Abstract: In order to improve the channel estimation performance of OFDM (orthogonal frequency-division multiplexing) based on compressed sensing, the problem of pilot pattern design was discussed. Analyzing the problem of traditional pilot design criterion, the Euclidean distance between the measurement matrix and the identity matrix was used as the definition of correlation of measurement matrix. Then a new pilot design criterion was gotten based on correlation minimization. At the same time, a cyclic substitution search algorithm based on the parallel trees was also proposed. At each cycle, only one pilot position of each branch was substituted. According to the pilot design criterion, the better ones were chosen in every branch first and among all the branches latter to avoid the locally-optimal but globally-incorrect selections. Simulation results demonstrate that the new pilot design criterion can obtain substantial improvement in sparse channel estimation, while the new search algorithm has better convergence.