Abstract: The three-component continuous waveform data of 22 reference stations in the earthquake early warning area of Inner Mongolia in January 2022 were selected, and the PSD and PDF distributions of the waveform data were calculated. The PSD distributions at frequencies from 0.01 to 50 Hz were counted, and the RMS and abnormal stations were analyzed. The variation characteristics of background noise under different frequencies and regions were studied according to high frequency band （≤1 s）, micro-seismic band （1 ~ 20 s） and long period section （≥20 s）. At the same time, the monitoring capability of seismic network was analyzed. The results show that the background noise level of high frequency band is related to population and industrial density. The noise level of the second type of ground pulsation is generally consistent. The observed noise energy of the first type of ground pulsation is weak, and the difference of the noise level of the geographical spatial distribution is small. In the long period of more than 20 s, the background noise is mainly affected by the natural environment, such as temperature and humidity, and has certain regional differences. The seismometer installed in the cave and underground can effectively reduce the influence of human activity interference sources, temperature and pressure around the station on the high frequency band and long period observation, so that the noise level in the high frequency band and long period observation is lower than that on the surface. Compared with high frequency and long period frequency band, the background noise analysis of microseismic frequency band can evaluate the status of station instruments such as swing and parameter error more effectively and accurately. After the repair of the fault stations, the background noise of the stations in the early warning area of Inner Mongolia is better than Ⅱ level, which significantly improves the monitoring capability of the seismometer network in the early warning area and plays a good guarantee role for the early warning project.