A Direct-Coupled (OTL) Headphone Driver Amp

Note (16/10 2004): a schematic of the final design has been appended at the bottom of this page.

Warning! This project involves potentially lethal voltages. The author takes no responsibility for any damage it may cause if undertaken, whether to body, mind,  property or anything else for that matter.

 

Introduction

Listening to a good pair of headphones can be a soothing experience when you’ve been chasing room modes and reflections from your speakers to the verge of insanity, or when your neighbours have been chasing you for making too much noise. But to truly experience the joy and emotion of a good recording, you need to treat the signal that drives the headphones with great care.

Quality 100-600 Ohm headphones have the potential to produce extremely clear sound, but they definitely benefit from a better signal source than your typical CD or preamp IC-based headphone output. Good current capability and a rather low output impedance are needed, but little voltage gain. If your preamp has a gain of at least 4 times, it will likely provide enough voltage (~2-8Vrms) to drive most ‘phones to deafening levels. (In fact, the direct-out ca. 2V from most CD-players is often loud enough.) To accomplish this, a powerful voltage follower is a good choice - the costly alternative is a power gain stage + transformer coupling.

On the condition that a unity voltage-gain stage is fine, I reasoned thus: a transistor or MOSFET voltage follower may need a pre-driver stage and voltage feedback to sound its best (IMLE); a well-dimensioned valve voltage follower usually doesn’t. A unity-gain valve headphone driver could be connected directly to your preamp’s (lowish-Z) line output, thus eliminating the need for another volume control and - potentially noise-introducing - gain/buffer stage to precede the output circuit. If you are the owner of a stepped attenuator or other quality volume control, you won’t want anything less for your headphone amp and will appreciate not having to buy/assemble another one for it. If you’re an audio "purist", you’ll appreciate less signal circuitry. (After all, very few listen to speakers and headphones at the same time so as to need two volume controls!)

 

About the circuit

The circuit (see schematic here) looks rather complex, but is actually quite simple; the complexity lies in the power supply (PS) and the noise reduction circuits, which make the amp very quiet. The basic signal path only contains two valves, three resistors, and two small capacitors neither of which is on the output – unusual for a valve amp. The lack of an output transformer or big coupling capacitor reduces cost significantly. A low impedance headphone is a tough load for any valve to drive, so the circuit must be able to deliver relatively high current with low distortion. The topology chosen is an Optimised White Cathode Follower (see Tube CAD Journal Oct -99, pp 4-8 for discussion and analysis). It is a kind of symmetrical push-pull stage in which the upper cathode follower (CF) creates an inverted signal at its plate to drive the lower plate follower without the need for a separate phase splitter valve. (An SE design is possible, but would be less powerful; see Variations below.)

 

Theory of operation

The neat thing about the White follower is that (like the SRPP) it works as a push-pull amplifier without needing an extra valve to serve as phase inverter for one of the valves in the pair. The little resistor in the plate lead of the upper valve (see figure below) creates a drive voltage for the lower valve which is always proportional to the voltage driving the upper valve, independently of the load’s characteristics. This is where it has an advantage over the SRPP, which needs one particular load impedance to work symmetrically (see Tube CAD Journal May 2000). The disadvantage is below-unity voltage gain, but we already decided we didn’t need any more. Also, the WCF can only work in Class A, but that is no problem with the humble power demands of headphones. (It would be if we were trying to drive speakers...)

For "purists", it might be noted that the WCF does rely on a kind of voltage feedback (or feed-forward, or both), but the path is very short, and unconditionally stable. Besides, the "damage" is already done, since we’re using CFs, which rely on current feedback, in the first place. If you feel a single-ended amp is more to your liking, you will find some on the ‘Net (e.g., Glass-Ware , Headwize, Headwize again, World Audio Design ). However, these all involve some flavour of feedback (either current or voltage or both)! A truly feedback-less design might have an SE (e.g. 437A or EC8020) combined gain/output stage with battery bias + OP xfmr (conventional or Parallel Feed ).

