The Emperor's New Amplifier TM
by Norman L. Koren
Updated June 4, 2003
November 3, 2002: I've added diagrams of the hand-wired circuit boards and more footnotes to the main schematic.
February 2003: Paul Heggeseth is building a version of TENA (brave soul). Check his site for updates.
| Schematic | Design philosophy
| Input stage/phase inverter
Toroidal output transformer | Bias servo and adjustment | Feedback | Class AB2 output and drivers
Power supplies | MOSFET regulators | Minimum feedback modification | Performance
Sound | Additional notes: underside and circuit board diagrams.
I spent much of 1997 designing and building my ultimate dream amplifier, which I named "The Emperor's New Amplifier"TM (TENA) for a quality it shares with the fabled emperor's wardrobe-- transparency. It was also an oblique reference to the marketing hype that pollutes high-end audio. I thought about commercializing it, especially when the lab where I worked announced it was shutting down, but I soon realized that marketing high-end tube amplifiers is not a reliable way to make a living. (I'm paycheck-addicted.) Counterpoint Electronics, a high-end audio manufacturer located in a 45,000 square foot building five miles from where I lived at the time, vanished overnight. Then I thought I might write a magazine article, but the move to Colorado distracted me.
TENA has been working reliably since 1997, and in a way it's ruined me as an audiophile. I've become contented. I'm so pleased with it I've abandoned the quest for audio perfection. (Photography keeps me busy enough.) Now the time has come to share it-- via the Internet. Brace yourself for a tour de force of amplifier design. It's arguably the most sophisticated vacuum tube amplifier ever. But be forewarned if you plan to build it: It's not a simple project. Absolutely not for beginners!
Notes on the schematic diagram
My goal was to design a high-powered wide-bandwidth amplifier with the finest sound and to keep it simple-- but not too simple. I rejected single-ended (SE) designs because of their low power and limited bandwidth. I wanted to see if I could get similar sound quality-- perhaps better-- with an efficient push-pull design. And I wanted to do it my way.
I aimed for wide bandwidth, but not too wide. Very low frequency 1/f noise and radio stations don't enhance music. Most of the coupling circuits have an RC time constant around 0.05 seconds, equivalent to a -3 dB frequency of 3 Hz-- low but not too low.
No net ac current is drawn from any power supply tap. Pairs of tubes draw equal but opposite current. This effectively removes the power supply from the signal path and reduces the need for voltage regulation. Nevertheless robust MOSFET regulators are employed for ultimate sonic refinement.
I avoid global negative feedback-- a loop from the amplifier output to its input. There are several local loops-- around one or at most two gain stages. Local feedback loops have numerous advantages and no adverse effect on sound quality. Stability is much easier to maintain. The extreme bandwidth of the output transformer allowed me to put a moderate amount of feedback in the output circuit while maintaining excellent bandwidth and stability.
I originally designed TENA to be switchable between triode mode (plate connected to screen grid) and ultra-linear (UL) mode (plate connected to output transformer 40% taps). Output modes are discussed in Feedback and Fidelity, Part 2. In conventional class AB1 operation, UL mode has much greater output power than triode mode, but it has higher output impedance and poorer linearity. In class AB2 operation (where grid current is drawn when the grid is driven positive with respect to the cathode) I got plenty of power in triode mode, so I eliminated the switch. (Duncan's Amp Pages has a nice little discussion of operating classes.) A triode behaves like a pentode with built-in local negative feedback. A pentode's screen grid suppresses feedback from the plate, but it can be restored by connecting it in triode mode (connecting the screen grid to the plate instead of a fixed high voltage). Many audiophiles despise negative feedback (see Feedback and Fidelity), but ask them which tube type they prefer...
Try as I might I couldn't keep it simple. Class AB2 operation requires individual cathode follower drivers, which requires extra power supplies. So it got complex, but every part has a purpose; none is superfluous. I did nothing because conventional wisdom told me to.
Components are operated conservatively-- well under maximum power dissipation and voltage ratings. This ensures maximum reliability.
I tried to avoid exotic, hard to get, or expensive parts. I chose the 6550C over "real" triodes because it's moderately priced, easy to find, rugged and has high power dissipation. It behaves like a real triode when connected in triode mode. Most "real" triodes have directly heated filaments which are difficult to integrate into feedback circuits.
Several aspects of the design philosophy implemented in TENA-- maximizing stability, minimizing RF interference (which can result in"gritty" sound and "listener fatigue"), and soft clipping-- are discussed in detail in Feedback and Fidelity. In writing this article I've become aware of a few things I may change if I have the time or make TENA into a commercial product, but I have no such plans at the moment.
