I suppose I'm asking for the short answer here, where would you use them? The short answer has to be "wherever the signal swings are biggest. The plate resistors for the PI are the prime candidates. The stage just before the PI is the next best. Use metal film for low noise in input stages, and carbon film in PI and output stages. This is really good news for people making their own amps or modding. You make the front end quiet with metal film and get tube squish from CC's in the later stages where the noise doesn't hurt you so much. Before changing those noisy plate resistors, check their value. If they're off spec (didn't they have like 20% tolerance or something) or have drifted, it might be possible that the off spec value is contributing to the uniqueness or tone of an amp. Replacing a drifted 150K or 80K plate resistor with a 100K might have some effect, I guess (maybe not?). Just replace the drifted one with as close a value as possible if you don't want to change the amp. So far we have: 1. high voltage across the resistor is necessary (in the range of 50 or 100V on up) 2. large signal swings across the resistor are needed (ideally, a large fraction of the static DC voltage} 3. only positions in the amp that have both high DC voltage and wide signal swings as in 1 and 2 will give you enough resistor distortion to benefit from; other places should be chosen for low noise and/or economy. 4. resistor power rating should be the minimum needed to work for a reasonable life in the circuit to maximize resistor distortion. Maybe a good guideline is that the dissipation should be selected to be as close to two times the average dissipation as possible. 5. as a corollary to the power guideline,we should be prepared to replace CC's every few years as the life at high temp makes them drift and get noisy(-er). I'd like to comment on #3. I support using a carbon composition resistor for a first stage. Someone might want to use an effects pedal to intentionally overdrive the first stage, so now there would be enough signal swing on the first stage to allow the Carbon Composition Tone Mojo to take place. As for #4, the benefit of using carbon composition resistors drops significantly for power ratings greater than 1/2 watt. Here are the numbers from Radiotron Designer's Handbook where the percent change is resistance of a 1M carbon composition resistor is measured for a given drop in applied voltage: 1/4 watt: = 2.1%/200V 1/2 watt: = 2.5%/300V 1 watt: = 1.3%/500V 2 watt: = 1.5%/500V For a given change in voltage, the 1/2 watt resistor changes its value 3 times more than the 1 watt resistor. These values are for a 1M resistor and the changes for a 100K resistor should be even less. I suppose there might be some benefit to using 1- or 2-watt carbon-composition resistors, but I would suggest that the amplifier should be designed to use 1/2-watt plate resistors. By the way, the JAN-R-11 spec for carbon composition resistors limits the voltage coefficient to .035% per volt for 1/4 and 1/2-watt resistors, and .02% per volt for 1-watt and above. I'm also wondering if it is useful to use carbon composition resistors in a phase splitter. My understanding is that second order harmonics generated by the phase splitter and following circuitry are canceled out by the push-pull output section. Also, I'm superstitious about purposely introducing distortion inside a feedback loop. In any case, the phase inverter equations for the typical Fender amp show that it is already unbalanced, even if the triodes and resistors are perfectly matched. However, I would guess that a split-load inverter would react differently to carbon composition resistors. The cathode output would not be affected by a change in its load resistor, but the plate voltage will. The second harmonic is generated only on one side of the push-pull circuit, so it shouldn't be cancelled out. Of course, if we want more second harmonics from our amplifiers, we could just lower the plate supply voltage and bias the triode closer to cutoff. But where's the Mojo in that ? I support using a carbon composition resistor for a first stage. Someone might want to use an effects pedal to intentionally overdrive the first stage, so now there would be enough signal swing on the first stage to allow the Carbon Composition Tone Mojo to take place. I suspect that if you're overdriving the first stage enough to cause the big signal swings, the tube distortion itself might well overpower the resistor distortion. It's a good place for some further work. In any case, we can definitely say that using them in the first stage will enhance the excess noise they carry. On the other points... Here are the numbers from Radiotron Designer's Handbook ... I would suggest that the amplifier should be designed to use 1/2-watt plate resistors. Or, looking at the numbers, even 1/4W where those can be used safely - it's change per volt is even larger. The