Ti Kan's Millett Hybrid Headphone Amplifier
See the companion STEPS
v1.2 power supply
- This low voltage
hybrid headphone amplifier, designed by
Pete Millett and published in the 11/02 issue of
AudioXpress magazine, was built by Ti Kan
revised PCB from n_maher and drewd.
- A companion outboard STEPS v1.2 power supply, set to 27V, is used
to power the amplifier. See my STEPS page
- For all RMAA and other testing an adjustable lab bench regulated
DC power supply (set to 27V) was used.
- Tubes used: GE 12AE6A, Westinghouse 12FM6 or Raytheon 12FK6.
- TI/Burr-Brown BUF634P output buffers, 221Ω bandwidth resistor.
- Vishay-Dale RN60D mil-spec 1% metal film resistors
- Wima MKP polypropylene capacitors
- Nichicon HE low-ESR high reliability electrolytic capacitors
- Alps RK27112A blue-velvet volume control potentiometer
- Enclosure based on a Hammond 1455N1602BK extrusion and custom-made
acrylic front and rear panels and top cover.
- A superbright (3.2 candela) 3V blue LED is mounted under each
tube socket to jazz up the appearance.
RMAA Test Results
RightMark Audio Analyzer
software, running on a Toshiba 2.8GHz Celeron laptop computer via an
M-Audio Firewire Audiophile
mobile interface running in 32-bit, 96KHz mode.
This test provides data and graphs of frequency response, noise, dynamic
range, total harmonic distortion, intermodulation distortion and
stereo crosstalk performance.
The following are test results of the Millett Hybrid amplifier with
TI/Burr-Brown BUF634P output buffers. This is the original design
configuration. The BUF634P's pin 1 bandwidth resistor was 221Ω.
GE 12AE6A, Westinghouse 12FM6 or Raytheon 12FK6 tubes were tested
in the amplifier, each at several bias points and load impedances
to explore the effects of changing tube bias and load.
Millett Hybrid with 12AE6A, 12V tube bias, 330Ω load
Millett Hybrid with 12AE6A, 13.5V tube bias, 330Ω load
Millett Hybrid with 12AE6A, 15V tube bias, 330Ω load
Millett Hybrid with 12AE6A, 16.5V tube bias, 330Ω load
Millett Hybrid with 12AE6A, 12V tube bias, 33Ω load
Millett Hybrid with 12AE6A, 13.5V tube bias, 33Ω load
Millett Hybrid with 12AE6A, 15V tube bias, 33Ω load
Millett Hybrid with 12AE6A, 16.5V tube bias, 33Ω load
Millett Hybrid with 12FM6, 12V tube bias, 330Ω load
Millett Hybrid with 12FM6, 13.5V tube bias, 330Ω load
Millett Hybrid with 12FM6, 15V tube bias, 330Ω load
Millett Hybrid with 12FM6, 16.5V tube bias, 330Ω load
Millett Hybrid with 12FM6, 12V tube bias, 33Ω load
Millett Hybrid with 12FM6, 13.5V tube bias, 33Ω load
Millett Hybrid with 12FM6, 15V tube bias, 33Ω load
Millett Hybrid with 12FM6, 16.5V tube bias, 33Ω load
Millett Hybrid with 12FK6, 12V tube bias, 330Ω load
Millett Hybrid with 12FK6, 13.5V tube bias, 330Ω load
Millett Hybrid with 12FK6, 15V tube bias, 330Ω load
Millett Hybrid with 12FK6, 16.5V tube bias, 330Ω load
Millett Hybrid with 12FK6, 12V tube bias, 33Ω load
Millett Hybrid with 12FK6, 13.5V tube bias, 33Ω load
Millett Hybrid with 12FK6, 15V tube bias, 33Ω load
Millett Hybrid with 12FK6, 16.5V tube bias, 33Ω load
The following two results were taken with OPA551PA opamps installed in place
of the BUF634P, and wire jumpers installed between pins 2 and 6 below
the DIP-8 socket. This operates the OPA551PA as a closed-loop voltage follower.