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.
I won't go into detail about modeling toroidal transformers-- Plitron
has a treasure
trove of fine articles on its Website, and I said a few things
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).
The cathodes (including the feedback winding center tap WHT) are referenced to VBB = -90V. This gives TU3 and TU4 the huge voltage swing required for the zero-gain cathode followers that drive the triode output stage. The large voltage swing is the reason that the 6SN7 was chosen over the 12AU7, which has a similar plate curve. If the output stage were operated in ultra-linear mode, it would have higher gain and wouldn't require as large a voltage swing on TU3 and TU4.
There is also a lesser feedback loop between the speaker windings (BLU, BLK) and the output tube cathodes. This loop provides less than 2 dB of feedback-- a very small amount. If I were an anti-feedback purist, I'd eliminate the main loop (to the cathodes of TU3 and TU4) as described below, but I'd keep this one, which qualifies as a local feedback loop because it involves only one amplification stage-- the output stage.
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.
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
|POWER SUPPLY DESCRIPTION|
|Main high voltage supply for output tubes (TU9-TU12). About 450V unregulated. Uses 4 MUR4100's in a bridge. NTC thermistor RV10 (Mouser527-3504-50) limits relay turn-on current. Uses two large filter capacitors CV1 and CV2 and choke (inductor) LV1. V+450 is the input to V+420V and VDR+. [ A circuit employing a 555 timer was designed to delay turn-on for about 20 seconds, giving tube cathodes time to warm up. The intent was to prevent a turn-on voltage surge which could potentially damage components such as filter and coupling capacitors. RT4 allows the supply voltage to gradually build up before the relay turns on. This protects the relay contacts by reducing the transient turn-on current. Its value (15k) was not highly optimized. In November 2002 I examined TENA, and found that the time delay circuit had never been implemented. Reliably is fine without it. Amazing what I've forgotten since 1997. ]|
|420V regulated supply for input stages (TU1-TU4). Could be considered part of the main high voltage supply. Regulated by IRF820 N-channel MOSFET with a modest heat sink.|
|Positive supply for driver tubes (TU5-TU8). About 200V. The IRF820 N-channel MOSFET regulator is controlled by a voltage divider (RD1, RV6) in the 420V supply. Requires a substantial heat sink. In reviewing the design I see it might be slightly perferable to to source VDR+ from V+420.|
|Negative supply for driver tubes (TU5-TU8). About -305V. Uses a separate isolation transformer (1:1 voltage ratio) with a voltage doubler. Unregulated.|
|-90V negative bias supply for output tubes, also connected to TU3 and TU4. Uses voltage doubler taken from power transformer bias winding. Regulated by IRF9610 P-channel MOSFET with a modest heat sink.|
|12.5V regulated supply for bias servo. Uses voltage doubler taken from filament winding and LM317T regulator.|
I tried and failed to keep it simple, but the sophisticated power supply is one of the keys to TENA's superb sound quality.
The 420V supply (V+420) on the bottom of the power supply schematic is a good example. Output voltage (V+420) is controlled by the gate of MV1-- set by the RV5, RV6+RD1 voltage divider and the RV7, CV4 low pass filter. RV7 can have a very high resistance (2.2 Megohms) because the gate has infinite impedance. This allows the use of a film capacitor instead of a leaky, noisy, unreliable electrolytic for CV4. When I hooked TENA to a pair of extremely efficient loudspeakers (normally driven by SE triodes), I heard a very slight hum. Because of this I would consider redesigning the low pass filter in two stages (adding an R and C) for a future iteration of TENA.
The primary function of voltage regulators is to maintain constant supply voltage independent load current. They are particularly useful when a large voltage drop is required, as in the VDR+ supply, which drops over 400V to 200V. TENA uses relative rather than absolute regulation. The reference for a relative regulator is a multiple of the line voltage. It it's done well, the reference voltage will change slowly and without transient noise in response to changes in the line voltage. An absolute regulator requires a fixed voltage reference. High voltage fixed references, suitable for vacuum tube circuits, have their share of problems. For example zener diodes can be plagued by noise and temperature coefficient issues. Since the amplifier's gain is hardly affected by the supply voltage, I prefer relative regulation; I know of no disadvantages. It's simple, reliable and quiet. Some other MOSFET regulator designs can be found in Duncan's Amp Pages and the Mods section of Welborne Labs' catalog.
Removing this loop would increase TENA's gain a bit much, so I apply local feedback (the good kind with no adverse side effects) to the cathode circuits of TU3 and TU4. To eliminate the main feedback loop, remove all conections from the output transformer secondary cathode feedback winding (ORN, VLT) and remove C3M and C4M. Reconfigure R3C and R3F as shown on the right. Changes to TU4 should exactly mirror the changes to TU3.