trick seems to be to use the lowest dissipation rating that doesn't excessively impact the resistor's life. And that's what rule #4 says. However, it does point to a quantizing effect that might explain some subtle differences in amplifiers. If a resistor is used at just the hairy edge of its safe voltage rating, the effect is maximized. Depending on the design of the amp, dropping the voltage to a stage to get the dissipation down just enough to use the next lower power rated resistor might let you get more resistor distortion by stepping to the next lower size. the JAN-R-11 spec for carbon composition resistors limits the voltage coefficient to .035% per volt for 1/4 and 1/2-watt resistors, and .02% per volt for 1-watt and above. ... if we could find JAN qualified resistors any more except in out of the way crevices 8-) My understanding is that second order harmonics generated by the phase splitter and following circuitry are canceled out by the push-pull output section. No. The output stage PP setup only cancels even harmonics generated within that stage. If it comes from the PI, it's not cancelled. Also, I'm superstitious about purposely introducing distortion inside a feedback loop. In any case, the phase inverter equations for the typical Fender amp show that it is already unbalanced, even if the triodes and resistors are perfectly matched. Superstitions aside, the feedback around the stage reduces the internally generated distortion by the feedback factor. That cooks out the carbon comp mojo you worked to get in there. This may be one (more) reason why the AC-30 sounds so different, even clean - there is no feedback loop to reduce resistor (or tube) distortion where most of it is generated. The cathode output would not be affected by a change in its load resistor, but the plate voltage will. The second harmonic is generated only on one side of the push-pull circuit, so it shouldn't be cancelled out. I answered KG's (great initials, those!) note first, and now that I read your post in detail, I do tend to agree. You're right, the cathode output will be much less affected by the resistor change, as it's acting as a cathode follower with unity feedback, so the actual resistor value is not what is driving the output. On the plate side, though, the resistor value is all that's there. The resistor change will only be generated by the plate side. However, there's an offsetting mechanism in the design. Split load phase inverters can only exist where the output tubes only need a smallish voltage to drive them, like with the EL84. If you did a diffamp PI, the EL84's would still only need the same amount of drive, and so the voltage effect would be limited to the same size as in the plate side of the split load - well, maybe 3db more, since it's generated in two plates. Which leads to another possible design-for-resistor-distortion trick. In the modern, enlightened design concensus developing here, we deliberately don't massively overdrive grids because of blocking distortion. However, massive, almost-full-power-supply swings are what maxes out the resistor distortion, and would also max out the tube's inherent nonlinearity. It seems like what we might want to do is to quickly blast the signal up to close to a 100V swing, then divide it back down with resistors to an appropriate level for the next grid. The distortion is made in the plate resistor swinging massively, and stays in the signal that's divided down to be amplified up in the next stage again. For EL84's you only need maybe 24V p-p to turn them fully on or off. Why not make the signal level in the PI driving them be 100V, then do a 4:1 divide down to drive the grids? This seems like another advantage of dividing down between stages beyond simply stepping out of blocking. It lets you max the soft distortion in the driving stage without stepping over into clipping. R.G. The reason I said that was that specifically for 6L6 , EL34, and other big pentode/beam tube outputs, the signal voltage almost qualifies. They're biased at maybe -30 to -40V, and the signal swings from 0V (just before grid clipping) to maybe -60 to -80V, so they get into the range where CC mojo might kick in. I have my suspicions that this might not help. Here's why. When one tube is at 0V, the other is at minus max, and also max distortion/squashing. On the other half cycle, the tubes swap. In a class AB amp with no feedback, that means both half cycles are treated to the same squashing, so the second harmonic distortion has been turned into symmetrical squashing by the (unfortunate) magic of push-pull AB operation. It's still soft squashing, so it may sound OK, even good. But it's not quite the same. Feedback around the output stage will reduce it as well. So - yes, it's an option, the excess noise doesn't have much gain to become evident through, but it will not necessarily have the same results as in a single ended preamp stage. R.G.