Seminar
Selected Topics in Sustainable Communication Networks
January 2024
Deep Learning optimized tunable Metasurface Parameters
Speaker: Johannes Müller
Reconfigurable metamaterials (MTMs) allow for manifold applications in Magnetic Resonance Imaging (MRI). Most importantly, the full control in the spatio-temporal domain enables the realization of arbitrary sensitivity patterns during the transmit (Tx) as well as the receive (Rx) phase of an MRI scan via (re-)shaping the magnetic field distribution. The optimization of linking the reconfigurable N degrees of freedom to a desired field profile, however, is a non-trivial problem. A one-dimensional prototype (N = 14) is presented that is digitally controlled in combination with deep learning (DL)-driven optimization. Both, the forward and inverse problem are addressed. The forward problem, involving the mapping of capacitance values to a desired magnetic field distribution, is well-posed and can be effectively solved through a variety of methods. The inverse problem, i.e., deriving capacitance values from observed magnetic fields, is tackled by deep learning, specifically through a Multi-Layer Perceptrons (MLPs) approach. The feasibility of training a neural network with simulation data is demonstrated. Deep learning-driven optimization of MTMs in MRI applications holds a huge potential for many applications in the future.
WUBBLE: Energy Efficient BLE Neighborhood Discovery Leveraging Wake-up Radio
Speaker: Damien Wohwe Sambo
In wireless sensor networks, much energy is wasted during connecting and control processes. This is particularly true for the neighborhood discovery process carried out in Bluetooth Low Energy (BLE) before the communications between devices. However, the fixed duration of this process has a critical energy cost when the number of neighbors is low or when only a few neighbors are required by the application. In order to address this issue, a novel protocol called WUBBLE is proposed to leverage wake-up radio technology to start and end the discovery process. Wake-up radio enables additional communications between devices with an ultra-low energy overhead. WUBBLE is validated by combining analytical analysis and experimental measurements, and results show that half of the energy could be gained to discover 90\% of the neighbors.
December 2023
A Utility function for Collective Perception Messages in VANETs
Speaker: Thenuka Karunathilake
Collective Perception (CP) services in Vehicular Ad-hoc Networks (VANETs) enhance safety by boosting awareness regarding both V2X-enabled and non-V2X-enabled objects. This is especially crucial during the transition period before the widespread adoption of V2X-enabled vehicles. Recent research affirms the capability of CP services to significantly enhance awareness.
However, these services come with certain drawbacks. One notable challenge is the requirement for additional channel capacity to transmit the generated CP messages (CPMs). Another challenge is the demand for processing power at the receiving vehicle. Our research focuses on the latter aspect, as understanding how to effectively utilize these received CPMs is essential. Consequently, we propose a utility function to assess the overall utility of received CPMs.
Sensing the Unknowns: A Study on Data-Driven Sensor Fault Modeling and Assessing its Impact on Fault Detection for Enhanced IoT Reliability
Speaker: Shadi Attarha
In the context of the Internet of Things (IoT), the effective operation of IoT applications heavily relies on the functionality of sensors.
These sensors are prone to failures or malfunctions due to various factors, including adverse environmental conditions and aging components within sensors. To mitigate the impact of faulty sensors on system performance, notable research has focused on employing machine-learning techniques to detect faulty sensor data. In this context, due to the scarcity of real faulty data records and challenges in generating them even in controlled environments, researchers often model faulty data to create synthetic datasets containing normal and abnormal data for evaluating fault detection models. According to our empirical findings, the present modelling approach for simulating defective sensor scenarios does not effectively reflect the complexity of real-world faulty sensor behaviours. As a result, greater research into sensor fault models is required to improve the efficacy of fault detection algorithms in actual applications. To close this gap, we compared existing fault models and developed a novel composite technique for modelling defective sensor behaviours that can more successfully represent real-world sensor behaviours.
Our goal was to see how different fault models affect the effectiveness of anomaly detection systems in real-world circumstances. Algorithms were trained on synthetic datasets produced from several fault models, and their performance in identifying real-world defective data was evaluated. We also offer a wide range of labelled datasets, including normal and pathological data obtained from real-world applications.
Self Monitoring Sensor Nodes for Critical IoT Applications
Speaker: Saurabh Band
Some IoT applications are deployed in a remote environment, making it difficult/very expensive to retrieve the sensor nodes if any problem occurs. Moreover, if the application is life-critical, the system must fulfill specific requirements, which might not be crucial for IoT systems in other day-to-day applications. These applications with similar conditions include monitoring systems for extraterrestrial habitats, mining sites, mobile base stations in remote areas, etc. Due to these reasons, it is important to have a robust and reactive system to any form of failure. In this talk, we discuss the problems such systems face and how to tackle them in real life.
November 2023
Where is the Wolf? About Animal Detection in the Wild
Speaker: Jens Dede
Artificial intelligence -- especially in the area of machine vision -- becomes more and more important. It helps to find, localize and identify all possible objects in images and videos. When working with automatic image capturing, for example, using camera traps, the detection becomes more challenging: Instead of dealing with high-quality, glossy images, as in most other applications, the camera trap images are often of poor quality. Blurry, dark, hidden or barely visible objects of interest hinder the detection. This presentation will discuss those challenges and present some possible solutions using the example of wolf detection.
Protocol parameter optimization in LoRaWAN with Q learning
Speaker: Piumika Karunanayake
LoRaWAN is a technology which enables to connect long distance nodes fulfilling the requirements of IoT. As thousands of nodes are connected to network servers through gateways, it is important to optimize transmission power and increase the packet delivery ratio by changing physical parameters such as spreading factor and code rate. Adaptive data rate mechanism is incorporated in LoRaWAN by monitoring the signal to noise ratio at the network server and optimizes transmission power and the spreading factor. In this work, Q learning agent is used to tune three parameters, transmission power, spreading factor and code rate with the object of minimizing the power consumption of end nodes with improving packet delivery ratio. In addition, performance observation window size is also changed with the original protocol to obtain better performance.
