We investigated a velocity analysis algorithm for the new array Ground Penetrating Radar (GPR) system 'Yakumo'. The common mid-point (CMP) data, which is used to estimate the vertical profile of subsurface velocity can be acquired easily with this system. However, each CMP data set acquired by Yakumo includes only 8 GPR traces with more than 20 cm trace interval. In order to estimate the velocity with this sparse GPR data, we proposed a reversible normal move-out (NMO) processing to remove aliasing and to interpolate the data with a new iterative f-k domain interpolation method to reconstruct the dense data before the velocity analysis. The proposed processing algorithm is applied to both the simulated and the measured data. With the CMP data acquired by 100 MHz antennas, the re-sampled data with more than 0.5 m gaps can still be well interpolated. And the artifacts in the velocity spectrum caused by the noise were well suppressed. We also applied the process to Yakumo CMP data and the estimated the velocity profile in a sandpit. Comparing to the velocity measured by Time Domain reflectometer (TDR), the error was less than 10%. With Yakumo system, we can obtain the CMP data together with the three dimensional GPR data, which greatly benefit the data acquisition.