Power amps

A whole tube power amp: phase inverter, output tubes, output transformer, speaker and the feedback loop around them. Pick the tubes and inverter (long-tail pair, cathodyne or paraphase), or one tube single-ended. Set the feedback, presence and resonance, then see how much the loop cuts the gain, how close it comes to ringing, how steady the output stays into a speaker, and which stage clips first.

Open the output tubes

Swipe sideways for the whole drawing. Tap a part to jump to its value. The red numbers are DC voltages to ground at idle.

Loop gain

At the speaker, one cycle

outputthe fundamental alone

Inverter outputs, one cycle

to tube Ato tube B

Harmonics at the speaker

evenodd dB below the fundamental, 2nd to 10th

All values

Reading the charts

Loop gain Picture the feedback loop cut open at the feedback resistor. Send a signal in at the cut, and measure how much comes back around through the inverter, the output tubes and the transformer. That ratio is the loop gain. Where loop gain is well above 1, the feedback network sets the amp’s gain, and distortion and the effect of the speaker’s uneven impedance both drop by about the loop gain. Where it is well below 1, feedback does next to nothing. The top chart shows the loop gain’s size in dB, and the bottom chart shows its phase.

Phase margin At both ends of the audio band, parts of the amp delay the signal, and the delay shows as a turn in phase. In the low bass the coupling caps do it. In the high treble the transformer’s leakage inductance and winding capacitance do it. If the loop gain is still 1 or more at a frequency where the phase has turned a full 180°, the feedback adds to the signal instead of subtracting, and the amp oscillates. Phase margin is how far short of 180° the phase is at the frequency where loop gain drops through 1. Under about 45° the amp rings on a square wave. At 0° it oscillates. If loop gain never reaches 1 at the top of the band, nothing rings there.

Response Volts at the speaker for each volt from the preamp, drawn twice: with the feedback loop closed and with it cut. With the loop cut, the feedback resistor stays connected to the speaker, so the load is the same. The gap between the two curves is the amount of feedback. Presence narrows the gap in the treble, and resonance narrows it in the bass.

Output impedance The impedance the speaker sees looking back into the amp. It is worked out from the output voltage into two different resistor loads. Damping factor is the speaker’s rated impedance divided by the output impedance. With little or no feedback, a tube amp’s output impedance is close to the speaker’s impedance or higher, so the speaker’s bass peak and treble rise come through in the sound. Feedback lowers the output impedance. The reactive load tool shows what a damping factor does to the voltage across a speaker.

Which clips first The inverter’s two outputs are drawn along with the speaker’s, each with its distortion figure. When the inverter runs out of room first, its two outputs flatten. When the output tubes run out of room first, their grids pull current on the peaks, and the speaker’s wave flattens while the inverter stays clean. Feedback hides the first few percent of either kind of clipping, and then the clipping comes on fast. With Clipping picked, Find full power stops just short of whichever part runs out of room first. The output tubes are judged as in the power load line tool. The inverter is judged by one side cutting off or its grid reaching 0 V. With feedback, THD at that point is usually 1.5 to 3 %. With a speaker load, distortion is measured on the speaker’s current, because current is what pushes the cone. Distortion measured on the voltage reads higher, because a speaker’s impedance climbs with frequency.

The three inverters A long-tail pair is two triodes that share a cathode resistor, with a second, larger resistor (the tail) under it. The preamp drives one grid, and the other grid is held still. The shared cathode moves with the first triode, and that drives the second triode the opposite way. A long-tail pair has gain and the most headroom of the three. Blackface Fenders use a 12AT7 with a small feedback leg. The brownface values come from the 6G7-A Bandmaster’s original schematic: a 7025, 47 pF across the plates, and a presence pot in series with .1 µF across the feedback leg. The Marshall and tweed Bassman values use a 12AX7. Their whole feedback leg is a 5 k pot with a cap from its wiper to ground. A cathodyne (split load) is one triode with equal resistors on its plate and its cathode, so the plate swings down as far as the cathode swings up. Its gain is slightly under 1 on each side, so a driver stage comes before it, and the feedback goes to the driver’s cathode. The cathodyne’s cathode idles at a high voltage, which leaves it less room to swing, so it clips sooner. The Princeton Reverb values come from the AA1164 schematic: fixed bias, and 2700 Ω of feedback into a 47 Ω leg. The 5E3 Deluxe values come from the 5E3 on the bench: no feedback, both 6V6GTs on one shared 250 Ω cathode resistor, and a supply that sags. A paraphase is two plain gain stages. The first stage drives tube A. A voltage divider sends a small part of the first stage’s output to the second stage, which inverts it for tube B. That small part has to be the right size to cancel the second stage’s gain. The match drifts as tubes age or the signal gets large, so the two sides become unequal. The paraphase values are typical, not from one amp. The blackface and brownface values start on a 4 Ω load with the feedback from the same tap. The Princeton values start on its single 8 Ω speaker. The Marshall values use feedback from the 4 Ω tap whatever tap the speaker is on, as most Marshalls do. Values not from a drawing are typical of the kind.