July 2023
Wednesday,
July 12
16:00 - 16:30
Processing of natural languages with limited resources for information retrieval systems
Speaker: Prof. Paul Dayang
Many African languages fall into the group of resource scarce languages. Therefore, there is a need to explore and build more specialised information systems that enable speakers of African languages to discover valuable information across linguistic and cultural barriers. Our aim is to lay some bricks towards the development of tools for the automatic processing of languages with low digital resources. At the middle of this work is the Natural Language Processing (NLP).
AI and Sustainable Agriculture in Sub-Saharan Africa: Connecting the dots to reach farmers!
Speaker: Prof. Jean Louis Fendji Ebongue Kedieng
Sustainable agriculture amounts to providing novel solutions to farmers and extension workers that use resources wisely and promote biodiversity while increasing the yield. Performing sustainable agriculture requires paying attention to the whole lifecycle of crops. Several processes can be improved by leveraging advances in AI, mobile phones, sensors, and drones. But agriculture in Sub-Saharan Africa (SSA) is generally performed in regions experiencing a lack of infrastructure such as road, electricity, Internet, in addition to weak purchase power and low literacy level of farmers that constitute a barrier to digital service adoption. Moreover, the lack of good quality data prevents the leverage of AI-based techniques, despite the efforts to produce relevant information on crops such as the PlantVillage project. Beyond developing models and evaluating their performance, AI-based solutions should overcome aforementioned limitations to reach farmers. This talk will attempt to identify some gaps and provide ideas to connect the dots to reach farmers.
Self-Monitoring Sensor Nodes for Critical IoT Applications
Speaker: Saurabh Band
Some IoT applications are deployed in a remote environment, making it difficult/very expensive to retrieve the sensor nodes if any problem occurs. Moreover, if the application is life-critical, the system must fulfill specific requirements, which might not be crucial for IoT systems in other day-to-day applications. These applications with similar conditions include monitoring systems for extraterrestrial habitats, mining sites, mobile base stations in remote areas, etc. Due to these reasons, it is important to have a robust and reactive system to any form of failure. In this talk, we discuss the problems such systems face and how to tackle them in real life.
June 2023
Wednesday,
June 28
16:00 - 17:00
WiFi Channel State Information (CSI) for Activity Sensing
Speaker: Prof. Dileeka Dias
Channel State Information in OFDM-based communications systems, Channel State Information (CSI) provides the channel frequency response at each of the sub-carrier frequencies.Thus, CSI contains rich information of the channel that incorporates detailed characteristics of the multipath environment. This makes it possible for sensing physical phenomena leading to activity detection, recognition and estimation in the environment between an OFDM transmitter and a receiver. WiFi, being ubiquitous and using OFDM in its physical layer, has recently been widely adopted for interesting sensing applications ranging from human tracking to activity detection to respiration estimation. The availability of low-cost commercial off-the-shelf (COTS) WiFI devices that enable CSI extraction open up a variety of powerful embedded applications at the edge.
The talk will provide an overview of CSI signal processing and sensing algorithms, and will extend to some recent work and experiences on their applications in contrasting real-world environments such as pedestrian detection and industrial coolers. Implementation of CSI-based applications in COTS WiFi devices will also be discussed. Emerging research areas related to CSI-based sensing will be presented.
Wednesday,
June 21
16:00 - 17:00
Physical layer parameter optimization in LoRaWAN with Q learning
Speaker: Piumika Karunanayake
LoRaWAN is a technology which enables to connect long distance nodes fulfilling the requirements of IoT. As thousands of nodes are connected to network servers through gateways, it is important to optimize transmission power and increase the packet delivery ratio by changing physical parameters such as spreading factor and code rate. Adaptive data rate mechanism is incorporated in LoRaWAN by monitoring the signal to noise ratio at the network server and optimizing transmission power and the spreading factor. In this work, Q learning agent is used to tune three parameters, transmission power, spreading factor and code rate with the object of minimizing the power consumption of end nodes with improving packet delivery ratio.
Wireless Compose-2: Experiment Results of an Ultra-Wideband Wireless Sensor Network with a Ballistocardiography Smart-Shirt for the ISS Columbus Laboratory
Speaker: Andre Luebken
The presentation summarizes the findings of Wireless Compose-2, a suite of experiments involving a robust wireless sensor network (WSN) designed for the International Space Station (ISS). Wireless Compose-2 showcases the benefits of low-power, low-weight, wireless data acquisition systems for various scientific and a medical experiment in space. The focus was to overcome the performance limitations observed in traditional narrow band RF systems used in typical spacecraft environments. Instead, an impulse-based ultra-wideband (UWB) solution with a suitable medium access scheme was used to enhance connectivity.
The first experiment measures the network performance of the novel UWB physical and MAC layers, demonstrating its data transmission rates and accurate ranging capabilities. The UWB technology outperforms existing WSN technologies, as demonstrated in a high-data rate transmission experiment supporting the medical part of the experiment.
The second experiment, BEAT (Ballistocardiography for Extraterrestrial Applications and Long-Term Missions), explores the potential of continuous wearable monitoring in space. By acquiring cardiac signals using a high-precision Ballistocardiography/Seismocardiography system, the potential for this health monitoring approach is demonstrated.
Finally, the energy harvesting capabilities on the ISS were investigated, specifically within the Columbus module. The potential power that can be harvested from internal light sources is characterized, providing valuable insights into the energy sustainability of WSNs in future space missions.
Empowering IoT Applications with Flexible, Energy-Efficient Remote Management of Low-Power Edge Devices
Speaker: Shadi Attarha
In the context of the Internet of Things (IoT), reliable and energy-efficient provision of IoT applications has become critical. Equipping IoT systems with tools that enable a flexible, well-performing, and automated way of monitoring and managing IoT applications is an essential prerequisite. In current IoT systems, low-power edge appliances have been utilized in a way that can not be controlled and re-configured in a timely manner. Hence, conducting a trade-off solution between manageability, performance and design requirements are demanded. This paper introduces a novel approach for fine-grained monitoring and managing individual IoT services, which improves system visibility and energy efficiency. The proposed method enables operational flexibility for low-power IoT edge devices by leveraging a modularization technique. Following a review of existing solutions for remote-managed IoT services, a detailed description of the suggested approach is presented. Also, to explore the essential design principles that must be considered in this approach, the suggested architecture and its building blocks are elaborated in detail. Finally, the advantages of the proposed solution to deal with disruptions are demonstrated in the proof of concept-based experiments.
