[FEA] Optimized convolve kernel
The convolve kernel, which convolves the Hilbert filter with the downsampled signal, is a pure time domain convolution leading to O(N*M) costs. Probably an FFT-based convolution is faster for larger data sets O(N * log(M)).