Abstract: An efficient decoding algorithm for the unitary space-time modulation is presented for the constellations created by a systematic method.First,the receiver calculates the inner products between the received signal and L/2+W (L is the number of constellations and W is the length of the window) constellations created successively.Second,the L/2 sums can be achieved by adding the L/2+W inner products every W ones,and the maximal one can be searched out.Based on the location of the maximal sum,the searching space of the maximum likelihood (ML) algorithm can be reduced to three constellations.Finally,the optimal constellation from the three constellations can be achieved.The proposed algorithm can significantly reduce the searching space of the ML algorithm.The smaller the value of W,the more the searching space can be reduced.Computer simulation results show that for the unitary constellations,the performance of the proposed algorithm can approach the performance as that of the ML algorithm at signal-to-noise ratio penalty of about 2.0 dB with a half of complexity of ML.