Abstract: The basic principle of fixed-point operation was introduced, quantitative error in fixed-point operation of fast convolution was analyzed, and the influence model of FFT points and system word length on output SNR was established when fixed-point processing of FFT adopted three algorithms of un-calibration, pre-calibration and stepwise calibration respectively, and the experimental verification was carried out. Experimental results showed that, when the un-calibrated algorithm was adopted, the output SNR first decreased and then increased with the increase of FFT points, and the longer the system word length, the larger the corresponding FFT points when the output SNR ratio began to increase. When the pre-calibration and stepwise calibration algorithms were adopted, the output SNR decreased gradually with the increase of FFT points, and finally approached to a certain value, which was not affected by the system word length. In the actual FFT design, different calibration algorithms can be selected according to FFT points and system word length, so as to meet the requirements of output SNR.