by Kevin Gilmore
(Project Editor: Chris Young)
The Blue Hawaii amp is my latest design in my search for the perfect amp to pair with my Stax Omega II headphones. The genesis for this hybrid electrostatic headphone amplifier occurred when I was in Hawaii on vacation, at a fancy hotel on Maui. Sitting at the bar on the beach, drinking “Blue Hawaiis,” I drew the schematic for the amp on a placemat. The design is my conception of the mysterious and rare Stax T2 amp, which I have never been able to find at anything resembling a rational price.
I searched out any information I could find on the T2 in an attempt to create my own version. I was able to determine that it used EL34s as output tubes in a grounded grid configuration, which is the lowest distortion tube output circuit known. It also used 6DJ8s as input tubes with some solid state in the second and third stages. My design uses the first and second stages from my solid state electrostatic amplifier coupled with a third FET stage and then the final grounded grid stage.
My design ended up with a fairly large amplifier pulling significant amounts of power which results in a very smooth and extended frequency range from DC to over 200khz (-3dB at 400khz). Of all my electrostatic amps, this one has the largest output voltage swing. This is not an amplifier for the timid, nor is it a good idea to build this as your first project, though some, however, have actually done so.
Figure 1 is the amplifier schematic. The entire amplifier has a differential topology from input to output to get a balanced input and for lower noise, less ground loop problems. The first stage is a differential amplifier with feedback directly from the output stage. It works equally well with both balanced and unbalanced audio input sources. The step attenuators from Goldpoint make good volume controls for this stage. The JFET device (Q1) is a dual JFET all on one wafer. It is known for extremely low noise and excellent matching, and is used in a number of expensive designs, such as the Nelson Pass amplifiers. Q17 is a current source that sinks 3mA.
Because the amp is totally DC coupled from input to output, drift in the input stage is a bad idea. Since the first two stages run in current mode, the JFET input is more linear than a pair of bipolar transistors. Dual transistors all on one wafer suitable for audio use are hard to find these days. The FETs steer current away from the current sources Q2 and Q3. Together Q2 and Q3 each supply 2mA or a total of 4mA. The Q17 current source takes away 3mA leaving 0.5mA in each of Q4 and Q5, but some of the sink current is coming from the output feedback, so each FET is actually using somewhere between 0.5mA and 1mA.
The approximate voltage gain of this stage is 5; this stage really runs in current mode. The unit was designed to work equally well in both balanced and unbalanced mode. For single-ended signals, ground either the + or – input and apply signal to the other. The much higher impedance of the JFET works better when one side is grounded for unbalanced inputs.
The second stage starts with a constant current source (Q2 and Q3). The current source feeds a common base amplifier (Q4 and Q5). The common base amplifier feeds a modified Vbe multiplier. I believe a famous designer is now calling this circuit a current tunnel. It’s the most linear way of translating the voltage down to the bottom rail. The voltage gain of this section is about 4. The basic idea of the first two stages is to supply the third stage with a very fast low impedance drive signal that is referenced to the bottom rail.
The currents flowing into the common base amplifier (Q4 and Q5) are the difference between what Q2 and Q3 are supplying and what the FET is taking away. The rest of the current goes down the tunnel to the vbe multipliers (Q6 and Q7) that convert the current back to voltage. The current sources in the second stage supply 2 mA each. With no signal, the FETs take 1 mA, leaving 1 mA going through the common base amplifier into the bottom transistors, which are wired as Vbe multipliers (like a zener diode in series with a resistor, except a lot less noisy). This generates the 13 volts (referenced to – rail) necessary to properly bias the third stage.
The third stage is another differential amplifier (Q13 and Q14) being driven via another constant current source (Q10 and Q16). The voltage gain is about 200. Q11 is the power supply for this stage and makes a 100 volt power supply with -400V as the reference. The power supply voltage for this stage is kept down to 100 volts to reduce the Miller effect and keep the frequency response up. The higher output impedance of this stage is lowered by the use of 2SJ79 transistors, which are used as zero voltage gain emitter followers. The use of FETs in this stage coupled with the current source further reduces the distortion and provides for a solid low impedance drive signal for the output stage.
