The lower trace shows classic crossover distortion, where one transistor imperfectly hands over to the other in a push-pull pair. Crossover in large-ish quantities (above 0.3%) is audible, roughening the sound.


A distortion of a sound is, in the broadest sense, any change that affects the original. Distortion that can be heard as a nasty ripping coarseness is a particular form that comes from non-linearity where the output, or what we hear, ceases to follow the input signal in fairly gross fashion. A measurement of distortion is supposed to tell us about the purity, or lack of it,  of the sound. It certainly does do this when there are large quantities (5% or more) of low order harmonics, which will add muddle. Just how much muddle depends upon the correlation of the harmonics to the original though; uncorrelated produces audible muddle; correlated produces a subtle change in timbre. Second harmonic above about 5% lightens timbre; third harmonic above 1% sharpens the sound (these are necessarily very approximate values).

So how distortion “sounds” is quite complex and below about 0.2% it becomes difficult to be certain distortion is audible. Digital distortions, where harmonics far from the fundamental exist and are totally uncorrelated to it, the ear detects as a coarseness or roughness in the sound. This situation exists with CD.

Amplifiers, however, have a similar but more insidious problem. Feedback produces dynamic skewing of the transfer function, and as high frequency signals rise in amplitude the distortion structure changes. This produces a pattern of high order harmonics changing in uncorrelated fashion with the music signal, an effect the ear readily detects. We look for this whilst testing by observing how the distortion spectrum changes with level.

Reducing feedback lessens this effect. Recently, amplifiers able to maintain a steady spectral pattern at high frequencies, comprising low order components, and they have an audibly neutral treble quality, free of character.


This is a spectrum analysis of the crossover distortion shown above. Distortion measures 0.29% at 1W (2V)  into 4 Ohms at 10kHz. Odd order harmonics predominate.

Our spectrum analyses of distortion show nine harmonics. If 2nd and 3rd dominate then the distortion is unlikely to be especially audible. If higher order harmonics exist then the chances of audibility increase.

Never to be forgotten is that the same conditions of non-linearity that produces harmonic distortion also creates rarely mentioned intermodulation distortion and it’s likely that this is the mechanism by which crossover distortion is heard. This is why our single published distortion figure measures distortion at just 1Watt, 10kHz, because it is a measure of crossover non-linearity.


Generally, if distortion at all frequencies and levels is below 0.2% or so in amplifiers it is not of major consequence in terms of sound quality. It may still subliminally stain the sound however.

It does appear that high feedback amplifiers with correspondingly low distortion may lack ‘air’ and stage depth in their sound. Valve amplifiers are unable to use high levels of feedback and seem to benefit in this manner. However, some designers of solid-state amplifiers dispute this, claiming high feedback is in itself not a problem, so much as how it is applied. 



Distortion is measured with the amplifier connected to both 8 Ohm and 4 Ohm resistive loads, the latter producing higher distortion due to the higher current draw.

The load resistors, custom built to our specification, use zero hysteresis (iron free) wire to avoid high frequency distortion from magnetic effects. This ensures our high frequency (10kHz) distortion measurements are representative of the amplifier under test, and are not influenced by the load. No test equipment earth connections are made to the loads; they are fully balanced.

We measure harmonic distortion at a low output of 1 Watt and just below full output (-1dBV), at 1kHz and 10kHz, into 8 Ohm and 4 Ohm loads. This is a spot measurement scheme that reveals an amplifier’s basic behaviour. With valve amplifiers we also measure distortion at 40Hz under the same conditions, to assess how well the transformer core can handle magnetisation.  Distortion from this source is predominantly third harmonic in nature. A good core will distort little up to full output, a poor one produce 0.5% or so at low power of a few watts and 5% or so at higher power, until saturation limits any further increase, sometimes well below full output in the midband. This is heard as ‘soggy’ bass.

The result published in the magazine shows distortion at 1 Watt, 10kHz into a 4 Ohm load, a realistic test yet one that yields the highest distortion figure and shows the presence of ‘crossover distortion’. It is also our quoted distortion figure and it can be up to ten times greater than the result at 1kHz that manufacturers commonly quote. Spectral content is shown by our Rohde & Schwarz UPL analyser and this also provides all distortion figures. A Hewlett Packard 8903B provides an interesting time domain picture of the distortion residual.

Other distortion measurements we may make are a sweep across the audio band to show distortion as a function of frequency, and intermodulation distortion in its various forms.


Whilst a distortion sweep is interesting, it cannot show spectral content, a drawback that limits its usefulness. Our spot checks show the common rise in high frequency distortion caused by falling open loop gain but provide a spectral picture.


Intermodulation distortion measures non-linearity using a slower method than harmonic distortion. It can be useful to illustrate a point, such as that high frequency non-linearity in Class D PWM amplifiers isn’t rendered inconsequential by low pass filtering at 20kHz, which eliminates high frequency harmonics. Strong intermodulation products are still measurable in the audio band.



Hi-Fi World, Powered by Joomla!; Hosted by Joomla Wired.