There are several widely used methods of measuring distortion in audio amplifiers, but probably the most useful is the direct comparison of the input and output signals, sometimes refered to as the bridge, or null method. This involves attenuating the output of the amplifier under test to the level of the input, and then adding or subtracting the input signal, depending on whether the amplifier is inverting or non-inverting respectively, so that the undistorted component of the output signal is cancelled leaving only distortion and noise.
There are several advantages to this method:
An ultra low distortion signal generator is not needed. The signal generator distortion and noise are cancelled along with the undistorted component. On one occasion I measured distortion components around 0.00001% using a signal generator with over 0.5% t.h.d. It was the noise level which eventually limited the accuracy, not the generator distortion.
Another advantage is that we are not restricted to using sine wave test signals, and with adequate amplitude and phase adjustment in the nulling circuit we can use a wide range of signals, including music, which will give a good indication of the performance in the intended application.
It is not intended to give a detailed description of the nulling method. This was the basis of my M.Sc. dissertation, which was 98 pages in length, with 44 references. I will merely present the relevant circuit diagrams, to show how the distortion traces published for my design were extracted.
The first diagram shows the basic circuit, which can be used for both inverting and non-inverting amplifiers, and also has the advantage of comparing the voltage difference between the output terminals with the voltage difference between the input terminals. In this way any earth line problems will not be missed, which is possible if we assume the output earth is identical to the input earth.
Adjustment of the variable components is difficult and takes a lot of time and patience. Simple preset pots are not good enough, and for ultimate results a network of fixed resistors needs to be built up to get as close as possible to the final value needed, and just a very small trimmer used for final fine adjustment. VR1 and VC are adjusted to give accurate common-mode rejection, and are set by applying a sine wave signal to both inputs V1 and V2 (with no test amplifier connected). With a 1kHz signal VR1 is adjusted to give zero o/p from A2, then using 20kHz VC is adjusted to give zero o/p. VR1 may now need to be reset slightly, so that there is zero o/p at both 1kHz and 20kHz. When testing an amplifier VR2 is determined by the gain of the amplifier, and is about 6k multiplied by the amplifier gain. Again accurate adjustment is needed.
VC1 and VC2 adjust to compensate for phase shift in the amplifier, and are widely different for different designs, so little guidance can be given. Two variable controls are shown, but this is only really needed for accurate compensation over a wide bandwidth, e.g. if testing with a music signal. For single sine wave tests VC2 can be omitted. For wide bandwidth testing the two capacitors may need to be similar or very different in value, depending on the nature of the phase shift in the amplifier, which may be close to a simple first order low pass filter, or may have something similar to a time delay element. (I found that an old 741 op-amp actually had more excess phase shift than a medium power class-B amplifier.) The real time delay in audio amplifiers is insignificant, the observed effect being just aditional phase shift for a given attenuation compared to a first order low pass response at low attenuation. The second order filter has greater phase shift for a given attenuation and can adequately match the amplitude and phase response of most amplifiers over a wide bandwidth, though it is usually easier to do this if any low pass filter at the input of the amplifier is disabled so that we are not also having to compensate for this. The output amplifier A2 is only amplifying the distortion voltage, so almost any low noise wide bandwidth op-amp will be good enough. R6 determines the gain of the whole instrument, and 620k is typical, giving gain about 100 relative to the input signals V1 and V2
A problem with the simple phase adjustment used is that although it can reasonably well cope with the response of an amplifier driving a resistive load, if we use a real loudspeaker load the complex impedance will give variations in output level and phase sufficiently large to swamp the small distortion signals we are interested in. To compensate for the speaker impedance would be very difficult, but it is still possible to reduce the effects to a low level if we instead concentrate on the output impedance of the amplifier, and compensate for the voltage drop across this impedance, which is really the only effect of the load impedance. The output impedance of the amplifier will generally be much simpler than that of the speaker, and a simple modification to the previous test circuit can be used in which Z0 is an impedance identical to the amplifier output impedance, and ZL is the load impedance. The rest of the circuit is as in the previous diagram.
Adjustment of the controls can then give the necessary nulling. In practice some variation on this basic circuit may be more convenient. In tests a distortion trace was clearly revealed for an amplifier loaded with a highly non-linear and frequency dependant load, the effects of which had previously totally hidden the distortion.
The inverting amplifier used in the test circuit must of course be designed for the minimum possible distortion, and although op-amps with very low distortion are available, I found that a discrete component design could easily give better results by a factor of 10 or more at 20kHz. A circuit I have used is shown next, and although I do not have accurate distortion figures for it, the 20kHz distortion is certainly well below 0.0001% and at 1kHz is better than 0.00001%. The 47k control adjusts the total current in the input differential stage to set minimum distortion. For the input stage alone minimum distortion is with equal currents through the two transistors, but of course this is not necessarily the minimum for the whole amplifier, and in principle a slight offset in the input stage may generate a small non-linearity sufficient to cancel opposite non-linearity in later stages. This is why I left this as an adjustment rather than just set the transistor currents equal using the usual current mirror techniques. In practice I found adjusting this control gave little change in distortion, but I never investigated further.
Finally, to give an example of the distortion traces obtainable by this method, here is a trace using an earlier, inferior design, which even so shows that noise level is the eventual limiting factor in obtaining useful information. The distortion level here is around 0.0006% at 2kHz test frequency using a fairly standard class-B amplifier. I used the single shot facility on the oscilloscope to improve clarity in this case.