TinyML for Developing Countries
Speaker: Gibson Kimutai
The Sustainable Development Goals (SDGs) are objectives and dreams that the United Nations have spelled out to end poverty, protect the planet, and ensure that by 2030 all people enjoy peace and prosperity. This talk showcases the implementation and deployment of TinyML in developing countries for the achievement of the SDGs. As computing in developing countries is scarce, we discuss how TinyML can be a probable solution to solving some of the existing challenges. We discuss the technologies supporting the TinyML implementation and provided a guide for the implementation of TinyML using a prior deployment. Last but not least, we provide opportunities and challenges for TinyML in developing countries.
May 2023
Automated Fault Detection Framework for Reliable Provision of IoT Applications in Agriculture
Speaker: Shadi Attarha
With the growth of Internet-of-Things (IoT), smart agriculture has become one of the most compelling IoT applications that supports crop management and better resource utilization. In this context, the quality of data gathered by widely distributed IoT edge devices has become critical to guarantee the accuracy of decisions in data-driven applications and cost-effectiveness. The data may be inaccurate and contain errors due to adverse environmental conditions or device faults. Supporting knowledge-based systems for monitoring and analyzing collected data to ensure the reliability of IoT services is vital. However, several challenges are encountered in fault detection for IoT applications, such as mimicking normal sensor behaviour by a faulty sensor, limited time and workforce. Also, the lack of labelled datasets containing both normal and real abnormal data points has affected the set of satisfactory data analysis methods. This work aims to propose a novel fault detection framework by utilizing a systematic feature engineering technique which is able to automatically identify abnormal data points, even nontrivial ones. The feature engineering technique helps to build a more reliable anomaly detection model, shortens the training phase, and preserves the sensor against unseen anomalies. Moreover, we provide collections of labelled datasets obtained from experimental situations from various sensors with and without sensor faults to evaluate our approach. The experimental results indicate that the proposed anomaly detection approach combined with the feature engineering technique outperforms established approaches, which are applied to the raw data without any features. It can be seen that extracting meaningful features is a pivotal step for having more precise anomaly detection.
Autonomic Network Management
Speaker: Andreas Könsgen
Increasing user demands, new applications such as M2M and new bearer technologies such as 5G result in a growing complexity of networks. The operation of the networks by the providers therefore becomes a challenging task. Autonomic Network Management (ANM) is an approach to automate the network operation and offload routine tasks from human administrators. Machine learning plays an important role in ANM since it identifies e.g. patterns in the data traffic which then are used to congestion, network faults or malicious attacks.
VANET Collective Perception Messages: Utilization Analysis
Speaker: Thenuka Karunathilake
The collective perception messages (CPMs) were introduced to increase vehicular safety by raising awareness about both V2X-enabled and non-V2X-enabled vehicles. CPMs are proven to be increasing awareness significantly but the CPMs tend to increase the network load resulting in increased latencies. In this presentation, we discuss how much of a percentage of received CPM objects are utilised in normal driving conditions.
Building a Reliable Monitoring System for Extraterrestrial Habitat
Speaker: Saurabh Band
The monitoring systems used for predictive maintenance is also prone to failure itself. This is matter of concern when these systems are used in life critical applications like monitoring extraterrestrial habitats. Thus, it important to address the faults that can occur in these systems and have a way to tackle it, especially in case of extraterrestrial habitats. In this talk we address the faults that can occur in this setup, and analyze how likely are these failures to shutdown these systems completely and how we can tackle these failures with help of backup devices.
April 2023
LoRa on Ice – live sea ice monitoring in Antarctica with open-source technology
Speaker: Jan Rohde
We are presenting the results of the implementation of our measurement system in Antarctica with which we are able to monitor sea ice parameters near the German Neumayer III station in near real-time. Additionally, we will give an overview about our developed hardware and the software used.
February 2023
DCP and VarDis: An Ad-Hoc Protocol Stack for Dynamic Swarms and Formations of Drones
Speaker: Andreas Willig
Coordinating a swarm or formation of cooperating drones requires both local communications (e.g. to avoid collisions with neighbor drones in the presence of disturbances like wind gusts) but also global communications (e.g. in implementing leader-follower swarm control schemes). Similar to vehicular communications, for local communications we use regular beaconing, e.g. at rates of 10 Hz. But what about global communications? In this work we propose a protocol stack in which both local and global communications rest on a single communications primitive, the frequent local broadcast of beacons. Global communications is achieved by piggybacking "crumbs" of data onto beacons and disseminating these crumbs throughout the entire network. Building on this approach, we propose the VarDis protocol, which offers the abstraction of a set of variables, for which VarDis aims to achieve fast and reliable consensus on their current value in potentially large multi-hop drone networks. We present results of a performance study of VarDis.
January 2023
Explainable IoT System for Life-Critical Extraterrestrial (Mars) Missions
Speaker: Saurabh Band
In recent years, much research has been done to settle a civilization on Mars. Among the
various required resources for this civilization, habitat is one of the crucial resources to
live on Mars. Currently, multiple institutes are trying to design and build a space habitat.
These space habitats are designed to provide a safe place to live during the initial missions
and are equipped with monitoring and life support systems to ensure astronauts’ safety.
This work explores the concepts to make habitat monitoring reliable and robust. We base
the work on three objectives to achieve this goal: 1. Design a device-level monitoring tool
for the hardware nodes, 2. Make the communication system reliable, and 3. Explore the
area of explainable IoT and propose a concrete architecture and Taxonomy to understand
the area for upcoming researchers better.
Detecting Wolves: Challenges in Image Recognition
Speaker: Jens Dede
After being almost absent for more than 100 years, wolves have returned to central Europe for at least two decades. Their continuously increasing population leads to more contact between humans, farm animals and wolves. Reliable detection of wolves from images and videos is beneficial for many applications. From general monitoring over multiple research studies to the deterrent of the animals can benefit from such kind of detection models. This talk gives an overview of the current status of the work: What are the challenges of detecting wolves, which frameworks are available and which pitfalls are showing up during the implementation?
