by=20 Kevin Gilmore
Grado headphones, for me, are terribly uncomfortable, but my amp sure = makes=20 them sound nice. (The ones in the picture are the SR225; I also have a = pair of=20 HP-1.) They are the only low impedance headphones out there that I know = of. Most=20 if not all of the standard opamp-only headphone amps have trouble = driving Grados=20 with any decent bass, because they want a high current output. This = amplifier=20 can output up to 0.5W into a 32-ohm load =E2=80=93 which is unbelievably = loud for the=20 Grados.
I built this headphone amplifier for dynamic headphones based on my = rules of=20 proper audio design. People who know my designs will realize that this = amplifer=20 is much more than just a headphone amp. It is a pure class A design = containing a=20 new never-before-seen servo loop that is not part of the audio signal = chain in=20 any way. I should patent it, but in any case, it is =E2=80=9Ccopyright = 2001 Kevin=20 Gilmore.=E2=80=9D
Kevin=E2=80=99s Rules of =
Proper Audio=20
Design
- Capacitors in the audio signal path are BAD. Even the best =
silver-mica or=20
poly caps exhibit non-linearities at low voltage levels. Capacitors =
belong in=20
power supply sections and nowhere else. Capacitors used to compensate =
an=20
amplifer generally mean that the amplifier is otherwise unstable, with =
poles=20
in the right half plane and is therefore a bad design.
- Transformers in the audio signal path are even worse: =
non-linearities in=20
the gain structure, parasitic capacitance between windings, impedance=20
problems=E2=80=A6. Transformers belong in linear power supplies and =
nowhere else.
- Ultra high open loop gain: REAL, REAL BAD!!! That basically means =
anything=20
with an opamp in it. Opamp circuits with open loop gains of 10,000 or =
more=20
require large amounts of feedback to make them usable. While this =
reduces THD,=20
the intermodulation products, and especially the transient =
intermodulation=20
products are much higher than they
should be.
- Servo loops MUST NOT be in the audio feedback loop. This rule is =
also very=20
important. Two of my favorite high-end audio electronics manufacturers =
put=20
servo loops into the minus input of their amplifiers. Most other =
manufacturers=20
that use servo loops do the same thing. opamps used for servo loops do =
not=20
have an output impedance low enough to make them suitable for this =
purpose.=20
Furthermore the dynamic output impedance of opamps adds =
non-linearities to the=20
audio when put in series with the gain resistor on the minus =
input.
Kind of makes designing ultra high quality audio stuff tough. My = design goals=20 in this amplifier were: keep the gain per stage down, keep all stages in = class=20 A, keep the differential front end from coming even close to clipping by = the use=20 of a current source. Because the amplifier has a low overall gain and = little=20 feedback, a servo helps to prevent DC voltages at the output. In = general, if the=20 open loop gain is kept down to eliminate all or most of transient=20 intermodulation distortion, then the amplifier circuitry has to be = extremely=20 linear and low distortion. Otherwise, you end up with something that = measures=20 and performs like crap.
The Circuit
The schematic for the standard (non-bridged output) version of the = headphone=20 amplifier is shown in figure 1. The open loop gain of the amplifer is = about 35.=20 Even with the feedback removed, the THD is less than .01%. = That=E2=80=99s important=20 because the more linear an amplifier is without feedback, the more the = THD, IM=20 and TIM distortions are reduced to unmeasurable levels with feedback = added.
Stage 1 is a dual FET fully-differential fully-balanced front-end. =
The idling=20
current is 2mA total per dual FET (1mA per FET) and 4mA for the complete =
front-end stage which consists of both dual FETs. The dual FETs generate =
the=20
bias which runs the second stage, and keeps it and the resulting output =
section=20
in class A at all times. The FETs are ultra low noise dual
units =
specifically=20
designed for audio uses. The total voltage gain of the first stage is =
50.
Stage 2 is the driver stage. It is a standard class A voltage = amplifier =E2=80=93 in=20 this case used as a voltage shifter. The voltage gain is 0.5 and the = idling=20 current 4.3mA.
The push-pull class A output stage is a series of paralleled emitter=20 follower, current buffers. The voltage gain is 0.9 and the current gain = is 75.=20 The idling current is 15mA per transistor (or 60mA for the 8 transistors = off the=20 +16VDC rail and 60mA for the 8 transistors off the -16VDC rail). I have = designed=20 the output section to run at what I have determined is the sweet spot = for these=20 transistors, which is 15mA each. Yeah, it gets hot; its supposed to get = hot (but=20 not hot enough to require heatsinks). It=E2=80=99s not possible to make = an amplifier=20 with an output impedance less than 0.1 ohm without throwing around a = fair amount=20 of current.