The 4th and final stage is a tube in grounded grid configuration (V1/Q8 and V2/Q15), similar to the common base amplifier in the 3rd section of my solid-state current-domain electrostatic amp. Q9 and Q12 are high compliance current sources and supply 25mA of bias current. Think of them as linear pull-up resistors for current (in fact, one builder has replaced the current sources with large resistors). The use of a current source here instead of load resistors acts to further linearize the output stage and reduce output distortion. V1 and V2 are the equivalent of common base amplifiers and do the entire rail-to-rail output voltage swings.
With feedback, the overall voltage gain of the amp is exactly 1000. The frequency response is kept high due to the low impedance cathode drive. The EL34s are biased at 10 watts and have an 800V voltage swing (by comparison, the output tubes of my original DC-coupled electrostatic amp are biased at 2 watts with a 600V swing), resulting in a frequency response well in excess of 100kHz into a 150pF load. (+0/-0.1dB).
A regulated power supply design is shown above. The ±15V supply is made with the standard 7815/7915 regulators. The high voltage supply is a pair of 400 VDC supplies, glued together at the output (P-channel MOSFETs are a lot more money than the equivalent N-channel MOSFET). In each section, beginning with a 460V raw supply, a PNP transistor (2SA1968) is used as a current source to feed the 400V zener reference. Then a N-channel FET is used as a high impedance, input voltage follower and outputs 400VDC. By the way, the same exact supply with a 350V zener reference string instead and a slightly smaller transformer (without filament windings) is what I use now for the solid state current domain headphone amp.
The bias supply is a voltage doubler with an adjustable reference. It has a range of about 350VDC to 650VDC. For low bias Stax headphones, put a 10M resistor to ground at the end of the 4.7M. to make the output voltage .66 times the voltage before the 4.7M, which puts it in the range for low bias.
Caution: This project involves working with high voltages, so be extremely careful! Keep one hand behind your back at all times. 800VDC across both arms might possibly stop your heart.
This amp was assembled on three printed circuit boards (two for each channel of the amp and one for the power supply) and housed in separate enclosures. A complete set of pc board patterns (pdf format) can be found above. They could be sent to just about any circuit board manufacturer to have boards made. The top of the board is almost all groundplane. All the parts, including the tubes, are mounted on these boards – the tubes are installed in pc-mounted ceramic tube sockets from Parts Express. The tubes must be exposed through the chassis. They dissipate 20W each (actually 10 to 12 watts of plate dissipation plus another 6.3V * 1.6A = 10 Watts of filament power).
It was so much easier to do a pc board for this amp, but if I were to make a prototype, I would again use the 0.1mm perf board; the layout look much like the circuit board. (Note: For a layout in a single chassis, see the interior view of Headamp.com’s Blue Hawaii amp below.) 99% of the wiring would be on the bottom, and it would be, therefore, rather flat. Mounting the tubes would be trouble though, and would cause mechanical problems. The tubes are fairly heavy and get stinking hot. Each chassis measures 12″ x 10″ x 3.5″. (Note: Headamp.com is selling the Blue Hawaii design in a single chassis measuring about 16.5″ W x 16.0″ D x 3.5″ H and may sell the Blue Hawaii pc boards. Please contact Headamp for more information.)
I have Mullard EL34 tubes, but keep them put away due to what I could sell them for if I wanted. I actually used the National Union tubes from Richardson Electronics which cost $11.50 US each.
All of the parts except the 2SA1968 have lots and lots of sources such as Digi-Key, MCM Electronics and Mouser Electornics. Only B&D; Enterprises has the 2SA1968 in the United States. In Japan and Canada, they can be ordered from Sanyo direct – the minimum order is 100 at a time, but then they are $1.25 each or so.
Q9 and Q12 are each made of six 2SA1968 transistors in parallel with one 2SA1968 as the driver. Matching the transistors is not required – unless one of the 2SA1968s is way off compared to the rest in which case it might get way too hot.
All resistors are 0.25W except where labeled. It is important to have all the pnp current source transistors correctly mounted to a large heatsink with silicon impregnated washers. If any one pnp transistor gets too hot it can short out the whole current source.
Standard tab heat sinks will do for the 2SK216 and 2SJ79 transistors, but the 2SA1968 and 2SC3675 transistors must be mounted a big heatsink (one for each channel), capable of dissipating 20 Watts of heat. I obviously fabricated them, but otherwise they can be obtained from Conrad Heatsinks cut to length. The IRFBC30 MOSFETs in the power supply must be heatsinked too: Mouser part number 532-529902b25.