Sensitivity Analysis on the communication parameters in IEEE 802.11p.
Speaker: Piumika Karunanayake
Sensitivity analysis provides the input parameters which mostly impact on a given metric. Although there are number of communication protocol parameters that influence on the performance of the protocol, it is important to identify which parameters impact the most. With that knowledge, improving the performance of the protocol in a given environment will not be challenging.
The selected protocol for sensitivity analysis is IEEE802.11.p and the selected application is vehicular network. Although there are many research studies conducted on changing one or two parameters together to improve the performance, carrying out a sensitivity analysis has not been conducted. There are certain methods to select the parameter combinations for the sensitivity analysis, since considering all combinations are time consuming. The selected communication protocol parameters for the analysis are minimum and maximum values of the contention window, AIFSN and transmission rate. However there are many other parameters but applicable for unicast scenario. In this work, only broadcast scenario is considered for analysis.
November 2022
Automated Fault Detection Framework for Reliable Provision of IoT Applications
Speaker: Shadi Attarha
With the growth of Internet-of-Thing (IoT), smart agriculture is one of the most compelling IoT applications that aids in crop management and better resource utilization. In this context, the quality of data gathered by widely distributed IoT edge devices has become critical to guarantee the accuracy of decisions in data-driven applications and cost-effectiveness. The data may be inaccurate and contain errors due to adverse environmental conditions or device faults. The supporting knowledge-based systems for monitoring and analyzing collected data to ensure the reliability of IoT services are vital. However, several limitations are encountered in fault detection for IoT applications, such as limited computational and power resources of edge devices, time and manpower. Furthermore, the lack of labelled datasets has affected the set of satisfactory data analysis models. The aim of this work is to enhance system reliability by assessing data trustworthiness and detecting abnormal sensor behaviours based on machine learning techniques and feature engineering. For this purpose, firstly, we provide collections of labelled datasets obtained from experimental situations with different real sensor faults. Secondly, with the help of feature engineering, the datasets are augmented with appropriate external data to improve the accuracy of data evaluation models. Finally, to overcome the challenge of detecting abnormal behaviours, we propose a method capable of combining the results of data analysis to assess sensor conditions. The experimental results indicate that it is possible to offer a time-efficient and reliable for fault detection.
Delay analysis on CPMs in VANETs
Speaker: Thenuka Karunathilake
The Vehicular Networks has become major research area because of the number of vehicles on the road increasing day by day. Therefore, number of fatal accidents are also tend to increase. It is found that majority of the accidents are caused by human error. One solution is to reduce human error is to introduce automated vehicles with self driving capability. However, market penetration of V2X enabled vehicles are slow causing longer transition period. During this transition period both V2X enabled and non enabled vehicles has to co-exit in the same network. In such networks, to improve safety the collective perception messages (CPMs) was introduced.
However, CPMs are periodic messages generated by all the V2X enabled vehicles, the communication channel can be overloaded. Therefore, in this speech we discuss the effect on delay of CPM messages caused by different market penetration rates and also by different CPM generation intervals using real world vehicular data set collected at the four-way intersection.
Local Clock Discipline in Mission-Critical Wireless Sensor Networks
Speaker: Andre Luebken
Wireless Sensor network technology has sparked some interest in the space industry as a means to partially replace the complex cable harnesses in spacecraft and launcher applications as well as for precise ranging applications.
In order to replace mission-critical sensor hardware, precise time synchronization is one of the major features such a wireless system needs to provide to be viable. The local time is derived from oscillators that need to be controlled to give an accurate representation of the reference time. A major problem here is the harsh environment these sensors are typically subjected to. High temperature differences and electromagnetic interference are the cause of large local oscillator drift between network nodes. Commonly used control mechanisms for this purpose do not work well under these conditions as they are designed to be used in stable environments to mainly correct for manufacturing tolerance between local oscillators.
This talk is designed to give an overview of the problem of local clock discipline and presents a simulation environment as well as results for different local clock control algorithms that are better suited for operation under extreme conditions.on adaptive protocol parameters are introduced using reinforcement learning.
June 2022
Adaptive Protocol Parameters for WSNs
Speaker: Piumika Karunanayake
Wireless Sensor Networks (WSN) are an infrastructure less network and widely used for multi-disciplinary applications. According to the requirements and the environment, the network is designed and the protocol is tuned to obtain the best performance of the WSN. In real world applications, all nodes in the network have a common protocol parameter set, irrespective of their position in the network. Tuning protocol parameters for each node manually is tedious and may not be practical for large number of nodes. In this presentation adaptive protocol parameters are introduced using reinforcement learning.
Collective Perception for Road Intersections
Speaker: Thenuka Karunathilake
Vehicular networks has become a major research field because of its promising applications mainly sefety related applications. However, market penetration of V2X enabled vehicles still quiet slow. Therefore, during this long transition period from conventional vehicles to V2X vehicles, collective perception was introduced to increase the road safety by transmitting perceived objects from local sonsors additionally to Collective Awareness Messages (CAM). In this talk, we will discuss the feasibility of using collective perception in a realistic intersection.
Detecting Wolves: Challenges in Image Recognition
Speaker: Jens Dede
After being almost completely absence for more than 100 years, wolves are returning back to Germany since at least two decades. Their continuously increasing population leads to more and more contacts between humans, farm animals and the wolves. The growing number of kills of farm animals – especially sheep, horses and goats – increases the demand for protection technologies.
Traditional fences which offer sufficient protection from wolves are inflexible and expensive. Therefore, alternative and smarter solutions have to be found.
The objective of the mAInZaun project is to develop a smart fence which detects wolves, alarms the owner of the farm animals and starts scare-off stimuli. For this, cameras continuously monitor the environment around the farm animals. If wolves are detected, the eviction will be started automatically.
This talk gives an overview of the current status of the work: What are the challenges of detecting wolves, which frameworks are available and which pitfalls are showing up during the implementation.
