Feedforward vs. Feedback Fixed-Parameter H2 Control of Non-Stationary Noise

Stationary random noise can be modelled as a wide-sense
stationary white noise filtered by a minimum phase filter. Such filter can be
used to design an optimal control filter minimising variance of the signal being
the effect of the noise and the secondary sound interference. However, in many
environments the noise is subject to change. For instance, some of the noisy
devices are switched on and off, speed of some rotors or fans changes, etc. As a
result contribution of different frequency components may significantly vary in
time. Solving the optimisation problem to update control filter is rather
avoided in on-line systems. In adaptive approach there are problems with
convergence or some unpleasant transient acoustic effects. In this paper, the
fixed-parameter approach to control is appreciated. Dominating frequency
components/bands can usually be distinguished for the acoustic environment.
Then, the idea of generalised disturbance defined by a frequency window of
different type can be applied. If a reference signal, correlated with the
disturbance to be reduced is available in advance, a feedforward structure can
be applied, and otherwise, a feedback structure is used. Spectral and
inner-outer factorisations are employed in order to cope with non-minimum phase
character of the acousto-electric plant. Efficiency of the proposed approach for
both control structures is verified based on the data obtained from an active
personal headset. The generalised disturbance based control systems are
confronted with the classical Wiener control systems designed for the given
disturbance.