The Stax SRC-5 headphone jack came from AudioCubes.com. Since the price has gone up to $19 each (I paid $10), it may be more cost effective to use the Allied jacks (see the current domain amp project article). Allied has a $25 minimum order, the cost of three pieces. Then they must be filed down on a lathe. Actually I am buying the male connectors from Allied, because no one else sells them. The male connectors are much easier to convert to standard Stax plugs. The power supply-to-amplifier connectors are the Amphenol military 12-pin connectors. The 4 connectors (two male and two female) were $130.
The custom Victoria Magnetics power transformer has these specs: 2 x 330VAC/150mA, 36VCT/100mA, 2 x 6.3V/5A (filament supplies). Everything with Victoria Magnetics is custom. I paid $110 for the transformer with shipping. They know about the Blue Hawaii design and will supply the correct transformer on request. For safety, I recommend a 2A/110Vac fuse located in the input line to the power supply.
Setup and Results
Test voltages (with the amp at idle) are shown in red on the schematic and are with respect to ground. To set up the amp, adjust the two pots in each channel of the amp. P1 controls the differential output voltage. Put a voltmeter between the 2 stators of one channel of the headphone and set this pot for zero. P2 controls the voltage with respect to ground. Put a voltmeter between any stator and ground and set for zero. Then repeat both adjustments a few times. The plates of both tubes should measure 0 volts with respect to ground when the pots correctly adjusted. Once the pots are adjusted, that’s it – there’s no change from headphone to headphone.
Setting the bias voltage depends on the headphones. For Stax headphones that can accept a high bias voltage, adjust the pot for 560V. I do not think that the Omega II headphones can be damaged by this amp unless the bias is set way too high. If the bias is set right, the outputs are close to 0V at idle, and all the LEDs are lit, then the amp pretty much has to be working correctly. Now if one or more of the outputs is stuck at +400V or -400V, then something is seriously wrong and needs to be fixed. An oscilloscope really helps.
Adjust the pot to 580 volts for Sennheiser HE-90 and HE-70 headphones or leave it at 560V. For Koss headphones, adjust the bias for 600V. To use the Sennheiser HE60 headphones with this amp, I made the adapter shown above. Those are RS-232 female connector pins that fit the HE60 pins perfectly. By hand I cut a circuit board with lands exactly 3.5mm apart put the pins on the HE60 connector, lay them down on the circuit board and solder. Then attach wires and a standard stax plug.
The amp can output 1500 V p-p measured stator to stator. At 800Vp-p, THD is less than 0.004% from 20Hz to 20kHz. The actual frequency response is 0 to 100kHz (-3dB at 150kHz) into an Omega II load. Compared to the sound of my previous tube amplifier, the bass is no longer tubby; it’s very sharp and tight. The high end is no longer rolled off, so female voices sound much more real. If the bias supply is reduced to 280V, the amplifier will drive all electrostatic headphones. I tried it last night on a pair of SRX’s. I never ever heard them sound so good.
Previously with a standard dummy head, I measured the SPL in Omega 2 headphones driven by this amplifier. With a drive signal of 800Vp-p per side, the resulting spl is 106dB. THAT’S LOUD! The amp can put out 1500 volts peak-to-peak, and thats louder! I just ordered a pair of Stax SR-001 MkIIs, which can reach up to 120dB. My ears distort before the amplifer/headphones do. It is quite loud at clipping, but the clipping is a hard clip with no oscillation or ringing. To use the amplifier with electret headphones, delete the bias voltage. And probably keep the output swing under 200V. Electrets phones when driven with this amplifier can probably get very very loud.
Several of my previous electrostatic designs are available in the Headwize Projects section. Comparing the Blue Hawaii to my all solid-state current domain amplifier, they really are more the same than they are different, but in general, the differences are the differences between tubes and solid state, such as a much smoother high frequency response, which in the case of the Blue Hawaii goes well beyond 500kHz. Additionally, the four times power consumption of the Blue Hawaii means a much stiffer and tighter bass response. Even though both are flat to zero and test similar, the BH bass is much more apparent and tighter.
c. 2004 Kevin Gilmore.