                                         

Optimisation of push-pull symmetry is the key to (theoretically) perfect function for this circuit (see figure above). On a positive pulse, the lower valve must decrease its conduction as much as the upper valve increases its conduction for symmetrical operation, and vice versa for a negative pulse. We won’t deal with the theory behind this here. I refer you to the aforementioned article in Tube CAD Journal. However, it may be noted that if the plate resistor is made too big, the amp will clip prematurely; if it is too small, the amp will approach single-ended operation (which may be OK, but will entail more 2nd harmonic, higher output impedance, and less power into a given load). The best push-pull action is achieved when the upper valve’s plate resistor is equal to the reciprocal of the valve’s transconductance (i.e., its output impedance at the cathode) at the op point. That way, both valves will contribute equally to the output voltage across any load impedance.

 

The Valve

The EL86/6CW5/CV5094 (a smallish power pentode) connected in triode mode was deemed suitable and chosen for this application (motivation: 1. TubeCAD Journal Jan -00 ; 2. they’re cheap; 3. EI made them before the war, and maybe will again soon; 4. I already had some NOS Mullards). The White follower works best with triodes when driving a reactive load (even dynamic headphones are slightly reactive?; correct me if I’m wrong), and the EL86 is unusual in being able to deliver the current we need in triode mode and at low plate voltage. It is a high-transconductance, low (g1g2) mu, low plate-resistance valve which also sounds fine and is plentiful. It has a predominantly 2nd harmonic character (like most valves do when driving a heavy load), and since the amp runs in class A push-pull, this (innocuous) distortion will be greatly cancelled (perhaps even too much for some).

Note: The EL86 is not a variety of the popular EL84, which is a bit wimpy in terms of current capability at low plate voltage/in triode mode to be ideal here. Confusingly, however, the PL84/15CW5 is an equivalent to EL86, with heater running at 15V/0.3A. Other equivalents are 8CB5, XL86/8CW5, LL86/10CW5, 45B5, UL84/45CW5, N119, N379. There’s no obvious penalty involved in using any of these different-heater versions, mostly advantages, as far as I can see (low price - from $1.50!, less heater losses - and at least my junk-box holds several 2x9/12/15/25/33V transformers, but no 6.3V ones!).

Here are the plate characteristics (pentode mode – sorry, I had no triode curves in right format; F. Philipse has PDFs) and a picture:

                                         Amperex EL86

 

[Note that the characteristics in CF mode look nothing like this, but more like the near-vertical lines of a low-mu, directly heated triode. Kind of like the 300B. ;-) ]

The working characteristics at our op point (100Vp-k, 50mA, -6V bias) are:

        mu=~8.5                 Gm=~12mS                 rp=~700 Ohms                 (conjectures, not measured!)

The total dissipation for each EL86 in this circuit is 5W, only 36 per cent of the maximum 13.75W rating (plate: 12W + g2: 1.75W max). This should vouch for long valve life. In addition, the heater voltage may be adjusted to ~6.0V (5 per cent "starved" operation). This may also contribute to longevity, and causes no significant change in characteristics.

 

Performance

The White follower runs without overall loop feedback since we don’t have any gain stage, so there will be some output impedance. However, this may be just as well. I haven’t heard it with overall feedback, so I can’t say. The resulting Zout is about 40 Ohms – low enough for headphones of 100 Ohms or more (mine are 120-Ohm Sennheiser HD590s and 300-Ohm HD580s, and both sound just beautiful). 300-600 Ohms may be ideal, but lower-impedance ‘phones above 30 Ohms might just work too, as they need less voltage swing. The amp can produce about 7V/70 mA rms (~0.5W) into a 100-Ohm load – much more than most other valve designs, and enough to cause hearing impairment - take care! Distortion has not been measured; it is of course below "obvious" audibility.

 

Problem: Hummm!

