Noise Modeling and Deep Learning Noise Suppression of Mud Signal

In Measurement While Drilling (MWD), the mud pulse signal is often used to transmit information, but the strong noise characteristic of the mud signal makes the signal demodulation difficult. Therefore, noise suppression has always been a critical algorithm in MWD systems. Due to the overlap between the noise and the communication signal spectrum, the performance of traditional noise suppression algorithms is arduous to improve. An accurate noise model is significant for noise suppression algorithms. However, the existing literature lacks the modeling of mud signal noise. Therefore, this paper proposes a noise modeling method based on probability distribution. This method decomposes the mud channel noise into three components: Pump Stroke Noise, Baseline Drift Noise and Gaussian Noise, then uses the actual drilling mud channel noise to establish noise probability distribution models. Besides, this paper proposes a mud signal denoising net (MSDnNet) with a convolutional neural network and conducts a simulation comparison experiment of noise suppression performance. The simulation results show that MSDnNet can improve the signal-to-noise ratio by about 30dB compared with the traditional adaptive noise suppression algorithms.