AMB β18 Stereo Headphone Amplifier

March 29, 2004

1280x960, 146KB
1280x960, 169KB
1280x960, 194KB
1280x960, 151KB
1280x960, 172KB
960x1280, 172KB
960x1280, 175KB
640x480, 41KB


After acquiring the excellent Sennheiser HD600 headphones, I realized that the headphone outputs found on most preamps, cassette decks and CD players simply couldn't do these headphones justice. After auditioning the HD600 via a $1600 Headroom Max, I became motivated to design and build a high-quality headphone amplifier that should not only extract the best out of the headphones, but has the looks to match too.

In keeping with my minimalist philosophy in audio, I decided that the only user controls will be a volume control knob and a power switch. Unlike the Headroom Max, there is no crossfeed or filter feature. There are two sets of input RCA terminals, internally wired in parallel. This is to allow the headphone amp to be inserted into the tape monitor loop of a preamp, or in between a preamp and a power amp without the need for Y-cables or adapters. Two output jacks allow simultaneous connection of two pairs of headphones.

I originally planned to use all-discrete components, but decided instead to try some of the recent crop of high performance IC opamps. Op amps were at one time considered lowly creatures not worthy in a high-end design (and many still subscribe to that belief). However, there are now new devices that offer truly remarkable performance that give all-discrete circuitry a run for the money. Combine that performance with reduced parts count, PC board real-estate savings and lower cost, these are solid reasons for me to go this route. In addition, due to standardized IC pin-outs, I thought it would be interesting to substitute different opamps to hear and measure their differences.

Most opamps do not work well driving headphone directly. The output current capability and output impedance characteristics of the better performing opamps are such that they tend to operate near or beyond the "edge" while driving reactive loads like headphones, particlarly those with low impedance. This is the reason why the headphone outputs on preamps, CD players, tape decks, etc., do not sound good. To protect the opamp from overcurrent distortion and damage, most of these designs use a resistor in series at their opamp outputs, reducing damping factor, introduce nonlinearlities in frequency response, and limiting the peak power delivered to the phones and consequently affecting the dynamic range. There are "power opamps" with high current output capability but none of them are known for sonic excellence, and they typically come in non-standard packages, making device swapping and comparison unfeasible.

I decided to use the opamp merely as a voltage gain stage, and use discrete power transistors in a push-pull configuration to provide the needed output current drive capability. This sort of output buffering topology is mentioned in Walt Jung's IC Op Amp Cookbook as well as his Audio IC Op Amp Applications book, and are used in various real world designs. However, rather than the usual bipolar power transistors, I decided to use power MOSFETs, because I have been a longtime fan of these as amplifier output devices. Ever since my good experiences with the Hafler power amplifiers from the late 1970s, I have been smitten by their excellent reliability, speed, and negative thermal coefficient characteristics. Moreover, they are noted by an absence of the harsh "solid state sound".

After sketching out a schematic diagram of what I wanted to build, and searching for parts information on the web, I came across the Headwize forums and an almost identical circuit to what I designed from Sheldon Stokes (a.k.a. SDS Labs). Sheldon even offers etched and pre-drilled PC boards for sale. I decided to use version 1.2 of Sheldon's board, which made my job easier. This board also includes the power supply section, simplifying matters even further. My circuit was largely identical to the SDS Labs design, except that I use somewhat different component values and parts. I have since made some parts changes to make it more like the SDS design, most of these are to make them fit the SDS board. The following paragraphs describes my amp in more detail.

The beauty of this design is in its simplicity and relatively low parts count, yet with the right choice of parts, it is capable of stunning performance (as amply illustrated in the actual measured specifications listed below). To this end, I chose quality parts with the goal of maximizing sound performance, while not going too far into the realm of costly "boutique" components that would simply make the amp too expensive.

