This paper studies the multichannel diffusion active noise control (ANC) systems, where a
group of controllers and error microphones are physically distributed at different locations
within an extended area and could communicate with each other in a distributed manner.
Since conventional diffusion ANC algorithms introduce the estimation bias via
communication between local controllers and degrade the overall performance, we aim to
develop a new diffusion filtered-x (Diff-Fx) based algorithm that incorporates knowledge
of spatial smoothness and balances between communication strength and estimation bias.
To further increase the convergence speed and save computational complexity, the
transform domain (TD) technique is employed. A detailed performance analysis is carried
out for the resulting Diff-TD-FxLMS algorithm. A variable spatial regularization (SR) rule
is derived from the analysis, which could adjust the diffusion strength automatically. The
performance of the proposed algorithm is also compared with conventional methods.

diffusion FxLMS,transform domain,multitask problem,network design,performance analysis