One of the nice things about bipolar power supplies is that noise on the respective rails is in anti-phase and therefore normally cancels at the output. However, the White Follower has one great disadvantage: instead of cancelling, it transfers the sum of the noise on the PS rails to the output – a "character flaw" particularly detrimental in a headphone amp, as headphones are so voltage sensitive. (Note, incidentally, that a single-rail White Follower would in practice be as noisy as a bipolar one with the same PS.) Explanation (see left figure below):

                    

Both rails see the same 100-120Hz AC hum component from the power supply, but in anti-phase: when the positive supply goes more positive, the negative goes more negative, and vice versa. As the upper valve has a plate resistor which is small in comparison with its plate resistance, most of the B+ noise will be transferred to the grid of the lower valve and, inverted, onto its plate. Conversely, only a small fraction ( ~1/9th) of the B+ noise will appear (non-inverted) at the upper valve’s cathode, as it is reduced by the plate resistor + the inverse of the valve’s mu. Thus, very little noise cancelling occurs. But it gets worse...

At the same time, (anti-phase) B- noise will be transferred non-inverted through the lower valve’s cathode to its plate through the cathode bypass capacitor; and since the grid leak resistor is very large and only has the parallel combination of the plate resistor and upper valve’s plate resistance (~75 Ohms) at the grid end to AC ground, virtually no B- noise will arrive at the lower grid to be amplified in anti-phase to the contribution from the cathode. As the two noise components – from B+ and B- – are in anti-phase, but the positive component is inverted at the lower grid, they will sum at the lower plate/upper cathode, i.e., the output.

This problem can be largely solved the crude way by regulation (which is what I did, since I had voltage to spare anyway) or very heavy filtering, but both of these are wasteful of power, cause even more heat, get bulky, and increase component count, thus cost. Wouldn’t it be nice if we could solve this problem more elegantly? Or, even if we do regulate, improve the PSRR further? Never a bad thing.

 

The Solution

Two solutions for eliminating noise presented themselves (see the answer to my letter in Tube CAD Journal July -00 pp 11-13 for a nice analysis): (1) changing the circuit by adding a gain + phase inversion stage into which additional B+ rail noise for the upper valve is injected to achieve cancellation at the output; or (2) bypassing the lower cathode to the B+ rail (also AC ground) instead of B- (right figure above), thus solving the problem where it occurs. Both are viable solutions, but are not problem-free. (1), while technically excellent, flouts our stated design goals – no more valves than necessary.* The problem with (2) is that the bypass cap would have to be big to maintain low-frequency p-p action (at least 1000uF) and high-voltage (probably 350V, as there are so few 250V electrolytics) – hence both bulky and expensive. (If the rails were lowered to, say, 95V, a 200V component could be used, but this is still pretty big.) In addition, a smaller cap to the B- rail (~47-100uF) might be necessary. Not a very neat solution!

                                         

However, as I was skimming through my collection of valve circuits and magazines, an idea came to me: what about using a P-channel MOSFET as a kind of voltage-follower series regulator/capacitance multiplier for the lower cathode’s bias? Because of the great PSRR of such a follower, the cathode would be immune to B- noise, while noise from the B+ rail could easily be injected into it from the MOSFET’s source (which also provides a low impedance to AC ground) via an RC network at its gate (see above figure). Since the gate draws extremely little current, a big resistor and small cap to B+ could be used. As a bonus, adjustment of the amp’s quiescent current would be easy to implement. This solution would be far cheaper and less bulky than the capacitor-only one. Problem solved! (If you loathe transistors, you can always try the bulky electrolytic-cap solution – I haven’t.)

*Note that I had not set this goal when I wrote the letter. In fact, the schematic had both a gain stage and overall feedback. It was only later that I decided that these features might not be strictly necessary (although still a nice option) in a headphone amp. 
[Addendum 26/4 2002: I am now using a version similar to the schematic in TCJ, i.e. with a VA front end and NFB. Whether this is an improvement is a matter of taste. It's somewhat  less "euphonic" or "soft", which I currently prefer.]

 

The Whole Caboodle

This is how the complete final circuit turned out:

 

            

                                                     The complete OTL Headphone Driver schematic


I cannot promise that the voltages will measure as indicated. The PS circuit has not been built exactly like this, although nearly (I didn’t have all the requisite components to hand), but this is the preferred version. See below for possible resistor adjustments.

