by Rudy van Stratum
[Editor: Although this headphone amplifier design has a direct-coupled output, please take note that the author cautions against using the DC output with low impedance headphones. The potential drifting of bias voltages at the DC output could damage headphones. DIYers who want to avoid this risk should use the AC-coupled output instead.]
I published a beautiful headphone design in 1995 in the Dutch magazine Audio & Techniek (which does not exist any more). It is a Single-Ended Push Pull (SEPP) OTL design a la Futterman, and the whole idea was to get a complete tubed amplifier without an electrolytic capacitor in the signal path. Therefore the design is quite complex, especially in the power supply. The charm of the Futterman is that it is the only design with tubes that can do away with the output cap completely (or it allows the use of non-polar MKP caps or even low voltage electrolytic caps, which can be tried in big quantities for low prices to find out the optimum).
In 1994 I built an electrostatic headphone driver with tubes for my Stax and Micro-Seiki MX-5 headphones. I used the design of Joseph Curcio as published in Glass Audio (volume 1, number 0, 1988). The sounds were fabulous and the effect was that I did not listen anymore to normal dynamic headphones.
Then I got the idea to build a headphone driver for dynamic headphones. Only then could I compare the electrostatic and dynamic types on a more honest basis. I wanted to use tubes as I’m ‘into’ tubes for a very long time (started with an old Quad II very long ago). The Curcio also used tubes throughout and did not need an output transformer. So the design criteria were: tubes, no output transformer (OTL), simplicity.
At that time I had just built an OTL amplifier for normal 8 Ohm speakers using only 3 tubes (EF86, ECC81 and one 6336A double-triode, published in Dutch magazine Audio & Techniek 1994). I had studied a lot of old material on the OTL subject. I decided to go for a ‘standard’ Futterman design using two ECC88’s per channel. And I thought at the time that I did not want an electrolytic capacitor at the output (you always can hear such a device, so when possible avoid it). Just when I was building the prototype there was an article on a schematic for dynamic headphones in Glass Audio. The article from Denzil Denner (Glass Audio volume 6, number 4, 1994) used one ECC83 and one 12BH7 and looked a bit like what I wanted to do. It used a Futterman-like SEPP output stage but had its own way to overcome the problems associated with it. There was an electrolytic cap at the output and I think with the 12BH7 it was optimised for use with mainly 600 Ohms headphones.
The schematic of my Futterman headphone amp design is shown in figure 1. A common-as-muck anode-follower amplifies the input signal (gain = 25) and the other half of the first ECC88 is a cathodyne phase splitter using relatively low resistor values of 22k. There is a lot of literature on the patent that Futterman got in 1954 about how to optimally feed the SEPP stage. This stage needs two out-of-phase inputs at the two grids. Because one half of the SEPP does amplify and the other does not, now the problem is how to get the two inputs of the right but unequal amplitudes. Denzil Denner in his Glass-1994-circuit solves this problem by using a simple voltage-divider for the lower half of the SEPP-stage. This will do and is the most simple solution but it will work optimally only for certain loads (which will vary anyhow). And what is more, Danner uses a feedback setting in his circuit that implies the use of an extra electrolytic at the input of the circuit (one electrolytic is bad enough). We must stay open minded however, I did not build the Denner-circuit and can not have a judgment on the sound quality it delivers (and to be honest: maybe electrolytics and output transformers are not that bad after all, but that is for another story).
Returning to the problem of how to drive the SEPP, the genius-like solution of Futterman was to ground the phase-splitter via the loudspeaker to ground. In the schematic you see the 22k resistor in the cathode of the phase-splitter go to the output point of the SEPP stage. The output is also fed back to the input tube by the 1 kOhm resistor.By the way, Futterman circuits are renown of their heavy use of negative feedback. In the original articles of the fifties Futterman is very proud of using 50 dB or more of negative feedback without the amplifier going into some form of oscillation. I must confess that the secrets of the workings of the Futterman-patent are not completely transparent to me yet, there still lacks a modern and clear explanation of it (have seen most of the literature on it, but .. ).
So far so good. Very simple. The power supply is the price you have to pay. Because I wanted DC operation at the output I need a symmetrical power suppy for the SEPP stage. So I have plus and minus 90 Volts to start with. Next I want to fiddle with the quiescent currents of the SEPP stage to regulate the DC offset. I use four separate windings of 4 Volt that after rectification allow me to regulate the bias of four tube-halves separately. Further, there is the normal high voltage supply for the first tube and an arrangement for the heaters (12 Volt regulated by a 7812 in a parallel series set-up).
