There are many headphones on the market which are difficult to drive from a standard output of a computer or MP3 player. In some devices (2nd gen iPod I’m looking at you) it is impossible to drive any headphones with decent bass response as the output stage is lacking in its ability to supply enough current. A headphone amplifier solves these problems by providing a nice and easy load for the source to drive.
My version of Kevin Gilmore’s Discrete Amplifier (KGDA) has evolved over the years. What started as a simple powersupply and amplifier has turned into a fully feature balanced amplifier with multiple inputs, complete with headphone output protection, digital potentiometers and a remote control. Each part of the circuit is explained in detail below.
The obvious place to start is the amplifier. The basis for the design is a pure class-A amplifier that is capable of driving 0.5W into 32Ω (the impedance of Grado SR225 headphones). The design requirements were:
- No capacitors or transformers in the signal path.
- Low open loop gain. (i.e. not an OPAMP) High gain means high amounts of feedback to make the amplifier usable, and thus also high amounts of transient inter-modulation. But hey at least harmonic distortion is low.
- No servo loops in the feedback loop. (i.e. no OPAMPs injecting noise into the -ve input of the amplifier).
With that in mind the schematic is shown above. The first stage is a dual FET differential and balanced front-end. This stage has a voltage gain of around 50 and biases the following stage into class-A.
The second stage is a standard push-pull class-A design with several emitter followers in parallel to provide loads of current gain to drive the most demanding loads. Each transistor and FET in the circuit is specifically designed for audio use.
The servo and current source for the FETs is interesting. Typical servos integrate the output signal and feedback the DC error into the -ve input of the amplifier, putting them directly in the feedback loop. In this design the servo instead changes the bias of the transistor current source for each FET. This works as the LED used to bias the current source is non-linear and will exhibit some current variance with voltage. A zener or other stable voltage for the current source would negate the already small effect the servo can have on the circuit.
An overdesigned amp needs an overdesigned powersupply. The powersupply used actually differs from the original circuit as it is the supply used in the KGDA Dynahi rather than the Dynalo powersupply shown here. The differences are minor and increase the current available for the balanced amps (more on that later). The changes are:
- Caps around each leg of the bridge rectifier.
- LM338s instead of LM317 / LM337.
- OPA548 instead of OPA541.
The circuit works by first creating a classic 24VDC regulated by the LM317/337 regulators. REF02 is a precision low noise 5V source. The OPA548 uses the REF02 as a reference voltage with a gain of 3.28 to give a 16.4V output. The second OPA548 takes the positive as a reference and is setup with a gain of -1 to give -16.4V
The rails in this powersupply thus always track each other. If something brings down the positive rail, the negative rail will match resulting in the disturbance not being passed into the main amplifier, not that it is so critical for balanced amps (more on that later).
A single ended amplifier drives the +ve terminal of a speaker while holding the -ve terminal at zero volts. A balanced amplifier on the other hand will drive the -ve terminal with the opposite signal of the positive terminal. This has several advantages:
- Double the voltage supplied to the speaker.
- All common mode signals are rejected.
- No noise over long cables.
- Current draw from +ve and -ve powersupply rails is always equal preventing an imbalance.
There are a few downsides:
- Headphones necessitate a 4-wire connector as the left and right drivers share a common point on a 3 pin connector.
- Double the current draw on the power supply.
- 4 pole volume control in a non-bridged amplifier.
Even though the KDGA provides bridging points for the amplifier as all my sources and pre-amps are balanced I instead opted to simply duplicate all the circuitry. The result is a 4 channel amplifier with a 4 gang volume control amplifying 4 different signals from the source.
For single ended sources I’ve added a small board with a BurrBrown DRV134. This component is a balanced instrumentation line driver and consists of two internal precision matched OPAMPs to convert a single ended signal to a balanced signal. It has a negative effect on the audio quality but it won’t get used often.
A balanced non-bridged amplifier creates a special problem for volume control. 4-gang potentiometers are expensive. A stepped resistor ladder is also quite expensive, especially when you only have 16 or 24 steps. More steps become prohibitively large and again expensive if precision resistors are used. There are a few opamp / gain based solutions as well but each add some additional distortion to the signal. Enter digital potentiometers.
Maxim produces the DS1802, a dual audio taper potentiometer with push-button control, with 64 steps!!!! These are precision potentiometers and come in a small dip package. Not wanting push-buttons on the front of my amp I used a bi-phase rotary encoder. These work by sending alternating pulses when they are turned. If pulse A is received before pulse B then we have a clockwise rotation, and likewise if pulse B is received before pulse A we have a counter-clockwise rotation. An oversized microcontroller provided by a friend’s spare parts bin decodes the bi-phase signal and pulses one output for volume up, and another for volume down.
The volume control board also has relays for selecting either the balanced input or the output of the DRV134 (non balanced input).
The use of digital potentiometers also added the easy ability to create a remote volume control. The remote signals in this case are decoded by another piece of hardware in my audio system with an Atmel microcontroller. The ATMEGA88 receives RC5 (manchester encoded) or NEC (mark-space encoded) remote signals. When the correct signal is received it is simply sent out via a DB9 connector to the correct device, in this case the volume control board.
Headphones are particularly sensitive to DC offset on the output. The small coils are easily fried and the lack of capacitive coupling on this amp means any fault in the amplifier would be passed directly to the output. The servo in the main amp works well to eliminate DC offset however it doesn’t do much against the initial power-on thump, and can’t protect against component failure.
The circuit works by summing all 4 outputs of the amplifier with an OPAMP and then running them through a low pass filter with a corner frequency of 1.5Hz. If any of the outputs have more than ±80mV DC present this will cause OPAMP to turn on either Q3 in the case of a positive swing, or Q4 in the case of a negative swing. Turning on Q3 or Q4 will result in the Q5/Q6 darlington pair being switched off and thus disconnecting the positive singals to each headphone driver. Once DC goes below ±80mV the OPAMP’s output will return to ground turning off both output transistors and the headphones are reconnected.
In the end we have an amplifier with loads of power to spare, excellent sound quality, low DC offset, headphone protection, and one hell of an awesome volume control. This project was started in 2003 as a simple single ended amplifier circuit and I’m not sure if it will ever truly finish.