I've seen 3 different recommendations for routing filter grounds to reservoir

[Gingertube]

I've posted this link before but worth looking at. It is an extract from Merlins excellent Designing Tube Preamps for Guitar and Bass book (2nd edition).
See the last page for grounding with multi channel amps.
http://www.valvewizard.co.uk/Grounding.pdf

It is basically a bussed ground. The input sockets are RF grounded via ceramic caps to chassis local to the input sockets which extends the chassis faraday cage shielding back along the guitar input cable shield. The signal ground chassis ground connection is made where the two channels meet.

I have also found (on one build) that the bussed earth to chassis connection gave lowest noise when done at the reverb recovery stage.

For single channel amps I do the signal ground chassis ground connection at the cathode of the input tube, always use isolated input jacks.

Cheers,
Ian

[nickb]

But, but... There are some that would argue too much filtering stiffens the power supply and changes the feel/tone in a negative way!

You will still get the important power stage sag as you only do this to the preamp supply. How big the difference is then is a function of how much headroom there was in the specific preamp design.

[pdf64]

As it hasn't been mentioned here yet, it's worth noting that the conductor between the rectifier and reservoir cap is the 'dirtiest' in the amp, a significant noise source, due to the high ripple current spikes required to keep the cap charged.
It should be kept short and away from any sensitive circuits.
Any connection to circuit common / 0V / chassis should be made from the '-' terminal of the cap, rather than the rectifier bridge (or HT CT if a bi-phase rectifier is used).
Considering how the above was implemented in vintage Fenders, it's surprising what a low residual hum level they generally exhibit

[Mike Sulzer]

Each stage can be thought has having a "negative input", the grid, and a "positive input", the bottom of Rk or whatever. So think of each stage as a differential amp, with a very low impedance positive input. These are inverting stages, and therefore the positive input is also the local ground for the output (like an op amp connected to invert). The output of one stage should feed the input of the next as directly as possible, and this means connecting the positive inputs together with the least resistive and inductive path possible, a shot wire, although fanatics might use a flat conductor.

Currents in other "ground" wires should not matter since they do not result in voltage across the "differential input" of the stage in question.

So this is number two, since the screens of the pentodes are signal inputs of a sort and should be referred to the same signal as the other inputs. But that is a detail that does not matter much.

[mikepukmel]

Oh man Ive been reading on this topic for a while. Thought I was mostly there. One point I got, from lots of excellent posts on MEF and also Merlin's book: First run the wiring from the end of the cathode components (resistor or resistor/cap or ground) to the filter cap that fed that tube, then run w wire from there to the star.

And Merlin's filter scheme looks a lot like the top row in the image, but he talks about splitting the resistor on both sides (send/return). Wish I knew enough to figure out what size components to use.

[mikepukmel]

As it hasn't been mentioned here yet, it's worth noting that the conductor between the rectifier and reservoir cap is the 'dirtiest' in the amp, a significant noise source, due to the high ripple current spikes required to keep the cap charged.
It should be kept short and away from any sensitive circuits.
Any connection to circuit common / 0V / chassis should be made from the '-' terminal of the cap, rather than the rectifier bridge (or HT CT if a bi-phase rectifier is used).
Considering how the above was implemented in vintage Fenders, it's surprising what a low residual hum level they generally exhibit

(Sorry I don't mean to hijack the thread, I hope the following question is more or less on topic)
So, take a vintage Fender, like the Deluxe or similar chassis. These amps, have a long wire carrying the dirty signal from the rect. tube to a junction on the board. From there, one wire goes up through the chassis to the reservoir caps, and another lead goes to the standby switch.

There is plenty of room in these chassis, to put the two reservoir caps right next to the power transformer. The dirty lead could be made very short.

It looks like a closer approximation to Merlin's ideal setup.

Another amateur question: Would the old tube amp designs be better if they made the main board wider (say 6" or so), then put the filter caps right on the board very close to the stages that get fed from the particular cap?

[Tom Phillips]

The old Fender layout can certainly be improved. The build method that was used at the time was judged good enough however by both Fender and their customers. Therefore, Fender just moved on with production.

[R.G.]

I always enjoy R.G.'s ground tutorials, but I wonder how many times he's had to cover the same material. If we gathered all his grounding posts and edited for duplicate content it would still be quite a tome,

Yeah, but then you'd find out how I contradict myself by thinking up the same theory and practice more than once.

[R.G.]

How about this. Given the opposing goals driven by the dual duty of ripple filtering and signal bypass it would seem desirable to eliminate one of those. The power supply ripple could be almost completely removed by using either very good RC filtering up front or more simply with a capacitor multiplier. With the ripple currents more less eliminated you can now choose your grounding scheme by considering only signal currents.

Aaaaaand now it gets complicated, at least in concept.

In an ideal world, the first filter cap makes pure DC. In the real world, you can only practically get down to maybe 2% to 5% ripple with practical sized caps, and in the guitar amp world, getting ripple too low causes loss of the valued "sag". At least it does to some ears.