Note that the upper order harmonic distortion and intermodulation distortion
components are reduced in magnitude, compared to the BUF634P results
above with the same tubes and bias points. I used 12AE6A tubes biased
to 13.5V for these tests because the object of the test here
were the buffer, not the tubes; and these GE 12AE6As exhibits the
lowest distortion amongst the different tubes in my collection.
With Discrete Diamond Buffers
The following results were taken with the
discrete diamond buffer
(a project by steinchen and n_maher)
installed in place of the BUF634P. As with the OPA551PA tests above,
I used the 12AE6A tubes biased to 13.5V for all these measurements.
This particular diamond buffer board was populated with 2N5087/5088
for the input and current mirror transistors, and MJE243/253 for the
outputs. 2N5486 JFETs are used for the adjustable current source.
These are all "standard configuration" parts as specified in the
diamond buffer schematic diagram.
With the discrete diamond buffers, the Millett Hybrid exhibits less
upper order harmonic distortion and intermodulation
distortion than the BUF634P. The THD and IM percentage numbers don't
tell the whole story. You must compare the graphs to see this. In fact,
these measurements produced virtually identical results to those with the
OPA551PAs (which has the benefit of being closed loop and thus very low
distortion). The tubes remain the dominant source of distortion, so the
differences in the output buffers is masked. What the results do tell us
is that both the OPA551PA and the discrete diamond buffers outperform
the BUF634P in this application.
I tested with the diamond buffers set to 15mA, 20mA, 25mA and 30mA
bias (only the first three are shown below, the fourth is no different
than the rest). The differences between these bias points were
insignificant, thus I recommend setting to between 15-20mA
for the lowest heat dissipation on the output transistors.
As expected, the frequency response, noise, and stereo crosstalk
measurements were not affected by the choice of output buffers.
Millett Hybrid 12AE6A + diamond buffers, 13.5V tube bias, 15mA buffer bias 330Ω load
Millett Hybrid 12AE6A + diamond buffers, 13.5V tube bias, 20mA buffer bias 330Ω load
Millett Hybrid 12AE6A + diamond buffers, 13.5V tube bias, 25mA buffer bias 330Ω load
Millett Hybrid 12AE6A + diamond buffers, 13.5V tube bias, 15mA buffer bias 33Ω load
Millett Hybrid 12AE6A + diamond buffers, 13.5V tube bias, 20mA buffer bias 33Ω load
Millett Hybrid 12AE6A + diamond buffers, 13.5V tube bias, 25mA buffer bias 33Ω load
Comparison: 12AE6A + diamond buffers, 13.5V tube bias, buffer bias 15mA, 20mA and 25mA, 330Ω load
Comparison: 12AE6A + diamond buffers, 13.5V tube bias, buffer bias 15mA, 20mA and 25mA, 33Ω load
Other Test Results
These were measured using a Wavetek 188 4MHz sweep function generator and a
Fluke 95 50MHz digital ScopeMeter.
- Voltage gain
With 12AE6A tubes, 330Ω load: 12.4 (21.9dB)
With 12AE6A tubes, 33Ω load: 7.2 (17.2dB)
With 12FM6 tubes, 330Ω load: 7.2 (17.2dB)
With 12FM6 tubes, 33Ω load: 4.2 (12.5dB)
With 12FK6 tubes, 330Ω load: 4.7 (13.4dB)
With 12FK6 tubes, 33Ω load: 2.7 (8.7dB)
- Maximum output voltage
(prior to onset of clipping, 12AE6A tubes, 27VDC supply, 13.5V bias)
5.85Vrms (16.5Vp-p) into 330Ω
2.85Vrms (8.06Vp-p) into 33Ω
- Maximum output power
(prior to onset of clipping, 12AE6A tubes, 27VDC supply, 13.5V bias)
0.1Wrms into 330Ω
0.25Wrms into 33Ω
- Frequency Response
(at 1Vrms output, sine wave, 12AE6A tubes)
1Hz - 535KHz, +0, -3dB (330Ω load)
10Hz - 535KHz, +0, -3dB (33Ω load)
- Output impedance
33.3Ω at 1KHz
- Current draw from power supply
With BUF634Ps installed, 180mA quiescent at 27VDC (includes the two
superbright LEDs at 10mA each), increasing to 240mA at maximum
output into 33Ω loads (both channels driven).
