First off, here are three versions of the schematic:
This update of the preamp took about 3 days of moderately frenetic work (10/21 to 10/23/99) and consisted of completely rebuilding the entire signal path and most of the power supply.
From this pic you can get a good idea of how I laid out the ground buss(es). It travels the length of the chassis, and is really two separate continuous pieces of household 12AWG solid core, stripped, tinned, and bent to shape. These two pieces are grouped into 1) the “noisy ground” which all of the power supply leads connect to; 2) the second, bigger ground which serves the stages. These two grounds, along with the chassis, are connected together at one point, which is the sleeve of the input jack itself. All local B+ decoupling caps are as close as possible to each tube, and the (-) lead of each cap is attached to the buss right along with the stages’ cathode connections. This is to localize the signal currents as much as possible on the buss. The input stage, the cascoded 6N1P, has its Rk connected right to the input jack sleeve, along with the grid leak resistor and it’s own decouping cap.
A close up of a new design product I’ve stumbled across–single point terminal lugs, with a threaded 6-32 base. You just spin ‘em right on the exposed threads of a bolt and you’ve got a 4KV flashover insulated node. They are VERY sturdy and can go wherever the hell you can make a hole.
Here she is, from stem to stern, in a seductive lanky overhead view. PT (bottom L), heater regulator (bottom R), can caps, 6N1P, 6N1P, 12AT7, 12AX7, 12AX7, 12AX7, 12AX7, at the top are the input 6N1P (R), and another 12AX7 (L). The plates are add-ons to mount more sockets, or to replace octals with 9-pins.
Here’s a shot towards the “noisy” end–the power supply. Notice the two line-side filters on the L, and the B+ FWDB with snubbers on the R.
This is the nasty shit going on next to the snubber capacitor. The vertical voltage scale is 50 mV/div. Notice the sharp spikes of HF, almost 150 mVpk.
The two pics of the scope trace tell a clear story. Keep in mind that the 1N4007 rectifiers that I am using are bypassed with a 1nF cap in series with a 150R resistor (a snubber network). The vertical scale is 10 mV/division, and the horizontal timebase is 20 µS/division for both shots. Yes, I said 20 microseconds. Take a close look at that left shot, which is the (-) terminal of the FWDB, before the CMC. See that damped sinusoidal waveform? That’s ringing at 50 kHz (the period is one division wide) of 40 mVpk. I would hate to see what this looks like WITHOUT the snubbers… I would say they’d easily be in the tens of volts. The shot on the right is the (-) lead right after the CMC, which connects to the “noisy” ground. I will try using a small HV cap across the FWDB, from (-) to (+), to minimize this even further. It should work well.
Schottky diodes, two of three units, using the chassis as a heatsink. They could easily pass 50A this way–but I’m not pulling quite that much heater current. It’s more like 3.6A or so.
This is the raw 8.1VDC rectified heater voltage, at the top of the input power supply cap. The vertical divisions on the scope are set to 0.1V/div. Notice there’s a good 0.2Vpk ripple on this rail. Not bad, but not that good.
Here’s the heater voltage on the downside of the 2N3055 series-pass transistor. Vertical scale is 10 mV/division, so we’ve got about 12 mVpk on the heater rail. Not too bad, especially considering there is NO capacitance hanging off this node yet! I am planning to add another 10,000 mics which should flatten this out even further. In any event, the series pass active filter element offers about 24 dB of hum reduction in this application.
That’s all I’ve got for now. Next mods include a LFO for phase shifting/tremolo duty, adding that extra heater rail capacitance, adding the small bypass cap across the FWDB, and perhaps adding an output tranny for better CMRR. During this time I will probably draw out a detailed schematic. If you ask me very nicely I may give you a copy.
Well I’ve got some stuff here recently, 10/1/99… Here’s the improved heater supply, with 2 mV of ripple on it (5 mV/div). This is after the addition of a 16,000 mic @ 16V cap across the heaters themselves.
PRELIMINARY RESULTS OF JFET CASCODE INPUT STAGE:
From the learned hifi dudes I know on the net, the jfet cascode is the way to go when you want low noise and high gain. Of course, these are desireable features to be had in a guitar amp as well, which lead me to marry the two in a crazy synthesis.
