Choosing Optimal Delays for Feedback Delay Networks

Feedback delay networks (FDNs) are often used in the context of artificial reverberation and are a class of sparse IIR filters. FDNs are based on a feedback loop with multiple channels containing delay elements, as well as a mixing matrix providing a connection between the channels. An example of a simple FDN is shown in Figure 1. For a practical FDN implementation, many parameters have to be chosen: the number of channels, the mixing matrix, the delays, several gains, and, in the case of FDNs that implement a frequency-dependent reverberation time, filters for each channel. Some of these parameters can be computed from room impulse response properties [2], while for others popular choices exist, e.g. using a Hadamard matrix as the mixing matrix.