Existing on-off-keying modulation in optical communication becomes inadequate to meet exponentially growing bandwidth demand. New receiver architectures and algorithms are being sought to confront this challenge. To achieve this goal, we propose innovative adaptive algorithms to simultaneously track optical polarizations and compensate any channel impairment without the need of any optical/electronic phase-locked-loop in Phase-Shift-Keying modulation for optical coherent receivers. These algorithms are demonstrated to be capable of coping with various impairments such as chromatic dispersion and polarization mode dispersion in optical channels to allow transmission of high speed data at 40 Gbps and beyond in even low-speed-grade long-haul fibers.

In similarity to its electronic counterpart, conventional coherent optical communication requires the difficult task of phase synchronization of a local laser with a transmit laser indiscriminately without regard of individual phase noise contributors. By studying carefully the characteristic of each phase noise contributor, we developed a novel coherent detection algorithm empowered by advanced digital signal processing techniques for quadrature-amplitude modulation (QAM) to enable practical realization of coherent communication without the need of phase synchronization. This algorithm enables 9 photons/bit sensitivity at a bit error rate 103at 10 GBaud/s for QAM4 with an aggregate 100-MHz linewidth, providing us a cheaper alternative to utilize regular lasers of linewidth up to 12.5 MHz for 2.5-GBaud/s transmission.

Currently existing subpixel motion estimation algorithms require interpolation of interpixel values which undesirably increases the overall complexity and data flow and deteriorates the estimation accuracy. We develop discrete cosine transform (DCT)-based techniques to estimate subpel motion at different desired subpel levels of accuracy in the DCT domain without interpolation. We show that subpixel motion information is preserved in the DCT of a shifted signal under some condition in the form of pseudophases, and we establish subpel sinusoidal orthogonal principles to extract this information. The proposed subpixel techniques are flexible and scalable in terms of estimation accuracy with very low computational complexity O(N/sup 2/) compared to O(N/sup 4/) for the full-search block-matching approach and its subpixel versions. Above all, motion estimation in the DCT domain instead of the spatial domain simplifies the conventional hybrid DCT-based video coder, especially the heavily loaded feedback loop in the conventional design, resulting in a fully DCT-based high-throughput video codec. In addition, the computation of pseudophases is local, and thus a highly parallel architecture is feasible for the DCT-based algorithms. Finally, simulation on video sequences of different characteristics shows comparable performance of the proposed algorithms to block-matching approaches.

We propose novel discrete cosine transform (DCT) pseudophase techniques to estimate shift/delay between two one-dimensional(1-D) signals directly from their DCT coefficients by computing the pseudophase shift hidden in DCT and then employing the sinusoidal orthogonal principles, applicable to signal delay estimation remote sensing. Under the two-dimensional (2-D) translational motion model, we further extend the pseudophase techniques to the DCT-based motion estimation (DXT-ME) algorithm for 2-D signals/images. The DXT-ME algorithm has certain advantages over the commonly used full search block-matching approach (BKM-ME) for application to video coding despite certain limitations. In addition to its robustness in a noisy environment and low computational complexity, O(M(2)) for an MxM search range in comparison to the O(N(2) . M(2)) complexity of BKM-ME for an NxN block, its ability to estimate motion completely in DCT domain makes possible the fully DCT-based motion-compensated video coder structure, which has only one major component in the feedback loop instead of three as in the conventional hybrid video coder design, and thus results in a higher system throughput. Furthermore, combination of the DCT and motion estimation units can provide space for further optimization of the overall coder. In addition, the DXT-ME algorithm has solely highly parallel local operations and this property makes feasible parallel implementation suitable for very large scale integration (VLSI) design. Simulation on a number of video sequences is presented with comparison to BKM-ME and other fast block search algorithms for video coding applications even though DXT-ME is completely different from any block search algorithms.

The feedback loop for temporal prediction of traditional implementations of an MPEG-compliant video coder requires conversion of images from the spatial domain to the transform domain and then back to the spatial domain due to the restriction of motion estimation schemes performed only in the spatial domain, severely limiting the throughput of the coder. However, it was shown that the low-complexity DCT-based motion estimation scheme (DXT-ME) can remove this bottleneck and thus reduce the overall coder complexity. In this paper, we develop efficient DCT-based motion compensation algorithms to complete the fully DCT-based motion compensated video coder. We demonstrate that the performance of this new coder structure is comparable to the conventional one while the overall coder complexity is lower.

Inter-pixel interpolation is required for most subpixel motion estimation schemes but it undesirably increases the overall complexity and data flow and deteriorates the estimation accuracy. We develop DCT-based techniques to estimate subpel motion at different desired levels of accuracy in the DCT domain without interpolation by establishing subpel sinusoidal orthogonal principles and showing that subpixel motion information is preserved in the DCT of a shifted signal under some conditions in the form of pseudo phases. Though applicable to other areas, the resulting algorithms from these techniques for video coding are flexible and scalable with very low complexity O(N/sup 2/) compared to O(N/sup 4/) for block matching methods. Importantly, DCT-based methods enable simplification of the heavily loaded feedback loop of conventional hybrid video coder design, resulting in a high-throughput, low-complexity fully DCT-based coder. Finally, simulation results show a comparable performance of the proposed algorithms with block matching methods.

The feedback loop for temporal prediction of traditional implementations of an MPEG-compliant video coder requires conversion of images from the spatial domain to the transform domain and then back to the spatial domain due to the restriction of motion estimation schemes performed only in the spatial domain, severely limiting the throughput of the coder. However, the availability of the low-complexity DCT-based motion estimation scheme (DXT-ME) that we proposed in the past can remove this bottleneck and concurrently reduce the complexity of the overall coder structure. We adopt an adaptive overlapping approach to further improve this DCT-based motion estimation scheme, as demonstrated in the simulation results. Furthermore, the analysis of stability of DXT-ME is presented and the case of a uniformly bright environment is also discussed. Finally, the parallel architecture of DXT-ME is outlined.

The rank revealing URV decomposition is a useful tool in many signal processing applications that require the computation of the noise subspace of a matrix. In this paper, we look at the problem of updating the URV decomposition and consider its implementation on a wavefront array. We propose a wavefront array for efficient VLSI implementation of the URV decomposition. The array provides a mechanism for accumulating the null space of the matrix as well as for handling rank changes during updating.

A new motion estimation scheme, discrete-cosine/sine-transform based motion estimation (DXT-ME) utilizing the principle of orthogonality of cosine and sine functions to estimate, in the transform domain, displacements from the motion information contained in the pseudo-phases of the images of moving objects, is proposed. The computational complexity of this method is only O(N2)+Odctfor an image of size N×N in comparison to O(N4), the complexity of the full search block matching approach, where Odctis the complexity of DCT/DST coding. Furthermore, incorporation of DXT-ME with DCT-based video standards achieves further savings of computations and makes symmetric codecs feasible. Unlike the pel-recursive algorithm, this scheme is not susceptible to noise. For complicated scenery or large moving objects, simple preprocessing is performed on images to extract the features of moving objects before applying DXT-ME. Simulation on some video sequences is presented to compare this scheme with the block matching method.