To overcome the disadvantage of side information transmission, an improved selective mapping algorithm is proposed for peak-to-average power ratio reduction of multiple-input multiple-output orthogonal frequency division multiplexing communication via underwater acoustic channels. The scrambling patterns are carried by comb pilot with different distributions. And taking advantage of neritic channel sparse characteristics, the scrambling patterns can be distinguished without side information. The results of experiment performed with numerical simulation and tank demonstrate that the proposed algorithm can distinguish between scrambling patterns without degrading the performance, abate the burst noise effect, and significantly enhance the quality of the system.