The disclosure describes techniques and systems for filtering noise from a physiological signal. In one example, one or more processors are configured to receive a signal indicative of physiological activity of a patient, wherein the signal comprises noise at one or more frequencies, and filter the noise from the signal according to a noise rejection model, wherein the noise rejection model predicts the noise at the one or more frequencies. The one or more processors may also be configured to, responsive to initiation of a blanking period for the signal, advance the noise rejection model in time during the blanking period, and, responsive to termination of the blanking period, filter, based on the noise rejection model advanced in time, the noise at the one or more frequencies from the signal.