Subspace system identification using a multichannel lattice filter

This work presents a new class of subspace algorithms for system identification, in which the projection of future data onto past data is performed by a multichannel least-squares lattice filter with parameterizations and channel arrangements particularly suited to subspace system identification. Because an adaptive lattice filter provides the core computational engine of the new subspace algorithms, the main computational burden of the algorithms can be executed in real time, as data is being sampled. This approach produces much faster system identification algorithms than existing batch algorithms for subspace identification.