Identifying linear models of systems suffering nonlinear distortions

The paper examines the effects of nonlinear distortions on the estimation of frequency domain models using multisine test signals. It is shown that odd order nonlinearities will always introduce an error, the magnitude of which depends on the interaction of different types of nonlinear components. The importance of minimising the signal crest factor is then illustrated. A novel wide band pilot test signal is proposed to establish the system bandwidth and detect nonlinearities. The elimination of the nonlinear effect for static polynomial and Hammerstein is then addressed, using multisines with prescribed spectra. A class of signals for directly measuring Volterra kernels are also described. Practical results are presented along with guidelines for signal design.