Presence and resonance Either control can go on any inverter. Presence connects across the feedback leg and reduces the treble that is fed back. Each kind of presence control is sized for a feedback leg of about 5 k. On a smaller leg, like the 47 Ω leg in a blackface or a Princeton, the control is scaled to fit, with a smaller pot and a bigger cap. That way it works over the same frequencies and doesn’t change the DC voltages. On 47 Ω it comes to 250 Ω and 10 µF. Styles that come with a presence control use the values from their amp. Resonance does the same job in the bass. It’s the SLO-100 depth mod, from a drawing on the SLOclone forums: .1 µF and a 1 M audio pot in series with the feedback resistor, with 4.7 nF across the pot. The .1 µF cap keeps the deepest lows out of the feedback. Turning the pot up adds resistance in the feedback path, while the 4.7 nF lets the mids and treble go around the pot. So only the lows are fed back less. Those values suit a 100 k feedback resistor, as in the Marshall-style amps. For a blackface’s 820 Ω or a Princeton’s 2.7 k the parts are scaled the same way: 10 k, 10 µF and .68 µF on 820 Ω. Without scaling, the cap would cut the feedback up to about 2 kHz, and the knob would lift the mids more than the bass. Scaled, the control raises the bass on every style. It also raises the output impedance in the bass, so the speaker’s bass peak comes through much more. Try it into the generic open-back or closed-back speaker and watch the output impedance chart. The pot is wired as a variable resistor (lugs 1 and 2) with an audio taper, so there is little change in the first half of its turn and most of the change near the top.

Inductor and gyrator resonance These two work from the other side of the loop. A branch runs from the feedback node to ground, across the feedback leg. The branch is 2.2 µF, 2.2 k, a 25 k reverse-audio pot (less resistance as you turn it up) and a coil, all in series. At the frequency where the cap and the coil cancel each other, the branch’s impedance is lowest, and it shunts part of the feedback to ground. Less bass is fed back around that frequency, so the bass comes up there. The inductor kind uses a 500 mH coil with a typical 40 Ω winding, centered near 152 Hz. The gyrator kind uses an op-amp circuit in place of the coil: a TL071 follower with .47 µF, 68 k and 47 Ω. It acts like a 1.5 H coil with 47 Ω in series, centered near 88 Hz. The gyrator is simulated as the real op-amp circuit, so its limits show. Well above resonance it looks like the 68 k resistor. The op-amp can run out of voltage swing or current. Above a few kilohertz its delay shifts the treble by about a tenth of a dB. Both kinds are sized for a 4.7 k leg. On a 47 Ω leg they are scaled down a hundred times. That leaves the gyrator with a 0.47 Ω output resistor, which no real TL071 could drive on a loud note, so treat that one as a what-if. Both lifts are broad. A smaller branch resistor or a bigger coil makes them narrower.

Supply The page opens with a supply that sags, on a power transformer sized for the tubes. Stiff holds B+ steady no matter how much current the tubes pull. Sags with the signal puts B+ behind a resistance and a reservoir cap, as in the power load line tool. That stands for the average effect of the power transformer and rectifier, with no ripple. The inverter is fed from the same supply chain, so its voltage sags too. As B+ drops, the output tubes and the inverter both have less room to swing, and a cathodyne on a lower supply clips sooner. The 5E3 values start with 650 Ω behind 16 µF. That matches the drop through the 5Y3GT in the full 5E3 on the bench, about 12 V for 18 mA more current. A solid-state rectifier sags much less. Its resistance is a few tens of ohms.

Inverter and preamp supply The inverter and preamp are fed through a chain of dropping resistors and filter caps. The chain starts at the screen node, or at B+ when the screens connect straight to B+. Each amp’s chain comes from its factory drawing: two 10 k steps on a Deluxe Reverb, 1 k and 4.7 k on a Twin or Pro Reverb, two 18 k on a Princeton Reverb, one 10 k on a 2203 or 2204, and so on. The preamp isn’t on this page, so a resistor that pulls the preamp’s idle current stands in for it. More resistors stand in for anything else on the chain (a reverb driver, a vibrato, a second channel). That way the inverter’s supply voltage, and how much it sags, come out as in that amp. With B+ set to sag, B+ still idles at the value you set with the chain connected.

Screens The screens can be fed from B+ through a dropping resistor or a choke to a filtered node of their own (B+2 on many drawings), straight from B+, or from a separate supply. Ultra-linear feeds each screen from a tap on its half of the primary. The tap’s share of the turns is set under Output stage. Triode ties each screen to its plate through a small resistor. Ultra-linear and triode both have less gain, power and distortion than pentode, so the same feedback resistor lowers the gain less. A choke has little DC resistance, so the screens idle closer to B+ and their voltage drops less when the amp is pushed. Ripple isn’t modeled, so the choke’s filtering doesn’t show. Only its inductance and resistance do. The Screen resistors menu isn’t shown for ultra-linear or triode. A resistor on each screen limits that screen’s current when its plate swings low. One shared resistor with no cap after it carries both screens’ current. The two screens’ currents at the note’s own frequency cancel in the shared resistor. The even harmonics are left, so the voltage drop grows as the amp is driven harder. Each amp starts with the screen parts from that amp. A choice in either menu stays when you change tubes, pairs or inverter, until Reset amp. On fixed bias, the amp is rebiased after a change to the same idle dissipation, as with Set bias. On cathode bias the cathode resistor stays the same and the idle current moves, as in a real amp.

Real amps Start from opens the power amp of a real amp on the bench, from its factory drawing or a redraw: the inverter, output tubes, transformers, rectifier and bias supply as drawn, with the supply running from the wall voltage. B+ has ripple and sags on a hard note as in the real amp, so the page updates slightly slower. Only part values can be changed on a real amp. Its switches, presence and screen wiring stay as drawn. Values the drawing doesn’t show are typical. A real amp saved to Your power amps is stored with the name of the amp it came from and the values you changed.