May 2022
The Living Habitat for Mars (Humans on Mars)
Speaker: Saurabh Band
Living on Mars will pose immense challenges for humans. One of the many reasons is the lethal environment on the planet. Therefore, the crew will have to live in a habitat that ensures its survival with the help of a life support system (LSS). Hence, it is of utmost importance that the crew can trust this system and that it is designed with the crew in mind. The Living habitat will be equipped with a robust sensor network which is intelligent enough to monitor the habitat as well as diagnose faults in the system itself.
Service Management for Enabling Self-Awareness in Low-Power IoT Edge devices
Speaker: Shadi Attarha
In the context of Internet-of-Things (IoT), efficient and flexible service management techniques are essential to improve performance and cost-effectiveness. In this regard, it is crucial to equip the IoT devices with tools that allow a flexible, well-performing, and automated way of efficient services provisioning. Current IoT low-power edge devices have been designed in a way that embedded services can not be monitored and re-configured during the run-time. Hence, finding a trade-off between design requirements, specific performance targets, and services manageability are necessary. The presented project focuses on the idea of service isolation and modularisation at the level of edge devices to observe IoT services and manage them under real-time requirements in extremely resource-constrained IoT environments.
Spatial models to describe indoor environments and to reason about object perspectives
Speaker: Zoe Falomir Llansola
On one hand the challenge is to show how to use spatial models to communicate about indoor environments. For that, first intelligent systems must recognize (i.e. using classical computer vision and machine learning algorithms) and locate objects in space. Then, addressing the following research questions is crucial: which kind of spatial features must the system describe? which kind of reference frames must use? Intelligent systems must have common grounding with users so that they can align spatial representations and communicate with each other.
Then, we will continue to discuss if there are aspects to improve, for example, in the classical algorithms we currently use for recognizing objects. For that, I will show some spatial reasoning tests about object perspectives which are applied to measure students’ intelligence. These tests involve spatial reasoning. And there are psychological experiments that show that people with better spatial reasoning skills are more innovative and successful in science, technology and maths. So, can artificial systems apply some spatial reasoning to evolve their objects' recognition methods to be more cognitive? or even more efficient?
IoT in Academic Institutions in Developing Countries: Lessons learned
Speaker: Marco Zennaro
This talk will present the general topic of IoT4D (IoT for Development). Lessons learned from using IoT in more than 30 workshops in academic institutions in Developing Countries and some success stories will be discussed. The final part of the talk will cover the latest evolution of IoT: Intelligence of Things.
April 2022
Machine Learning for Vehicular Networks: challenges and means of solution
Speaker: Minette Zongo Meyo
Road congestion in urban traffic can sometimes be paralyzing. And the increasing volume of road vehicles has made transportation efficiency to become a challenge. Intelligent Transportation Systems are expected to make everyday vehicular operation safer, greener, and more efficient. Machine learning-based platforms for transportation are a valuable solution to achieve this. In fact, ML models can successfully make use of historical data from IoT devices to either control congestion or provide efficient route planning solutions to drivers platforms. However, some challenges hinder the applicability of ML in the transportation domain. This presentation describes some of them and provide research topics worth exploring. A traffic flow prediction system using ML models is explored at the end of the presentation.
Energy-Driven Computing: Rethinking the Design of Energy Harvesting Systems
Speaker: Geoff Merrett
Energy harvesting computing has been gaining increasing traction over the past decade, fuelled by technological developments and rising demand for autonomous and battery-free systems. Using energy harvesting instead of batteries introduces numerous challenges to embedded systems, not least the transition from an energy-limited source (which can provide virtually unlimited power) to a power-limited source that is highly unpredictable and dynamic (both spatially and temporally, and with a range spanning many orders of magnitude). The typical approach to overcome this is the addition of intermediate energy ‘buffer’ (a small battery or supercapacitor) to smooth out the temporal dynamics of both power supply and consumption. This has the advantage that, if correctly sized, the system ‘looks like’ a battery-powered system; however, it also adds volume, mass, cost and complexity and, if not sized correctly, unreliability. In this talk, I will present a different class of computing to conventional approaches, namely energy-driven computing, where systems are designed from the outset to operate from an energy harvesting source. Such systems typically contain little or no additional energy storage (instead relying on tiny parasitic and decoupling capacitance), alleviating the aforementioned issues. Examples of energy-driven computing include intermittent systems (which power down when the supply disappears and efficiently continue execution when it returns) and power-neutral systems (which operate directly from the instantaneous power harvested, gracefully modulating their consumption and performance to match the supply).
Collective Perception for Road Intersections
Speaker: Thenuka Karunathilake
Vehicular networks has become a major research field because of its promising applications mainly sefety related applications. However, market penetration of V2X enabled vehicles still quiet slow. Therefore, during this long transition period from conventional vehicles to V2X vehicles, collective perception was introduced to increase the road safety by transmitting perceived objects from local sonsors additionally to Collective Awareness Messages (CAM). In this talk, we will discuss the feasibility of using collective perception in a realistic intersection.
Detecting Wolves: Challenges in Image Recognition
Speaker: Jens Dede
After being almost completely absence for more than 100 years, wolves are returning back to Germany since at least two decades. Their continuously increasing population leads to more and more contacts between humans, farm animals and the wolves. The growing number of kills of farm animals – especially sheep, horses and goats – increases the demand for protection technologies.
Traditional fences which offer sufficient protection from wolves are inflexible and expensive. Therefore, alternative and smarter solutions have to be found.
The objective of the mAInZaun project is to develop a smart fence which detects wolves, alarms the owner of the farm animals and starts scare-off stimuli. For this, cameras continuously monitor the environment around the farm animals. If wolves are detected, the eviction will be started automatically.
This talk gives an overview of the current status of the work: What are the challenges of detecting wolves, which frameworks are available and which pitfalls are showing up during the implementation.
May 2022
The Living Habitat for Mars (Humans on Mars)
Speaker: Saurabh Band
Living on Mars will pose immense challenges for humans. One of the many reasons is the lethal environment on the planet. Therefore, the crew will have to live in a habitat that ensures its survival with the help of a life support system (LSS). Hence, it is of utmost importance that the crew can trust this system and that it is designed with the crew in mind. The Living habitat will be equipped with a robust sensor network which is intelligent enough to monitor the habitat as well as diagnose faults in the system itself.
