Robust and computationally efficient anomaly detection in videos is a problem in video surveillance systems. We propose a technique to increase robustness and reduce computational complexity in a Convolutional Neural Network (CNN) based anomaly detector that utilizes the optical flow information of video data. We reduce the complexity of the network by denoising the intermediate layer outputs of the CNN and by using powers-of-two weights, which replaces the computationally expensive multiplication operations with bit-shift operations. Denoising operation during inference forces small valued intermediate layer outputs to zero. The number of zeros in the network significantly increases as a result of denoising, we can implement the CNN about 10% faster than a comparable network while detecting all the anomalies in the testing set. It turns out that denoising operation also provides robustness because the contribution of small intermediate values to the final result is negligible. During training we also generate motion vector images by a Generative Adversarial Network (GAN) to improve the robustness of the overall system. We experimentally observe that the resulting system is robust to background motion.

We present novel bounds on the capacity of binary channels with independent and identically distributed insertions, deletions, and substitutions. The proposed bounds are obtained by exploiting an auxiliary system where the channel is the same as the one in the system of interest, but the receiver is provided with (partial) genie-aided information on the insertion/deletion process. In particular, we show that, when this information is revealed, we obtain a memoryless channel whose capacity, evaluated by means of the Blahut-Arimoto algorithm, gives an upper bound on the capacity of interest. We also show that capacity lower bounds can be derived as well, by exploiting the same auxiliary system and resorting to suitable information-theoretical inequalities. In most scenarios, the proposed bounds improve the existing ones, and significantly narrow the region to which the actual capacity can belong.

In this paper, we consider the optimal permutation of encoded symbols that will result in the maximum burst erasure efficiency for a given low-density parity-check (LDPC) code under belief propagation (BP) decoding. A novel algorithm is proposed to permute columns of the parity-check matrix to achieve high burst erasure correction efficiencies. This method can be applied to any linear code to increase its burst-erasure efficiency while retaining its random-error correction capability. Results show that the algorithm outperforms previous methods.

We derive the capacities of a class of channels, either memoryless or indecomposable finite-state, that also suffer from bursts of erasures. For such channels, we analyze the performances of low-density parity-check (LDPC) codes and code ensembles under belief propagation (BP) decoding, using density evolution (DE) techniques. Although known LDPC codes perform well in non-erasure-affected channels, their performances are far from the capacities when both random errors and erasures are present. We show that enhancing the codes' erasure handling using published methods beats, in some instances, the BP thresholds. However, to achieve capacity, codes must be constructed to tackle both effects simultaneously.

We propose to use various trellis structures to characterize different types of insertion and deletion channels. We start with binary independent and identically distributed (i.i.d.) insertion or deletion channels, propose a trellis representation and develop a simulation based algorithm to estimate the corresponding information rates with independent and uniformly distributed inputs. This approach is then generalized to other cases, including channels with additive white Gaussian noise, channels with both insertions and deletions, and channels with intersymbol interference (ISI) where the latter model is motivated by the recent developments on bit-patterned media recording. We demonstrate that the proposed algorithm is an efficient and flexible technique to closely estimate the achievable information rates for channels with insertions and/or deletions with or without intersymbol interference when i.i.d. inputs are employed while we also provide some notes on the achievable information rates when Markov inputs are used. We emphasize that our method is useful for evaluating information rates for channels with insertion/deletions with additional impairments where there does not seem to be a hope of obtaining fully analytical results.

In this work, we propose to use trellis structures to characterize channels with insertions, deletions and substitutions. We start with binary-input binary-output channels with either insertions or deletions, and develop a simulation based approach to estimate the corresponding information rates. This approach is then generalized to the case where there exist both insertions and deletions. We show that, the proposed algorithm is an efficient and flexible technique to closely estimate the information rates for channels with insertions, deletions and substitutions.

We consider the problem of timing recovery in magnetic recording channels based on MAP estimation of the timing information. The read and write clocks are modeled as random walk processes that allow for slowly varying phase and frequency offsets of the clocks. We propose a new timing error detector (TED) that provides sufficient statistics about the instantaneous timing error. Using the clock models and the new TED, the MAP estimates of the sampling times are derived. This method is shown to be more robust than the conventional algorithm based on the Mueller and Muller TED and easily implementable for a small additional complexity.

In this work, we use various trellis structures to characterize insertion/deletion channels with intersymbol interference (ISI). We propose to use a two-element state vector to represent the channel memory and possible insertions/deletions, and develop a suitable trellis to describe the communication process. Using a simulation based approach, we compute the corresponding information rates. We show that, the proposed algorithm is an efficient and flexible technique to closely estimate the information rates for insertion/deletion channels with ISI.

Heat-assisted magnetic recording is a promising approach for enabling large increases in the storage density of hard disk drives. A laser is used to momentarily heat the recording area of the medium to reduce its coercivity below that of the applied magnetic field from the recording head. In such a system, the recording materials have a very high magnetic anisotropy, which is essential for the thermal stability of the magnetization of the extremely small grains in the medium. This technology involves new recording physics, new approaches to near field optics, a recording head that integrates optics and magnetics, new recording materials, lubricants that can withstand extremely high temperatures, and new approaches to the recording channel design. This paper surveys the challenges for this technology and the progress that has been made in addressing them.

