A 6EM7/6EA7 Headphone Amplifier
A Conventional, Single-Ended Topology Developed as a "Test-Bed"
My earliest perception that there was something "different" in listening to a valve based, single-ended headphone amplifier, was the odd realisation that I was typically using the headphones ten or more times as often as I otherwise would, and then for ten or twenty more times as long on a typical session. later I heard this effect referred to as a reduction in listening "fatigue". I found that the concept of listening "fatigue" in this sense had little if any meaning until I actually experienced it. Later I became easily able to pick the enjoyable, warm sound of an even order dominant distortion characteristic. Early versions of this amp, with distortion figures probably reaching double figures, probably aided such discernment!
The 6EM7/6EA7 Valve
I'm not entirely sure how I happened upon the 6EM7/6EA7. I'm pretty sure it was as simple as mucking around with the few valves I actually had on hand, and looking up a Philips Miniwatt databook I'd been given, for any valve that looked "interesting". Seeing the application reference for this particular valve struck a chord with me; I knew well that the vertical output stage for a television bares considerable similarity, albeit broadly speaking, to an audio output stage, typically of single ended topology.
Further I'd also already long had an interest in what are called "compactrons" (at least in the US); multiple but dissimilar valves in a single envelope. So here was the 6EM7/6EA7, representative of what might be termed a "family" of similar dissimilar-dual-triode valves intended for television vertical deflection stages. The "family" also loosely includes other octal based valves 6GL7 and 6DN7, and the novar or noval based 6CM7, 6CS7, 6CY7, 6DE7, 6EW7, 6FD7, and 6GF7A. Actually, comparing spec's, I'd hesitantly say the octal based 6EM7/6EA7 is to it's family as the noval based 6BM8 is to its triode-pentode family; probably about the best rated overall, all things considered. So it's the more worthwhile type to work with. The 6EM7/6EA7 is also relatively inexpensive and is generally well regarded and recognised, (though not in the same league as the 6BM8 of course!).
1; 6EM7/6EA7 RCA datasheet;
click on the thumbnail to see a larger image.
The spec's for the 6EM7/6EA7 are shown in Figure 1. The small-signal triode is not unlike, say, a 12AX7, with quite reasonably high µ (of 68), operating at relatively low anode current (1~2mA or so). gm is low, at 1.6mA/V, and ra is high, at 40 kOhms. Originally this section would presumably have been used as the vertical oscillator section in a television application. The power triode section on the other hand certainly isn't short of muscle, with a 10W anode dissipation rating, and an ra of just 750 Ohms. gm is relatively high, (at 7.2mA/V), µ is 5.4 and anode current typically as high as 50mA. These ratings list comparably alongside those of cathode follower favourite, the 12B4, and were it directly heated one might usefully compare spec's with the 2A3 too.
In practice I've found the 6EM7/6EA7 easy and tolerant to deal with, and certainly sonically very agreeable. I do have one reservation. The dissimilar triodes require very different supply voltages. The small-signal triode is run off typically a 300V B+, drawing merely a mA or two. The output triode uses a B+ somewhat less than 150V, at up to 50mA. A single supply needs sufficient VA rating to "hold up" at 300V to provide B+ for the input stage running at say 1.5mA, while feeding a 15~20W 5 kOhm resistor dissipating 4~5W to provide B+ for the output stage. I found this quite awkward to construct, initially; approximately 50% of the power consumed by the amplifier is wasted as heat dissipated by this big dropper resistor - a crude and clumsy implementation.
There are however many other options for the power supply. This is especially so if a custom-made supply transformer is envisaged. Ideal perhaps is a single-secondary transformer (not considering heater winding[s]), at, say, a much lower secondary voltage of Vac = 120Vac~150Vac or so. Rectified and smoothed, this would supply the output stage directly. The type of filter used (L or C input) and minor additional series resistance, would control B+ for a given Class A mode standing current. Meanwhile a doubler across the same transformer winding would give a low current B+ at substantially higher DC for the input triode. I think this would be the simplest effective arrangement and much better than what I actually used.
