Abstract: Sparse representation of signals is of great significance in many applications. In this paper, sparse representation for chirp echoes in broadband radar on an orthogonal dictionary is proposed. The stretch processing of broadband radar is reformulated in matrix form, and an orthogonal dictionary is established. Combining the orthogonal sparse representation with theory of compressed sensing, a novel sampling mechanism for chirp echoes called randomly selecting can be obtained. Simulation results show that the performance of the sparse representation is better than that in the redundant dictionary consisting of Gabor atoms. Furthermore, sparse representation on the orthogonal dictionary is much more computationally efficient. Real data experiment validates the feasibility of randomly selecting sampling mechanism for chirp echoes.