(A) Schematic of a simple feedback controller with intrinsic gain, and time delay, . (B) A Smith Predictor with intrinsic gain, , time delay and calibrated internal feedback loop. (C,D) Rearrangements of the two model architectures to allow derivation of the cursor response function, , and force amplitude response, . Note that both act as ‘comb filters’ and exhibit delay-dependent submovement peaks. However the architectures predict different relationships between intrinsic gain and force amplitude response. (E) Feedback gains inferred from experimental data assuming the simple feedback controller architecture, for 5 s windows early, middle and late in each trial. Thick line shows average over 8 subjects. Note that feedback gains for different delay conditions become less similar as the trial progresses. (F) Feedback gains inferred from experimental data assuming the Smith Predictor architecture. Feedback gains for different delay conditions become more similar as trial progresses. (G) Delay-dependence of feedback gain (mean-squared difference between delay conditions) inferred from the two architectures. The analysis used a 5 s sliding window through the entire trial. Shading indicates s.e.m. over 8 subjects. (H) Phase delay of feedback gain at 1 Hz inferred from Smith Predictor architecture through trials with 0 and 200 ms delay. (I) Average time lag between cursor and target through trials with 0 and 200 ms delay (and no spatial perturbation).