The power supply features a Amveco toroidal power transformer, International Rectifier 4GBU06 bridge rectifier and "snubber" capacitors. It is regulated with Fairchild LM317T and LM337T regulator ICs with adjustable voltage using Bourns 25-turn cermet trimpots. These are adjusted to supply ±12VDC. The original SDS design was fixed at ±15V, but I needed to reduce this to accommodate the Analog Devices opamps which have lower supply voltage limits. The voltage regulator ICs feature capacitor decoupling on the adjust pins for improved ripple rejection, as well as provide a smooth power-up without transients. There is 37,600µF worth of Nichicon and ELNA low ESR electrolytic capacitors in the power supply section, before and after the voltage regulators. Two of these capacitors are mounted on the chassis rather than onboard. There is more current reserve here than many power amplifiers in the 100-watt per channel class!

The IC opamp is a dual unit, and is socketed to be easily swappable. So far I have tested with the following list. The first four are the modern preferred units, the rest were older generation models that I also tested for comparison purposes. I will try other opamps in the future and update this page. See the data sheets of these opamps by clicking on the part number links.

  1. Analog Devices AD8066AR (FastFET™, 180V/µS SR, 115MHz GBP)
  2. Analog Devices AD8620AR (JFET, 60V/µS SR, 25MHz GBP)
  3. Texas Instruments/Burr-Brown OPA2132PA (FET, 20V/µS SR, 8MHz GBP)
  4. Texas Instruments/Burr-Brown OPA2134PA (FET, 20V/µS SR, 8MHz GBP)
  5. National Semiconductor LM6172IN (bipolar, 3000V/µS SR, 100MHz GBP)
  6. Signetics/Texas Instruments NE5534AP (Bipolar, 9V/µS SR, 10MHz GBP)
  7. Texas Instruments TL072CP (JFET, 13V/µS SR, 3MHz GBP)
  8. NTE 858M (JFET, 13V/µS SR, 4MHz GBP)
  9. NTE 889M (JFET, 6V/µS SR, 2MHz GBP)
  10. NTE 778A (Bipolar, 0.5V/µS SR)

The two Analog Devices opamps have SOIC-8 (surface mount) package. They are each mounted to a SOIC-8 to DIP-8 adapter.

One concern about using opamp ICs is the power supply voltage limits, particularly the ±12V for the Analog Devices units. This restricts the maximum available output voltage swing, which may be inadequate with some high impedance headphones. However, upon testing with the HD600 (which is a 300 ohm high impedance model), I found that this amplifier provided enough voltage output to achieve very loud playback levels without clipping. If using headphones with much higher impedance, or those that are particularly inefficient, I recommend using only the opamps that can handle ±15V or more of supply voltage, and adjust the power supply regulators accordingly to increase this amplifier's output voltage capability.

An important modification from the "stock" SDS Labs design is the elimination of the R12/R62 resistors, and the substitution of R10/R60 with 4.7mA CRDs (current regulating diodes; the Vishay-Siliconix CR470). This is simple and does not require altering the PC board layout, but should only be done if the output devices are MOSFETs. It is not appropriate with bipolar output transistors. The modification has several important benefits:

  • The opamp is now biased to source 4.7mA of current at its output at all times, causing it to operate in class A. This is regulated by the CRD, and is the entire amount of current flowing through the Vbe multiplier. (In the original design, the opamp may sink a small amount of current due to the output being asymmetrically biased toward the positive supply rail, but that amount of current varies with power supply voltage and the bias potentiometer setting, as well as the relative tolerances of the resistors tied to the two supply rails. It is rather ineffective at keeping the opamp operating in class A. With this circuit change, the current is is now independent of these variables).
  • Sourcing current out of the opamp is usually better than sinking, because it causes the NPN transistor in the opamp's complementary push-pull output stage to be "on" rather than the PNP transistor. The NPN is usually more linear. (If you want to make a similar modification to your SDS Labs amp, but want the opamp to sink current, eliminate R10/R60 and substitute R12/R62 with the CRD).
  • Having a steady and regulated amount of current flowing through the output MOSFET biasing section improves the circuit's ability to drive the MOSFETs. This is because the MOSFETs' gates are capacitive in nature.
  • The power supply rejection ratio (PSRR) of the output buffer is significantly improved as a result of not having the passive resistors tied directly to the supply rails. This is perhaps the most important benefit.

