Optimal and suboptimal binary inputs for system identification

A heuristic random search technique for finding a near-optimum binary input sequence to a linear dynamical system is presented. The problem requires the optimization of a multimodal real function that depends on a string of binary integers. The algorithms combine a global random search procedure with a local (neighborhood) search which examines all sequences within a prescribed Hamming distance. The algorithm is applied to the determination of the sequence of air and oxygen breaths that are optimal for estimating lung parameter values. Simulation studies show that the algorithm finds an optimum input sequence 10 bits in length in 100% of the trials, and 20 bits in length in 97% of the trials. Near-optimum values are also located with strings 30 and 40 bits in length using approximately 1000 iterations