Abstract: Frequency radar is one of the conventional instruments for observing horizontal wind fields in the mesosphere and lower thermosphere region. It calculates the correlation functions of signals received by multiple antennas based on the full correlation analysis method and retrieves wind fields according to the amplitude characteristics of the correlation functions. In actual observations, instability in the system hardware and interference in the spatial detection region can cause short-term distortions in the amplitude characteristics of the correlation functions, leading to abrupt short-term variations in the retrieved wind fields and causing long-term statistical results to deviate from physical characteristics. Existing work primarily focuses on denoising received signals, but the widely adopted polynomial fitting method is sensitive to noise. Short-term noise fluctuations can propagate during retrieval and cause abrupt wind field variations. Therefore, there is an urgent need to introduce noise-level-independent metrics to guide the denoising process. Inspired by MST radar, this paper introduces the antenna contribution value as a parameter metric into MF radar and proposes the MF-AH algorithm to address amplitude distortions in correlation functions. The algorithm is validated through numerical simulations and real MF radar data. Results show that, compared to the traditional polynomial fitting method, the MF-AH algorithm effectively mitigates short-term abrupt wind field variations caused by reduced signal-to-noise ratios within the range of −5 to 20 dB. The retrieved wind fields are more consistent with the physical laws and observational characteristics of the middle and upper atmosphere.