In my case I stuck with a number of quite oppositely structured (relatively high voltage, low current) transformers, simply because they were on hand, and had integral heater windings aplenty. Not really particularly good reasoning, but there we are; what is actually available on-hand governs much of what I use and do! I really didn't want to use one of the several suitable 120~150Vac transformers I also had, as in each case I'd have to go looking for an additional heater supply transformer. In retrospect, this alternative might well have been a better long-term choice. The power supply I ended up with (Figure 2) used a 300-0-300 secondary transformer, the biggest little power supply choke I had, and a tri-section electrolytic that although quite old and not given any reforming, has never missed a beat.
2; Power Supply Components;
click on the thumbnail to see a larger image.
Right from the outset I wanted a really big chassis, that I could easily work on, revamp and add bits to, and then at the end of each period whip back together again, convey from workshop to study, and listen to. I intended to use this chassis to start trying several different topologies experimentally (see Figure 3). The wooden sides and ex-VCR lid chassis took little time to make, (you guessed immediately, right?), but proved very effective in this role. The metal chassis unscrews from the wooden sides with four big screws, then I flip it upside down and use the same screws and holes to sit the chassis upside-down for working on. It all works out really quite well, and I'm determined to employ this methodology again in future.
Figure 3; Test-bed Chassis Internal; click on the thumbnail to see a larger image.
This project headphone amp actually started out as a 6BM8 based topology, then involved an EL36/6CM5 pentode, ( - intended originally for horizontal output stages in televisions -), in triode mode and then a bit of both, before the 6EM7/6EA7 thing happened! After that things got progressively simpler if anything; quite opposite to the experiments with totem topologies for the input stage, that I'd originally intended. I'm hoping to return the next version, Version 7, back in this direction (see Figure 4 and Figure 5). This article is about "Version 6". Version 7 started out, in turn, as a smaller, tidier packaged remake of Version 6, because Version 6 is something I'd like to retain for a while, I'm so pleased with it. I thought I had now better build a cleaned-up version of it.
But already Version 7 is becoming something else; right now I'm contemplating dropping the transistor constant current source load on the input valve, and turning that part into a Kimmel mu-follower. That's a topology which has long particularly interested me. I'm hoping it will retain the high gain and good PSRR features of the Version 6 constant-current-source-loaded input stage.
4: Version 7 New Chassis;
click on the thumbnail to see a larger image.
Figure 5: New Chassis Underside; click on the thumbnail to see a larger image.
Early versions with the 6EA7/6EM7 simply used both triode sections as cascaded anode-loaded stages. These versions worked quite well, and had a lovely warm sound that spoke of absolute lashings of 2nd harmonic distortion; probably into double-figures percentage-wise, for all I know! The biggest hassle I found was controlling the B+. I wasn't using an active regulator on the supply, just a choke and electrolytics and series power resistors. The transformer regulation itself was very poor. So every time I changed circuit values during development, (especially the bias on the output stage), the load on the supply changed, so the B+ changed, so the operating points of both input and output stages deviated from that planned. Round and round in circles I went!
Nevertheless, with iterative swapping of power resistors in the B+ chain I generally got the thing where I wanted. However I also continued to be plagued by hum problems.
I spent hours and hours chasing hum. I put a 56 Ohm + 56 Ohm resistive divider across each of the two 6.3Vac heater windings my transformer provided, with the centre-point of the divider grounded. No change. It didn't matter which bit of these networks was made the ground point either. I star-earthed absolutely everything religiously, using 15A mains wiring wire and a single big chassis-earth bolt. No change. I beefed up the capacitance across the supply B+ at each stage. This produced some change, but it took enormously large values to achieve worthwhile improvement. I fitted a metal shield between mains transformer + filter choke, and the output transformer area of the chassis. No change at all. Nothing.
The breakthrough came when I accidentally wrecked one of my little output transformers. I was drilling a small bolt-hole into the chassis too closely to it. In one of the stupidest, clumsiest, most senseless mistakes I've ever made, I allowed the spinning chuck of my power drill to tear down one side of one of my more valuable little output transformers. many tens of severed wire ends left the transformer irrecoverable. Stunned and horrified at such gross carelessness, I collected my bat and my ball and walked away from the whole project for quite some time. When I picked up again I started fiddling around with a replacement transformer, and for a while had the amp powered up with the replacement transformer loose and temporarily wired on the workbench. Hum miraculously gone! But when I mounted the new transformer back up in place on the chassis; the hum was back with a vengeance! Clearly the mounting position of the output transformer was the key.