I had previously been using a Svetlana 6n1p in cascode configuration. This started out as a fixed bias incarnation, but on the advice of Randall Aiken I tried a self bias arrangement. His suggestion was to simply tie the upper grid to its cathode through a high value resistor, and place a cap from grid to ground. The bias voltage is thereby developed INSIDE the tube itself due to it’s finite internal cathode resistance. This value is approximately 1/gm.
As he suggested, it did open the sound even more, retaining some of the conventional plate loaded stage’s warmpth and non-linearity. Strictly speaking the lowest thd came from the rigidly fixed upper grid, but who cares about that here?
When connected like this (tubes stacked on top of one another, top tube’s grid held at fixed potential, signal input to bottom tube’s grid, top tube’s plate loaded and signal output) then you are isolating the input and ouptut circuits from one another.
In a normal, plate loaded tube, because the plate and grid are out of phase with one another, as the grid goes negative, the plate goes positive, and vice versa. You can see how the plate voltage will have an effect on the current through the tube, such that when the plate voltage is high the plate current will tend to rise, and vice versa. This is a measure, or more precisely a function of the internal plate impedance of the tube. A pentode, as you may well know, has a very high plate impedance due to the electrostatic screening effect of the screen grid. Thus the plate voltage has very little to do with the plate current. We know how pentodes can have VERY high values of mu for this very reason–the pentode’s plate doesn’t have much of a “say” in the plate current. A triode, on the other hand, is not like this, exhibiting some lower level of plate impedance.
In a nutshell, the lower your plate impedance is, the lower your gain is going to be, since the plate voltage will thereby have a larger bearing on the plate current. There are other factors like transconductance to contend with: a high transconductance tube can have high gain AND low plate impedance. Remember that mu=gm*rp. The mechanism of plate impedance is not hard to understand, since as the tube tries to turn on harder, the plate voltage will drop, and thus the tube is NOT able to turn on quite so hard. It is a sort of degenerative feedback, this plate impedance, and it is what makes the triode stand alone as such a low distortion device. This mechanism does NOT happen to anywhere near the same degree in SS devices that I am aware of.
Now, in the cascode connection, as I’ve said, the input and output circuits are isolated from one another. The bottom tube is operating with a fixed voltage across it, since it’s plate is attached directly to the upper tube’s cathode, and the upper tube’s grid is fixed, thereby fixing the cathode’s voltage as well. The upper tube will keep a relatively stable votlage from grid to cathode, and that keeps the cathode’s potential very stable. So the plate voltage of the lower tube will be essentially unchanging.
When operated in this way, the lower tube is merely changing the CURRENT that it is passing. Since the plate voltage is fixed, the measure of mu never enters the equation. There is no voltage gain at all from this stage. The only parameter of immediate consequence for the lower tube is the transconductance, which is a measure of how much the plate CURRENT will change when you vary the VOLTAGE on it’s grid. This plate current is fed directly into the cathode of the upper tube.
Now, the upper tube does not care what kind of mu it has either. Since the grid is not changing in potential, the grid/cathode voltage is not changing. Instead a varying number of electrons is fed into its CATHODE, and these electrons have to go somewhere. When a greater number flow into the cathode, you can think of it almost as if the cathode is made more negative… there is now a surplus of electrons. Saying that the cathode is made more negative with a surplus of electrons is functionally the same thing as saying the grid has been made more POSITIVE. And as you know, a positive grid leads to an increase in plate current. On the other half of the cycle, as the lower tube shuts off, there will be a deficit of electrons flowing into the top tube’s cathode. Therefore it will have fewer electrons, and the effect is similar to the cathode being made more positive, which is, of course, like the grid being made NEGATIVE, which shuts off the current flow through the tube.
Now, if the upper tube’s plate voltage is allowed to droop too much during turn on, you will have a form of distortion, since the electrons that have collected on its cathode will not be drawn with such force towards the plate. But it will not be very drastic–as you can imagine you cannot have a continual build up of electrons without some kind of current flow occuring, since the effective grid bias on that top tube will continue to change (as I just described above).