I used Philips E88CC SQ vacuum tubes (special quality gold pin). Other good brands include Siemens, Telefunken, some Russians like Sovtek, Chinese Golden Dragon etc. There are a lot of good tubes around; I prefer New Old Stock of European brands.
Matching the output tubes is not necessary, because the 4 separate pots can correct a lot of mismatch (which of course does not benefit the maximum output power and THD). Matching is best done by exchanging tubes and looking at the voltages at different points in the circuit. But, most of the time, matching is commercial marketing to sell more expensive tubes. I tried more than 10 tubes in the circuit and they all worked all right. Differences between halves within a tube are not worth the expense and trouble in this circuit. It is better to use 4 tubes from the same same brand/year (or 2 of the same for left/right input and 2 of the same for left/right output). The check is whether the DC output voltage is around 0 volts. If not, then swap the tubes from one place to another and try again. Note: matching the output tubes will not prevent DC drift.
The resistors are all standard Philips metal film 1 Watt, except where otherwise noted. The bias adjustment pots vary the currents of upper and lower half of the SEPP output stage (the 2K pots in the power supply). They are CRUCIAL. In effect, they are operating in a floating sense and are not attached to ground anywhere. You can use any pot, but I used Bourns precision pots that work just fine with their 10 rotations.
All electrolytic capacitors are from Philips (blue colored). The signal caps (0.47uF) are ERO MKT, 630 Volts. The output cap (for the ‘safe’connection) can be either a 100uF, 100V MKP or MKT capacitor (the very big one that is black colored in the photo is an Audyn 100uF MKP) or a high quality 330uF/16V electrolytic. MKT capacitors have a polyester dielectricum; MKP uses a polypropylene dielectricum. MKP caps are about twice the price of MKT. Nevertheless, it depends on brand and make which of the two sounds better; everyone has to judge for himself. More than once, I had MKT caps that did sound better than the more expensive MKPs. For electrolytic output caps, I mostly use Philips electrolytic caps of 16V, 25V or 35V types. These are very fine for the money (blue tubular caps).
I ordered a custom-made power transformer with 7 windings. Or several transformers together can do the job. For example, you can use the combination of a standard tube transformer for the high voltage and filaments, a toroidial transformer (70-0-70 VAC) for the bipolar supply and two cheap transformers that give 4 VAC for the bias section (such as those used for toy trains and the like). Maybe you can even use batteries to supply the bias voltages. I think you can use a simpler bias arrangement for both channels (figure 3) that uses just two 4VAC secondaries, but I simply don’t know. I can only assure the good workings of figure 2, because that’s is the one that I used for many years.
To rectify the ±90VDC supply and the 220VDC high voltage supply, I used 1N4007 diodes. To rectify the bias set-up voltages, I used small bridges of type B40C800 (total of 4 pieces for 2 channels). For the heater supply, I used a bridge rectifier of 10 Amperes that I fixed to the bottom of the chassis (B40/10 and the like).
I used a cooling plate for the 7812 voltage regulator (TO3 version) that I mounted to the chassis. I used a general power supply fuse, around 500mA for 230VAC or 1A for 120VAC. I think this fuse is too slow to have any effect if something is going wrong (it can handle 110 Watts!). I always take a fair margin and take slow fuses; this is not the safest way, however. For safety, one can take a fast fuse and probably 250-500 mA will do (just try and see; if the fuse stays in, it is okay). The on-off switch is a standard toggle switch.
I did not encounter any instability during the construction process. As seen in the photo, I started with the power transformers, then a separate euro-card with all power supply parts (in a logical order) and then another euro-card for the amplifier circuits. I kept the signal wires as short as possible and hard-wired all components to each other using the back-side of the Euro-card. For wiring I used mostly OFC-copper wire.
I made the chassis myself using all kinds of aluminium parts. The dimensions are around 44 x 12 x 10 (cm) (all internally measured). The chassis has a perforated cover not shown in the photos.
Biasing the Output Stage
The first 2 stages are auto biased. So the only variables are bias 1 and 2 of the output stage. Using an ECC88 the optimal bias for the minus pole is around 1.5 Volts leading to 15 mA of current in the lower half of the ecc88. So BIAS 2 = 1.5 Volt. For the plus pole the optimal bias is around 2 Volts leading to a quiescent current of 12 mA for the upper half of the ECC88. The difference in quiescent currents between plus and minus just makes the amplifier putting out 0 mV of offset at the output.
