No description available.

The advent of turbo codes in 1993 represented a significant step towards realising the ultimate capacity limit of a communication channel, breaking the link that was binding very good performance with exponential decoder complexity. Turbo codes are parallel concatenated convolutional codes, decoded with a suboptimal iterative algorithm. The complexity of the iterative algorithm increases only linearly with block length, bringing previously unprecedented performance within practical limits. This work is a further investigation of turbo codes and other concatenated schemes such as the multiple parallel concatenation and the serial concatenation. The analysis of these schemes has two important aspects, their performance under optimal decoding and the convergence of their iterative, suboptimal decoding algorithm. The connection between iterative decoding performance and the optimal decoding performance is analysed with the help of computer simulation by studying the iterative decoding error events. Methods for good performance interleaver design and code design are presented and analysed in the same way. The optimal decoding performance is further investigated by using a novel method to determine the weight spectra of turbo codes by using the turbo code tree representation, and the results are compared with the results of the iterative decoder. The method can also be used for the analysis of multiple parallel concatenated codes, but is impractical for the serial concatenated codes. Non-optimal, non-iterative decoding algorithms are presented and compared with the iterative algorithm. The convergence of the iterative algorithm is investigated by using the Cauchy criterion. Some insight into the performance of the concatenated schemes under iterative decoding is found by separating error events into convergent and non-convergent components. The sensitivity of convergence to the Eb/Ng operating point has been explored.

Evaluation of soft output decoding for turbo codes is presented. Coding theory related to this research is studied, including convolutional encoding and Viterbi decoding. Recursive systematic convolutional (RSC) codes and nonuniform interleavers commonly used in turbo code encoder design are analyzed. Fundamentals such as reliability estimation, log-likelihood algebra, and soft channel outputs for soft output Viterbi algorithm (SOVA) turbo code decoding are examined. The modified Viterbi metric that incorporates a-priori information used for SOVA decoding is derived. A low memory implementation of the SOVA decoder is shown. The iterative SOVA turbo code decoding algorithm is described with illustrative examples. The performance of turbo codes are evaluated through computer simulation. It has been found that the SOVA turbo code decoding algorithm, as described in the literature, did not perform as well as the published results. Modifications to the decoding algorithm are suggested. The simulated turbo code performance results shown after these modifications more closely match with current published research work. / Master of Science

No description available.

International Telemetering Conference Proceedings / October 30-November 02, 1995 / Riviera Hotel, Las Vegas, Nevada / This paper discusses an error-correcting scheme applied to a telemetry system over HF radio channel. According to the statistical properties of transmission error on HF radio channel, the scheme uses one important diffuse convolutional code, which is threshold decoded and corrects the random or burst errors. The operation of this code is explained, and a new method for word synchronization and bit synchronization is proposed. Coding and decoding, word synchronization, and bit synchronization are all activated by software program so as to greatly improve the flexibleness and applicability of the data transmission system. Test results of error-correcting are given for a variety of bit-error-rate (BER)s on HF radio channel.

Image representation is a key component in visual recognition systems. In visual recognition problem, the solution or the model should be able to learn and infer the quality of certain visual semantics in the image. Therefore, it is important for the model to represent the input image in a way that the semantics of interest can be inferred easily and reliably. This thesis is written in the form of a compilation of publications and tries to look into the Convolutional Networks (CovnNets) representation in visual recognition problems from an empirical perspective. Convolutional Network is a special class of Neural Networks with a hierarchical structure where every layer’s output (except for the last layer) will be the input of another one. It was shown that ConvNets are powerful tools to learn a generic representation of an image. In this body of work, we first showed that this is indeed the case and ConvNet representation with a simple classifier can outperform highly-tuned pipelines based on hand-crafted features. To be precise, we first trained a ConvNet on a large dataset, then for every image in another task with a small dataset, we feedforward the image to the ConvNet and take the ConvNets activation on a certain layer as the image representation. Transferring the knowledge from the large dataset (source task) to the small dataset (target task) proved to be effective and outperformed baselines on a variety of tasks in visual recognition. We also evaluated the presence of spatial visual semantics in ConvNet representation and observed that ConvNet retains significant spatial information despite the fact that it has never been explicitly trained to preserve low-level semantics. We then tried to investigate the factors that affect the transferability of these representations. We studied various factors on a diverse set of visual recognition tasks and found a consistent correlation between the effect of those factors and the similarity of the target task to the source task. This intuition alongside the experimental results provides a guideline to improve the performance of visual recognition tasks using ConvNet features. Finally, we addressed the task of visual instance retrieval specifically as an example of how these simple intuitions can increase the performance of the target task massively. / <p>QC 20161209</p>

