Table 1.

Processing steps (left column) and options tested (right column). The complete sequence of all combinations listed below was performed for decoding continuous 2D wrist position and again for decoding continuous 2D wrist velocity.

Build spatial filters on 90% of the data from a given recording session.	11 different spatial filtering methods tested (CAR, PCA, 8 versions of CSP, and unfiltered) as described in section 2.2.4.
Select a subset of the calculated spatially- filtered signals to use for decoding.	For each filtering method tested, use the best 4, 8, 12, 16, or 32 of the calculated spatial filters as described in section 2.2.5.
Calculate power in different bands as a function of time from each spatially-filtered signal in the specific subset of filtered signals being tested.	Regardless of the spatial filtering method used, the following set of power bands were calculated from each spatially filtered signal: 8–12, 12–18, 18–24, 24–35, 35–42, 42–70, and 70–200 Hz. Power was then log transformed (see section 2.2.6).
Smooth resulting power data over time.	Smooth each power signal by taking a moving average of the calculated power over 1 (unsmoothed), 3, or 5 100 ms time steps as described in section 2.2.6.
Using the same 90% training data, calculate position or velocity decoding functions from the smoothed power data and the simultaneously- recorded kinematic data.	Calculate 2D decoding functions using Kalman filtering or linear regression as described in section 2.2.6.
Apply spatial filters, power calculations, power smoothing, and decoding functions to the remaining 10% of the data to generate predicted position or velocity data.	Assess accuracy of the different signal processing options by calculating the coefficient of determination (R2) between the actual and predicted continuous wrist movement values as described in section 2.2.7 (X and Y components calculated separately and averaged together)
Repeat the above process using the next 10% of the data for testing and the remaining 90% of the data for training.	All options are repeated ten times to get 10 different performance measures for each combination of options.