 

Ear and Headphone Protection

As drawn, the circuit has no satisfactory failure protection. Reverse diodes on the output limit excursion to +/- 12.7V (note polarity!) and the 10M grid resistor to ground might reduce output offset a bit (by causing grid current to flow from ground) in case the DC servo opamp fails. Note that the headphones must be unplugged during power up (wait at least a minute) and power down – I measured an offset of >12V on start-up! However, for better protection, the circuit below is recommended (from Tube CAD Journal Jan -00, p 15). If you don’t include it you are on your own – it’s your ears and headphones! Unfortunately, it doesn’t seem to be possible to avoid a relay in the signal path. I suggest you use a good one made for audio, such as a mercury reed relay (unfortunately no longer available new because of the toxic mercury), or a good vacuum-sealed relay with gold contact points. For this circuit, a small separate 2x12V transformer with a simple PS giving +/-12V must be used. (The heater supply cannot be used here, because it is negatively biased to avoid exceeding rated cathode-to-heater potential.) This can also power the DC servo. Also connect the reverse diodes on the output to the +/-12V rails, as indicated. This PS should be left permanently on (not switched along with the amp) or at least switched on before the HT supply. For the opamps, probably any dual, low current consumption units will do, e.g., OP 275. (For 600-Ohm ‘phones, the rails/zeners could be raised to 15-18V, if desired).

                 

                                       Output protection circuit (from Tube CAD Journal, Jan 2000)

 

Construction issues

Apart from the servo and protection circuits (and optionally, parts of the PS), the circuit should be point-to-point wired, and built into a stable metal or metal/wood cabinet. The PS voltage-dropping resistors (220R/6W) and/or MOSFET voltage divider (39K/115k) and/or heater PS resistors (1R2/6-10W) may have to be adjusted on test. All gate- and grid-stopper resistors are 1k and must not be left out. The 10 and 22uF caps should be close to or mounted onto the valve pins (upper plate and lower cathode of each channel). Further bypassing with smaller values is an option. The V1 plate to V2 grid coupling cap should have as short leads as possible. Preferably star-ground all ground returns to one point (onto a stand-off). A line filter on each primary wouldn’t hurt anything, either. Don’t forget fuses: a common primary one and one for HT secondary (not drawn – find value experimentally). Heater and HT transformers/windings could be turned on simultaneously, as the HT voltage is too low to cause much damage to the cathodes (I guess – but see next section on slow-start).

Quiescent current is set by measuring voltage drop across the 82-Ohm plate resistors on each channel; it should be about 4.1V. Check that output offset hovers around not more than +/- 1-2mV before plugging in the ‘phones.

 

Components

I feel ordinary solid-core transformer-winding wire is fine for hook-up, but YMMV. All resistors unmarked for power are 0.6W metal film (or better in the critical locations: the grid leaks). Gate and grid stoppers can be smaller (0.25W). The power resistors are wirewounds (I used RCD series 100 or 235), except the 82-Ohms, which are 2W metal film (or bulk foil). Using non-inductive types is unnecessary, as the values are so small. All non-polarised caps should be polypropylene (SCR MKP or better) or paper-in-oil, 250V minimum, except those in the snubber (CRC filter before the rectifier), which should be ceramic 4kV. The 330n caps (which reduce ripple on the rails) may have to be mounted last, after noise has been adjusted out. These can be paralleled with much larger caps (e.g. 100uF/low-leak/low-ESR electrolytics) if slow start-up is desired. The PS electrolytics should be good low-ESR 250V types. The 47uFs can be 100V. The MOSFETs must be handled with great care (static sensitive!) and the PS ones must be mounted and isolated on a large heat sink (min. 3oC/W@10W). The 12V Zeners should be at least 0.5W. The servo opamp must have low voltage offset/low gate current; the LF412 is recommended, but any double JFET-input opamp with low offset will work – I used an OP275. To feed the servo, a separate +/- 12V supply is a better and safer solution than the one shown. It could also feed the protection circuit outlined above.