The following shows the waveform response of the Millett Hybrid amplifier
equipped with 12AE6A tubes, biased to 13.5V and BUF634 output buffers.
In all graphs except the Lissajous waveform, the top trace is the input
and the bottom is the output. These were measured using a Wavetek 188 4MHz
sweep function generator and a Protek 6510 100MHz oscilloscope.
The tests were done with the amplifier volume control set to maximum,
and the output level is adjusted to slightly below the threshold of
clipping using the function generator's amplitude control.
The graphs show that the amplifier inverts absolute phase.
There is significant slewing at the leading and falling edges of the
100KHz square wave, but no ringing or overshoot. The 100KHz
sine, triangle, and Lissajous graphs also show some amount of phase delay
between the input and output. The triangle wave peaks show visible rounding.
Despite the -3dB frequency response of over 500KHz, this is obviously not
a "fast" amplifier.
1KHz square wave
10KHz square wave
100KHz square wave
100KHz sine wave
100KHz triangle wave
As expected, this amplifier exhibits high total harmonic distortion.
The intermodulation distortion is also high. The 12AE6A is the best
performer in the bunch in terms of distortion, and the 12FM6 the worst.
The 12AE6A pair I have is also better matched to each other than the
other tube types tested. In all cases it was a bit surprising that
the THD contains strong even and odd order harmonics. While
the bias setting causes some change to the relative strengths
of the harmonics, the effect is small as compared to the overall
magnitude of distortion.
The voltage gain varies dramatically depending on the tube type used.
The 12FK6 tube may not provide enough gain for some low-output source
and high impedance headphone combinations. The maximum output voltage
swing is dependent on the setting of the bias adjustment. For maximum
swing the bias should be set to one half the supply voltage.
The clippihg behavior is classic tube-gear soft. When mildly overdriven
this amplifier will not sound overly harsh. However if the bias is
set far away from one-half the supply voltage, the clipping will be
This amplifier has a low noise floor. When used with a ripple-free
DC power supply, the spectrum is devoid of hum or noise spikes.
The tubes are somewhat microphonic, so vibrations should be avoided.
The 22Ω output resistors cause the effective gain to drop by
4.7dB when 33Ω dummy loads are added (compared to 330Ω
dummy loads). This also makes the RMAA noise floor appear to be 4.7dB
lower because the input test signal level has to be increased to
make up for the loss.
The 33Ω load also causes the amplifier distortion to rise
significantly. The stereo crosstalk also suffers with the low
impedance load, primarily due to signal ground pollution as a result of
increased output current return to ground. This makes sense because
the difference in stereo crosstalk between 330Ω and 33Ω loads
is about 20dB over most of the audio band, which a factor of 10.
The bass response is down 0.5dB at 20Hz when loaded with 33Ω
dummy load, due to the high-pass filter characteristics of the 470µF
output coupling capacitor. Otherwise, the frequency response is nice and
flat across the audio band.
The high frequency response of this amplifier is quite extended,
-3dB at 535KHz (I only tested this with the 12AE6A). This is in
stark contrast to the rolled-off highs found on many tube amps.
The measurements show an amplifier's technical strengths and weaknesses,
but it doesn't describe the sound. I am happy to report that despite
the measurements, this neat little amplifier is fine-sounding,
in a euphonic kind of way. It doesn't break new ground in
dynamics, resolution or soundstage, nor was it ever intended to. This
is a different kind of audio project, simple, safe and easy on the wallet.
The sense of nostalgia when examining the vintage NOS tube boxes
is a pleasure to behold. Moreover, no solid state device evokes
the visual enjoyment that tubes provide, especially when displayed in
their full glory with a clear-top case, and lit from below
with blue light! :)