Service Management for Enabling Self-Awareness in Low-Power IoT Edge devices
Speaker: Shadi Attarha
In the context of Internet-of-Things (IoT), efficient and flexible service management techniques are essential to improve performance and cost-effectiveness. In this regard, it is crucial to equip the IoT devices with tools that allow a flexible, well-performing, and automated way of efficient services provisioning. Current IoT low-power edge devices have been designed in a way that embedded services can not be monitored and re-configured during the run-time. Hence, finding a trade-off between design requirements, specific performance targets, and services manageability are necessary. The presented project focuses on the idea of service isolation and modularisation at the level of edge devices to observe IoT services and manage them under real-time requirements in extremely resource-constrained IoT environments.
Spatial models to describe indoor environments and to reason about object perspectives
Speaker: Zoe Falomir Llansola
On one hand the challenge is to show how to use spatial models to communicate about indoor environments. For that, first intelligent systems must recognize (i.e. using classical computer vision and machine learning algorithms) and locate objects in space. Then, addressing the following research questions is crucial: which kind of spatial features must the system describe? which kind of reference frames must use? Intelligent systems must have common grounding with users so that they can align spatial representations and communicate with each other.
Then, we will continue to discuss if there are aspects to improve, for example, in the classical algorithms we currently use for recognizing objects. For that, I will show some spatial reasoning tests about object perspectives which are applied to measure students’ intelligence. These tests involve spatial reasoning. And there are psychological experiments that show that people with better spatial reasoning skills are more innovative and successful in science, technology and maths. So, can artificial systems apply some spatial reasoning to evolve their objects' recognition methods to be more cognitive? or even more efficient?
IoT in Academic Institutions in Developing Countries: Lessons learned
Speaker: Marco Zennaro
This talk will present the general topic of IoT4D (IoT for Development). Lessons learned from using IoT in more than 30 workshops in academic institutions in Developing Countries and some success stories will be discussed. The final part of the talk will cover the latest evolution of IoT: Intelligence of Things.
April 2022
Machine Learning for Vehicular Networks: challenges and means of solution
Speaker: Minette Zongo Meyo
Road congestion in urban traffic can sometimes be paralyzing. And the increasing volume of road vehicles has made transportation efficiency to become a challenge. Intelligent Transportation Systems are expected to make everyday vehicular operation safer, greener, and more efficient. Machine learning-based platforms for transportation are a valuable solution to achieve this. In fact, ML models can successfully make use of historical data from IoT devices to either control congestion or provide efficient route planning solutions to drivers platforms. However, some challenges hinder the applicability of ML in the transportation domain. This presentation describes some of them and provide research topics worth exploring. A traffic flow prediction system using ML models is explored at the end of the presentation.
Energy-Driven Computing: Rethinking the Design of Energy Harvesting Systems
Speaker: Geoff Merrett
Energy harvesting computing has been gaining increasing traction over the past decade, fuelled by technological developments and rising demand for autonomous and battery-free systems. Using energy harvesting instead of batteries introduces numerous challenges to embedded systems, not least the transition from an energy-limited source (which can provide virtually unlimited power) to a power-limited source that is highly unpredictable and dynamic (both spatially and temporally, and with a range spanning many orders of magnitude). The typical approach to overcome this is the addition of intermediate energy ‘buffer’ (a small battery or supercapacitor) to smooth out the temporal dynamics of both power supply and consumption. This has the advantage that, if correctly sized, the system ‘looks like’ a battery-powered system; however, it also adds volume, mass, cost and complexity and, if not sized correctly, unreliability. In this talk, I will present a different class of computing to conventional approaches, namely energy-driven computing, where systems are designed from the outset to operate from an energy harvesting source. Such systems typically contain little or no additional energy storage (instead relying on tiny parasitic and decoupling capacitance), alleviating the aforementioned issues. Examples of energy-driven computing include intermittent systems (which power down when the supply disappears and efficiently continue execution when it returns) and power-neutral systems (which operate directly from the instantaneous power harvested, gracefully modulating their consumption and performance to match the supply).
Energy-Driven Computing: Rethinking the Design of Energy Harvesting Systems
Speaker: Geoff Merrett
Energy harvesting computing has been gaining increasing traction over the past decade, fuelled by technological developments and rising demand for autonomous and battery-free systems. Using energy harvesting instead of batteries introduces numerous challenges to embedded systems, not least the transition from an energy-limited source (which can provide virtually unlimited power) to a power-limited source that is highly unpredictable and dynamic (both spatially and temporally, and with a range spanning many orders of magnitude). The typical approach to overcome this is the addition of intermediate energy ‘buffer’ (a small battery or supercapacitor) to smooth out the temporal dynamics of both power supply and consumption. This has the advantage that, if correctly sized, the system ‘looks like’ a battery-powered system; however, it also adds volume, mass, cost and complexity and, if not sized correctly, unreliability. In this talk, I will present a different class of computing to conventional approaches, namely energy-driven computing, where systems are designed from the outset to operate from an energy harvesting source. Such systems typically contain little or no additional energy storage (instead relying on tiny parasitic and decoupling capacitance), alleviating the aforementioned issues. Examples of energy-driven computing include intermittent systems (which power down when the supply disappears and efficiently continue execution when it returns) and power-neutral systems (which operate directly from the instantaneous power harvested, gracefully modulating their consumption and performance to match the supply).
February 2022
Deep-Space Communication
Speaker: Dr.Andreas Könsgen
This presentation gives an overview about deep-space communication, i.e. primarily the communication to control the spacecraft themselves and to offload payload such as measurement data from observation missions. The presentation explains the conditions and facilities for deep-space communication, gives some examples for research on the topic and an outlook into possible future developments.
January 2022
Offloading an Energy-Efficient IoT Solution to the Edge: A Practical Solution for Developing Countries
Speaker: Gibson Kimutai
Agriculture contributes to the economies of many developing countries. Tea is the most popular crop in Kenya as it contributes majorly to her economy. Among the various stages of processing tea, fermentation is the most important as it determines the final quality of the processed tea.