An additional NP0 multilayer ceramic decoupling capacitor is added and wired directly across the supply pads of the opamp, which helps amplifier stability with the ultra wideband opamps.

Overall voltage gain is set to slightly less than 5 (13.6dB), different than the original SDS design which has a gain of 2 (6dB). With some high impedance headphones the original gain is not sufficient to achieve adequate listening volumes with certain sources. At this gain, comfortable listening levels can be achieved with most line input devices on the Sennheiser HD600, while the volume control is near center, to provide a good range of control.

The original Harris RFP15N05/RFP15P05 power MOSFETs specified have been discontinued. Some SDS builders used the Hitachi/Renesas 2SK213/2SJ76 series, but these have incompatible pin-out assignments to the Harris and International Rectifier devices. While the 2SK213/2SJ76 have low input capacitance (which is desirable), they have low transconductance and have a rather low maximum drain current rating (0.5A). Moreover these devices are not consistently stocked and available from vendors in the USA.

I chose to use the IRFZ24N (N-channel) and IRF9Z34N (P-channel) power MOSFETs from International Rectifier. These are 5th-generation HEXFET™ devices for linear use, excellent for this application, and are available from Digi-Key and Newark Electronics.

The MOSFETs are biased to run with 80mA of quiescent current via the adjustable Vbe multiplier based on a Motorola 2N5210 small-signal NPN transistor, and a 25-turn Bourns cermet trimpot. The MOSFETs and voltage regulator ICs are cooled by AAVID/Thermalloy extrusion heat sinks. The MOSFETs each dissipate about 1 watt while the voltage regulators slightly less than 2 watts, making the heat sinks warm to the touch, but not too hot. The heavy bias causes the output stage to operate in class A at any reasonable volume levels (even into low impedance headphones). This operates the output MOSFETs in their "sweet spot", without any crossover distortion.

All resistors are precision metal film units, most of them are Vishay-Dale RN55 series (mil-spec), which have very low noise and superb temperature stability. HF compensation are via NP0 multilayer ceramic capacitors for their excellent high frequency performance, and power supply decoupling is redundantly done with solid tantalum, metallized polypropylene, and NP0 multilayer ceramic capacitors from Vishay and Kemet. This assures stable operation of the voltage regulator ICs and a low power supply rail impedance over a wide frequency band. The amplifier is direct coupled at the input and output, as well as in the negative feedback loop.

The inputs feature Vampire RCA jacks, the outputs are via Neutrik locking phone jacks. These all have gold contacts. The volume control is a high quality stereo potentiometer from Noble. The IEC AC power entry and switch module is from Corcom, and there is an obligatory blue LED power-on indicator on the front panel.

The chassis/enclosure is home made. The base is aluminum, and is the foundation on which all other pieces are attached. The side walls are black glossy acrylic, the front and rear panels are clear acrylic, painted to have matte black surfaces but retains "frosted", translucent edges. The top cover is clear acrylic with a cutout to allow the heatsinks to vent. Virtually all fasteners are socket head machine screws. The enclosure was much more time-consuming and took far more effort and expense to build than the electronics. I think the result speaks for itself.

Attention is paid to the internal wiring to ensure that there is no ground loop. The aluminum base plate is connected to the signal ground via a single-point to serve as a partial shield. All wiring are twisted tightly into pairs where appropriate to improve noise induction immunity, and relatively heavy gauge wiring is used (18 gauge stranded copper for power and output, 20 gauge stranded copper for line level signals) to ensure low impedance. I did not use silver/teflon or other fancy wiring, because I do not believe that the short runs within the amp would make a measurable or audible difference. I added inline Molex KK series modular plugs to all lines leading to the main circuit board, to make board removal a simple "unscrew and unplug" operation, without the need to de-solder anything. Heat-shrink tubing is used to cover all soldered connections at the front and rear panel jacks, potentiometers, etc.