Remember though that the ACTUAL voltage, on the upper tube, from its grid to its cathode, will not change much at all. I only mentioned the “effects” to give a better understanding of what’s going on. By now, you should be able to see why they call the cascode a current mode of operation. Only the lower tube’s Vgk (as the input signal voltage), and the voltage on the upper plate (as the output voltage) really change during operation. The upper assembly of a cascode connection is a TRANSIMPEDANCE device, since you feed it a CURRENT signal, and it outputs a VOLTAGE signal. Really, a plate resistor is also a TRANSIMPEDANCE device, but the active device, when paired with the plate resistor above it, completely isolates the bottom tube from the output voltage. That’s the main benefit here.
One downside to the cascode connection is the output impedance is fairly high. Remember you can’t get something for nothing. Just like adding an unbypassed cathode resistor will raise the plate impedance of a conventional plate loaded stage, the same thing happens here. This time the cathode “sees” an impedance equal to the plate impedance of the bottom tube, which can be rather high. As a result, the output impedance of the cascode circuit is approximately equal to the plate load resistance.
In the lastest design, I have replaced the bottom transconductance device with a VERY low noise JFET, the 2sk170. The added benefit of this change is that the jfet has a quite high transconductance of 22mS, compared to the 6n1p’s gm of 7.5 mS, and the 12ax7′s of 1.4 mS. This transconductance is the operative parameter for the lower device, and thus nets me a LARGE ultimate voltage gain. Here’s a picture:
Adding the jfet freed up a single 6n1p stage, which I had contemplated using as an active current source to load the cascode. I had also considered using it as a mu-follower, which is a modified form of the SRPP stack. The only issue here was that the voltage gain from the stage was going to be HUGE (calculated gains of >4000x!) so that would have meant immediate supply-limited clipping.
What I finally settled on was a good old cathode follower, to lower the output impedance down to <200 ohms. This keeps the bandwidth very high and allows me to disregard any loading issues downstream.
If you’re paying attention you’d have noticed that the upper tube’s grid IS held at a rigidly fixed potential. This had to be done in order to keep the Vds of the JFET within reason.
Ultimately, the circuit measured 180Vp-p square wave output for a square wave input of 1Vp-p. This yields a voltage gain of 180x, or approximately 45 dB. That’s NICE gain from a single tube socket and a cheap jfet! For comparison, you can get that kind of voltage gain from an ef86, run from 300V B+, and a 220k plate resistor. The MAJOR difference is that the “downstream” load impedance of the ef86 can only be about 500k, whereas the cascode circuit with cathode follower will EASILY drive 2k load impedances. Admittedly, you COULD just run the jfet off of the ef86 as a source follower, and get low Zout. Thanks to Steve Conner for pointing that out.
As for circuit parameters:
tube type: Svetlana 6N1P
jfet type: 2SK170
cascode tube rp: 47k
rk (follower): 47k
rs (jfet): 100R
cascode tube grid bias: 6.1VDC
cascode tube vp: 190VDC
cascode tube vk: 10VDC
input Z: 2m2
measured voltage gain: 180x
output Z: <200R
output coupling cap: 220nF
I heartily recommend this circuit to everyone! My thanks to Jeremy Epstein for the jfet.
Yes, it’s MORE MODS TO THE PREAMP!!!
Below is a shot of the new and improved power supply. I was rather unhappy with the noise present from the latter half of the preamp, the stages which provide dry/fx mixing, final volume setting, and phase splitting. What’s the quickest way to get a nice, quiet power supply? ACTIVE REGULATION! Luckily I had some IRF840′s on hand, and some heatsinks. I moved some of the power supply nodes around, so that the first cap after the rectifiers is 100uF, with nothing tapped off of it. Next comes the MOSFET regulator, configured as a simple series-pass element. Here’s a picture:
On the left-hand side you can see the small heatsink, gate capacitor, and MOSFET. The heatsink is anchored to the chassis with–believe it or not–hot glue. Now, before you start thinking I’m insane, I have both calculated the dissipation AND observed the actual temperature of the heatsink to be quite low, even after hours of operation. The temperature does NOT rise enough to cause a softening of the hot glue. Yes, you’re right–I SHOULD have anchored it properly with bolts, but this seems to work well enough, and I am lazy. I blew out two MOSFETs before I realized that I did NOT have a gate protection zener in place. Essentially, this zener functions to keep Vgs within a few volts. Otherwise, without the zener, as the supply is turned off and begins to drain down, the source voltage drops much faster than the gates, and fzzzt, you’ve toasted the gate insulation. Ahh well, live and learn. A quick trip to Radio Shack produced two 6.2V zeners, and, wired cathode to cathode, they prevent Vgs from ever going above 7V. So what’s the end result, you ask. Well, as measured before the MOSFET regulator, the B+ rail looked like this:
The vertical scale is 5V/div. That means there’s a good 5Vpp ripple on that rail… not horrendous, considering that there’s about 500VDC at that node, but not good if you’re looking for silence! After the regulator, though, things look considerably better:
Here, the vertical scale is 5mV/div, yielding a ripple of about 2mV on the rail. Now we’re talking! In terms of ripple reduction, we’ve got almost 70dB less noise on the clean side of the regulator. That’s a level of filtration that you can certainly hear–or NOT hear, if you want to get picky about it.
