The error feedback method for roundoff noise reduction is applied to a
time-varying multiple-input multiple-output (MIMO) system. The optima
l error feedback coefficients are derived for a general Mth-order stru
cture, which reduces to M distinct optimization problems. The develope
d theory is applied to a signal processor implementation of a 4th-orde
r time-varying AC motor model. By employing 2nd-order error feedback.
the roundoff noise can be cut down very efficiently with only 8% incre
ase in program execution time, which is only a fraction of the costs o
f a double-precision implementation. Although the model is time varyin
g, keeping the error feedback coefficients fixed results in almost equ
al reduction as with time-varying coefficients.