In essence, what I built is a small power amplifier specifically for use with headphones. As such, it has very low output impedance and ample output current drive capability, and is suitable for use with low impedance and high impedance headphones.

Even though I tested with some opamps with bipolar inputs, they should not be used with this amplifier due to the purely direct-coupled design. The high input bias current of the bipolar opamps causes unacceptably high output DC offset, and it varies considerably as the volume control setting is changed. With the FET opamps, the output DC offset is less than a couple of millivolts on each channel.

The broadband noise floor is also very low, however there is some AC mains noise, likely to be magnetically-induced due to the proximity of the power transformer to sensitive circuits on the PCB layout. This is one area of the amp that could use some improvement.

The compensation capacitor values are "tuned" here for best performance with the ultra-fast AD8066AR. With this opamp, the resultant square wave step response shows no ringing, and very little rounding or slewing of the rising and falling edges (as indicated by the excellent rise time and slew rate measurements, see below). With certain other opamps there is a very small but well damped overshoot. This can be tuned-out by increasing the compensation capacitor values, with a resultant drop in slew rate, increased rise time and reduced frequency response. I opted to leave things as is because there is no apparent ill effect to sound quality, because no real world music program material would have such fast transients.

I found that the AD8620AR opamp provides the best sound, even though the AD8066AR measures slightly better in terms of speed and bandwidth. The OPA2132PA and OPA2134PA perform well, but with some perceptible blurring of definition.

I personally did extensive listening comparisons between this amp and the Headroom Max and the PPA. Moreover, at a recent local Head-Fi gathering, A-B style listening comparisons to a fully-loaded PPA, Meier Audio Corda HA-2, Ray Samuels Audio Emmeline HR-2, and other headphone amplifiers by attendees produced very positive comments, It validated that this amplifier could stand amongst the better headphone amplifiers.

Actual measured specifications

  • Measured with the AD8066AR opamp except where noted.
  • These specifications do not represent that of a "stock" SDS Labs headphone amplifier, due to various modifications, changes to components and opamp choice.
  • Equipment used:
    • Wavetek 188 4MHz sweep function generator
    • Protek 6510 100MHz dual trace oscilloscope
    • Fluke 95 50MHz digital dual trace ScopeMeter
    • Beckman HD100 digital multimeter
    • e-Cube CF-968L mini PC (2.4GHz Celeron/1GB RAM) with M-Audio "Transit" USB Mobile Audio Upgrade, running RightMark Audio Analyzer (RMAA) v5.3 and other FFT software
    • 33 ohm and 220 ohm 5W dummy load resistors
    • Denon Audio Technical CD C39-7147
    • NAD C520 CD player
    • Sennheiser HD600 and HD50 headphones

Voltage gain x4.8 (13.6dB), volume at maximum
Input sensitivity 208mV input for 1V output, volume at maximum
Input impedance 47.6K ohms, volume at maximum
Output impedance Less than 0.2 ohm, 20Hz-20KHz
Maximum output voltage Per channel, 1KHz sine wave, prior to onset of clipping:

220 ohm load
15.3Vp-p (5.4Vrms) (AD8066AR)
14.7Vp-p (5.2Vrms) (NE5532AP)
14.1Vp-p (5.0Vrms) (OPA2134PA)
14.1Vp-p (5.0Vrms) (AD8620AR)
13.3Vp-p (4.7Vrms) (OPA2132PA)
13.3Vp-p (4.7Vrms) (LM6172IN)
13.0Vp-p (4.6Vrms) (NTE889M)
12.4Vp-p (4.4Vrms) (TL072CP)
12.4Vp-p (4.4Vrms) (NTE858M)
12.4Vp-p (4.4Vrms) (NTE778A)

