Biological signal conditioning for system identification

The use of mathematical models to characterize physiological systems attempts to summarize how a system behaves in terms of its internal structure. Thus the models should be isomorphic to the underlying physiological structure, and the model parameters must be estimated from input-output observations of the system. The model must be identifiable, in the sense that a unique parameter set exists and that parameter set is resolvable in the presence of noise. This paper considers these identifiability issues with specific emphasis on the effect of biological observation noise in degrading the parameter estimation process. It is shown that conditioning the output signal via design of the experimental input can minimize this degradation. These concepts are applied to metabolic and respiratory system studies.