Inspired by Douglas Self’s book on audio electronics (‘Small Signal Audio Design’) and his power amplifier design consisting of a massively parallel array of NE5532 op-amps, I set out to design a headphone amplifier – dubbed the ‘NE-XT‘.
The amplifier only utilises NE5532 op-amps (44 in total!) and is designed to run from a single-rail DC power supply. It is also fairly inexpensive, coming to about 20 EUR per amplifier (including PCB manufacturing, components, and shipping). Additional costs arise for the enclosure, volume potentiometer, as well as for any jacks and switches you wish to use.
The resulting design is low-noise and low-distortion, easily capable of driving anything from 8 Ohm earbuds to 600 Ohm studio-grade headphones with plenty of gain reserves.
The technical headphone amplifier specifications are as follows:
|Input Impedance||10 kOhms|
|Output Impedance||0.08 Ohms (+ 10 Ohm short-circuit current limiting resistor)|
|Bandwidth (-0.1dB)||20 Hz to 20 kHz|
|Gain||15 dB max.|
|Equivalent Input Noise (EIN)||-120 dBu|
|THD (1 kHz, 150 Ohm Load @ 100mW (average))||0.001%|
|Power Consumption (Idle)||2.5 W|
The headphone amplifier can be divided into four main sections: power supply, input buffer, voltage amplification stage (VAS), and output driver. All sections utilise NE5532 low-noise op-amps in various configurations, the workings of which are detailed below.
The golden rules of this design were: paralleled op-amp sections to reduce noise and enhance driving capability, as well as low-impedance design to reduce the Johnson noise generated by each resistor.
All following schematics will show only one channel. For stereo, of course, there needs to be a duplicate of each section for each channel.
The power supply consists of reverse polarity protection (via a P-channel MOSFET to give minimum Rds(on)), a fuse, as well as an LM317 regulator which gives a fixed output voltage (the main rail voltage of the headphone amplifier – which of course can be increased) for a large range of input voltages. Various decoupling capacitors (electrolytic and ceramic, of varying sizes) are included for regulator stability and general power supply decoupling.
Since the amplifier is designed to run off of a single DC rail and op-amps generally require two rails (+ve and -ve) to function, the op-amp inputs need to be biased to half the supply voltage to ensure maximum voltage swing capabilities. The op amp supply pins are thus tied to VCC and ground.
Biasing is achieved via the use of op-amp ‘bias generators’ shown below. These allow filtering of the bias voltages and are also relatively unaffected by loading due the paralleled unity-gain voltage followers, compared to simple biasing via voltage dividers at each op-amp input. The paralleled are summed via two 10 Ohm resistors.
The input buffer provides a fairly high impedance to the source connected to the input of the headphone amplifier and prevents any loading effects for the following volume control stage.
R1 and C1 form a low-pass filter, preventing any RF from getting into the system and being (audibly) demodulated. The cut-off frequency is set to 1.6 MHz, however the source impedance will further reduce this. Assuming a maximum source impedance of 1kOhm, this will be at 145 kHz, which is still well out of the audio band.
R2, R3, and C2 form a high-pass filter and set the lower -3dB frequency of the headphone amplifier, as well as the input impedance at AC frequencies (roughly 10kOhm). R3 also biases the op-amps inputs at half the supply voltage for maximum possible voltage swing. C2 prevents any DC bias voltage appearing at the source, as well as preventing any DC offset at the source appearing at the op-amp inputs.
The use of paralleled op-amp sections is a re-occurring theme in this design and helps (rather counter-intuitively!) reduce the noise and increases the load driving capabilities. The outputs of both op-amps are combined via the 10 Ohm resistors.
Voltage Amplification Stage (VAS)
The voltage amplification stage or volume control stage provides all of the voltage gain in the amplifier. The voltage gain available in this headphone amplifier can be varied anywhere from effectively completely muted to approximately 15dB (5.6x linear), which is plenty for most applications.
Ideally, the control law of the volume stage should follow a logarithmic curve, as this best matches the way we hear. A simple volume control could simply be a ‘log’ potentiometer, where first terminal is connected to the input and the wiper to the output. However, these apparently logarithmic potentiometers are not truly logarithmic at all, but instead approximate a logarithmic function by using two linearly changing track resistances of differing slope.
Additionally, using a passive potentiometer as the sole volume control would mean all the 15dBs of gain would have to be applied AT ALL TIMES before reaching this potentiometer, as the potentiometer would act to decrease the signal level. Clearly, a better solution must be sought out…
In comes the Baxandall active volume stage, named after the late Peter Baxandall. Peter Baxandall is probably most famous for his active tone control.
A simplified, block-diagram form of the volume control is shown here.
It consists of a LINEAR taper potentiometer of total resistance R, a unity gain (and high input-impedance) buffer, and an inverting amplifier of gain k. In this form it is a bit hard to see how the volume control functions, so let’s redraw it in a more convenient form – that of a potential divider.
I have omitted the unity gain buffer and assumed it is part of the inverting amplifier, as its only function is to provide a high input impedance.
