Passive intermodulation (PIM) interference has become a key problem to restrict the performance of high-sensitivity communication systems, especially under fast time-varying channels. To address this problem, this paper proposes an extended model-based PIM interference cancellation method. Zadoff-Chu(ZC) sequences are designed as training sequences in transmitters. The channel model is decomposed into a linear combination of time-varying basis functions, which can be the discrete basis expansion model (BEM).The time-invariant basis coefficients are extracted by using the least-squares estimation. The reconstructed PIM interfererence signals are eliminated in receivers to achieve efficient suppression. The experimental validation shows that under the Rayleigh fast time-varying channel with a maximum Doppler shift 650 Hz, the proposed method can significantly improve the PIM interference suppression performance by more than 10 dB of signal-to-interference ratio (SIR) gain for bit error rate (BER) of \(10^{-2}\) . The results show that this method can significantly improve the PIM interference suppression performance in fast time-varying scenarios, which provides an effective solution for the anti-PIM design of 6G air-to-space integrated networks.