Wavelet Denoising and Conditioning of Neural Recordings
Luca Citi, Silvestro Micera · 2013
This chapter presents wavelet-based denoising algorithms as a preprocessing stage before spike detection and sorting. The first part of the chapter overviews wavelet-based denoising algorithms. The dyadic wavelet transform is compared with a timeinvariant approach, showing that the latter is best suited to the denoising of neural signals. The second part of the chapter shows a sample application with eletroneurographic (ENG) signals recorded from the sciatic nerve of rabbits while the experimenter stimulated the paw of the animal. The wavelet-based denoising is compared with a traditional band-pass filter in two cases: when followed by spike sorting and when followed by traditional rectified bin integration (RBI). The results illustrate the benefits of wavelet denoising over standard band-pass filtering and demonstrate that there is an even more marked improvement when the subsequent step requires signals with high signal-to-noise ratio (SNR), such as in the case of spike sorting.