The servo circuit is new: most of the servo designs (Mark Levinson = and Krell,=20 for example) put the output of the DC servo back into the =E2=80=93 leg = of the=20 amplifier. I just do not like this. That puts the noise and = non-linearities of=20 the opamp inside the audio loop.
My servo feeds back to the current sources for the dual FETs in stage = 1. Like=20 all servos, it is an integrator. Due to the large (relatively) = integration=20 capacitor and the 1 meg resistor, the frequency of this filter is 0.05 = Hz. With=20 even a decent opamp, the servo=E2=80=99s noise is in the tens of = microvolts, and does=20 not affect the operation of the current sources significantly.
The servo opamp in this amplifier measures the DC at the output, if = any,=20 integrates it and applies it to the midpoint of the two LEDs. The LEDs = do have a=20 slight change in voltage with respect to current, about 3 or 4%, and = that is=20 enough to make the servo work. Notice that if the transistors or the = resistors=20 are very poorly matched, the servo will not work because its total = control range=20 is at most 10%. Most standard servos (such as the Mark Levinson or Krell = servos)=20 have a much wider range.
For high impedance headphones, a little DC will not hurt the phones. =
With the=20
low impedance Grados, even 0.1VDC over a long period of time will =
definitely=20
damage and/or change the sound. If all the parts are hand matched, the =
power=20
supplies are exactly the same and all the resistors are really good =
quality, the=20
amp should be stable and should not drift. In that case, the servo could =
be=20
omitted or replaced with a 20K trimmer pot wired from +16VDC to -16VDC, =
with the=20
wiper going to the DC adjust pin. The prototype uses 0.05% tolerance =
resistors,=20
and I hand-matched the transistors. The output DC is less than 6mV and =
has=20
stayed absolutely stable for the few months I have had the unit.
Figure=20
2
The amp gives about 0.5W into 32 ohms. In the classical definition of = class=20 A, the top transistors would be sourcing 120mA and the bottom = transistors=20 sinking 0mA. However the two 3k resistors in the second stage actually = prevent=20 the total shutdown of either transistor bank by keeping the opposing = stage at an=20 absolute minimum of 5mA. In any case, 0.5W into Grados is unbelievably = loud.=20
The balanced bridge output version of the amplifier (figure 2) is for = those=20 headphones that can be wired as dual mono (see the add= endum=20 for instructions on converting a pair of standard Grado SR-80 headphones = into=20 dual mono headphones). It has twice the voltage swing, twice the slew = rate and 4=20 times the output power (competes with the $2600 HeadRoom balanced Max=20 amplifier).
The ultra-regulation of the power supply (figure 3) is so over the = top and=20 unnecessary that most, if not all, people building this amplifier would = not even=20 notice the difference. However there are a number of benefits to this = design.=20 First its a dual tracking design. Because the open loop gain of the amp = is low,=20 its common mode rejection due to the power supply is not great. However = if both=20 the + and =E2=80=93 voltage rails move up and down the same amount, = there is no bias=20 drift.
Because of the pre-regulators, the total line/load variations are = under=20 0.0001%. The fast capactors allow the power supply to react rapidly and = in a=20 controlled manner with highly reactive loads like the Grados. The opamp = outputs=20 are active in both directions; they can push or pull to keep the power = supply at=20 exactly the right point. The power supply design was an attempt to come = up with=20 the absolutely best power supply I could. It is even quieter than = batteries.
Construction
Download high = resolution=20 images of PCB pattern and layout
The prototype was built with point to point wiring, keeping it tight = and=20 tiny. The layout is exactly as shown on the schematic, which is also why = the=20 board is only one layer. Although the amplifier is easy to build without = a PC=20 board, I designed one for the amplifier. Each board (you need 2 for the = standard=20 amplifier or 4 for the bridged version) is 3.3=E2=80=B3 x 3.8=E2=80=B3. = In a few months, I may=20 redo the board in one of the systems where I can ship the file over the = internet=20 and get boards back (like expresspcb.com).
The servo=E2=80=99s total control range is at most 10%, so the some = of the=20 servo-related parts must be closely matched for optimal operation of the = servo.=20 The 500 ohm resistors, 1.6V LEDs, bias transistors and second stage = transistors=20 must be matched to within 0.5% or even 0.25% for optimal operation of = the servo=20 (the dual FETs are already matched). To match the LEDs, put one in = series with a=20 10k resistor, hook it to 15VDC and measure the voltage across it. Do it = to a few=20 of them and pick the closest match.