I haven't tried this yet. I'll report on it if and when I do. Simulated frequency response is virtually unchanged. One significant advantage: You can use the Plitron PAT 4006 output transformer, which is a part of their current product line.
Few tube amplifiers come close and none get better-- 1 dB down (±0.5 dB) at the output (SPKR) at 9 Hz and 80 kHz with no irregularities at any of the intermediate stages. Perfect match between phase inverter outputs below 10 kHz. No more than 0.2 dB difference above 20 kHz. Response is stable even under difficult loads. A 10 kHz square wave has only a slight overshoot and no ringing when TENA is loaded with a 2 uF capacitor (a similar load to a large electrostatic speaker) in parallel with a 5 ohm resistor. Most competitive "high-end" amplifiers show severe ringing under these conditions.
The PSpice model for soft clipping is rather crude; actual onset is more gradual than the simulation indicates. Below 57 Watts RMS, distortion is vanishingly low. It increases gradually above 57 Watts, but is of very low order. That's why I tell people the output is "about" 80 Watts RMS. The advantage of a design with gradually increasing low order distortion, as opposed to an extremely linear design that clips abruptly, is discussed in Feedback and Fidelity.
Output impedance is around 0.9 ohms. The power output and damping factor are sufficient to drive nearly any loudspeaker.
Koren Vacuum tube audio page | Photography
Feedback and Fidelity | Improved vacuum tube models for SPICE simulations
TENA has three hand-wired perfboard circuit boards (Radio Shack 276-1395).
PC boards would be preferable, but the perfboards have been reliable. The
scales are in inches. The drawings were done in Corel Draw, which probably
isn't ideal, but it was on my computer and I knew how to use it. The illustrations
show the component side of the boards. The wiring on the opposite side
is shown as thick gray lines. When I built thems I printed out these diagrams
along with mirror images so I could see the wiring side in the proper orientation.
The ruler shows inches.
GND is ground.
The original Corel draw files can be downloaded by shift-clicking on TENAboards.zip.
**** SMALL SIGNAL BIAS SOLUTION TEMPERATURE = 27.000 DEG C
NODE VOLTAGE NODE VOLTAGE NODE VOLTAGE NODE VOLTAGE
( 1C) 9.0105 ( 1G) 7.6727 ( 1P) 235.3900 ( 2C) 109.9600
( 2G) 108.6600 ( 2P) 302.9800 ( 3C) -70.1920 ( 3G) -82.5550
( 3P) 179.6100 ( 4C) -70.1920 ( 4G) -82.5550 ( 4P) 179.6100
( 5C) -49.6790 ( 5G) -62.7190 ( 6C) -49.6930 ( 6G) -62.7340
( 7C) -49.6930 ( 7G) -62.7340 ( 8C) -49.6930 ( 8G) -62.7340
( 9C) 1.1049 ( 9S) 438.3000 ( 10C) 1.1033 ( 11C) 1.1033
( 12C) 1.1033 ( 12S) 438.3000 ( A_6) 1.1080 ( A_7) 1.1080
( BLK) 0.0000 ( BLU) 0.0000 ( BRN) 438.6700 ( B_6) -62.7350
( B_7) -62.7350 ( B_8) -62.7350 ( CRF) 1.1094 ( GRN) 438.6700
( INP) 0.0000 ( ORN) -82.5560 ( VA6) 6.6810 ( VB2) -102.4900
( VB3) 435.7900 ( VB4) 367.8200 ( VB6) 10.5180 ( VBB) -82.5560
( VBS) -115.0000 ( VD-) -305.0000 ( VD1) 208.8900 ( VIO) 438.6700
( VLT) -82.5560 ( VOP) 12.5000 ( YEL) 438.6700 ( SPKR) 0.0000
( VBIN) 446.0000 ( VDR+) 204.9200 ( VDR-) -257.3600 (V+420) 413.8200
(V+450) 438.6700 (BIAS_5) -62.7190 (BIAS_6) -62.7350 (BIAS_7) -62.7350
(BIAS_8) -62.7350 ($N_0001) 7.6727 ($N_0002) 7.6723
December 8, 2003
|Images and text copyright (C) 2000-2012 by Norman Koren. Norman Koren lives in Boulder, Colorado. Since 2003 most of his time has been devoted to the development of Imatest. He has been involved with photography since 1964. Designing vacuum tube audio amplifiers was his passion between about 1990 to 1998.|