In this work, we proposed a novel method to investigate the underlying rapid pressure-to-flow dynamics induced by changes of arterial CO2. Autoregulation was modeled as a multivariate system. The instantaneous effect of CO2 to cerebral blood flow velocity (CBFV) was removed adaptively by the recursive least square (RLS) method from CBFV. The residue CBFV and arterial blood pressure (ABP) were then filtered by a Gaussian-modulated sinusoidal pulse filter, in order to optimize the time and frequency resolution when estimating the instantaneous phase difference between the signals using Hilbert transform (HT). The results indicate that the effect of CO2 on dynamic autoregulation is slower than on CBFV.