I’ve put in small 4.7 Ohm resistors in the SEPP where you can measure the idle current. Ideally you measure 55 mV and 70 mV over these resistors to get 12 mA for the upper half of the SEPP (that goes to the plus 90 Volt) and 15 mA for the lower half of the SEPP (that goes to the -90 Volt supply). The corresponding biases for the ECC88-halves are around -2 and -1.5 Volts respectively. The best way to go is to set up both biases to these voltages and tune one of the biases till you have 0 mV offset at the output. (Note: why the currents of upper and lower halves are not the same? Because if you try to do this you get several volts on the output, a characteristic of this Futterman solution. In a normal 8-Ohm Futterman amplifier this can not easily be seen because the phase-splitter draws almost no current compared to the big SEPP output tubes of the like of EL519’s).
If you use the output with the cap there is no need to adjust anything at all afterwards. The DC output however drifts away around several tens of millovolts and needs to be used carefully. I mainly used this output to check the sound during short periods, not recommended for using ‘blindly’ at any time.
I did not use a volume pot in the amp itself. I use a separate tube pre-amp to do the job. Adding a volume pot into the amp itself is of course no problem. I would use an Alps 100k dual pot (a blue one, around 25 dollars); these work very fine.
I built this amplifier in 1994 and published the schematic in 1995. It worked the first time and I have listened to it for 7 years very nicely. The sound is very good and you can put in any phones you like (32 Ohms or higher). The headphones shown in the top photo are Yamaha YHD-3 (300 ohms). I used during several years the following ones: Sennheiser HD580 precision (300 Ohms), Sennheiser HD433 (40 Ohms). The best results I got with Sennheiser HD600 and AKG K-240 phones.
As an amateur I have limited capabilities to measure things, but I guess the output impedance of the design is about several ohms due to feedback and the SEPP-arrangement. Because it is a real push-pull topology, it can swing out more than 20 mA easily. I don’t have measuring equipment, but made these calculations of output power:
|30 Ohms||13.5 mW in class A, 24 mW in class B|
|100 Ohms||45 mW/80 mW|
|400 Ohms||180 mW/320 mW|
|600 Ohms||270 mW/480 mW|
I never had any problems with phones going not loud enough. For the power hungry phones you could think of using the very special E288CC tubes that doubles the capacity of the now used ECC88/6922/6DJ8. The very expensive and scarce E288CC can be plugged into this circuit without modification (you have to set-up the idle currents anew though).
Later I bought the X-cans of Musical Fidelity, just to have something comparable. These X-cans can put out 100 mW at 40 Ohm and I wonder how they do that with only one ECC88 (2 halves in parallel as a cathode-follower). I wonder how they do that with only one ECC88 (now we know that the ECC88 is only used in the input stage and the an opamp is used to drive the cans).
I never could compare the loudness of both drivers with each other because I did not want to become deaf. But the Futterman-circuit has the credits for sound quality. It just sounds more open and relaxing (tube-like if you will). But I have to confess that I think the guys at Musical Fidelity did a magnificent job for this kind of money and this kind of simplicity. The Futterman sounds slightly better on almost all respects than the X-Cans. The differences are not dramatic, but the Futterman sounds more tubey or lush or flowing of airy, whatever. Of course, it is all tube you hear and no IC’s in there.
Although this Futterman amp has both DC- and AC-coupled outputs, the quiescent voltage of the DC output can drift by up to 500mV, which may be too much for most headphones. The design is in principle, though, a DC design. I think that listening through the DC output is possible with higher impedance phones (which I did for many hours and nothing wrong happened), as the design is now. Furthermore, adding a simple DC-offset regulator in the feedback loop or a high voltage stabilized power supply can improve on these things.
The difference in sound quality between the MKP capacitor-coupled output and DC output is marginal. I would say that the MKP-cap adds a small effect of harshness in the highs and it robs a little of the overall dynamics and liveliness of the sound. It is better than an electrolytic capacitor though, that makes the sound ‘darker’ and more ‘shut in’, less open, more ‘boomy’ in the lows. The MKP-solution is of course an option for all capacitor-coupled amps, but do not forget that my MKP cap is only 100 Volts (there are versions of 100mF/250 and 400V but these are very very huge and pretty expensive).
3/3/2003: Alex Cavalli submitted a version of the Stratum headphone amplifier incorporating several optimization techniques based on the work of John Broskie and circuit simulations in OrCAD PSpice. The simulation files he used can be download here. He writes:
1. The way the output of the phase splitter is connected to the SEPP stage causes the triodes to operate in grounded cathode mode (see Totem-Pole Output Stage by John Broskie). This gives the amplifier a Zo of about 300 ohms. The better way to wire the phase splitter to the SEPP stage is to cross the connections. This operates the triodes as cathode followers, lowering the Zo to about 65 ohms. This is better for low Z phones.
