by Sheldon D. Stokes
A couple of years ago, I bought a set of the now-famous Grado SR-60 headphones. I use them when I travel or when I don’t want to bother anybody with music at home. But my reference system doesn’t have a headphone jack except on my CD player, and it doesn’t have a volume control. And if that wasn’t bad enough, the headphone amp isn’t very good quality. So I needed a good quality headphone amp that I could use for all my sources.
A problem with the Grado head-phones is their low impedance. They present about a 32 ohm load on the ampli-fier driving them. My first thought was to use tubes, and drive the headphones with a cathode follower. but with the 32 ohm impedance, this isn’t very practical unless you’re building a large OTL amp. So I decided to build a solid state headphone amp. The circuit shown here is an adaptation of Walt Jung’s headphone amp described in the Audio IC Op-Amp Applications book. From what I’ve heard, this book is out of print.
The changes I’ve made in Walt Jung’s circuit are that I have replaced the bipolar output transistors with MOSFETs, and I changed the biasing method for the MOSFETs. I also designed the power supply. But the essence of the circuit remains unchanged. The heart of the circuit is a dual op-amp. Which drives a pair of push pull output MOSFETs. The output from the MOSFETs is included in the feedback loop of the op-amp, so the distortion is very low. The gain of the amp is set at a value of 2, and it includes an input pot for volume adjustment. [Editor: To increase the gain, increase the value of R11. For example, R11 = 10K will set the gain to 10.]
The headphone amp circuit is basically a simple non-inverting op-amp gain stage with external buffering (figure 1). I find that for op-amps to sound their best, they should not be operated at the edge of their drive capabilities. Many commercial headphone products use op-amps directly to drive a pair of headphones. While it can be done this way, I have found that adding a buffer to the output of the op-amp reduces the harshness and stridency dramatically (two common complaints about op-amp based designs). I rarely use op-amps for serious audio design, due to what I consider to be questionable sonic merits of op-amps. But in this circuit, the op-amp is running at a very low gain setting, and is not driving the headphones directly. These two factors make this headphone amp a very neutral and musical device. The headphone amp also uses a low impedance, regulated power supply sourced from a 25W toroidal transformer (figure 2). For such a small amplifier regulation is practical and the sonic benefit is quite noticeable. This amplifier is direct coupled as well, so there isn’t any capacitors in the signal path.
The amplifier uses a zener diodes to provide the bias current for the output FET’s. The output stage is biased fairly heavily for such a small amplifier. I find that FET’s sound their best when they have fairly high bias currents. This amp will run class A up to two watts. Each device is dissipating a watt of power at idle, and should be mounted on a heat sink (I used an AAVID TO-220 Power Cooler, Digikey #HS132-ND). The biasing portion of the circuit also has the provision for limiting the amplifier output by using a pair of LED’s (per channel). If voltage swing gets too large, the LED turns on and the output signal is reduced. I don’t use the LED’s in my prototype of this circuit, they shouldn’t degrade the sound quality of the amplifier, but I left them off just because I didn’t think I’d need them, and I try to keep as little in the signal path as possible.
This amplifier is powerful enough to also drive an efficient set of speakers. It produces about 4 watts of power before clipping. This amp clips asymmetrically (at 10V p-p), as the gates of the MOSFETs are not driven in their potential center. The reason I have done this is that I have found that op-amps sound the best when they sink a bit of output current. Many folks put a resistor on the output to one of the voltage rails to “bias” the op-amp into class “A” operation. By tying the output of the op-amp to one of the gates instead of the middle of the two gates, the op-amp output is sitting at about 4.5 volts above ground potential, and thus is “biased” into class A. This may seem to contradict what I said earlier about needing to buffer the output of the op-amp so it doesn’t drive difficult loads. In this case, the op-amp is still only driving the gates of the MOSFETs, and the load it’s output stage sees is essentially the same as if it were connected to the center of the MOSFET gates.
The asymmetrical clipping is not a problem for a headphone amp because your ears will bleed long before the MOSFETs clip. If you are going to use this design for driving speakers, you should tie the op-amp output the center of the MOSFET gates using a pair of 450 ohm resistors (figure 3). Then the amp will clip at 15V p-p. I listened to this amp driving my Quad ESL’s and it’s quite promising. This amp sounds very good. If you own a set of Grado headphones and are driving them with an op-amp based headphone amp, I believe that the buffering the op-amp with a pair of MOSFET’s (per channel) will result in much improved sound quality. I haven’t heard a more transparent sounding headphone amp available at a reasonable cost.
The circuit board layout and population guide are shown in figures 4 and 5. The schematic only shows parts for one channel. On the circuit board, parts for the other channel are offset by 50 (e.g., R10 for channel 1 becomes R60 for channel 2). All of the parts, including the toroidal transformer, are available from Digi-Key Electronics and Radio Shack.
The International Rectifier MOSFETs that I used originally don’t always work. I’m now using Harris MOSFETs: RFP15N05 for the N channel and RFP30P05 for the P channel. But a wide array of the MOSFETs from the same Harris series will work as well. You can use others as long as you choose well-matched complimentary pairs with decent power handling and similar specs.