ALL TUBE INPUT STAGE
Although I had spent a good deal of time tweaking the 2sk117 / 6n1p hybrid input stage, I was never satisfied with its ultimate performance. Yes, the input impedance was very very high. Yes, the voltage gain was high too, and the noise was low as well. But there was another issue…
JFETs are no different than BJTs in some respects, and one of those is unit-to-unit variability. As a result, the Idss, or drain current when gate is shorted to source, varies considerably from specimine to specimine. I had a JFET that was mailed to me gratis by Jeremy Epstein, a net-buddy of mine, who picked up a buttload of them so that he could match them himself for hifi use. The JFET he sent me was one that didn’t have a match, a loner in the world of transistors. So, since I only needed one for a mono guitar, he sent one to me.
The problem I ran into was a matter of bias voltage. I had originally used some small (<100 ohms) source resistance to generate SOME negative Vgs to bias the JFET. I didn’t realize it at the time, but in reality the actual guitar signal was quite a bit hotter in voltage than I originally thought. This lead to a clipping of the current through the JFET. Obviously it was one-sided, giving a nice 2nd harmonic distortion to the cascode’s output signal. I might have never noticed, except I could hear it on the clean channel–there was almost a sound like paper on the strings, a slight hint of distortion, that just would NOT go away, no matter what I did downstream. I knew that the input stage had to be clipping somewhere, and it took a while to find it. To rectify the situation, I tried at first to alter the bias on the JFET. I began by raising the potential on the upper tube’s grid, which would raise the cathode’s potential and thus that of the drain of the JFET. I had to keep in mind, though, that the Vds had to be limited to <40V. That really didn’t give me much to play with. Besides, due to the quite high impedance of the JFET drain, the increase in Vds did not produce the increase in Id that I had hoped. My attentions turned to the source circuit. I tried different values of source resistance, to no avail. I then tried tying the source resistor to a negative supply rail of approximately -30V. Obviously the Rs had to be increased as well, but it didn’t make any difference… it still insisted on clipping at some point or another. I had pots everywhere, gator clips running all over the place, and my patience was running out. After all, I KNEW FOR A FACT that the self-bias cascode using all 6n1p’s sounded damned good. So eventually, I returned to that circuit. Before I did, though, I tried a couple of wacky ideas. One was to keep the JFET in the bottom half, but change the upper tube from a cascode to a constant current source. I also tried wiring it as a mu follower. I even tried using a Svetlana SV84 as a preamp tube. [NOTE: they are all pretty damned microphonic, so don't even bother trying it at home!] None of them satisfied me, so the tried and true 6n1p cascode went back in. And let me tell you, it still sounds good. I added a bit more filtration to the power supply node to the input stage, decoupling it further from the supply. Remember, the supply rail itself is now actively regulated, so it’s VERY quiet. I can barely tell it’s on… there’s just the shot and Johnson noise to clue me in. The moral of the story is that the JFET cascode input stage MIGHT have worked if I had a heap of JFETs to keep plugging in there. As it was, I had one, and I didn’t have the patience to sit there and fiddle around with it any longer, even if I DID have others to try. There was almost TOO MUCH GAIN from that JFET cascode, and it was hard to decide what to do with it–do I shunt it to ground, do I use a high series impedance, etc etc. Every choice has its sonic consequences. Thus endeth the story of the JFET cascode input stage. Here’s a copy of the PREAMP SCHEMATIC ; it’s a bit outdated now, but not by much.
Ok, that old schematic is out of date… here’s the latest (1/13/03).