Presently, the process of monitoring is done manually by tea tasters by tasting, smelling, and touching tea which compromises the quality of tea. In this paper, a deep learner dubbed “TeaNet” is deployed in Edge and Fog environments for real-time monitoring of tea fermentation. We power the
system using a Photovoltaic (PV) energy source to overcome the challenge of unreliable power supply from the grid. Further, the energy consumption of the solution is reduced by applying duty cycling, where idle components are designed to sleep. We used the Analysis of variance (ANOVA) and Post-hoc for data analysis. From the results, Edge registered the lowest latency
compared to the Cloud and Fog environments. During deployment of the energy-optimized model, 50.6559Wh amount of energy was saved.
Imperceptible Sensor Systems for Healthcare Applications
Speaker: Prof.Björn Lüssem
Flexible, stretchable, sometimes even “imperceptible” sensor systems have been intensively studied for their application in the healthcare and wellness sector. Although first systems were based on conventional, silicon-based electronics, the field has experienced a recent boost by the introduction of so called mixed organic conductors.
Mixed organic materials conduct ionic and electric charge equally well, which enables completely new design principles for electronic devices used, e.g., in highly sensitive organic biosensors or neuromorphic devices. In particular, the strong coupling between ion and charge transport observed in these organic mixed conductors has made them a key driver in novel organic bioelectronics based on Organic Electrochemical Transistors (OECTs), with various technological demonstrations in the healthcare sector that include in-situ measurements of brain activity, collection of electrocardiograms, and the tracking of eye movement.
In this presentation, recent advances in the use of flexible and almost imperceptible sensor systems for healthcare monitoring or treatment are presented. An emphasis is put on Organic Electrochemical Transistors, whose working mechanism is reviewed. Current bottlenecks for device optimization are summarized, stressing the need for advanced two-dimensional device modeling and a targeted design of improved semiconductors, electrolytes, and contact materials.
Detecting Wolves: Challenges in Image Recognition
Speaker: Jens Dede
After being almost completely absence for more than 100 years, wolves are returning back to Germany since at least two decades. Their continuously increasing population leads to more and more contacts between humans, farm animals and the wolves. The growing number of kills of farm animals – especially sheep, horses and goats – increases the demand for protection technologies.
Traditional fences which offer sufficient protection from wolves are inflexible and expensive. Therefore, alternative and smarter solutions have to be found.
The objective of the mAInZaun project is to develop a smart fence which detects wolves, alarms the owner of the farm animals and starts scare-off stimuli. For this, cameras continuously monitor the environment around the farm animals. If wolves are detected, the eviction will be started automatically.
This talk gives an overview of the current status of the work: What are the challenges of detecting wolves, which frameworks are available and which pitfalls are showing up during the implementation.
December 2021
Mobile Road Side Units for VANETs
Speaker: Thenuka Karunathilake
The number of vehicles on the road is increasing day by day and due to this, the number of fatal accidents on the road is also increasing. One solution for safer roads is to deploy Vehicular Networks (VANETs). The main application of VANETs is enhancing vehicular safety by enabling Vehicle-to-Vehicle communication and Vehicle-to-Infrastructure communication. A major issue for safety-related applications in VANETs is the very short latency requirements because of the high traveling speeds of vehicles. In order to meet these latency requirements, VANETs has introduced the new networking component 'Road Side Unit (RSU)'. However, due to large investment cost-related RSU, the expected level of deployment was not achieved. As a solution for this, the idea of mobile RSUs was introduced. In this speech, we are going to talk about the current status of mobile RSUs in VANETs.
November 2021
Context-aware Enhancements to Epidemic Routing in Opportunistic Networks
Speaker: Vishnupriya Parimalam
Opportunistic networks enables the devices to communicate as and when the opportunity rises. This property of OppNets has been explored in routing approaches in a similar operating manner as the traditional infrastructure networks. Specifically, context-aware routing approaches have been the major focus of OppNets in the recent literature. However, the potential of OppNets also extends to data dissemination in destination-less networks. Forwarding approaches in such a network need context such that the data dissemination is favored without the necessity to reach a particular set of users or a particular destination. Hence, the context-awareness need to be defined differently for destination-less OppNets as compared to OppNets in destination-oriented scenarios.
Opportunistic Networking - Augmenting the Networks of Future
Speaker: Suvadip Batabyal
The future networking architecture will include both terrestrial and the aerial networks with a multitude of devices communicating amongst themselves using different communication technologies. With such an exponential growth in the dimension of network and co-existence of varied wireless technologies, the primary challenge will be the effective and efficient usage of scarce spectrum for a high bandwidth communication. One of the naive technologies which helps in alleviating this problem through spectrum reuse is device-to-device communication underlay cellular networks along with the notion of multi-hop communication paradigm. However, the future networks will also need to service high speed mobile devices (upto 300kmph) which is yet another challenge that needs to be solved. In such scenarios, the opportunistic networking paradigm can improve network availability through opportunistic link establishment. Opportunistic networking paradigm aims to provide an infrastructureless communication by exploiting device proximity. In this presentation we will discuss how the opportunistic networking paradigm can augment the networks of the future and help in improving
service availability and quality of experience for the end user. We shall look at some of the use-cases where opportunistic networking may be employed and how it can be incorporated into modern networking architectures.
On Communication Networks for Distributed Control in Smart Grids
Speaker: Leonard Fisser
Energy grids and especially distribution grids are in rapid change. The shift to green energies introduces a multitude of new components, such as batteries, electric vehicles (EV) and photovoltaic (PV).The fact that these components express highly dynamic behavior and service requirements are high, forces Distribution System Operators (DSO) to implement active operation management.
In this seminar, we present our current work on distributed control in future distribution grids from the view of communication networks. We focus on our current DFG project nick-named OUREL. After a brief introduction to the control-theoretic aspect of our project, we have a look at the communication network side, highlight key challenges and how we can approach them. We present our work on an all-to-all flooding protocol, designed to effectively disseminate periodic status updates in Smart Grid topologies. Afterwards, we have a look at advanced topics which are part of my PhD-work and include Topology Modeling, Network Coding and Age of Information. If times allows for it, I would like to give you a short demo of our currently work-in-progress full system real-time emulator.
