Mud pulse telemetry system plays a critical role in modern oil drilling engineering. However, the valid signal is complicated to be obtained because of the main disturbing of pump noises, especially in ultra-deep drilling operations. Therefore, four model-based noise suppression plans for the mud pulse signal pump are mainly proposed in this paper. First, three kinds of pump noise state-space models (linear time-invariant model, linear time-varying model, and nonlinear model) are constructed based on the detailed analysis of the pump noise characteristics. Then, the standard Kalman filter based on the linear system and the extended Kalman filter and unscented Kalman filter algorithm adopting the nonlinear system are respectively used to reconstruct and filter the pump noise. Moreover, the wavelet threshold denoising method based on a new power threshold function is specifically adopted to filter residual random noise after the removal of pump noise. Simultaneously, several noise-containing mud pulse signals with different SNRs (Signal-to-Noise Ratios) in the stable/unstable pump noise states were simulated. Besides, the noise suppression capability experiments of the four plans are deeply conducted. The results indicate that all the four plans can suppress and even remove the stable or unstable pump noise from the original signals: plan 1 and plan 2 have similar denoising effect, which can improve the SNR of signals with stable and unstable pump noise of different intensities by approximately 5–25 dB and 5–14 dB; while the plan 3 can increase the SNR of signals by about 4–20 dB and 2–8 dB; and plan 4 can increase that by about 4–20 dB and 1–8 dB, respectively. Under the comprehensive consideration of denoising ability, computation intensity and convergence speed, it is suggested that plan 1 integrating the time-invariant linear model and the standard Kalman filtering reveals superior performance. Furthermore, using the processing method consisting of plan 1 and wavelet threshold denoising method, the SNRs of the denoised signals can reach 6–7 dB even if the original SNR is as low as −30 dB. The proposed method is applied to the actual deep well mud pulse signal processing. The effective downhole pulse transmission waveform is successfully obtained, providing a highly reliable input signal for subsequent data restoration and contributing to an essential and practical significance for improving mud pulse transmission quality.

泥浆脉冲遥测系统在现代石油钻井工程中起着至关重要的作用。但是，由于泵噪声的主要干扰，很难获得有效信号，尤其是在超深钻孔操作中。因此，本文主要针对泥浆脉冲信号泵提出了四种基于模型的噪声抑制方案。首先，在对泵噪声特性进行详细分析的基础上，构建了三种泵噪声状态空间模型（线性时不变模型，线性时变模型和非线性模型）。然后，分别使用基于线性系统的标准卡尔曼滤波器和采用非线性系统的扩展卡尔曼滤波器和无味卡尔曼滤波器算法来重构和过滤泵噪声。此外，具体采用了基于新的功率阈值函数的小波阈值去噪方法，对去除泵浦噪声后的残留随机噪声进行滤波。同时，模拟了几种在稳定/不稳定泵噪声状态下具有不同SNR（信噪比）的含噪声泥浆脉冲信号。此外，还对这四个方案的噪声抑制能力进行了深入的实验。结果表明，这四个方案都能抑制甚至消除原始信号中的稳定或不稳定泵浦噪声：方案1和方案2具有相似的去噪效果，可以提高不同强度的稳定和不稳定泵浦噪声的信号信噪比。分别降低5–25 dB和5–14 dB；方案3可以将信号的信噪比提高约4-20 dB和2-8 dB；计划4可以分别将其增加约4–20 dB和1–8 dB。综合考虑去噪能力，计算强度和收敛速度，建议将时不变线性模型与标准卡尔曼滤波相结合的方案1具有较好的性能。此外，使用由方案1和小波阈值去噪方法组成的处理方法，即使原始SNR低至-30 dB，去噪信号的SNR也可以达到6-7 dB。将该方法应用于实际的深井泥浆脉冲信号处理。成功获得了有效的井下脉冲传输波形，为后续的数据恢复提供了高度可靠的输入信号，并为提高泥浆脉冲传输质量做出了重要而实际的贡献。