New Vacuum Tube Amplifier

Features

• Output: 4-6550's in triode-mode class AB2 push-pull parallel. About 80 watts RMS per channel.
• No global negative feedback. Several local loops with limited negative feedback.
• Ultra-wide bandwidth Plitron toroidal output transformer.
• Servo to maintain precise dc-balance in the output circuit.
• MOSFET-regulated power supplies. Relative rather than absolute voltage reference.
• Designed using extensive PSpice computer simulation.
• Constructed as a pair of monoblocks.

Schematic diagram


Notes on the schematic diagram

• The power supply has been simplified-- Power transformers and rectifiers have been omitted and some parts have been omitted from the MOSFET voltage regulator circuits: 1N5242 zener diodes between the source and gate and 10k resistors in series with the gate. These parts serve as protection in case of accidental short circuits, but don't affect the operating point. The full power supply schematic is shown below.
• 6SN7's are used instead of 12AU7's for the driver tubes. They have the same plate characteristics, but they have higher maximum plate voltage (450 vs. 330 V) and greater plate dissipation (3.75 vs. 2.75 W per section). Think of the octal-based 6SN7 as a 12AU7 on steroids.
• The NODESET blocks (lower right) initialize dc levels in the bias servo so the simulation runs properly. They don't exist as physical entities.
• All resistors are 1/2 watt unless noted, and with the following exceptions. R3P, R4P and R5C through R8C (68K) are 2W. R9C through R12C (20 ohm) are 1W: I used 2-10 ohm resistors in series to make them. R9S through R12S (200 ohms) are 2W.
• Capacitor voltage ratings: It never hurts to go over the minimum, though the capacitors will be larger and may cost more. I often use caps with higher voltage ratings because I have them on hand or found them at a good price in an electronics surplus shop. Here are some minimum ratings: C1G: 100V. C2G: 400V (600V would be better; necessary without the time delay); C3G and C4G: 400V (600V would be better); C3M and C4M: 400V; C5G through C8G: 600V; CBS2, CBS4, CBS6 and CBX1 through CBX3: 100V. The voltage ratings of many of the power supply capacitors are shown on the circuit board wiring diagrams, below.
• Some of the feedback connections may be a little hard to trace. ORN goes between the 20 ohm output transformer secondary feedback winding and R3F near TU3. Similarly, VLT goes to R4F near TU4. BLU and BLK on the output transformer secondary speaker winding go to the output tube cathode circuits.
• RLS (5 ohms) is a simulated speaker load.
• CRF: All three nodes are connected together.
• BRN and VIO are not used. They are the ultra-linear taps. In the original version of TENA there was a switch to select output tube screen grid connections between BRN and VIO (UL mode) and GRN and YEL (triode mode).

Input stage/phase inverter

Input stage TU1 is a simple voltage gain stage with local negative feedback, derived from the R1B, R1C voltage divider. It is capacitively coupled to split load phase inverter TU2. The capacitor has an unusually low value-- 0.01 µF-- because TU2 has an exceptionally high input impedance-- several Megohms. The advantage of capacitive coupling is that it allows the voltage level in TU2 to be set for maximum output and it allows the ac current in TU2 to be precisely equal to, but 180 degrees out of phase with, the current in TU1. The net ac current drawn by these two tubes from V+420 is therefore zero. This is an effective way of isolating the audio signal from the power supply, which doesn't need to supply ac current. In conventional designs ac signal often has to flow through electrolytic capacitors, which are grungy leaky devices with memory-- harmful to audio quality. I designed TENA to draw zero net ac current from all power supply outputs (easy to do in a push-pull design), at least up to the power level where one of the output tube pairs starts cutting off.

Toroidal output transformer

We chose the Plitron toroidal transformer because of its exceptional bandwidth: -3 dB at over 200 kHz, the result of high primary inductance (the good stuff) and low leakage inductance (the bad stuff-- kind of like HDL and LDL cholesterol)-- much better than can be achieved with a conventional EI transformer. High bandwidth is important because output transformers have an intrinsic second order rolloff, which can make them unstable in the presence of negative feedback unless careful phase compensation is applied (see Feedback and Fidelity). Phase compensation reduces the bandwidth, which is not a problem with the Plitron toroidal transformers. But this bandwidth comes at a price-- toroidal transformers are much less tolerant of dc-imbalance than EI transformers; they may saturate at dc imbalances as low as 8 mA. (I don't know the exact number; I never simulated it.) You would have to set the bias of each tube individually, and then you'd have to worry about how the tubes age. So I designed a bias servo circuit to maintain perfect dc-balance under all circumstances except outright tube failure.

The Plitron PAT 4006CFB 100 Watt toroidal output transformer is not currently listed on Plitron's website, but I've heard (June 2003) that it is available. Contact Norman Woo. The closest models are the 4006, which lacks the special feedback winding, and the 2100-CFB which has a higher primary impedance. The minimum feedback version of TENA (below) works with the 4006.

