Queuing methods in reliability theory, computer and communication systems
Using the methods of queueing theory we aim to investigate the performance measures of complex info-communication systems including wireless networks, sensor networks, and cognitive radio networks with the help of analytical, numerical approaches combined with tool-supported analysis. The research carried out by the group is focused on the so-called finite-source queueing systems. Its origin dates back to the machine interference problem, sometimes called the machine-repairman model. It was later applied for the performance modeling of time-shared terminal and repairable complex systems, just to mention some.
Modeling of complex systems
The properties of many-particle systems cannot be analyzed analytically, therefore one has to apply the tools of statistical physics. Several systems in our natural, technical, and sociological environment are based on complex networks (i.e. social-, transport-, computer-, cooperation- and metabolic networks). Complex networks have special characteristics: scale-free and ’small-world’ properties etc. The topological modeling of them and computer simulation of their processes are important and useful in science.
Computer simulation of spreading processes
In this topic, we investigate diffusion phenomena that is present in many fields of science. First, we use computer simulations and analytical calculations during our work. The observed phenomenon may be information spreading on a social network, the spreading of a new technology, or even an information packet on a computer network. in our work usually we use discrete time and several different network topologies. Our results are published in international journals and at high noted conferences.
Development of Ultrasound-CT
We have studied different experimental UHCT configurations the necessary basic instrumentation (uC, NI PXI, FPGA controlled), algorithms and software environments resulting sufficient experience for further work towards using experimental arrangements with a higher number of 3D arranged transmitter-receiver pairs in fluid as a coupling medium.
Development of Distributed Sensor Network for meteorological observations and numerical weather Prediction Calculations
The prediction of weather generally means the solution of differential equations on the base of the measured initial conditions where the data of close and distant neighboring points are used for the calculations. It requires the maintenance of expensive weather stations and supercomputers. However, if weather stations are not only capable of measuring but can also communicate with each other, then these smart sensors can also be applied to run forecasting calculations. This applies the highest possible level of parallelization without the collection of measured data in one place. Furthermore, if more nodes are involved, the result becomes more accurate, but the computing power required from one node does not increase. Our Distributed Sensor Network for meteorological sensing and numerical weather Prediction Calculations (DSN-PC) can be applied in several different areas where sensing and numerical calculations, even the solution of differential equations, are needed.
e-Health and Ambient assisted living systems
Our research aims to assist elderly or sick people in everyday independent activities using the latest assistive technologies based on the Internet of Things. The system consists of three major components: an intelligent ambient assisting living system, the health and activity monitoring and recognition system, and an assistive robot providing personal assistance. This is a complex support system, which must have learning and adaptive behavior, and for this reason, we plan to use artificial neural networks.
Intelligent embedded systems design and applications
We designed a hardware-software co-design platform based on FPGA, developed for fast prototyping of embedded systems using hardware modules that can be easily connected and “driver” modules that can manage I/O devices and sensors basic behavior. Using neural networks (NN) to add learning capabilities and adaptive behavior is essential for an intelligent system and the use of FPGA is an important feature in terms of their hardware implementation. The designed architecture allows the insertion of intelligent interfaces based on neural networks created with this type of module. This platform is based on low-cost general-purpose FPGA boards without the need for hardware design. Among possible applications are intelligent computer peripherals enabling people with any kind of handicap to use a computer and communicate, as any kind of industrial or domestic device with embedded and hidden intelligence at user for prosthetic, automotive, “domotic” and automation fields where the trend is to produce easy-to-use devices.
Analysis and modeling of low Earth orbit (LEO) satellite broadband communication services
Satellite network architectures, radio channel mechanisms, QoS characteristics (data rate and latency), cognitive radio techniques, environmental regulation (space debris, international spectrum use), and data security risks.
Performance analysis of the QUIC (Quick UDP Internet Connections) transport mechanism
TCP/UDP/QUIC comparison, congestion control solutions, connection setup, packet loss and recovery methods, QoS solutions, characterization of the encryption used, data protection, and tracking capabilities. The combined effectiveness of HTTP/3 and QUIC for Content Delivery Networks (CDN). QUIC and the Internet of Things (IoT) for low throughput networks. Modeling interoperability problems between mobile networks and QUIC. QUIC adoption processes for popular services.
Analysis of supercomputers (HPC) and parallel resource use
Perform highly complex problems, simulations, and Big Data analysis quickly and efficiently. Parallel computing architectures and algorithms: use of distributed and parallel algorithms, handling heterogeneous systems: sharing CPU and GPU for processing. Combining GPUs and CPUs. Scalability and performance optimization on HPC: network latency, synchronization costs, data sharing, dynamic load balancing strategies, and power consumption optimization. Fault tolerance techniques in parallel systems, checkpointing, and error recovery to save and recover long-running jobs. Supercomputing and cloud integration: scaling resources between cloud and supercomputing infrastructures. Hybrid HPC systems design techniques.
Wireless Sensor Networks (WSN) analysis for environmental monitoring, industrial automation, healthcare applications, and smart cities
Energy efficiency and sensor node battery optimization: energy efficiency to optimize communication protocols, data collection, and network reliability and performance. Analysis of sleep modes and wake-up strategies to maximize lifetime. Data compression and aggregation, decentralized data processing techniques. WSN network reliability and fault management: fault tolerance and self-repair, redundancy and resource optimization. WSN security and privacy: secure communication protocols, privacy and anonymity. Scaling of WSN systems and large-scale networks, hierarchical network design techniques. New communication technologies and integration of WSN: 5G, B5G, 6G, mesh networking techniques, modeling.
Investigating the throughput performance of multipath communication
The current Internet communication environment allows only a single path for data transmission in a communication session. The single path assumption is quite acceptable for systems, which use a single connection interface, or a “single exit point” to the Internet. On the other hand, a lot of currently used devices have factory built-in multiple network interfaces: RJ-45 for the wired network, RF interface for the Wi-Fi wireless network connection, and mobile phone data transfer connection interface (e.g. 3G, HSDPA or LTE). The single path communication technology is not able to use the advantages of the multiple interfaces. The communication performance (e.g. throughput) could be highly improved if the networking environment would be able to support the usage of multiple paths for a communication session.
Implementation of artificial neural networks in field-programmable gate arrays
The research aim is to design hardware implemented artificial neural networks (ANN) using reconfigurable hardware (FPGAs). For hardware implementation, we can choose between HDL description of the NN, high-level synthesis tools (HLS), and a new method that I have developed using the System Generator tool by Xilinx. This method allows the easy generation of hardware description Language (HDL) code from a system representation in Simulink. This HDL design can be synthesized for implementation in the Xilinx family of FPGA devices. In this way is possible to create “application specific neural networks” in an easy and fast time to market way. We have implemented and trained in many different ways Feed-Forward ANN and also competitive neural networks. Use case: Application of hardware implemented ANN in activity pattern recognition.