This article analyzes the problems existed in the MMSE-RISIC algorithm in detail and proposes an improved MMSE-RISIC equalization algorithm for SCFDE whose frame structure applied Unique Word (UW). This algorithm adds a noise prediction module to utilize the correlation between the estimated noise in UW and the estimated noise in date. The noise in data after MMSE equalization is predicted and eliminated by the estimated noise in UW. The estimation accuracy of RISI will also improve. Finally, the parameters based on the 300 km troposcatter channel in north China are utilized for making a simulation. The results show that the improved MMSE-RISIC equalization can lower a bit of error rate (BER) under condition of low signal to noise ratio (SNR). And SNR gains about 2 dB when BER=10 -3 compared with the MMSERISIC equalization.