“…Indeed, if one considers a sampling rate of 20 kHz and windows of about 2 ms around each spike, then one would wind up with 40-dimensional space, which could potentially be too sparsely populated to define dense clusters of spikes. Features are often extracted using spike minimum [42], spike maximum [42], spike width [42], peak-to-peak amplitude [9,87], peak-to-peak time interval [87], amplitude later in the waveform [8,71], principal component analysis (PCA) [1,15,23,26], PCA on the major portrait radius of the reconstructed phase space portrait of spikes [11], the discrete wavelet transform [41,57], the wavelet packet transform [34], autoregressive coefficients from spikes [51], the integral transform [91], discrete derivatives of the spike [24], positive and negative peaks of Brought to you by | University of Manitoba Authenticated Download Date | 6/10/15 11:13 AM the first derivative of the spike [51,88], and the positive and negative peak of the second derivative of the spike [88].…”