Study On Carrier Frequency Offset Estimation For OFDM/OFDMA-based Wireless Communication Systems

Orthogonal frequency division multiplexing (OFDM) technology possesses many features, such as efficient spectrum utilization, robustness against frequency selective fading, and has become one of the critical techniques to support high speed data transmission. Meanwhile, orthogonal frequency division multiple access (OFDMA) has become one of the most popular technologies for the future wireless communication systems, in which multiple users are permitted to transmit data simultaneously via the frequency division multiplexing. However, OFDM/OFDMA is very sensitive to carrier frequency offset (CFO). The CFO may destroy the orthogonality of the subcarriers, which would result in the intercarrier interference, the multiple access interference, signal distortion and performance degradation, etc. Therefore, CFO has become the main obstacle to application of OFDM/OFDMA. Therefore, CFO estimation problem has been one of the most valuable topics in wireless communication systems.In this dissertation, the performance of CFO estimation are analyzed systematically, and the CFO estimation algorithms for OFDM systems and OFDMA uplink systems are studied. Focusing on the frequency spectrum utilization, computational complexity and estimation accuracy, etc., for OFDM and OFDMA uplink systems, several effective and achievable CFO algorithms are proposed.Aiming at the low estimation accuracy caused by only using the eigenvalues, a CFO approach based on reconstructing CFO information matrix is proposed. A matrix including the CFO information is defined according to the covariance matrix of received data, and then the CFO is obtained by using not only the eigenvalues but also the eigenvectors of the defined matrix, which improves the estimation accuracy efficiently. Without any training sequences and pilots, the proposed algorithm can obtain high spectrum efficiency. Meanwhile, the proposed algorithm can keep good performance at low SNR.The existing CFO estimation algorithms for OFDM systems deal with the computation of complex matrices, such as the eigendecomposition (EVD), the inverse etc., which causes unattractive computational complexity. To solve this problem, a CFO estimation algorithm, named as Unitary-ESPRIT, is proposed. By utilizing the inherit property of the received data and the time smoothing technique, a Hankel matrix is constructed. Via the unitary transformation, the EVD and the inverse of the complex matrices can be avoided, which reduces the computational complexity significantly. Meanwhile, the proposed algorithm incorporates the time smoothing technique and the forward-backward averaging technique, which leads to an improved estimation performance. Furthermore, the proposed blind algorithm satisfies the requirement for real-time processing with low computational complexity and high frequency spectrum utilization.The peak-searching-based CFO algorithms have favorable estimation accuracy and super resolution. However, the complex data processing may cause unattractive computational complexity and cannot ensure the real-time implementation, especially in OFDMA uplink systems. Therefore, via the unitary transformation, a fast CFO estimation algorithm based on MUSIC is proposed. Via the periodic property of received signals, the data matrix can be rewritten as the product of a Vandermonde matrix and a user-signal matrix. Since it incorporates unitary transformation, the EVD of the complex matrix and the complex-value spectrum computation can be avoided, which will obtain a low computational complexity. The property of centro-hermitian CFO steering vector and the forward-backward averaging technique are used to improve the estimation performance.The proposed algorithm can provide high spectrum efficiency and better estimation performance which is quite significant to further popularize MUSIC.To further improve the popularization of the peak-searching-based CFO algorithms, a CFO estimation algorithm based on compressive sensing theory is proposed. The CFO sparse representation and the redundant auto dictionary are given by using the sparse property of the CFO distribution and the covariance matrix of received signals. Via the compressive sensing theory, the proposed method converts the CFO estimation problem into a convex optimization problem, and it can effectively estimate the CFOs of all users by finding the sparse vector. The presented method not only ensures the super resolution performance, but also can avoid the eigenvalue decomposition in the subspace methods, and is more practical.Compared with the peak-searching-based CFO algorithms, the close-form CFO estimation algorithms, such as ESPRIT, have lower computational complexity for uplink OFDMA systems. However, they only take advantage of the eigenvalue to estimate the CFOs, which causes poor estimation performance. To solve this problem, a CFO estimation method is proposed by reconstructing a CFO information matrix. The CFOs are estimated by using not only the eigenvalues but also the eigenvectors of the defined CFO information matrix, which improves the estimation accuracy efficiently. To further improve the performance of the close-form CFO estimation algorithms, via the idea of blind source separation, a CFO estimation algorithm based on joint diagonalization is proposed. A whitening matrix is firstly derived by using the covariance matrix of the received signals. In order to obtain the CFO steering matrix, the spatially covariance matrices which are whitened by the whitening matrix are sent to the joint diagonalizer. The proposed algorithm exploits each vector of the estimated steering matrix to estimate the CFOs, which improves the estimation performance effectively. Aiming at the unattractive computational complexity and the estimation performance of the close-form algorithms for uplink OFDMA systems, an effective CFO estimation algorithm, named as Unitary-ESPRIT, is proposed. Via the utilization of the unitary transformation, the complex computations are converted to the real-valued processing, which can lead to a reduced significantly computational complexity and an improved estimation performance.