Mounting the output transformers under the chassis, (not alongside the mains transformer and the DC choke up on top), and orientated at right angles to the power transformer, was the answer. It wasn't long before I'd seen over and over again in internet discussion notes and textbooks that even a valve novice should know not to orient mains and output transformers the same way and position them alongside each other. <Sigh>.
The hum was now very much reduced and quite tolerable. But it was still there, and I don't like it at all. The complete elimination of it, for all practical purposes, came with the adoption of the constant current source load for the input stage. Clearly, a benefit of this topology is vastly improved PSRR. Later, the addition of loop negative feedback improved things still further.
Had these later improvements not occurred, I almost certainly would have eventually side-tracked to develop a B+ regulator as an alternate countermeasure. This is an area I've long wanted to experiment in. I have in mind applying my diverse collection of TV power transistors. Those typically used for TV horizontal output stages tend to have more than adequate Vceo, (and certainly current handling and frequency response capability). But their gain is appallingly low and I imagine this would make them unwieldy to design with. There are, however, a number of similar devices intended mainly for switch-mode power supply applications. Vceo is down in the 400V~600V region, and gain is correspondingly a little better. I suspect that these devices might be interesting to design around.
A lot of fun to listen to for long periods, the early versions of my 6EM7/6EA7 headphone amp ultimately were deemed to possess considerable overkill in the distortion department! That, and the hum problems referred to, lead to the constant current source load for the first stage. This, in letting the first stage run at high gain, yielded the opportunity for a little loop negative feedback, and resulted in a nice balance of clean spec. design, (finally a frequency response flat 20Hz to well past 20kHz!), but still with warm sonic qualities (I have little doubt the distortion characteristic remains partially present, and H2 dominant!).
How the First Stage Works
click on the thumbnail to see a larger image.
The schematic of the Version 6 amp is shown in Figure 6. You might also refer to the characteristic curves for the input and output triodes shown at Figure 7 and Figure 8 respectively.
Figure 7; Va vs. Ia Curves for the Input Triode;
click on the thumbnail to see a larger image.
The first stage is an anode loaded triode voltage amplifier, (typically abbreviated to "PLT" for Plate Loaded Triode), with a constant current source ("CCS" hereafter) load. This type of loading confers high gain and low distortion. The operating point of this stage was selected for reasonably good linearity. The intended strategy was to let the output stage alone confer the amplifier's sonic signature, and go for a "clean" input stage, rather than engender more complex interactions by involving the first stage too. The small signal triode element of the 6EM7/6EA7 is relatively high gain (µ = 68) and this CCS loading gives a stage gain approaching that figure, (at least in theory!).
The CCS itself is based around a 2SA1626 transistor. I started by considering using the well known MJE350, but decided it'd need more base current (and hence more current sent down the voltage divider at its base), than I was comfortable with. I didn't have many suitable alternate low power PNP transistors with 400V or so Vceo, so was quite please to find I had several pre-loved 2SA1626's squirreled away! The alternative would have been to see what P-channel MOSFET's I had, and design the CCS around one of those.
The CCS loading for the first stage triode appears as though it were a very very large value load "resistor" operating from a very very high "B+". Signal voltage swings at the valve's anode appear insignificant relative to this "B+" (because it is apparently so very high) and so the voltage drop across this load "resistor" never appears to significantly change; so the current through it therefore appears to be effectively "constant".
