Mismatch robustness of linear prediction and its relationship to coding

Source mismatch problems of linear least square predictors are discussed in terms of spectral distance measures and process distance measures. It is shown that the linear predictors are robust for several stability measures. The relationship of the predictor mismatch to linear predictive coding is also discussed, and a simpler alternative to Chaffee's rate distortion coder is given. In the Appendix, topological implications of distance measures are studied in order to supplement the robustness theorems in the text.