In succeeding stages, earlier than the PI, ripple is already pretty low. You can do a quick and dirty estimate by approximating the ripple reduction by the amount each series resistor/shunt cap in the chain acts as "voltage divider" to 120Hz ripple. This is a crude analogy, but it works for understanding. A 10uF cap has an impedance at 120Hz of 1/(2*pi*120Hz*10uF) = 133 ohms. So a 1K series resistor with a 10uF cap is a drop at 120Hz of 133/(1000+133) = 0.117, so it cuts any 120 hz content by a factor of 10, or 20db. Using a 2K resistor instead of 1K makes for a 23db loss, and using a 1K and a 20uF cap also gives a 23db loss. It's common to see resistors up to 10K for earlier decoupling stages, so you can see up to -40db loss in 120Hz ripple. (Higher ripple components are reduced proportionately more.)

When you use several of these in series you get additive db losses. The first dropper from first filter cap to the screens can have much more than this, because it usually has an inductor, not a resistor, but from there to the PI, you get one ripple reduction. From there to the mid stages and reverb, another loss; from there to preamps, another loss. The losses in ripple ADD. So 20db ripple reduction to the PI, another 20db or more to the mid stages, and easily another 40db to the preamps. You get -80db ripple reduction. Somewhere south of -60db and you're down to difficult to hear at all.

That means that the successive decoupling stages play bigger role in reducing stage to stage interaction than they do ripple. The ripple is already pretty well suppressed.

There is no particular reason that an input stage has to have only one cap. What happens if the first filter cap, the one right after the rectifiers and transformer center tap (if rectifiers are not a bridge) is replaced by two caps, and the two caps are separated both on + and - terminals by resistors? The truly ugly current pulses of rectification and filtering are held inside the first of the two caps, and the resistors and second "first" cap act as a second filter, with the ripple content reduced by Xc/Rtotal. So you can knock the ripple down a bit there, which sequestering the ripple pulses in the first cap. That frees you to do things like shielding the rectifier and CT return wires, stuff like that. You can attach the (-) lead of the second-first cap to the signal ground and have some more isolation of the ripple pulses. The resistors reduce ripple some.

Tube amp makers in the 50s and 60s could not do this economically because filter caps were EXPENSIVE. You can because they're now relatively cheap. A 47uF 450V cap is in the range of $2.50 to $4.00 (Mouser, today, just looked it up) which is cheaper than your time to install it.

Dealing with sag gets complicated, because it is yet another interactive part of the complex grounding/ripple/hum/sag power supply question. This is because sag is not a clear concept, at least not to me. I'd have to go spend more time thinking and modelling to get a better answer. The idea is that sag comes from the power supply dropping a little under load. Probably true, but is it caused by the supply to the OT dropping, the supply to the screens dropping, the supply to the PI, the supply to the preamp stages? It is probably confounded with increased distortion in one or more of the above as a result of a fixed stage bias moving around as the power supply moves around, and causing not just smaller signal, but possibly slightly more distortion either independently of or in concert with the lowered output ability. And remember, results on one amp or even a few amps don't tell you the whole story. Remember the blind men and the elephant.

There are ways to force sag to be anything you'd like it to be. But they require knowing what to do and what it affects to do them well. Just dinking around with filter cap values and placement is a form of easter egging.

You could do a capacitor multiplier, and in fact that is a GREAT idea for preamp stages. A simple setup would use a power MOSFET to do this. But some thinking shows that a capacitive multiplier cannot really store energy like a cap does, it only uses the power device to fake a capacitor's low impedance for smoothing. The actual energy has to come from some earlier cap or the multiplier sags out of the linear range. Even better, you could just put a zener in for the cap and get a fixed regulated voltage for your preamp stages. Almost the same circuit. You get fixed voltage. If you want a varying voltage, use the cap, not the zener.

This works poorly for stages that use a lot of power. Effectively, you have to build a power "regulator/multipler" that can run the whole stage power, and have a bigger cap to store the energy for it, and lose the voltage needed for headroom to make the multiplier/regulator work right.

On the other hand, you can FORCE sag any time you want by messing with the DC voltage on the reference to the multiplier/regulator and just making the output voltage sag any way you want it to. You wind up spending efforts on circuits to get control. In a world of cheap power parts and other electronics, the major cost to this is that you have to go learn how to do it.

[pdf64]

...take a vintage Fender, like the Deluxe or similar chassis. These amps, have a long wire carrying the dirty signal from the rect. tube to a junction on the board. From there, one wire goes up through the chassis to the reservoir caps, and another lead goes to the standby switch.

There is plenty of room in these chassis, to put the two reservoir caps right next to the power transformer. The dirty lead could be made very short...

Yes, that would be better.
Don't forget the black wire coming back from the doghouse, that carries the same nasty current signal as the red wire, so it's a double win.
I don't think it actually causes much trouble in a Fender because the wires are routed away from sensitive parts of the circuit.

