Introduction

Artificial reverberation is a challenging application in signal processing because it is necessary to approximate large systems (such as concert halls) having hundreds of thousands of poles and zeros in the audio band. Instead of pursuing explicit models which are prohibitively complex, it is necessary to find alternative abstractions which can be implemented at reasonable cost and which capture the salient psychoacoustical attributes of natural reverberation. An important practical requirement is a stable numerical implementation of sparse, high-order, nearly lossless linear systems. This paper addresses this and related issues.