Alpha matting is an important topic in areas of computer vision. It has various applications, such as virtual reality, digital image and video editing, and image synthesis. The conventional approaches for alpha matting perform unsatisfactorily when they encounter complicated background and foreground. It is also difficult for them to extract alpha matte accurately when the foreground objects are transparent, semi-transparent, perforated or hairy. Fortunately, the rapid development of deep learning techniques brings new possibilities for solving alpha matting problems. In this thesis, we propose a residual convolutional grid network for alpha matting, which is based on the convolutional neural networks (CNNs) and can learn the alpha matte directly from the original image and its trimap. Our grid network consists of horizontal residual convolutional computation blocks and vertical upsampling/downsampling convolutional computation blocks. By choosing different paths to pass information by itself, our network can not only retain the rich details of the image but also extract high-level abstract semantic information of the image. The experimental results demonstrate that our method can solve the matting problems that plague conventional matting methods for decades and outperform all the other state-of-the-art matting methods in quality and visual evaluation. The only matting method performs a little better than ours is the current best matting method. However, that matting method requires three times amount of trainable parameters compared with ours. Hence, our matting method is the best considering the computation complexity, memory usage, and matting performance.

With the emergence of Convolutional Neural Network (CNN) models, precision of image classification tasks has been improved significantly over these years. Regional CNN (RCNN) model is proposed to solve object detection tasks with a combination of Region Proposal Network and CNN. This model improves the detection accuracy but suffer from slow inference speed because of its multi-stage structure. The Single Stage Detection (SSD) network is later proposed to further improve the object detection benchmark in terms of accuracy and speed. However, SSD model still suffers from high miss rate on small targets since datasets are usually dominated by medium and large sized objects, which don’t share the same features with small ones. On the other hand, geometric analysis on dataset images can provide additional information before model training. In this thesis, we propose several SSD-based models with adjusted parameters on feature extraction layers by using geometric analysis on KITTI and Caltech Pedestrian datasets. This analysis extends SSD’s capability on small objects detection. To further improve detection accuracy, we propose a two-stream network, which uses one stream to detect medium to large objects, and another stream specifically for small objects. This two-stream model achieves competitive performance comparing to other algorithms on KITTI and Caltech Pedestrian benchmark. Those results are shown and analysed in this thesis as well.

Quantum error correction codes were introduced as a means to protect quantum information from decoherance and operational errors. Based on their approach to error control, error correcting codes can be divided into two different classes: block codes and convolutional codes. There has been significant development towards finding quantum block codes, since they were first discovered in 1995. In contrast, quantum convolutional codes remained mainly uninvestigated. In this thesis, we develop the stabilizer formalism for quantum convolutional codes. We define distance properties of these codes and give a general method for constructing encoding circuits, given a set of generators of the stabilizer of a quantum convolutional stabilizer code, is shown. The resulting encoding circuit enables online encoding of the qubits, i.e., the encoder does not have to wait for the input transmission to end before starting the encoding process. We develop the quantum analogue of the Viterbi algorithm. The quantum Viterbi algorithm (QVA) is a maximum likehood error estimation algorithm, the complexity of which grows linearly with the number of encoded qubits. A variation of the quantum Viterbi algorithm, the Windowed QVA, is also discussed. Using Windowed QVA, we can estimate the most likely error without waiting for the entire received sequence.

In this thesis, we take an application of multiple symbol differential detection (MSDD) technique in direct-sequence code division multiple access (CDMA) system. It is well- known that MSDD is an effective noncoherent demodulator which outperform the conventional M-ary differential phase shift keying (MDPSK) in additive white Gaussian noise (AWGN) channel. Take MPSK demodulator into consideration, the performance of MSDD based on noncoherent demodulation approaches the performance of coherent demodulation. However, there is little research about MSDD in frequency-selective fading channel. We are now combining the MSDD and Rake receiver to be the signal demodulator. In conventional, there are two kinds of Rake receivers. One is coherent demodulator. Another is noncoherent demodulator. For coherent demodulation, it needs to have channel estimation at each path. The advantage is that the performance will be improved. On the other hand, the disadvantage is complexity and operation will increase. On the contrast, for noncoherent demodulation, it is the performance degradation and complexity simplification. In this thesis, We suggest a multiple symbol differential detection on Rake receiver for CDMA system. From our computer simulation, only for hard decision, the performance is improved and the improvement is proportional to the number of multipath and the number of the length of multiple symbol. This will not happen in conventional MDPSK. However, from our observation, the improvement of performance is degrading as the number of multipath increase. Thus, we employee the technique of Viterbi decoding differential detection (VDDD) to demodulate the differential sequence. By the property of decision interval, the VDDD can obtain additional improvement.