This paper studies bit-patterned media recording channels when there exist written-in errors. Using a signal processing point of view, we consider an abstract model for the overall recording channel, incorporating errors occurred during the write process. Based on the resulting channel model, we develop various trellis structures, and propose several detection algorithms for data recovery. Furthermore, we investigate the achievable information rates for patterned media recording systems with different levels of write error probability, which provide an information theoretical performance assessment for such systems.

We present a family of equalizers and targets for certain inter-symbol interference (ISI) channels with the property that the performance of a maximum-likelihood (ML) or maximum a posteriori (MAP) based detector for the target channel is also simultaneously optimal for the equalized channel. In particular, the MMSE Decision feedback equalizer (DFE) belongs to this family of filters. Although, these solutions are infinite impulse response (IIR) filters, we can achieve good performance using finite impulse response (FIR) filters. We present an algorithm for designing equalizers and FIR targets that minimize the probability of sequence detection error.

Timing recovery is crucial for magnetic recording systems. A conventional timing recovery loop (also known as the phase-locked loop) consists of a timing error detector (TED), a loop filter, and a voltage controlled oscillator (VCO), all of which process the samples in a sequential manner. This sequence of operations in the timing recovery loop performs well if the timing error is a small fraction of the bit interval. However, in the cycle-slip regions, the timing error is comparable to the bit interval, and the loop fails. In this paper, we represent the timing error in magnetic recording systems using a discrete Markov model that does not confine the timing error to only small fractions of the bit interval. By utilizing such a model, we derive an optimal baud-rate processing unit that does not perform tasks in sequence, but jointly. The derived unit has a similar structure as the classical first-order phase-locked loop (PLL). Simulation results show that the new detector outperforms the standard Mueller and Muller phase-locked loop. This performance gain is substantial if the timing error process is extremely noisy or if there is residual frequency-offset. For moderately low-noise timing errors without residual frequency-offset, the improvement over the Mueller and Müller phase-locked loop is just marginal.

A (full-complexity) per-survivor iterative timing recovery scheme, which jointly performs timing recovery, equalization, and error-correction decoding, was recently proposed [1] to deal with the problem of timing recovery operating at low signal-to-noise ratio. Although it outperforms other iterative timing recovery schemes, it has very high complexity. In this paper, we propose a reduced-complexity per-survivor iterative timing recovery scheme to make it implementable, in real-life applications. Simulation results indicate that for low to moderate complexity, the reduced-complexity scheme provides a better performance than the full-complexity scheme.

The paper gives a brief description of the data storage technologies that are expected to appear in the near future, and the challenges these new systems bring to designing robust read channels. We focus on perpendicular, heat assisted magnetic recording and probe storage systems.

We propose a new iterative timing recovery scheme based on per-survivor processing that jointly performs timing recovery and turbo equalization on partial response channels with error-correction codes. The scheme embeds the timing recovery process inside the Bahl, Cocke, Jelinek, and Raviv (BCJR) equalizer, using per-survivor processing. This per-survivor BCJR equalizer then iteratively exchanges soft information with an error-correction decoder. Results indicate that the proposed scheme yields a better performance than a conventional receiver that performs timing recovery and turbo equalization separately, especially when the channel encounters severe timing jitter noise. We also present evidence that suggests that the proposed scheme can correct a cycle slip much more efficiently than the others.

A conventional receiver performs timing recovery and equalization separately. Specifically, conventional timing recovery is based on a phase-locked loop that relies on the decision provided by its own symbol detector. We propose a new timing recovery scheme based on per-survivor processing (PSP) that jointly performs timing recovery and equalization for uncoded partial response channels. In the proposed scheme, each survivor of the Viterbi algorithm maintains its own estimate of the timing offset, and this estimate is updated according to the history data associated with the survivor path. As compared to conventional timing recovery at BER = 10/sup -4/, the proposed scheme can provide a 0.5 dB gain in SNR.

It is possible to obtain either exact realizations or useful approximations of linear systems or matrix-vector products arising in many different applications, by synthesizing them in the form of repeated or multi-channel filtering operations in fractional Fourier domains, resulting in much more efficient implementations with acceptable reduction in accuracy. By varying the number and configuration of the filter blocks, which may take the form of arbitrary flow graphs, it is possible to trade off between accuracy and efficiency in the desired manner. The proposed scheme constitutes a systematic way of exploiting the information inherent in the regularity or structure of a given linear system or matrix, even when that structure is not readily apparent.

The extended versions of common Laplace and Fourier transforms are given. This is achieved by defining a new function f/sub e/(p), p/spl epsiv/C related to the function to be transformed f(t), t/spl epsiv/R. Then f/sub e/(p) is transformed by an integral whose path is defined on an inclined line on the complex plane. The slope of the path is the parameter of the extended definitions which reduce to common transforms with zero slope. Inverse transforms of the extended versions are also defined. These proposed definitions, when applied to filtering in complex ordered fractional Fourier stages, significantly reduce the required computation.