The far left diagram shows the potentiometer split into two, as it simply is a potential divider with two varying resistors, with the ‘potentiometer setting’ indicated as alpha, which can vary from 0 to 1. The inverting amplifier is now in parallel with the ‘lower’ resistor. There are now two possible ‘extreme’ cases to consider.
Middle diagram: If the potentiometer is turned completely to one side, alpha is zero and thus the top resistor is R and the bottom resistor is effectively a short. Since the short is in parallel with the high input impedance amplifier, the short dominates and – remembering basic potential divider theory – there is no voltage drop across the amplifier (all of the input voltage is effectively dropped across R). No voltage at the input terminal of the amplifier means the signal won’t be amplified and Vout will be zero. This is the case of ‘infinite’ attenuation.
Far right diagram: However, if the potentiometer is turned completely to the other side, alpha is 1 and the top resistor is now a short, with the bottom resistor being R. Now all of the input signal voltage is apparent at the input terminal of the inverting amplifier! And the output voltage will simply be Vout = -k * Vin, i.e. the maximum possible amplification.
By going through the math, it turns out that the relationship between voltage gain and potentiometer setting (alpha) is:
Gain = – (alpha * k) / (1 + k * (1 – alpha)).
Importantly, the gain is NOT a function of the total potentiometer resistance. Additionally, note that the output is inverted.
This gives an approximate but good fit to a logarithmic response – by using a linear-taper potentiometer. Furthermore, the noise level at the output is not much reduced compared to amplifying the signal by the maximum gain, and then attenuating via simple potentiometer.
The complete volume control schematic is shown below, including decouplings capacitors, biasing resistors, and so forth. I have chosen to follow the volume control by a unity gain inverting amplifier, to ensure that the output is in phase with the input. This is merely a distinct preference and completely optional.
Capacitors C10, C12, and C14 are compensation capacitors to ensure HF stability. They essentially form a low-pass filter and thus limit the high-frequency response. The maximum gain is set by R14/R12 (and R18/R16). As is usual in this design, op amp sections are paralleled and their outputs summed to improve noise performance.
Note also that there is no coupling capacitor connected to the third pin of the potentiometer. This is due to the fact, that C7 and C8 already perform AC coupling duties (also to ensure no crackling sound when rotation the wiper of the potentiometer) and thus no DC current can flow through the potentiometer.
At last, we arrive at the output stage of the amplifier – to me the quirkiest part of the design. The circuit diagram is shown below.
At the top left, unity gain buffers are shown which act as pre-drivers to feed the output drivers that follow and provide an interface between the volume control stage and the output stage.
The output drivers are 12 paralleled unity gain voltage followers which have their outputs summed by 1 Ohm current-sharing resistors. A single NE5532 op-amp can quite comfortably drive 13V (peak) into a 600 Ohm load with minimal distortion. Headphone amplifier voltages are comparatively much lower and combined with the fact that we have 12 op-amps driving one channel means the distortion will be vanishingly low for nearly any expected volume level.
Finally, the output is fed through a current-limiting resistor, which limits the short-circuit current, should the output be tied to ground. Then, a very large, electrolytic coupling capacitor to block any DC voltage from reaching the headphones. The capacitor is this large at it also acts as a high-pass filter in combination with the headphone impedance. For 250 Ohm headphones (quite typical for studio headphones), the -3dB cut-off is at 0.3Hz – well below the audio band and thus will not cause significant distortion. The final ‘bleed’ resistor to ground ensures a (dis-)charge path for the capacitor.
PCB Manufacturing and Assembly
The PCB layout and routing was done in KiCAD and resulted in a board with dimensions of 80mm by 60mm, so fairly compact given the number of op-amps.
JLCPCB manufactured the board and assembled the ICs, diodes, and some capacitors – which all look great. You can see a top-down view of the assembled board below. The ‘dodgy’-looking soldering of (mainly) the resistors was my work (as well as all through-hole components), as JLC unfortunately don’t stock thin-film resistors and only could have supplied thick-film ones. Unfortunately, thick-film resistors aren’t great for audio, since they have a significant voltage coefficient. I thus chose to leave the resistors for me to assemble myself. Additionally, any capacitors with signal voltages running through them should be of C0G type, since many other ceramic capacitor dielectrics have capacitance that varies with frequency – thus producing distortion.
And the rear view of the board:
I bought this very inexpensive Teko enclosure (for roughly 4 EUR). In hindsight, I probably should have spent more money on it, as it had several gaps at the enclosure boundaries. However, I went along with it, drilled some holes, and mounted the board, as well as connectors, switches, and LEDs.
This is the ‘raw’ enclosure.
I typically use draw.io to draw up the drill locations for my front and rear panels. Then, I print them out and try to align them with my enclosure before drilling.
Once drilled, I mount the PCB using PCB stand-offs (to avoid contact between PCB pads and the case), attach the various connectors, and wire everything together. Not the neatest job in the world – but thankfully this case has a lid..
And the rear view…Now, the final product is sitting on my desk, works surprisingly well, and has replaced the headphone amp I had been using for the past couple of years! (Notice the cheap enclosure and the gaps between the panels…)