Does anyone have a reason to use one of these over the other? (A) has the advantage of not needing a negative rail, of running on a lower PS, and of not needing the PS to be of any particular voltage, and (B) has the advantage of needing a few less parts. Beyond that, does anyone have experience with performance differences between the two?
Attachments
do you want input signal common mode rejection
the "long tailed" part of "long tailed pair" refers to the common mode independance of the bias - how good an aproxiamation to a ccs it is - a large value R to a negative supply can be an adequate "ccs"
if the R is too small you get larger common mode signal on the plates
bjt ccs can run down to 1 V or less - so depending on signal, tube bias you can get the "long tail" bias independance even without a negative supply
the "long tailed" part of "long tailed pair" refers to the common mode independance of the bias - how good an aproxiamation to a ccs it is - a large value R to a negative supply can be an adequate "ccs"
if the R is too small you get larger common mode signal on the plates
bjt ccs can run down to 1 V or less - so depending on signal, tube bias you can get the "long tail" bias independance even without a negative supply
A and B operate on completely different loadlines.
Putting CCS in tail forces balance but you are still operating across resistive load so the slope of the load line is determined by the load resistor (obviously).
Putting CCS in plate circuit gives you horizontal load lines. However, since you have CCS plate loads there is hardly any signal across the tail resistor to enforce balance so balance will be degraded. Distortion per phase will be excellent, though.
Putting CCS in tail forces balance but you are still operating across resistive load so the slope of the load line is determined by the load resistor (obviously).
Putting CCS in plate circuit gives you horizontal load lines. However, since you have CCS plate loads there is hardly any signal across the tail resistor to enforce balance so balance will be degraded. Distortion per phase will be excellent, though.
There is a way to get away with using both- if the tail current source is a bit sloppy (say, a single transistor CCS) and the plate load CCS are very tight (say, a cascode CCS), surprisingly, one can get excellent balance and CMR while realizing maximum gain and linearity. I didn't believe it myself until I was urged to breadboard an example and it worked just fine. There will be more about this in the soon-to-appear 4th edition of Valve Amplifiers.
Yes, but then you start adding common-mode gain. A similar issue came up in another thread. Diff-mode gain is roughly mu. Common-mode gain is roughly Ranode/Rtail (factor of 2 somewhere too?). For a phase splitter you need low common-mode gain, so if CCS are used in the anodes then you need sloppy ones here and a good one in the tail. SY says the opposite, so one of us is wrong!
At risk of hijacking the thread SLIGHTLY (sorry), but still on the topic of differential pairs, could anyone explain this phenomenon:
In a simulation I built a bog-standard long tailed pair with 6SN7s, sane anode resistors and a CCS tail. Given a single ended signal on one input and earthing the other produces exactly what you'd expect - a fairly low distortion differential output on the anodes. This being a Class-A circuit the distortion will fall with the level of the input/output to nearly immeasurable in the 10s of mV output range.
Why is it when you feed both grids the EXACT same signal (ie connected together) you get a very low residual (fair enough, imperfect CCS, capacitance, etc) BUT this residual has a surprisingly high percentage of distortion despite the signal being so low in amplitude? Even in simulation with perfectly matched tubes.
In a simulation I built a bog-standard long tailed pair with 6SN7s, sane anode resistors and a CCS tail. Given a single ended signal on one input and earthing the other produces exactly what you'd expect - a fairly low distortion differential output on the anodes. This being a Class-A circuit the distortion will fall with the level of the input/output to nearly immeasurable in the 10s of mV output range.
Why is it when you feed both grids the EXACT same signal (ie connected together) you get a very low residual (fair enough, imperfect CCS, capacitance, etc) BUT this residual has a surprisingly high percentage of distortion despite the signal being so low in amplitude? Even in simulation with perfectly matched tubes.
Hey, the big dumb blonde one spent a whole day finding out that this is a most excellent way to make a vacuum tube flip flop! Reducing the quality of each CCS can reduce the instability but I had the best luck with a pure CCS in the tail. Adding CCS's with parallel resistors in the plate circuit improves the gain a bit but too much resistance makes the circuit unstable. The tubes were 6SN7's and the CCS's were 10M45's.
Soon.
Sorry, that's the most I can say. Other than it's going to be awesome, and everyone should buy multiple copies so he can keep bringing excellent Scotch along on our trips.
That would be a common mode signal. About the only thing that does appear at the output are the distortion products the VTs generate themselves, and therefore, aren't suppressed by CMRR.
You will also tend to see more distortion products since CM "amplification" is no longer balanced.
Solid state LTPs do likewise.
I have always been very wary of loading a triode with a CCS. The distortion spectrum of a triode changes with load, and in fact I feel that adjustment of load and operating point of any triode stage is the firstest bestest way to get the sound I want. I can't imagine that a near zero load would be advantageous.
Of course, since the plate load appears in parallel with the input R of the following stage as the composite load of the tube, I suppose things could be adjusted there.
P
Of course, since the plate load appears in parallel with the input R of the following stage as the composite load of the tube, I suppose things could be adjusted there.
P
Well, it's an advantage if you want to maximize linearity. If the load tends toward infinity, the dependence of gain on rp goes to zero. A horizontal load line minimizes the effect of curve bunching as the plate voltage swings toward B+. Granted, not everyone wants linearity to be maximized...
Hi SY
Very interesting information, and opposed to the circuit Morgan proposes in the 3rd edition, namely 'poor' CCS (single MJE350) at the plate, and good CCS (cascoded BJTS) in the tail. Have you also tested the performance of this circuit with wildly unmatched tubes, just like your experiments described here?
Regards, Erik
You are essentially viewing the non-linearity of your CCS. Much of the time this does not matter, as long as the impedance is high enough, but sometimes it could be an issue. My guess is that when people talk about CMRR there is an assumption that the residual common-mode signal is just a tiny copy of the input CM signal, not a highly distorted version of it.
A good CCS needs either to have an almost infinite impedance, or a high and linear impedance. High and non-linear might not be good enough, yet I suspect that this is what some CCS offer.
Nope, I just threw together a bench proto a couple years ago to see if the circuit worked because the use of CCS everywhere seemed... counterintuitive (this was a different circuit than the one shown in VA3). I satisfied myself that the circuit was stable and never went back to it. Hmmm, perhaps it's best to go back to my lab notebook and see exactly how it was set up.
In simulation before I've experimented with CCSes in the plates, each in parallel with a 1Meg resistor and a CCS in the cathode. The CCS has a much higher impedance than 1Meg in theory and thus the balance is set by the resistors (and only providing a light load to the triodes). It does work but if you plot the frequency responses of both phases you end up with a funny mismatched frequency response, presumably due to component capacitances which become significant when the load is so light. I was getting a mismatch of up to 0.5dB at some frequencies. The grid on the undriven tube ends up shunting some of the signal and reducing the 1Meg load
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.