A new structure for adaptive linear-phase filtering

In many applications, including time-varying modeling and adaptive noise canceling, it is desirable to adaptively perform linear-phase filtering to prevent any phase distortion in the observed data. Much attention has been devoted so far to the purely algorithmic aspect of this problem, i.e., how to design an appropriate adaptation algorithm. It is also of interest, however, to measure the extent of the structural influence, i.e., what system structure to choose to "optimize" the performance-to-cost ratio. This paper presents a different parameterization of the linear-phase filtering problem, naturally leading to a new, very efficient filter implementation. The resulting realization is then used as the system structure for an adaptive linear-phase filter. When associated with suitable stochastic gradient algorithms, this new adaptive structure is found to exhibit a tracking performance competitive with that of linear-phase lattice/ladder realizations, for a computational complexity 33 to 40 percent lower. It can therefore be used advantageously in virtually all applications of adaptive filtering. Simulations performed in the context of adaptive linear-phase modeling show good agreement with the theoretical analysis.