2. The cathode current for the phase splitter is sourced in the lower triode of the SEPP. This puts its operating point >3mA higher than the upper triode. This is not a terribly important problem, but it does limit somewhat the current swing before the amp leaves class A mode because the upper tube will cut off before the lower tube does and/or the lower tube will exceed maximum current capacity before the upper one does. So, while this is not super critical, the design is not getting the best from the output stage because it is not balanced. This fix is to source the phase splitter cathode current from the negative supply. This is done by removing the 1K resistor to ground and replacing it with a 27K resistor to the negative supply. Now both SEPP triodes have the same quiescent current.
3. The quiescent current as described in the article is probably too high. The maximum cathode current for 6DJ8/6922 is 20mA. If the lower tube is set to idle at 15mA then it can really only swing up to 20mA before exceeding this maximum rating. In the push-pull arrangement, this is a maximum of 10mA into the load (3mW peak into 32 ohms). Design maximums can be exceeded, but if they are exceeded regularly, tube life will likely suffer. Now that the idle current is balanced between both SEPP triodes, it is better to adjust both bias supplies so that approximately 10mA flows through each tube. This would be 470mV across each test resistor (with some normal variation needed to adjust the DC output to zero). This bias point permits the output stage to swing 20mA (13mW into 32 ohms) before leaving class A and without exceeding maximum ratings on the peak current. This change will also decrease the distortion some because both tubes are at the same operating point and are supplying the same amount of current thereby taking better advantage of the natural even-harmonics-cancellation of the SEPP stage.
4. The amp has a lot of gain, exceeding maximum currents with less than 0.1V at the input. We can burn some of this extra gain in a feedback loop that will lower both the distortion and the Zo. For the values shown, Zo goes to 10 ohms. For a 32 ohm load at 20mA peak, distortion goes from 1.4%% (with lots of harmonics) in the original design to 0.4% (with many less harmonics) in the modified design. The inputs are .08V peak and .25V peak respectively at 1KHz. Simulations are using PSpice 9.2. (Note that even in this example, the original design is exceeding maximum current for both triodes). Even though simulated distortion values don’t really take into account all possible distortion mechanisms, these results indicate that the changes should bring about a comparatively better distortion figure.
5. This amp can have a direct-coupled output. The same type of adjustments apply, except as I explained in my remarks that the idle current should be set for about 10mA. For example, adjust the lower bias pot until about 47mV appears on the lower triode’s plate resistor. Then adjust the upper bias pot until the DC offset is zero. This will set both tubes to about 10mA idle current. Of course, idle current in the upper tube can be measured using its plate resistor. These changes, however, won’t affect the DC drift much. But, they will help some after the tubes get broken in.
Rudy van Stratum replies:
The modified circuit is a real addition to the discussion. The text of Alex clearly shows the differences. A few important remarks however are in place:
1. Right. These Zo’s are without feedback. Remember that my design uses quit a lot of feedback and therefore has a lower Zo than 300 Ohms. One of the goals of the design was exactly copying the Futterman-patent into a headphone design, not done before to my knowledge. This includes the special way of arranging feedback via the 1k resistor.
2 and 3. Also right. I think the 470mV across the test resistor must be 47mV. And again: the difference between idle currents at bottom and top are a logical consequence of the Futterman-design.
3. The modification of Alex does away with the Futterman concept almost completely. Removing the 1k resistor is removing the feedback. Alex replaces this one resistor in effect with 2 new ones: R10 and R15 (thereby getting the output cap into the feedback loop). Alex does not mention the distortion figures and gain-factor and Zo etc with the 1k resistor added in my design. Therefore the comparison between circuits is misleading. My design has not a lot of gain, I guess that 1V of input is not a problem (comparable gain with the X-cans of MF).
Good suggestions and probably a better overall result (to be tried by future builders).
My design has as one of the goals the replication of Futterman’s patent, therefore a slight imbalance between idle currents of top and bottom as a logical consequence. The Futterman circuits are very famous for their outstanding sound quality – see, for example, the articles and books of Rosenberg (in search for musical ecstacy, 1994).
The feedback arrangement of 1k is crucial to the Futterman design. When making comparisons please take the measurements with the feedback loop connected in my design. In effect, the solution of Alex makes it a completely new design. It has to be seen whether the differences in Zo and distortion are that great in the end and how the circuits behave in real life circumstances (of course I can not judge on that because nobody has ever heart the circuits alongside each other).
c. 2003 Rudy van Stratum.