Amplifier Distortion Measurements


One of the factors which can affect the audible performance of audio amplifiers is the tendency for the audio signal patterns to become distorted by various non-linearities. As a result, it is common for reviews to provide values which are intended to indicate how much a given amplifier may distort the waveforms. Unfortunately, experience has indicated that many of the measured results don’t give a very reliable indication of the audible results. The purpose of this document is therefore to have a fresh look at distortion measurements, and to suggest an alternative approach that might prove more useful when dealing with amplifiers that are used to reproduce music.

Conventional measurement techniques

The most common method for indicating the amount of non-linear distortion introduced by an audio amplifier is the Total Harmonic Distortion (THD). For measurements of this type, the test waveform is a nominally pure single-frequency sinusoid and the measure is the summed level of any components produced at harmonics of the chosen frequency. In some cases a total Distortion + Noise level is measured, and this may then include any effect due to background noise or mains hum/buzz. In other cases the output is processed so that only power at (or near) the harmonics of the test waveform are included in the measured result.

The main advantage of the THD measurement is that it is easy to define and carry out. However the resulting values are often uninformative. It is quite common for people listening to amplifiers to form the view that – when they attempt to rank amplifiers in terms of audible performance – the measured THD values do not correlate at all well with audible performance. Hence THD seems to be a poor guide or predictor of amplifier performance.

A common alternative is some form is two-tone Intermodulation Distortion (IMD) measurement. This may be carried out in various ways by chooing variour pairs of frequencies with various relative levels. Given two input frequencies, and , this method then either measures the summed level of all components, , (where n and m are any non-zero integers) or at a chosen sum/difference frequency. This method can be more revealing than THD of some specific effects, but still seems to be a poor guide to overall performance when an amplifier is used with musical waveforms.

For a period during the late 1970’s and early 1980’s reviews also sometimes employed squarewave test-signals and examined effects due to slew-rate limiting and “Transient Intermodulation Distortion” (TID). These effects could arise due to amplifiers being unable to cope with signals whose rates of change exceeded their capability. In particular, when an amplifier was asked to drive low-impedance or reactive loudspeaker loads. The main reason for tests of this kind was the realisation that musical signals can contain transient signal patterns which might reach relatively high power levels, and require rapid changes in the output voltage and current. Simple sinewave test signals in the audio band may not always fully cover what was required for musical patterns, hence the need for an alternative.

Unfortunately, square-waves are not actually much more like music than is a simple sinewave. As a result, although such tests are useful for some purposes, they also do not seem to be a useful guide to musical performance in many cases. In addition, in terms of square-waves the problem of TID was largely ‘solved’ by appropriate design without coming to a much better agreement with distortion measurements and perceived performance.

Another test used at times is to assess the amount of distortion produced when the amplifier is driven into clipping. This is useful in some contexts. However a fairly obvious point is that this is best avoided altogether by ensuring the amplifier is not asked to output levels beyond its ability! Once clipping is avoided, its percived quality when clipping becomes of little interest. That said, where clipping does occur, it can be useful to assess the details of the waveform shape changes this produces.

Various other methods for assessing amplifier nonlinearities have been proposed over the years. Two example which are of particular interest here are:

The Belcher method employs a pair of maximal-length pseudo-random noise generators, arranged to have different repeat-durations with no simple relationship. The result is a subtle form of intermodulation distortion test. We can then specify two discrete combs of frequency components for each noise generator, and expect that any other components which arise will be due to amplifier imperfections. The main advantage of this method is that the test waveform has a complex spectrum that can symultaneously contain a series of components, spread across the audio band. In this respect it is more like some typical music than a simple sinusoid. This method has been claimed to give a better guide to musical performance than THD or two-tone IMD. However, despite the complex spectrum, the test waveform tends to differ from typical music in other ways. As a result, it may not probe some types of amplifier imperfection which might sometimes be significant.

The Hirata method is based upon noting that musical waveforms often contain transient events which are brief, and asymmetric in shape. Recognising this, Hirata devised test waveforms which are asymmetric and quasi-transient, but which have a mean (average level) of zero – i.e. no d.c. component. Waveforms of this general type are very interesting since they can probe effects whereby the linearity of an amplifier (or other audio component) is sensitive to medium-term statistical or short-term asymmetries in the waveforms. Unfortunately, as with the Belcher method, Hirata’s proposal in the above article is based upon waveforms that have some related drawbacks. For example, the Hirata waveforms were essentially a group of rectangular pulses. The resulting spectra then depend upon the sharpness of the pulse edges, etc.

The purpose of what follows is to outline some of the amplifier effects which might need to be probed to identify non-linearities which affect musical reproduction. Then go on to consider an alternative method that might provide measured results more relevant to musical performance.

Next page.