Convergence results for the sample matrix inversion (SMI) canceler algorithm in nonstationary noise are presented. Exact results are given for the convergence rate of the average output noise power residue of the canceler normalized to the quiescent average output noise power residue for a two-input canceler (one auxiliary), and lower and upper bounds are derived for cancelers with two or more inputs under the assumption that there is no internal noise. These bounds are a function of the number of independent samples processed per channel (main and auxiliary), the number of auxiliary input channels, and the external noise environment. The external noise environment is modeled as a single interfering source that is conditionally Gaussian, with a power level specified at each sampling time instant. Furthermore, this model is generalized in the sense that a joint probability distribution function is defined for the power levels over a canceler processing batch. This leads to the capability of modeling and evaluating the SMI canceler in a variety of nonstationary single interference scenarios. >