Introduction

An ideal sound system is one in which the original signal is reproduced without any change to the signal, except its amplitude. Since such change is inevitable, these questions must be asked:

• Which types of distortion are most audible?
• Which parts of the system introduce the most distortion?
• How can distortion be reduced?

Distortion perception

Recent work by Earl Geddes on distortion perception challenges the notion that %THD (Total Harmonic Distortion) and %IMD (Intermodulation Distortion) measurements provide an audible indication or how much a signal is distorted. A surprising result is that these numbers mean very little when it comes to perception.

This downloadable PDF article on distortion perception is essential reading for understanding the perception of distortion: [Distortion Perception] Gedlee Website

One study shows a group of people who can't really tell if higher %THD sounds more distorted or not.

A [previous version of the website] contained downloadable audio samples, where different values of %THD, %IMD and Earl's new Gedlee metric are shown. There is a clear audible connection between perceived distortion and his new metric, but not with the other measures.

The Geddes work puts the audio enthusiast in a difficult position, making it more difficult to correctly interpret any measure of distortion.

Speakers - the worst offenders

While we may debate issues of what a listener can perceive, what is fairly obvious is that the most significant distortion occurs when electrical energy is converted to acoustic energy. In comparison to speakers, all other components may be considered as almost perfect.

Published measurements of some subwoofers, for example, show well over 50% THD. While this may not literally sound as bad as it looks, the original signal is being significantly altered. It is important to distinguish between pure harmonic distortion measurements that look at the magnitude of the harmonics in the frequency domain (i.e using FFT or other means) and THD+Noise measurements which just band-reject the fundamental and look at the rms level of everything left, including noise.

Types of distortion

There are two basic types of distortion - linear distortion (aberrations in the frequency, phase and time response) and nonlinear distortion where extra frequency components are added. Most commonly discussed are THD (total harmonic distortion) and IMD(intermodulation distortion), which are nonlinear distortions.

When distortion measurements are made on speakers, harmonic distortion components are often separated, and 2nd, 3rd, 4th and 5th order components are shown. If a 20 Hz tone is played, such a chart will show that 40 Hz information is added (2nd order harmonic distortion), as well as 60 Hz (3rd order), 80 Hz (4th order) and 100 Hz (5th order). General consensus seems to be that odd and higher order distortions are more audible. What should be obvious is that this is undesirable in a subwoofer, and high distortion can invalidate the common claim that "you can't localise sound from a subwoofer because the wavelengths are too long!" For this claim to be true, the distortion needs to be low. It may be true at low SPL, but alter with the increased distortions at high SPL. The 5th order harmonic distortion will be clearly audible and give clues to the location of the subwoofer, and possibly interfere with imaging as well, especially if it is located in a corner to the side of the mains.

It should be understood that in speakers, all the different performance measurements are interrelated. For example, if a driver has a hump in the frequency response, it is an indication that there will be distortion figures associated with that depending on what kind of distortion testing is conducted and the type of signal used. If sine waves are used to conduct distortion testing, test results will probably be better. This is because residual energy from other frequency contents do not affect the results.

Aiming for low distortion

There are a number of mechanisms to reduce distortion in speakers. These include:

• increasing cone size and/or multiple drivers per passband to reduce excursion
• shorting rings to reduce 2nd order harmonic distortion
• advanced motor designs to increase symmetry of the magnetic field
• motor designs such as Adire's XBL^2 motor with dual gaps to achieve flat motor strength curve
• voice coils with modified winding to achieve a ruler flat motor strength curve
• improved suspension systems to match the linearity of flat BL motors
• low inductance motor systems
• servo controlled drivers
• horn loading
• bass loading designs included bass reflex and transmission line enclosures
• diaphragm damping

One of the simplest ways to reduce nonlinear distortion is by increasing headroom. A system which can handle 120 db will likely have lower distortion at 100 db than one which can't go past 105 db. Of course, this system may have to make compromises in linear distortion in order to achieve this. It is up to the designer to choose a compromise that seems to best fit.

Note also that distortion products, just like the undistorted sound, vary in evident loudness according to the Fletcher-Munson curve, even though the response may be flat. Hence, for any given order of distortion, the distortion will be most audible around 3.5kHz. Furthermore, as the human hearing drops off after approximately 20kHz, non-linear distortion is - theoretically - undetectable above 10kHz. Second-order distortion (the lowest order) at 10kHz is 20kHz, so anything above can't be heard.

Another way of looking at some aspects of distortion are the Cumulate Spectral Decay and Time Energy curves. Basically a driver is supposed to transfer electrical energy into acoustic energy, but in the process some energy is stored in the cone and suspension system in the form of resonances, which sometimes are referred to as breakup modes in the cone. This energy is partially tranferred into acoustic energy over a longer period of time if the suspension system cannot dissipate it, thus when it is mixed back with the oncoming signals, is may or may not show up as distortion depending on test method and test signal used. This is why one cannot draw conclusions on speaker quality only from distortion figures, and there seems not to be a direct relationship.

Previous Version Note

A previous member wrote began this article, and the original is included following this note.
As my take on this topic may be different to the original, I'll leave it here:

Original Version

A loudspeaker must be linear
A loudspeaker is not a musical instrument.It must reproduce sound with no distortion.
<this is just a template , which should be filled with accurate information when i can soon>

Distortion types ::

1. harmonic
2. Intermodulation distortion(non harmonic)

Loudspeaker Distortion ::

1. Suspension
2. Magnetic circuits
3. Cone structure

Link

[Diagnosis and Remedy of Nonlinearities in Electrodynamical Transducers] Klippel AES 2000 paper

---