Research on reconstructing original data from data that can only be partially observed due to noise or missing data has been ongoing for many years. Such problems are generally referred to as matrix estimation problems. The problem can be formulated when the data to be estimated can be defined as matrix variables in the problem of reconstructing partially observed data. When the properties of the target matrix are unknown, a common approach is a method called matrix rank minimization, which is known for its high estimation accuracy in various fields such as audio, image, and wireless communication. However, this method assumes that the data belong to a linear subspace, and if this assumption is not satisfied, the estimation accuracy significantly deteriorates. Therefore, in recent years, this assumption has been extended to manifolds, and various methods based on this assumption have been proposed. This paper reviews the progress of these methods and describes the latest techniques in this field.

We consider design of sparse IIR filters based on the least absolute shrinkage and selection operator (LASSO) for reducing the computation cost of IIR filters. The design algorithm is summarized as follows. First, we solve a constrained quadratic programming problem where the constraint is the positive realness with maximum pole radius. Next, after fixing the denominator coefficients, we apply the LASSO to the filter design problem so that the numerator coefficients include zero. Finally, we solve a constrained quadratic programming with zero coefficients constraints. Moreover, we confirm the effectiveness of the proposed method by using two design examples.

In this paper, an algorithm for array shape calibration is proposed. Our method uses two pilot signals. In the proposed method, we assume that the direction of arrival (DOA) of one pilot signal is available, but the DOA of the other is unknown. Then, we estimate the unknown DOA of the pilot signal and the location errors of a microphone array simultaneously. To estimate the unknown parameters, we firstly optimize a cost function derived from the phases of the signal subspace that is obtained from observed signals. The optimal solution obtained from the cost function is a function of the unknown DOA. Second, assuming that the location errors of a microphone array follow the Gaussian distribution, the DOA of the pilot signal is estimated by minimizing the squared sum of the obtained optimal solution. Simulations and the experiment in a real environment validate our algorithm.

In this paper, we propose a method of embedding images into a speech signal. An image is embedded in the high-frequency band above 16kHz of the speech signal to be difficult to perceive for humans. In the proposed method, the image is embedded after being binarized by the error diffusion method. When the image is embedded, we apply an iterative phase restoration method to avoid noise generation. We also investigate a compensation method for an extracted image whose quality degrades due to transmission channel characteristics. Simulation and real-environmental experiments showed that the proposed method significantly improved the quality of the extracted image on the receiver side.

Sound source localization is carried out based on TDOA (Time-Difference-Of-Arrival) between microphones. When the TDOA is detected from a phase difference between two microphones, an ambiguity of phase difference often occurs under a high resolution situation and thus the estimation accuracy degrades significantly. In this paper, a propagation delay characteristic is approximated directly using FIR (Finite Impulse Response) filter, in which an impulse response is described by sinc function and an adjustable parameter is just a TDOA. Although there is not the ambiguity of phase difference in this procedure, an appropriate TDOA has to be searched using any optimization because the TDOA takes a real value. This falls into a non-linear optimization problem and there are many local minima in the search space therefore, a local minimum stagnation often occurs when applying a normal steepest decent algorithm for the search. In the proposed method, TDOA is estimated by introduction an avoidance method of local minimum stagnation. The validity of the proposed method is shown from several experimental results.

In recent years, defect detection and classification using machine learning as an alternative to visual inspection has been studied. In this paper, we propose a method for defect detection by taking the difference between pseudo-images generated using CycleGAN and the original images. Compared to the conventional detection method using binarization, our proposed method can detect defects independent of the shooting environment, thus significantly reducing the risk of overlooking defects.

In this paper, we investigate the analytical definition of the threshold voltage of nanosheet MOSFETs using the quantum drift diffusion (QDD) model. The QDD model is a device simulator that can simulate the quantum confinement effects in the inversion layer of advanced MOSFETs. We have analyzed three-dimensional characteristics of a nanosheet MOSFET with a sheet width of 10 nm, a sheet thickness of 4 nm, and an equivalent oxide thickness of 0.35 nm in the QDD model. We have investigated the analytical definition of threshold voltage in the nanosheet MOSFET using the Q_TH definition proposed by Takeuchi et al. and the 2ψ_B definition that has been widely used in the past. For the nanosheet MOSFET, the threshold voltage in the Q_TH definition is 0.40 V, and the threshold voltage in the 2ψ_B definition is obtained to be 0.19 V. The threshold voltage by electrical definition is 0.36 V. The difference is 0.04 V by the Q_TH definition, and 0.17 V by the 2ψ_B definition. In nanosheet MOSFETs, the threshold voltage obtained by the Q_TH definition is closer to the threshold voltage obtained by the electrical definition than the threshold voltage obtained by the 2ψ_B definition.

Inexperienced drivers, such as the elderly and novice drivers, are prone to cause traffic accidents due to human error. Autonomous driving is expected to reduce traffic accidents by assisting their recognition, judgment, and operation. However, it is only effective in situations where drivers can easily make decisions, such as driving on highways, and is still challenging in urban areas. In this paper, we focus on safe and efficient autonomous driving in situations where multiple moving obstacles simultaneously exist, assuming an urban intersection. Since it is difficult to construct a driving model for such a situation, we introduced a reinforcement learning method that does not require a driving model. This paper proposes a collision-avoidance problem as a multi-objective sequential decision-making problem. We propose a method for learning a non-convex Pareto front concerning safety and speed using the multi-objective reinforcement learning algorithm, Pareto-DQN. The proposed method's performance through computer experiments is verified in a T-intersection environment. We confirmed the acquisition of multiple Pareto-optimal driving policies that could not be achieved using conventional methods with linear scalarization. The proposed method is helpful for system designers because it provides a more detailed representation of the driver's non-convex preferences.

Pedestrian Dead Reckoning (PDR) is used as a method to estimate pedestrian's positions when the satellite positioning is not available. Previous studies on PDR used experimentally determined parameters and models. In this paper, in order to avoid the use of experimentally determined parameters and models, we propose a method to construct a PDR system suitable for pedestrians individually by utilizing the satellite positioning data effectively. PDR generally requires three steps: walking detection, step length estimation, and heading angle estimation. The proposed method constructs a walking false detection prevention system and estimates the average step length when the satellite positioning is available. Once the satellite positioning becomes unavailable, the PDR system adaptively constructed for the individual pedestrian starts to work. From experimental results, we can see that the proposed method works well and can provide accurate positioning results even when the satellite positioning is not available.

Managing disruptions on railway networks with high-density train services can be challenging for rescheduling crew members. It is crucial to reschedule their duties flexibly to ensure reliable services. However, conventional approaches to crew rescheduling can make it difficult for resource dispatchers to find appropriate solutions as they provide fixed objectives and constraints. To address this, we have developed a user assistance system that enables dispatchers to compare feasible crew plans and select one that suits the current situation. The system employs two techniques, zero-suppressed binary decision diagrams and satisfying trade-off methods, to achieve this goal. Testing on a medium-sized railway line showed that the system outperformed the conventional approach in terms of obtaining a user-preferred crew plan. The results indicate that the system is effective in interactive comparison situations, assisting resource dispatchers in improving their tasks and ensuring reliable railway services even during disruptions.

As a part of the smart lander for investigating moon (SLIM) project, we are studying the method to estimate the three-dimensional position of the lander accurately. The conventional method based on image matching finds the correspondence between the known map and captured images using the local features detected from these images and then estimates the lander position by extracting essential local features. By adjusting the hyper-parameter appropriately, the conventional method shows an adequate estimation accuracy for the SLIM project. However, the conventional method is not robust against changes in map image and hyper-parameter value. As a robust estimation algorithm, we propose a Bayesian self-localization method that simultaneously estimates the three-dimensional position and the hyper-parameter according to the observed data. In this paper, we design a Bayesian model for self-localization by regarding the self-localization problem as a constrained optimization problem. Moreover, we derive the variational Bayesian algorithm for Bayesian estimation of the lander position. Our numerical simulations targeting two essential regions on the moon for the SLIM project demonstrate that the proposed and conventional method can adequately estimate the lander position and that the proposed method is more robust against changes in map image and hyper-parameter value.