The CCS source fixes the first stage current at around 2.2mA. The value of this current is basically one diode drop across the value of resistor in the 2SA1626's emitter circuit;
Vak[1N4148] / 220 Ohms = Ie = Ic[1N4148] = Iak[triode 1] = 0.5V / 220 Ohms = 2.25mA
In the CCS circuit, the 820 kOhm resistor sets up a forward biasing current through the two 1N4148 diodes of approximately;
270V / 820 kOhm = 330µA
This current is designed to be a notional 10 times that expected as base current of the transistor, thus swamping changes or variations at the transistor base, and thereby establishing a solid reference. With this current the 1N4148's are only just "on" so their forward voltage drop is relatively low, at around 0.5V. The current is determined such as to be significantly larger however, than the base current of the transistor. The base current is;
Ic / Hfe = Ib = 2.25mA / 80 = 28µA
The load-line produced on the Ia versus Va curves for the first stage triode, is effectively horizontal. I've marked this on Figure 7. Ia stays always at the fixed level set by the constant current source. The resistor(s) in the valve's cathode circuit, (820 Ohm + 82 Ohm = 900 Ohm), then set the value of bias voltage appearing at the cathode;
Vk = Ia.Rk = 2.2mA x 900 Ohm = 2V
Following the Ia = 2.2mA horizontal line on the Ia versus Va curves until it intersects with the curve for a grid bias of -2V, predicts a quiescent voltage at the valve's anode of around 210Vdc. In practice the current Ia measured close to 2.3mA, the cathode voltage was 2.05Vdc and the anode voltage was 190Vdc. This is about as close as I've ever gotten spec. sheet design and actual practice to agree. I gather however that those "well versed in the art" expect much better correlation!
Negative feedback is inserted at the first stage triode's cathode. The cathode resistor is divided to facilitate this in about a 10:1 ratio. This later was chosen simply to minimise the effect of the split on the open loop gain of the stage, since only the 820 Ohm portion can be AC bypassed (with the 470uF capacitor), in this circuit arrangement. The value of feedback resistor, 390 Ohm, was then found empirically (using a 5 kOhm pot in its place temporarily) such as to give an overall amplifier gain of 4 (+12dB). I had found, by previous experience, that this level of closed loop gain would be about right for the type of headphones being used and typical CD signal source levels.
It is interesting to apply the theoretical operational amplifier formula for amplifier gain. This formula assumes that the open loop gain is very very high indeed, which the gain of this amplifier certainly is anything but, so some error would naturally be expected;
Av(cl) = (390 Ohm + 82 Ohm) / 82 Ohm = 5.8
This compares well enough to the practical value of Av(cl) = 4.
The open loop, (i.e. no loop negative feedback), gain, Av(ol), of the whole amplifier had previously been measured at about 17 (+24.6dB) from CD player input to headphone output. With closed loop gain now set at 4 (+12dB) the difference, (12.6dB) is the loop gain, a measure of the amount of negative feedback applied in this manner. This level would generally be considered to be fairly mild. It certainly isn't enough to start yielding stability problems, I observed in practice. I was even lucky enough (this time!) to select the correct "end" of the output transformer secondary to take negative feedback from, first time! (Well, one has a 50:50 chance, huh!),
The Output Stage
Figure 8; Va vs. Ia Curves for the Output Triode;
click on the thumbnail to see a larger image.
I must 'fess up to the fact that selecting an operating point for the output stage was something of a rough and ready affair. I had been given a collection of "odds-n-sods" little speaker transformers - most of them NOS. They were formerly spare parts stock of a retired TV repairman. Each is good for no more than a Watt or two, (down hill and with a tail-wind), and design frequency response was probably next to appalling. I was hoping that the very light loading that headphones would provide, plus a little loop negative feedback, would give me a full frequency bandwidth. This proved easily correct, and I really had little trouble achieving adequate frequency response or better.
Black and white valve TV's frequently had single-ended audio output stages. So I new there was not going to be a problem using these little transformers with 20mA~30mA DC through them (again especially with the light loading of the headphones). My main problem was the nominal primary impedance. My collection of transformers seemed to divide into two groups. Firstly there were those with a turns ratio such as to give a primary Z of 13 kOhms ~ 27 kOhms off an 8 Ohm speaker load. I guess that these are intended for single ended pentode output applications - a 6BM8 for instance.
The other small group of transformers gave a primary Z of 450 Ohms ~ 750 Ohms for an 8 Ohm speaker load. I figured these were way too low to be intended for pentode drivers. For a while I thought that they may in fact be vertical output transformers, but they seem way too small for that.
