Preface
When you cannot measure it,
when you cannot express it in numbers,
your knowledge is of a meager
and unsatisfactory kind;
it may be the beginning of knowledge,
but you have scarcely, in your thoughts,
advanced to the stage of science.
William Thomson, Lord Kelvin (1894)
At the beginning of the 1990s, several research groups around the world were working on adaptive filters for hands-free telephones. At that time the first experimental DSP boards were available and well suited for audio experiments. While many research groups focused only on simulations, a few also implemented real-time experiments with such DSP boards. While such boards needed to be programmed in assembler, it was very tedious to make a program run and at the same time very time consuming, and many posed the question, “Why would experimental work be worth at all?” Why would it not be good enough to perform Matlab® simulations? In simulations one can be so much faster, have everything under control, and measurement precision is not an issue at all. However, experiments showed that simulations were not capturing the true nature of the problem. Some designs sounded artificial owing to their phase response degradation, some showed smaller Echo Return Loss Enhancement (ERLE) but simply sounded better. The “Quality of Experience” as we call it today was more important than the performance metric based on a Euclidian distance of ERLE. Also, rapid changes in the environment as they occur when the local speaker ...