R5 and R55 are needed only if no potentiometer is used (e.g., it’s driven from a pre-amp or the variable outputs of a CD player). C16 and C66 provide feedback compensation for the op-amp and improve the amplifier’s stability. They are not necessary with the parts I’ve specified, but don’t hurt anything. If left in place, they allow the dual op-amps to be switched out to try different ones without fear of stability problems. R12 and R62 are not necessary when using MOSFETs, but are necessary if the output devices will be bipolar transistors.
9/1/1998: The schematics in figures 1 and 3 incorrectly labeled pin 7 of the opamps as V+. Corrected.
10/2/1998: Reversed R7 and R11 in figure 1.
3/1/1999: Corrected R7a, R7b labels in figure 3.
9/22/1999: Updated figure 1 to version 1.1: R8/R9 corrected to 1W; R5 now listed as optional; added R12; added C16 as optional; changed Q1/Q2 to Harris MOSFET types. Circuit board layouts (figures 4 and 5) also updated to version 1.1.
11/23/1999: Added more information on parts selection.
12/13/1999: This idea came from Jeffrey Baker: remove the diodes in the original MOSFET bias network and replace them with 4 LEDs (now D3 – D6). He and several other DIYers reported that the original MOSFET bias components resulted in excessively high idling currents and distorted sound. With the new bias network, the MOSFET gates are held 8V apart, and the idling current I(d) is between 40-80 mA. Sheldon Stokes has approved this modification. However, the PC board layout still reflects the old design. Baker writes:
I finally finished [the SDS Labs headphone amplifier] and am quite happy with the results. The buzz is gone and there are no other oscillations that I can measure on my admittedly lame scope. As a bonus, the amp emits an eerie green glow. Subjectively, it sounds better than every other headphone amp I have on hand. These are the modifications I made to the design:
– The power supply (up to and including the regulators) is in an external chassis.
– D3, D4, D5, and D6 have been removed and replaced with 4 2V LEDs in series.
– Rp was replaced with a series-type stepped attenuator, one for each channel.
– Fairchild N-Channel MOSFETs and Phillips ECG P-Channel MOSFETs were used.
Also added gate resistors (Rg) to both MOSFETs in figures 2 and 3. With some brands of MOSFETs (such as those by International Rectifier), the gate resistors help to keep the amplifier from oscillating. They should be installed as close to the MOSFET gate leads as possible.
7/7/2000: The Harris RFP15N05 MOSFET has been discontinued. The Intersil HUF75309P3 N-channel MOSFET is reported to be a good substitute.
7/14/2000: Because different brands of MOSFETs can have different biasing requirements, the author has incorporated an adjustable bias scheme (Vbe multiplier) in the amplifier’s output stage, so that DIYers can can set the MOSFET idle current via a trimmer pot. The amplifier schematic in figure 1 has the MOSFET gates biased at 8V. The Vbe multiplier replaces D3 to D6. With the 20K pot shown, the multiplier can provide a bias voltage range of about 0 to 12V. A 15K pot would have a range of about 0 to 10V. The NPN transistor can be any general purpose type: 2N2222, 2N3904, 2N4401, etc. The author writes:
- A 15K pot (or 20K) will give a max of 16 diode drops. For 8 volts across the FETs (just conducting), you’ll need 13 diode drops. To set the bias, I’d turn the pot so that it is as low a resistance as possible. Then with a voltmeter across one of the two 10 ohm resistors hooked to the sources of the FET’s, I’d slowly turn the pot up until I read 0.6 – 0.8 volts across the resistor for an idling current of 60 – 80mA. I’m planning to build another amp for my office, and I am going to try out different FETs. I’ll also add the Vbe multiplier to the PC layout.
5/22/2002: Sijosae (from the forums) built this mini version of the SDS amplifier, which he uses with his Philips HP-890 headphones. The wood frame of the chassis has been covered with a natural wood veneer and varnished. The power supply is a separate 24V commercial unit. The SDS amp is also powerful enough to drive very efficient DIY speakers, also shown below.
The speakers are 8″ full-range drivers mounted on large 5-gallon plastic water bottles. Sijosae used a jig-saw to cut off the neck (to mount the speaker) and the bottom of each bottle. The interior of the speaker is stuffed with a roll of plastic foam wrapped in felt. The back of the speaker remains open.
The speakers are located on top of a bookshelf. A small brick placed inside each bottle adds mass to the enclosure and keeps the speakers from rolling away. The speakers cables are made from LAN cables and electrical clips. He says “The sound of the speakers is good. But with my system, headphone sound is much better. I think all DIYers love their own work. Therefore in spite of many flaws in my speaker, I love sound of mine.”
5/22/2002: Algar_emi (from the forums) completed this SDS headphone amplifier. At first, the amp oscillated. He had omitted the 100-ohm gate resistors on the MOSFETs. When he put them in, the oscillation went away. Here are the actual performance measurements of his amp:
Output DC offset: +1mv and -2mv DC
Noise (Max volume, no input) (Vrms): 0.2mV
S/N (Max volume): 64dB
Crosstalk (Max Volume in 32ohms): 68dB
Gain Voltage: x2
Zin: 100K ohms
Max Vinput (for load 32 ohms): 2.88Vrms or 5.8dBm in 600 ohms
Pwr Out (load 32 ohms, max volume, Vin max 2.88V): 1W
THD: lower than 0.05%
Heatsink Temp at full power: 42 deg C
Supplies ripple noise (Full load): <2mV RMS