I used a Tektronix curve tracer to match the transistors. Figure 4 = shows two=20 circuits for matching the beta (gain) of the NPN and PNP transistors = using just=20 a voltmeter. Simply measure (and match) the voltage of each = transistor=E2=80=99s=20 collector with respect to ground. It should be in the range of 10V to = 11V for=20 the NPNs or 5V to 6V for the PNPs.
There are no substitutes for the FETs. The PNP and NPN transistors in =
the=20
prototype were the MPS8099 and MPS8599. Onsemi has discontinued them, =
but there=20
are still lots available. I am buying all my transistors now over the =
web from=20
MCM Electronics. I have =
had too=20
much trouble with everyone else. The 2SA1015 PNPs are $0.46 each; the =
2SC1815=20
NPNs are $0.41 each. The 2SJ109 p-channel dual FET is about $6.50; its =
n-channel=20
counterpart is $5.90. By the time you add it all up, it=E2=80=99s still =
under the=20
minimum
MCM order, so you=E2=80=99ve got to buy something else.
The high-speed 5uF capacitors in the power supply are not critical. = They=20 satisfy the =E2=80=9Clunatic fringe=E2=80=9D in audio. These capacitors = are rated for a slew=20 rate (dv/dt) of about 4 times a standard capacitor. They cost $7.50 from = the Illinois Capacitor = Company.=20 Similarly, the opamps in the power supply are not critical. In the = prototype=20 power supply, I used the Apex PA09, which has a 400V/uS slew rate and = costs $167=20 each (remember, I am crazy). Normal people should use the Texas = Instruments=20 OPA549 (10V/uS) that cost $11 each. The LM3xx regulators are heatsinked=20 (dissipating at least 3W each, 6W for the bridged version of the = amplifier).=20
The enclosure is the Mod.U.Line by Precision Fabrication Technologies = Inc.=20 (part number 03-1209-BW), available from Newark Electronics. They are = about $85=20 a piece these days. The aluminum enclosure is easy of use, easy to = punch, and=20 keeps a decent paint job. The headphone jack is a cheap Radio Shack one. = Although it works fine. I have since gotten some Neutrik connectors = which I will=20 put in at some point. In theory, there is a ground loop between the RCA = jacks on=20 the back plate and the headphone jack on the front plate. Although the = 60 Hz hum=20 is about 110 db down, the Neutrik jacks are isolated and will make this = problem=20 go away.
The amplifier=E2=80=99s output is voltage limited (not current = limited) because the=20 second stage runs out of voltage swing =E2=80=93 typically about 6V rms. = For a 32 ohm=20 load, the maximum output power is 1.125W (0.1875A). For a 300 ohm load = such as=20 the Sennheiser HD600, the maximum output power is 0.125W (0.02A). To = increase=20 the maximum output voltage to, say, 10V rms, try increasing the power = supply to=20 =C2=B120VDC and change the 500 ohm resistors in the current = sources/sinks to 600 ohms=20 (or =C2=B124VDC and 700 ohms =E2=80=93 but careful not to fry the output = transistors).
The bridged version will output 4.5W into the 32-ohm Grados and 0.5W = into the=20 300-ohm Sennheiser HD600. At full blast, the amp does drop out of class = A, but=20 at levels that will fry your ears anyway.
The Result
Adjustments: In the power supply, adjust the top 20K trimmer = pot for=20 +24 volts at the tap after the LM317. Then adjust the bottom 20K trimmer = pot for=20 -24V at the tap after LM337. (Note: the =C2=B148V taps are shown for = reference only;=20 they are not actually used by the amplifier.) If the servo has been = replaced=20 with a 20K trimmer pot, adjust the trimmer so that the output measures = 0VDC.=20 Anything under 25mV is fine.
I listened to the balanced HeadRoom Max at The Home Electronics Show. = The low=20 frequency bass slam is absent from the unit. When I listened to the = BlockHead=20 ($3333) and compared it to my unbalanced/balanced amp, the same thing = happened.=20 Without the kaboom, it=E2=80=99s just not as much fun to listen to.
This amp gives the Grados and Etymotic Canalphones a fuller and much = more=20 upfront sound than the built-in headphone jacks on various players = =E2=80=93 the bass=20 has a snap to it that it never used to have. And the image moves from = around=20 your head to the center of your nose. The Etymotics tend to sound kind = of thin=20 and distant with other amps. In general, I have to say that this amp = produces=20 bass that is much more solid =E2=80=93 similar to putting the microphone = in front of a=20 bass violin instead of inside it.
c. 2001, Kevin =
Gilmore.
From The Homepage of =
Kevin=20
Gilmore. Republished with permission.