Algorithm steps

The algorithm steps can be summarized below:

1. Use non-stationary autoregression to compute $a_n(t)$ with $b_n(t)=d(t-n\Delta t)$ ($\Delta t$ is the time interval) according to equation 2;
2. Find the roots $Z_n(t)$ of each polynomial according to equation 7;
3. Compute instantaneous frequency $f_n(t)$ according to equation 6;
4. Compute local phase $\phi_n(t)$ according to equation 5;
5. Fit the data to compute the amplitude corresponding to each frequency component $\hat{A}_n(t)$ and the fitted denoised data $\hat{d}(t)$ according to equations 4 and 8.