Real-time density monitoring during pavement compaction process is crucial to ensure asphalt concrete (AC) construction quality. Ground penetrating radar (GPR) may be used for continuous real-time AC density estimation using AC and its components dielectric constants. However, the effect of surface moisture, resulted from roller water sprays, jeopardizes the predicted AC density accuracy. While algorithms were developed to remove the surface moisture effect, they are only applicable to 2 GHz GPR antenna. This study proposed a novel algorithm to remove surface moisture effect based on "frequency-select effect." This algorithm is applicable to most of the GPR antenna central frequencies. The Al-Qadi, Lahouar, Leng (ALL) density prediction model was used in this study. Static field tests were performed using both 2 GHz and 1 GHz air-coupled antenna at the Illinois Center for Transportation (ICT) of UIUC. The GPR signals were processed through proposed algorithm using power spectrum of the surface reflections. There is a high correlation between AC dielectric constant and density; hence, the dielectric constant ground-truth values were used to estimate the errors using the proposed algorithm. The proposed algorithm was found to be efficient and accurate; errors of dielectric constant are less than 3% for the 2 GHz case and 5% for the 1 GHz case, respectively.