M. R. P. Thomas
Fundamentally, loudspeaker design has changed very little in the past 30 years. The late 70s saw the introduction of electromechanical modelling techniques which remain the basis for the design and analysis of loudspeakers today. Aside from the use of lighter, stronger materials in the construction of drive units, the area which has undergone the most development is the design of equalisers. Classically, loudspeakers use a passive analogue filter network to compensate for a non-flat frequency response, but they are far from ideal and are prone to variation with temperature and age. Recent years have seen the use of linear DSP equalisers which conform to much tighter specifications. However, there are two major flaws in the design of even the latest equalisers. Firstly, it has long been known that at extreme voice coil excursion a loudspeaker is a nonlinear device, though no documented attempt at producing a nonlinear equaliser has yet been proposed. Secondly is the use of swept sine tones to characterise frequency responses, which can sometimes yield different results to methods where excitation energy is spread over a wider bandwidth (which is more akin to music or speech). Swept sines generally provide no phase information, which is an area of increasing interest in loudspeaker equalisation. This project investigates the analysis of loudspeaker frequency responses using a Maximum-Length Sequence (MLS) or M-Sequence Decorrelation technique, which uses pseudorandom noise to yield both amplitude and phase information. The measurements are used to model the frequency response over a wide range of amplitudes, fitting a Volterra-Series approximation and defining a level-dependent equaliser.
|Institution||Master's Thesis, Imperial College London|