Bias servo and adjustment

The time-averaged (low pass filtered) dc current of an output tube operating in class AB fixed bias is relatively constant at low power levels but increases at high power levels. For this reason a fixed voltage cannot be used as a reference for biasing the output tubes. One tube (TU9, driven by TU5) operates at fixed bias, and its low pass filtered cathode voltage (CRF) is used as the reference for biasing the other tubes.

The bias servo is illustrated in the lower left of the schematic. It uses the LM324 quad op amp-- cheap but perfectly adequate. Inputs U1A, U1B and U1C of the LM324 compare cathode voltages 10C, 11C, and 12C with reference voltage CRF, which is the voltage on cathode 9C low pass filtered with RBS2 = 33k and CBS1 = 10µF ( located near U1B on the schematic). The LM324 outputs control the P-channel MOSFETs, each of which controls a voltage divider between VBB (-90V) and VOP (+12.5V) to deliver the appropriate bias voltage to the driver grid circuits (BIAS_6, BIAS_7, and BIAS_8). This measures between -45 and -50V in my amplifiers, which operate at 60 mA plate current. Audio purists please note: the servo operates at extremely low frequencies; the op amp and MOSFETs are well outside the audio signal path.

A single potentiometer, RB5 (in the VBB supply, bottom center), controls the bias current directly in TU9, and all the other tubes indirectly through the servo. Bias current may be measured across any of the 20 ohm resistors R9C-R12C as E/20. They should all be the same if the servo is working properly. 1 to 1.2 volts is a good nominal value, corresponding to 50 to 60 mA per tube (70 mA was used in the Dynaco Mark III). Increasing the current increases power consumption and reduces tube life and output power, but moves you closer to Class A (where both tubes always conduct).

Class AB2 output stage and drivers

Class AB2 differs from the more common class AB1 in that the output stage grid is driven positive-- it draws grid current-- at high power levels. Class AB2 has no advantage for output tubes operating in pentode mode and little advantage for ultra-linear mode. But it results in a huge power boost for output tubes operating in triode mode. You can get almost as much power out of class AB2 triodes as you can out of class AB1 pentodes.

If you try to do operate in class AB2 with conventional capacitive coupling, the coupling capacitor starts charging as soon as grid current is drawn. This drives the grid negative-- toward cutoff, and it recovers with the RC time constant of the coupling capacitor and grid resistor. To operate successfully in class AB2, the output stage must be either transformer or direct coupled. I chose direct coupling because interstage transformers are expensive and have limited bandwidth.

The direct coupled drivers are the source of much of TENA's complexity. Because the quiescent grid voltage of each output tube must be set individually to control its quiescent (dc) current, one driver tube (TU5-TU8) is required for each output tube (TU9-TU12). Cathode followers (CF's) were chosen because they have low output impedance and can source the needed output tube grid current. The cathodes have to be somewhere near -50V to properly bias the output tubes. This means the CF must be driven by voltages outside the range of conventional power supplies, hence the need for VDR- and VDR+: the price of perfection. In reviewing the design I find that the driver tubes may be operating a little too conservatively-- dissipating only 0.78 W (of a 6SN7 maximum of 3.75 W). I've discussed driver dissipation under PSpice output, below. I may increase VDR+ from 205 to around 250 V by increasing RD1 from 470k to 680k. This would reduce the power dissipation in MOSFET MD1.

Output tube grid stop resistors R9G-R12G play an important role in TENA's soft clipping. When power levels become high enough level for grid current to be drawn, a voltage drop across these resistors gradually limits the plate current. Soft clipping consists of low order harmonics which have much less adverse effect on sound quality that the high order harmonics characteristic of abrupt clipping. But total harmonic distortion for soft clipping amplifiers tends to be higher. Yes, lower harmonic distortion doesn't mean better sound. See "The great harmonic distortion scam" in Feedback and Fidelity. TENA oscillated when the grid stop resistors were removed. This was the only performance feature PSpice didn't catch. The reason is that the output transformer model is somewhat simplified-- it's extremely difficult to model its distributed capacitance.

Power supplies

The time delay circuit (U3 (the 555B chip), Q1, Relay_SPDT_nb, RT1, CT1, CT2, RT3, D1, RV1, and RT4) has apparently never been implemented. RT4 should be replaced by a straight wire; VBIN is connected directly to NTC (negative temperature coefficient; 50 ohms cold; Mouser527-3504-50) thermistor RV10.

The precise values of most of the capacitors in the power supply, particularly CV1, CV2, CB1, CB2, CD1 and CD2, are not critical. In many cases they were determined by parts availability. If the values are 2 uF or under they are film capacitors. If they are over 2 uF they are electrolytics.

Depending how you count there are two (power transformers), four (rectifier circuits) or six (voltage levels). All use fast recovery rectifier diodes. All except VDR- are taken from the mighty Plitron 854710 toroidal power transformer, which I can't seem to find in their catalog. Toroidal power transformers perform well, but they have less of an advantage than toroidal output transformers-- you don't need wide bandwidth for 60 Hz. The CL80 inrush current limiter limits turn-on current in the tube filaments.