For logging while drilling (LWD) systems, it is necessary to adjust the working state of the downhole tools in real-time according to different operating conditions. In this paper, on the basis of the characteristics of LWD systems, a mud pressure-apperceived downlink system was examined. For the design of this system, a signal acquisition and processing board was created based on a piezoelectric ceramic sensor to acquire the mud pressure signal. The error sources of the downlink command sending process were analyzed, and an error accumulation compensation processing algorithm was proposed to improve the recognition success rate of the downhole system. Moreover, to reduce noise interference on the characteristics of the mud impulse signal, a filtering algorithm was proposed based on the empirical mode decomposition method, and the success rate of instruction issuance was determined by identifying feedback instructions. Field tests were conducted to further improve the system, the results of which suggested that the system had good mud adaptability, high recognition success rate, and a certain application value.