October 2021
Impact of digital technologies for future industries
Speaker: Dr. Thushara Weerawardane
The process of the industry revolutionised over years from mechanisation to cyber physical systems. The productivity, cost and speed changes within the production process from material to finish product in time varying physical system. Globally, many industries face numerous challenges of handling uncertainties in supply & demand variation of the supply chain & logistics for both raw material and finish products. During manufacturing processes, the reliability of equipment & plant plays key roles to minimise the downtime from the performance and efficiency perspectives. The rapid development in technology and automation created vast improvement in manufacturing systems by providing high quality product and services to a competitive market environment during the last century. Modern technology advancements have created a cyber physical environment in the industry through the integration of automated systems by providing reliable and ubiquitous connectivity among manufacturing processes and systems. With development of industrial internet of thing (IIOT) and AI based technologies, data becomes main fuel for many digital industries. The real-time information availability over the manufacturing process and systems, creates efficient and high quality production platforms. AI technologies provide predictive maintenance and prescriptive analytics within manufacturing industries to create cost effective reliable systems. Currently many industries are moving towards the process of digital transformation within an integrated automated cyber-physical system in order to be competitive in the market.
A Multi-agent-based simulation system for crowd evacuation in the fire emergency environment
Speaker: Gayamini Shanmuganathan
Evacuation and personal safety are major concerns from wide areas or indoors under emergency. Every disaster, whether man-made or natural, we face many unexpected problems, for instance, fatal accidents, and property damage. In some moments like this, panic among pedestrians beings is frequent. Everyone wants to save their lives. As a result of panic, some people lose the energy of thinking. It can also trigger conflicts among pedestrians. In emergencies there is less time to react; This can lead to a large loss of life, since many people who are caught may unaware of the exit to get out from there and are more likely to run in the wrong direction. To prove this, several investigations indicate that a substantial number of deaths occur due to wrong decisions residents make within the available evacuation time. Besides, the conclusion from the past analysis relieves guiding residents during evacuation proves to be more effective because it decreases the average escape time thereby increasing the chance of survival in a fire emergency. However, widespread fire disaster has the highest occurrence of frequency in several concentrated short periods among disasters. To provide a better solution to the issue, systems have been developed to alert or warn pedestrians in the presence of a fire emergency, so they can act promptly. Several approaches are done by past researchers to simulate the crowd evacuation, Nevertheless, the concept of autonomous agents has been bringing into play successfully to investigate collective human behavior during an emergency evacuation over the past decades. Similarly, artificial intelligence is strongly making its footprint in all disciplines. Our first and foremost objective is to evacuate residents safely in the shortest possible time in the event of a fire. With the support of reinforcement learning, which is a subfield of AI, we have done the pedestrian evacuation simulation system in the fire emergency environment. Several steps were taken to develop our system to achieve efficiency in an evacuation, in other words, safely evacuate residents within a short time. When considering a large-scale environment, it has many features, and only a few selected features are designed by ArcMap to enforce as the foundation of the system environment. We adopt the multi-agent concept to train the agents in the NetLogo environment. IIOne important point is that the moment we start training, any resident can only attend the training if they become pedestrians, and they will then share their experience with
others. Also, the deep reinforcement learning algorithm leads them to achieve the optimal policy. Besides, the A* algorithm directed the pedestrians to designated shelters. As well as fire dynamics are created by fire agencies. These results demonstrate empirically that the proposed simulation system is effective with time efficiency and the system has a strong capability to describe, represent, and explain the reality of evacuation.
July 2021
Early warning system for landslides using Wireless Sensor Networks
Speaker: Piumika Karunanayake
Landslide is a natural disaster which causes a considerable damage to the natural habitat,
environment, economy and other resources. Due to the randomness of the event, monitoring,
predicting and controlling are major challenges associated with landslides. Yet, developing an
accurate prediction mechanism with an effective early warning system has become a need of
the hour since the damages and the losses occurred due to the landslides are intolerable. There
exist expensive, advanced mechanisms deployed to predict the possibility of occurring
landslides by using satellites and radar systems with artificial intelligence capabilities.
Comparing with the existing high-end systems, a cost effective wireless sensor network which is
capable of identifying the underground movements and soil conditions is introduced as a
practical solution. Yet, dealing with a large number of sensor data and identifying the
correlation of the variables and the occurrence of a landslide is difficult. Hence in this work,
machine learning is used to predict the occurrence of landslide with a set of sensor data
gathered for a period of two years on a land which is identified as prone for land slide in Sri
Lanka. After developing three models for prediction, one model was selected as its performance
measurements are better compared to other two including an accuracy of 99.8%. An prototype
of warning system is also built which takes the model output and display a web based warning
message. Although the developed machine learning model is site specific, similar approach can
be implemented in other landslide prone areas to improve the efficiency of the disaster
management system in Sri Lanka.
Collecting Data in Ubiquitous Infrastructures: How to Engage Communities and Make Sense of Large Volumes of Data
Speaker: Catia Prandi
In this talk, I will present some case studies where crowdsourcing/crowdsensing, and participatory sensing were applied to ubiquitous infrastructures to investigate how to gather data regarding relevant issues (such as urban accessibility and environmental sustainability). In presenting these studies, I will focus on how the systems have been designed and evaluated using HCI methodologies, and I will point to strategies for involving users, such as gamification/gameful experiences and data visualization, in collecting data and making sense of large volumes of data to benefit the different communities.
June 2021
LDACS: Self-organizing air-to-air communication
Speaker: Sebastian Lindner (Research Fellow at TUHH)
The Single European Sky Air Traffic Management Research (SESAR) program is a joint undertaking to overhaul and modernize European air traffic management. During all phases of flight, modern digital data links shall realize the Future Communications Infrastructure (FCI). The air-to-air (A2A) component of the FCI is the L-band Digital Aeronautical Communications System (LDACS) A2A mode, which is currently in early stages of development.
In this talk, we would like to present the challenges such a mobile network must tackle and give an overview on the self-organizing medium access control (MAC) that is being researched. Also, an offshoot project towards a Machine Learning-based predictive MAC that realizes coexistence with legacy communication systems in the same frequency band will be discussed.