This paper presents an architecture named K-filter able to model non-linear systems both memoryless and with memory. The most general version of the k-filter applies to any non linearity but sometimes at the cost of holding a considerable computational load, specially when the memory of the non-linear system increases. Thus, the paper is basically devoted to present how different simpler versions of the original k-filter can be obtained taking into account symmetrical properties of the input/output relation of the non-linear system to model. The theoretical points along with the simulation results will show how these symmetrical considerations simplify the k-filter without making worse the performance.

Peer Reviewed