Ideally a power triode in single-ended is supposed to see about two times its ra as anode load - this being the speaker load reflected as transformer primary impedance. This may have meant possibly a 16 Ohm speaker load was originally intended, with a triode with ra of 450 Ohms ~ 750 Ohms. In my case I intended using 32 Ohm (nominal) headphones. These transformers would translate that to 3 kOhm or more on the primary side. My 6EA7/6EM7's have an ra of about a quarter of this, 750 Ohms or so, so these transformers looked eminently "suitable".
I knew from the typical applications suggested in the 6EA7/6EM7 data sheets that I'd be looking at a voltage across the valve, Vak, of around 100V or so. Add 20V or so on the cathode as self-bias, and drop 20V or so over the Rdc of the output transformer primary, and there was a B+ for the output stage of something a little less than 140Vdc~150Vdc or so.
Current through the stage? Well, the more the merrier it seemed to me, looking for linearity from the output triode's Va vs. Ia curves. Conversely less was better in terms of power supply loading and Idc through the output transformer primary. 30mA just seemed like a sensible, albeit high, figure, although the datasheets indicated typical operation of the valve right up to 50mA or so. (Somehow I doubted that my little speaker transformers would stand anything like 50mA continuous Idc and still convey any useful signal at all!). 100V at 30mA on the valve was 3W. That should be more than enough for the few tens of milliWatts I'd be looking for at the headphones!
Construction proceeded normally enough, other than the various trials and tribulations earlier reported. And that's how this project ended up. It was certainly good fun at both the design and constructional, and at the listening levels!
Single ended, and in this case valve, topologies offer unique sonic characteristics and open up a festering debate between the engineers and the subjectivists. The sonics are beguiling, and much sought after by audiophiles across the world. Demand has escalated the prices of valves suitable for single-ended driving of loudspeakers; the 300B and 845 fetch outrageous prices in the audiophile market and there must be good reasons for this.
My quest for the single-ended sound has been rewarded at very low cost with a headphone amplifier. I have heard single-ended amplifiers of both valve and hybrid topologies, and feel that the headphone single-ended amplifier discussed here offers precisely the same listening experience, but at ridiculously low cost. It remains an expensive exercise exploring these sonics with amplifiers designed to drive speakers, and as the powers rise, so do the costs, very steeply. Typically low amplifier powers, such as the 10W max of a single-ended 300B, require relatively high sensitivity speakers, 96dB/W/m or more, to achieve adequate listening levels with acceptable dynamics, and even at these low powers, the valve implementations are very expensive. Alternatively, high-powered single-ended valve amplifiers, perhaps capable of driving 87dB/W/m speakers, are very expensive, and employ very dangerous voltages. On the other hand, headphone amplifiers are fun, cheap to construct, easy to modify, and every bit as sonically marvellous as their high power cousins, and high sensitivity speakers, often criticised for their peakiness, are rare and also expensive. For me, the choices are clear.
Distortion in single-ended topologies is assymetric, and therefore even-order. This leads me to the inescapable belief that we humans like a little even-order harmonic distortion. We do not enjoy intermodulation products, and we patently dislike odd-order distortion, but we hanker after the warm, emotional romance of H2 and a little H3 while at the same time agreeing with the engineers that the best amplifiers have the lowest distortion. This is the nub of the argument, one I will leave well alone......
It is often the case in the engineering world that the most inefficient solution offers the greatest pleasure; after all, most of us prefer large displacement slow speed motors in our cars for instant response and lazy motoring, so this preference is nothing new. Indulgence always has its compensations.......
While the headphone listening experience is partially flawed - each ear is presented only with single channel information, and centre-stage sound-images are placed in the middle of or immediately above ones head - our brains quickly account for and compensate these aspects, at least to some degree. This amplifier offers a huge opportunity to be up-close and personal with the single-ended romance, and it could deliver the same marvellous sonics for you and at ridiculously low cost!
The generous and thoughtful assistance of dear friend and artful amplifier designer, Mr Hugh Dean, is gratefully acknowledged in the preparation of this text. I am particularly thankful also for the enthusiastic and mindful parts acquisitional skills of friends Mr Rod Humphris and Mr Darryl McNab, without whose transformer "findings" the project would have been a lot slower to get off the ground!