Note that the PT HT winding CT should connect to the -ve lead from the reservoir cap, and then the that lead should extend to connect to the chosen circuit common 0V star point.

...Would the old tube amp designs be better if they made the main board wider (say 6" or so), then put the filter caps right on the board very close to the stages that get fed from the particular cap?

Yes, I think that would naturally lend itself to better 0V practice.

[Malcolm Irving]

... In succeeding stages, earlier than the PI, ripple is already pretty low. You can do a quick and dirty estimate by approximating the ripple reduction by the amount each series resistor/shunt cap in the chain acts as "voltage divider" to 120Hz ripple. This is a crude analogy, but it works for understanding. A 10uF cap has an impedance at 120Hz of 1/(2*pi*120Hz*10uF) = 133 ohms. So a 1K series resistor with a 10uF cap is a drop at 120Hz of 133/(1000+133) = 0.117, so it cuts any 120 hz content by a factor of 10, or 20db. Using a 2K resistor instead of 1K makes for a 23db loss, and using a 1K and a 20uF cap also gives a 23db loss. It's common to see resistors up to 10K for earlier decoupling stages, so you can see up to -40db loss in 120Hz ripple. (Higher ripple components are reduced proportionately more.)

When you use several of these in series you get additive db losses. The first dropper from first filter cap to the screens can have much more than this, because it usually has an inductor, not a resistor, but from there to the PI, you get one ripple reduction. From there to the mid stages and reverb, another loss; from there to preamps, another loss. The losses in ripple ADD. So 20db ripple reduction to the PI, another 20db or more to the mid stages, and easily another 40db to the preamps. You get -80db ripple reduction. Somewhere south of -60db and you're down to difficult to hear at all.

That means that the successive decoupling stages play bigger role in reducing stage to stage interaction than they do ripple. The ripple is already pretty well suppressed.
... .

Very true, but let's not forget that each gain stage is amplifying whatever ripple is present at its input by 30dB or whatever, and of course these dB are also additive for a sequence of gain stages.

[nickb]

You could do a capacitor multiplier, and in fact that is a GREAT idea for preamp stages. A simple setup would use a power MOSFET to do this. But some thinking shows that a capacitive multiplier cannot really store energy like a cap does, it only uses the power device to fake a capacitor's low impedance for smoothing. The actual energy has to come from some earlier cap or the multiplier sags out of the linear range. Even better, you could just put a zener in for the cap and get a fixed regulated voltage for your preamp stages. Almost the same circuit. You get fixed voltage. If you want a varying voltage, use the cap, not the zener.

This works poorly for stages that use a lot of power. Effectively, you have to build a power "regulator/multipler" that can run the whole stage power, and have a bigger cap to store the energy for it, and lose the voltage needed for headroom to make the multiplier/regulator work right.

On the other hand, you can FORCE sag any time you want by messing with the DC voltage on the reference to the multiplier/regulator and just making the output voltage sag any way you want it to. You wind up spending efforts on circuits to get control. In a world of cheap power parts and other electronics, the major cost to this is that you have to go learn how to do it.

When I suggested it I thought it was obvious that it was only practically applicable to the preamp. I guess I should have spelt it out.

[trobbins]

We seem to be venturing in to the world of hi-fi when aiming to suppress supply ripple voltage to lower and lower levels, and whether that ripple voltage then begets a ground link ripple current that then ingresses the signal. At least hi-fi allows global or multistage feedback to further suppress the effect of hum and ripple that enters within the feedback loop.

I find it very hard to measure any meaningfull level of rms ripple on a preamp stage rail with my Kiethley 197, which has a 200mVFS range. I haven't yet set up to measure that hum spectrum using a soundcard to get better comparative levels of attenuation as the typical ladder RC filter extends back to the input stage - I'd guess there must be some blogs or threads going in to that. I sort of like the idea of taking a comparative output signal measurement for when a known resistance is added to a particular 0V inter-stage link in config#1, to see how much contribution a parasitic level of wire resistance would likely add.

[Gregg]

I'm not an expert in theory but from experience and from what I've seen in different amps I can say that many grounding approaches will work, some of them as far from the star ground method as you can get. In PCB amps star ground is almost never used especially in multi channel "multi switching" amplifiers. It's mostly bus ground and sometimes all caps are clustered together at the power supply section of the PCB which is a violation of the theory but it works - the amp is quiet, doesn't oscillate etc.
The same applies for where to ground to chassis. Both at the input jack and first reservoir cap will work. Sometimes the second is quieter.

[Malcolm Irving]

We seem to be venturing in to the world of hi-fi when aiming to suppress supply ripple voltage to lower and lower levels, and whether that ripple voltage then begets a ground link ripple current that then ingresses the signal. ...

True enough, but I've read about a very high gain guitar amp (think it was a Dumble) that claimed to have an overall gain of near 1 million. That's a gain of 120dB !

EDIT: In high gain amps the signal is limited by clipping at various stages, but the buzz, hiss and hum start at a much lower level than the guitar signal but are then not clipped.