33 ohm load
12.7Vp-p (4.5Vrms) (AD8066AR)
12.4Vp-p (4.4Vrms) (AD8620AR)
11.6Vp-p (4.1Vrms) (OPA2134PA)
11.6Vp-p (4.1Vrms) (OPA2132PA)
11.6Vp-p (4.1Vrms) (NTE889M)
11.6Vp-p (4.1Vrms) (NE5532AP)
11.0Vp-p (3.9Vrms) (LM6172IN)
10.7Vp-p (3.8Vrms) (TL072CP)
10.7Vp-p (3.8Vrms) (NTE858M)
10.7Vp-p (3.8Vrms) (NTE778A)
Power output 600mW continuous power per channel into 33 ohms
Frequency response +0dB, -3dB, ref. 1Vrms output at 1KHz:
0Hz - 1.4MHz (AD8066AR)
0Hz - 1.2MHz (LM6172IN)
0Hz - 1.2MHz (OPA2134PA)
0Hz - 1.1MHz (AD8620AR)
0Hz - 900KHz (OPA2132PA)
0Hz - 630KHz (NE5532AP)
0Hz - 540KHz (NTE858M)
0Hz - 450KHz (TL072CP)
0Hz - 270KHz (NTE889M)
0Hz - 260KHz (NTE778A)
Rise time 10KHz square wave, at maximum output voltage, 10% to 90%:
0.25µS (AD8066AR)
0.4µS (AD8620AR)
0.4µS (LM6172IN)
0.5µS (OPA2132PA)
0.5µS (OPA2134PA)
1.0µS (NTE858M)
1.0µS (TL072CP)
2.0µS (NE5532AP)
5.0µS (NTE889M)
8.0µS (NTE778A)
Slew rate 10KHz square wave, at maximum output voltage:
77V/µS (AD8066AR)
45V/µS (AD8620AR)
45V/µS (LM6172IN)
33V/µS (OPA2132PA)
33V/µS (OPA2134PA)
16V/µS (NTE858M)
16V/µS (TL072CP)
8V/µS (NE5532AP)
3.5V/µS (NTE889M)
1.9V/µS (NTE778A)
Total Harmonic Distortion 0.0013%, 1KHz (RMAA)
Intermodulation Distortion 0.013%, 60Hz/7KHz 4:1 (RMAA)
Noise level -90.5dBA (RMAA)
Stereo crosstalk -76.5dB (RMAA)

RMAA Test Results

RightMark Audio Analyzer software, running on a Toshiba 2.8GHz Celeron laptop computer via an M-Audio Transit USB mobile interface running in 16-bit, 44KHz mode.

This test provides data and graphs of frequency response, noise, dynamic range, total harmonic distortion, intermodulation distortion and stereo crosstalk performance. There is no difference in the RMAA results when the amplifier output is unloaded, or loaded with 330Ω. The noise floor shows the induced AC hum components, fortunately they are mostly -85dB or lower in amplitude. This makes the noise inaudible on most except the most sensitive low-impedance headphones.

Oscilloscope waveforms

The following shows the waveform response of the β18 amplifier with the AD8620 opamp. In all graphs except the Lissajous waveform, the top trace is the input and the bottom is the output. These were measured using a Wavetek 188 4MHz sweep function generator and a Protek 6510 100MHz oscilloscope.

The square wave graphs show that there is minimum slewing and ringing at the leading and falling edges. The 100KHz sine, triangle, and Lissajous graphs also show very small amount of phase shift between the input and output. Within the 20Hz to 20KHz audio band, there is no measurable phase shift.

1KHz square wave

10KHz square wave

100KHz square wave

100KHz sine wave

100KHz triangle wave

100KHz Lissajous

Schematic Diagram

This reflects the actual components and values used, and can be used to compare and contrast that of the original SDS Labs design.

Epilogue and Futures

This project had sparked an idea that I could improve upon the design to make an even better amplifier. I have experimented with some designs and one of these, called the , has been presented to the DIY community. The M³ is inspired by the β18 as well as the PPA amplifier. The M³ web site has all the details. See also this Head-Fi forum thread for the original announcement and further information about the M³.

M³ is a joint project of Morsel and myself. As some of you might know, Morsel is well known in the headphone amplifier scene. Her key roles in the PPA, Pimeta and Meta42 headphone amp projects add a new dimension of expertise to the fold.