System identification with neural network controlled resonator-banks

The application of the neural-network-controlled resonator-bank structure for system identification is described. The proposed structure consists of a feedforward neutral network and a resonator-based digital filter. The neural network section serves as an associative memory that controls the dynamic transfer characteristics of the generic linear filter structure. Previous research using FIR filtering characteristics has shown that the structure is capable of learning structurally different dynamic transfer characteristics and associating them with patterns in the auxiliary input signals. The authors summarize the properties of the structure and demonstrate its application in the identification of a time-variant nonlinear dynamics