In this paper, we investigate the timing and carrier

frequency offset (CFO) synchronization problem in decode and

forward cooperative systems operating over frequency selective

channels. A training sequence which consists of one orthogonal

frequency-division multiplexing (OFDM) block having a tile

structure in the frequency domain is proposed to perform synchronization.

Timing offsets are estimated using correlation-type

algorithms. By inserting some null subcarriers in the proposed tile

structure, we propose a computationally efficient subspace decomposition-

based algorithm for CFO estimation. The issue of optimal

tile length is studied both theoretically and through simulations.

By judiciously designing the tile size of the pilot, the proposed

algorithms are shown to have better performance, in terms of

synchronization errors and bit error rate, than the time-division

multiplexing-based training method and the computationally demanding

space-alternating generalized expectation-maximization