Oxygen concentration detection plays an important role in industrial p rocess control.In this paper,tunable diode laser absorption spectroscopy (TDLAS) is used for online m onitoring of oxygen concentration.The experimental system employs vertical cavity surface emitting lasers (VCSEL) with output wavelength of 760nm as light source and adopts the radio-type structur e with optical path of 60cm.Continuous measurement for 72hours shows that the system has a high a ccuracy in short time, but slow drift comes out when the time is longer than 4hours.Based on the anal ysis of first harmonic signal and second harmonic signal,this paper argues that the changes of backgro und noise causes measurement error,and the composition and characteristics of background noises o f the system are carried out and discussed too.After analyzing the system signal from the perspective of frequency domain, we propose that under the premise of the amplitude-frequency characteristic of the system′s background noise keeping stable,cross-correlation algorithm could be applied t o deduce the background noise in real-time.The test of 5% oxygen for 60hours points out th at after data processing, the maximum relative error is reduced from 3.2% to 1.82%.This indicates that th e proposed method could effectively improve the system′s performance.