cause came to mind.

[Steve Conner]

unfortunately, non-microphonic, non-noisy 12AT7's are as scarce as hen's teeth

That's funny. My hi-fi amp uses them in the front end, and I never had much trouble. I bought four NOS RFT ECC81s from Watford Valve and selected the best two. They all seemed to work fine, hum from the filament was the main problem.

[jimboyogi]

When you turn down the pot you are basically adding resistance in series with the signal!
As a quick test, do you get maximum hiss when the reverb pot is at maximum, or when the pot is somewhat less than maximum? (I'm just curious really, as we already know most of the noise is coming from the transducer or grid stopper, so the pot is presumably not a big issue).

I tested this, the hiss (and all of the noise) is at maximum when the pot is at maximum.

I have followed advice and gone for a 100R grid stopper, and upped the grid leak to 47K. The frequency response with this grid leak is still good enough, so no need for a treble shunt cap from plate to ground. It really isn't very noisy at all any more, although I can't say for sure if these 2 recent changes have made much difference.

Question - are all reverbs a bit noisy when turned up to wet surf levels

[raiken]

That's funny. My hi-fi amp uses them in the front end, and I never had much trouble. .

I suspect your hi-fi doesn't quite have the gain my Tomcat does...

[Steve Conner]

Randall: Yes. I don't think 12AT7s were specified for low noise, and the first triode position in a high-gain amp is a big ask. (Ideal job for a JFET or LND150 if you ask me. )

Jimbo: It depends on the drive level into the tank, but it wouldn't surprise me if that were the case. I've only built one amp with reverb, I used considerably more drive than Fender did in their combo amps, and it was noisier with the reverb turned up to what I considered a good maximum level. There was more white noise and a little hum pickup from the transducer coil.

(Again, a solid-state recovery amp might help. You can easily get a transistor with better noise figure than a 12AX7, especially into the lowish impedance of the reverb transducer.)

[jimboyogi]

Randall: Yes. I don't think 12AT7s were specified for low noise, and the first triode position in a high-gain amp is a big ask. (Ideal job for a JFET or LND150 if you ask me. )

Jimbo: It depends on the drive level into the tank, but it wouldn't surprise me if that were the case. I've only built one amp with reverb, I used considerably more drive than Fender did in their combo amps, and it was noisier with the reverb turned up to what I considered a good maximum level. There was more white noise and a little hum pickup from the transducer coil.
(Again, a solid-state recovery amp might help. You can easily get a transistor with better noise figure than a 12AX7, especially into the lowish impedance of the reverb transducer.)

Steve, sounds exactly like what I am finding. When you say that it depends on the drive level to the tank, am I correct in thinking that the higher the drive level, the lower the noise? My load lines for my driver stage tell me that at full input to my driver stage before clipping/grid current (~12v pp) I should be feeding the tank with ~ 2.8 mARMS. The specified nominal drive current for this tank (type F) is 2 mARMS. It certainly sounds pretty wet

[Steve Conner]

The higher the drive level, the higher the output from the tank, so the less recovery gain you'll need for a given "wetness". Hence, less noise.

My recovery amp gain was ultimately limited by vibrations feeding back from the speaker into the tank. I couldn't get the reverb knob up quite full without it starting to howl. Recently I swapped the coupling cap between recovery amp and mixer for a smaller one, to cut a lot of bass from the reverb. That fixed that, but I think I went too far, from 22nF down to 1000pF.

Where did you get the drive current figure? I've never seen any. My build used a 12AU7 in self-split push-pull, feeding a small Oxford Electronics OT. It was pretty loud through headphones, I think it put out about 0.3W into 8 ohms.