During FFT computation, energy leakage and picket-fence effect are the most critical error sources derived from FFT algorithms; meanwhile, data truncation and rounding during the butterfly calculation implemented by fixed-point DSP can make even more significant error due to finite wordlength. Based on some established FFT spectrum correction algorithms, an improved double-windowing method is presented, and an extended-precision FFT scheme for fixed-point DSP application is put forward. The feasibilities of the two reformative methods are discussed theoretically, and the validities are testified by the simulation results