Error-correction power amp: practical design.
One of the earliest practical implementations of an Hawksford error correction power amp I know of is a MOSFET output design by Robert Cordell. Bob's design uses error correction in the output stage and overall negative feedback and achieves 0.001%THD up to 20kHz and rated power. His original paper is quite large and takes some time to download, but is worthwhile to read up on error correction and the many issues of power amplifier design in general. There are similar vintage commercial designs by Tandberg and others as well.
My own experimental design is aptly called paX - for power amplifier, eXperimental.
There are two main goals I have with this design:
- Try to eliminate dissipation-dependent temperature effects on the biasing of the output stage;
- Make the amplifier global-feedback free and instead use error correction to linearize the transfer function (after all, this IS Linear Audio ;-) ;
The first goal is the reason I use Sanken output darlingtons. They have built-in diodes that track the temperature of the output transistor die. Because the diodes are used to set the bias voltage, any Vbe change of the outputs due to temperature fluctuations are accurately mirrored in the bias voltage. This way, firstly the long-term average bias is stabilized. It also stabilizes the bias short-term when the output die changes temperature due to signal cycles. These last changes are sometimes called 'memory effect': the bias change lingers after a high power cycles and appears to 'remember' the high temperature. Needless to say that these bias changes are unwanted because they change the amplifier operating conditions during a signal cycle which is a source of amplifier non-linearity.
However, for this to work, you have to drive the
diodes with a current: only then does their diode drop track the output
transistors' Vbe's. Because the
diode drops are not always the same in absolute terms, there is a small series
pot to set the quiescent current in the output stage to 40mA with a diode
current of 2.5mA. That is the condition for accurate temperature compensation.
The output devices being darlingtons, only a small
drive current is necessary. I use an open loop buffer to drive them (I use only open loop topologies in this design lest I be accused of
using some 'hidden' feedback).
Both the current sources as well as the input buffer are supplied through a floating +/-15V supply. This supply is bootstrapped so it can rise above Vout, so it will not limit the maximum output level.
I did some initial measurements on the prototype that are shown here:
- Comparing THD vs. freq with and without EC at low level (0.5W in 8 ohms;
- Comparing THD vs. freq at 50W in 8 ohms;
- Checking THD vs. output level.
Comments?
Current projects:
