Toward intelligent wireless communications: Deep learning - based physical layer technologies

Advanced technologies are required in future mobile wireless networks to support services with highly diverse requirements in terms of high data rate and reliability, low latency, and massive access. Deep Learning (DL), one of the most exciting developments in machine learning and big data, has recently shown great potential in the study of wireless communications. In this article, we provide a literature review on the applications of DL in the physical layer. First, we analyze the limitations of existing signal processing techniques in terms of model accuracy, global optimality, and computational scalability. Next, we provide a brief review of classical DL frameworks. Subsequently, we discuss recent DL-based physical layer technologies, including both DL-based signal processing modules and end-to-end systems. Deep neural networks are used to replace a single or several conventional functional modules, whereas the objective of the latter is to replace the entire transceiver structure. Lastly, we discuss the open issues and research directions of the DL-based physical layer in terms of model complexity, data quality, data representation, and algorithm reliability.     

Optimal Antibody Purification Strategies Using Data-Driven Models

This work addresses the multiscale optimization of the purification processes of antibody fragments. Chromatography decisions in the manufacturing processes are optimized, including the number of chromatography columns and their sizes, the number of cycles per batch, and the operational flow velocities. Data-driven models of chromatography throughput are developed considering loaded mass, flow velocity, and column bed height as the inputs, using manufacturing-scale simulated datasets based on microscale experimental data. The piecewise linear regression modeling method is adapted due to its simplicity and better prediction accuracy in comparison with other methods. Two alternative mixed-integer nonlinear programming (MINLP) models are proposed to minimize the total cost of goods per gram of the antibody purification process, incorporating the data-driven models. These MINLP models are then reformulated as mixed-integer linear programming (MILP) models using linearization techniques and multiparametric disaggregation. Two industrially relevant cases with different chromatography column size alternatives are investigated to demonstrate the applicability of the proposed models.     

Trajectory optimization and positioning control for batch process using learning control

Efficiency and accuracy are critical in the motion control of a batch process. This paper proposes a new intelligent motion control method for a batch process based on reinforcement learning (RL) and iterative learning control (ILC). The proposed learning-based motion control method enables the system to learn from its previous experience. The motion control method can be divided into two parts: (1) RL-based trajectory optimization and (2) ILC-based positioning control. Experiments were conducted to demonstrate the effectiveness of the proposed method. The results indicate that the proposed method not only reduces the process time effectively while ensuring system stability, but also achieves excellent positioning accuracy.     

Hybrid water treatment cost prediction model for raw water intake optimization

In order to reduce the total cost of a dual source drinking water treatment plant operation, a comprehensive hybrid prediction model was built to estimate the necessary chemicals dosage and pumping energy costs for alternative source selection scenarios. Correlations between the water quality parameters and the required treatment chemicals were estimated using historical data and the expected pH variations associated with each chemical addition, which was based on the Caldwell–Lawrence diagram. The pumping energy costs were also estimated using a data-driven approach that was based on historical plant data. The research has practical implications for water treatment operators seeking to minimize plant operational costs through selecting raw water intake volumes for their treatment plant based on multiple source options and offtake tower gate levels. Future research seeks to better link current and future water treatment dosage cost predictions directly to water quality measurements taken from vertical profiling systems.     

A KPI-based process monitoring and fault detection framework for large-scale processes

Large-scale processes, consisting of multiple interconnected subprocesses, are commonly encountered in industrial systems, whose performance needs to be determined. A common approach to this problem is to use a key performance indicator (KPI)-based approach. However, the different KPI-based approaches are not developed with a coherent and consistent framework. Thus, this paper proposes a framework for KPI-based process monitoring and fault detection (PM-FD) for large-scale industrial processes, which considers the static and dynamic relationships between process and KPI variables. For the static case, a least squares-based approach is developed that provides an explicit link with least-squares regression, which gives better performance than partial least squares. For the dynamic case, using the kernel representation of each subprocess, an instrument variable is used to reduce the dynamic case to the static case. This framework is applied to the TE benchmark process and the hot strip mill rolling process. The results show that the proposed method can detect faults better than previous methods.     

Linking bottom-up intonation stylization to discourse structure

A new approach for intonation stylization that enables the extraction of an intonation representation from prosodically unlabeled data is introduced. This approach yields global and local intonation contour classes arising from a contour-based, parametric and superpositional intonation stylization. Based on findings about the linguistic interpretation of the contour classes derived from corpus statistics and perception experiments, we created simple prediction models for the partial generation of intonation contours from discourse structure defined by discourse segment boundaries and the information status of nouns within these segments. The predicted intonation contours were evaluated by human judgments of adequacy that yielded a high accordance.     

Improving partial mutual information-based input variable selection by consideration of boundary issues associated with bandwidth estimation

Input variable selection (IVS) is vital in the development of data-driven models. Among different IVS methods, partial mutual information (PMI) has shown significant promise, although its performance has been found to deteriorate for non-Gaussian and non-linear data. In this paper, the effectiveness of different approaches to improving PMI performance is investigated, focussing on boundary issues associated with bandwidth estimation. Boundary issues, associated with kernel-based density and residual computations within PMI, arise from the extension of symmetrical kernels beyond the feasible bounds of potential inputs, and result in an underestimation of kernel-based marginal and joint probability distribution functions in the PMI. In total, the effectiveness of 16 different approaches is tested on synthetically generated data and the results are used to develop preliminary guidelines for PMI IVS. By using the proposed guidelines, the correct inputs can be identified in 100% of trials, even if the data are highly non-linear or non-Gaussian.     

High Ash Char Gasification in Thermo-Gravimetric Analyzer and Prediction of Gasification Performance Parameters Using Computational Intelligence Formalisms

The coal gasification is a cleaner and more efficient process than the coal combustion. Although high ash coals are commonly utilized in the energy generation, systematic gasification kinetic studies using chars derived from these coals are scarce. Accordingly, this paper reports the development of the data-driven models for the gasification of chars derived from the high ash coals. Specifically, the models predict two important gasification performance parameters, viz. gasification rate constant and reactivity index. These models have been constructed using three computational intelligence (CI) methods, namely genetic programming (GP), multilayer perceptron (MLP) neural network (NN), and support vector regression (SVR). The inputs to the CI-based models consist of seven parameters representing the gasification reaction conditions and properties of high ash coals and chars. The data used in the modeling were collected by performing extensive gasification experiments in the CO₂ atmosphere in a thermo-gravimetric analyzer (TGA) using char samples derived from the Indian coals containing high ash content. Values of the two gasification performance parameters were obtained by fitting the experimental data to the shrinking unreacted core (SUC) model. It has been observed that all the CI-based models possess an excellent prediction accuracy and generalization capability. Accordingly, these models can be gainfully employed in the design and operation of the fixed and fluidized bed gasifiers using high ash coals.     

Data-driven stochastic subspace identification of flutter derivatives of bridge decks

Most of the previous studies on flutter derivatives have used deterministic system identification techniques, in which the buffeting forces and the associated responses are considered as noises. In this paper, one of the most advanced stochastic system identification, the data-driven stochastic subspace identification technique (SSI-DATA) was proposed to extract the flutter derivatives of bridge decks from the buffeting test results. An advantage of the stochastic method is that it considers the buffeting forces and the responses as inputs rather than as noises. Numerical simulations and wind tunnel tests of a streamlined thin plate model conducted under a smooth flow by the free decay and the buffeting tests were used to validate the applicability of the SSI-DATA method. The results were compared with those from the widely used covariance-driven SSI method. Wind tunnel tests of a two-edge girder blunt type of Industrial-Ring-Road Bridge deck (IRR) were then conducted under both smooth and turbulent flows. The identified flutter derivatives of the thin plate model based on the SSI-DATA technique agree well with those obtained theoretically. The results from the thin plate and the IRR Bridge deck helped validate the reliability and applicability of the SSI-DATA technique to various experimental methods and wind flow conditions. The results for the two-edge girder blunt type section show that applying the SSI-DATA yields better results than those of the SSI-COV. The results also indicate that turbulence tends to delay the onset of flutter compared with the smooth flow case.     

Direct multivariable controller tuning for internal combustion engine test benches

Dynamical test benches are typically used in the development phase of engine systems and require tracking controllers with a high performance. Unfortunately, during such a work the components or operation parameters of the engine system are changed very frequently, making the use of classical model based control approaches very time-consuming. Against this background, this paper proposes a direct data-driven design approach for multivariable control of rotational speed and shaft torque of an internal combustion engine at a test bench based on an extended version of a recently introduced method for non-iterative direct data-driven tuning of multivariable controllers. This extension allows employing data collected in a closed-loop experiment in the direct identification of the controller parameters. The effectiveness of the proposed approach is shown on a test bench equipped with a production light duty Diesel engine. A comparison with the industrial state-of-the-art controller is provided on both a dynamically challenging test and a typical driving cycle as measured on an instrumented vehicle with the same internal combustion engine. The results confirm that the new method recovers the performance of the well-tuned industrial control, but can be developed in a fraction of the time as no explicit model of the system is needed.