For colored interference noise in digital ultrasonic signals at ultrasonic frequency band, an adaptive canceller was designed based on a transversal filter and the least mean square adaptive filtering algorithm. An adaptive filtering algorithm was proposed combining the fixed and adaptive step lengths. The proposed method could track automatically the changes of noise characteristics by adding a special probe to receive the environment noise without manually setting the parameters and the expected signals of the adaptive filter. By combining adaptive step method and fixed step method, the proposed method has good tracking and steady-state properties. Experiment results show that proposed method filters effectively out the colored noise located outside the frequency band of the target ultrasonic signal in real time and the signal-to-noise has improved up to 16 dB and the time complexity is O(n). The method has been successfully applied to gas ultrasonic flow measurement and filters out the colored noise in real time automatically and effectively without the manual intervention.