 

Variations

                      

I feel this circuit works very well as it is. However, the obvious tweak for the SE enthusiast to try would be to convert it to single-ended operation (disadvantages were noted above). This is rather easy: just remove the cap from upper plate to lower grid; the lower valve is now a current source. You should also replace the IRF9610 in the lower cathode lead with an 82–120 Ohm resistor (no bypass; see left figure above) to increase the effective load impedance. You can leave the upper plate resistor in – it won’t do any harm. I felt the sound became somewhat brighter/harder. In the right figure above, the lower valve is a pentode-connected current source (well, almost – g2 won’t follow the cathode perfectly, but certainly well enough). If you go for the SE version, try this one; I felt the sound improved quite a lot (surprising, perhaps, as the headphone load will usually dominate strongly). It’s actually a worthy competitor to the White version. Note: I wouldn’t recommend switches to change configuration, except maybe for initial comparisons – until you decide which way you like best. The amp must anyway be off during switching.

             

There are even more possibilities (figures above). Deriving a separate voltage source for the upper g2 to run in SE pentode or ultra-linear mode (with pentode CS as above) might sound very good – and kill the noise! Voltage can be taken from before the IRF610 reg. in the positive PS. Another idea is to keep the White follower, but remove the lower cathode bypass cap; this gives slightly asymmetric PP. I initially did this by mistake (forgot to bypass!), and found the sound euphonious and pleasing, but a bit "larger than life". Another idea is to put a 1-2.2uF cap across the plate resistor (or from plate to ground). This would reduce the signal to the lower valve as frequency increases, so that higher frequencies are amplified increasingly single-ended. This could have been the ultimate solution – warm and super-detailed SE mids and highs and firm, powerful P-P bass? – if it were not for the fact that output impedance would increase with increasing frequency as SE is approached, causing an ultimate treble roll-off of about 1-2dB. This isn’t "technically" adequate, but who knows, perhaps it sounds good with CD? Unfortunately, the load for the upper valve in the SE region will be low with bypassed lower cathode; see above remark about pentodes’ superiority as CSs. (Note: this tweak is just not compatible with low-impedance ‘phones, like Grados; they would suffer severe roll-off.)

This amp uses an active servo to keep the output at 0 volts DC. Some people feel that servos affect the sound adversely. One reason might just be that the break frequency is set too high. This servo has a rather low break frequency (-3dB@~3Hz), and in addition, a 2nd-order characteristic, instead of the usual 1st-order one. Both should contribute to reducing sonic imprint as much as possible. If you still want to get rid of it, there is always the traditional, cumbersome cap solution; proceed as follows: (1) remove the DC-servo assembly and input coupling cap; (2) replace the 10M resistor with a 470k one; (3) arrange output coupling as in figure below – use HQ polarised electrolytics (OSCON, Cerafine, Black Gate, etc.), not bipolar caps. (The protection circuit is best retained, but must be inserted before the coupling capacitors.)

                                               

 

Coda: sonic impressions

In one word: nice. Warm, but quite clear; "colourful" and nuanced, but not coloured; no hardness, but quite revealing of source quality; good transient response ... all my subjective impressions of course. (Did I hear you say "Proud Poppa Syndrome"?). I previously made a MOSFET driver of very similar construction (except in was SE – as in the above tweak); this sounded noticeably thinner and less "dimensional" (in its defence, it should be mentioned that FETs need a lot of quiescent current to reach their most linear region, and it may not have been high enough in my amp). 
   Your sonic ideals may of course differ from mine, but I doubt that you will be entirely dissatisfied with this amp. (Still, if you are, the PS and protection circuit - not to mention the valves – are perfect for many other projects.) If you do attempt it, drop me a line; I’m curious to hear your opinion, whether good or ... erm, good. Watch those high voltages, have a ball, and don’t let smoke of any colour or odour stop you!

                                                             Unidentified glowing object? 'Well, this dude sure ain't no EL86!'

– Morgan L.

Addendum 16/10 2004: Here is the schematic of the final version, drawn from memory. There is a slight error: the 100-ohm resistor on the upper EL86 screen grid should go directly to the upper anode, rather than to B+ (100V). (In my version, the loctal equivalent 7N7 was used in the 6SN7 position.)

Back