An Intelligent Approach for Mathematical Morphology on ManyCore Architectures
Mathematical Morphology (MM) is a non-linear branch of image processing, and computer vision, based on geometry and on the mathematical Theory of Order. It was developed by Matheron and Serra in the 1960s, and its initial applications were in biomedical and geological image analysis problems. Later on, it has proved to be a powerful tool for computer vision tasks involving binary and grey-level images for noise suppression, enhancement, edge detection, skeletonization, segmentation, pattern recognition, multiscale image processing, statistical analysis, and optimal design of morphological filters, just to name a few examples. Finally, in the 1980s its theory was generalized for Complete Lattices, allowing it to be applied to colour images.
The basic operators of Mathematical Morphology are dilation and erosion, which can be combined algorithmically to form more complex operators. However, this combination is not trivial, even for a specialist. Moreover, it is very computationally intensive. In the literature, dedicated hardware architectures using FPGAs have been developed, in which the acceleration of the morphological operations is obtained by fine-grain instruction decomposition. Generally, in practice, these architectures are however not flexible for more complex algorithms involving image/video processing.
Many-core Architectures (MA) provide large (generally 2D) arrays of computational nodes, where each node contains a powerful processing unit. These architectures represent an attractive choice for developing more complex applications for MM. However, this scenario introduces some conflicts. For example, MM tends towards very fine-grained operations connected in a sequential fashion, whereas MC needs to balance communication in a 2D mapping scenario. These conflicting requirements represent an ideal scenario for the application of multiobjective optimization.
Thus, this talk will provide an overview about MM, some dedicated hardware for its acceleration, and discuss the possibility of using intelligent systems for generating appropriate mappings for MA-based systems, for Image and Video applications.
Low Phase Noise Signal Generation utilising Oscillators, Resonators & Filters and Atomic Clocks
Microwave Lecture on behalf of the IEEE MTT Society
Oscillators and clocks are used in almost all electronic systems. They set the timing of operations and clock elements as required. The phase noise, jitter & stability of these oscillators often sets the ultimate performance limit. Oscillators requiring low phase noise are used in communications, control, RADAR and navigation systems and also as flywheel oscillators for atomic clocks, particle accelerator systems and Very Long Base Interferometry (VLBI) systems.
This talk will initially discuss the theory and design of a wide variety of oscillators offering the very best performance. Typically, this is achieved by splitting the oscillator design into its component parts and developing new amplifiers, resonators and phase shifters which offer high Q, high power handling and low thermal and transposed flicker noise. Key features of oscillators offering the lowest phase noise available will be shown, for example: a 1.25GHz DRO produces -173dBc/Hz at 10kHz offset and a noise floor of -186dB and a 10 MHz crystal oscillator shows -123dBc/Hz at 1Hz and -149 at 10Hz.
New compact atomic clocks with ultra-low phase noise microwave synthesiser chains (with micro Hz resolution) will also be briefly described to demonstrate how the long-term stability can be improved.
New printed resonators (and thereby filters) demonstrate Qs exceeding 540 at 5GHz on PCBs and > 80 at 21GHz on GaAs MMICs. These resonators produce near zero radiation loss and therefore require no screening. L band 3D printed resonators demonstrate high Q (> 200) by selecting the standing wave pattern to ensure zero current through the via-hole and new ultra-compact versions (4mm x 4mm) have been developed for use inside or underneath the package. Alumina based resonators demonstrating Qs >200,000 at X band have also been produced. Tuneable versions (1%) have recently been developed.
Jeremy presented the first course on oscillators including a lab class at the IEEE International Microwave Symposium in Boston in 09. This was repeated in 2010, 2011. A battery powered lab kit offering 5 experiments with full theoretical and simulation support was provided. The kit also produced the state-of-the-art performance with flicker noise corners around 200Hz. The methodology behind this course will be described. Theory and 5 experiments on the same day was part of the reason for success. This course is being run again in Boston in June 2019.
The next generation of oscillators will offer orders of magnitude improvement in performance. Our current attempts to do this will be described.
Healthcare robotics – from intelligent design to AI
This presentation gives an overview of healthcare robotics portfolio at Bristol Robotics Laboratory that spans from surgical to assisted living robots as well as robotic systems to support radiotherapy. The fast and widespread use of robotic technologies in healthcare applications aims at improving patient outcomes and reducing NHS costs. The robotic technologies include soft sensors and actuators, complex robotic system design and control, haptics, machine learning and sensor processing.
Space vs Medicine: Advances in User-Driven Robotics
Advances in Space Engineering and Technology have dramatically revolutionised the way we work and live today. Similarly, intelligent robots have revolutionised terrestrial assembly and servicing processes. Innovative Design Engineering solutions are enabling Space robots to undertake unmanned operations in Earth orbits and on the surface of the Moon, Mars and beyond. In this presentation, Dr. Saaj will talk about her journey through space and her vibrant portfolio of challenging Design Engineering projects for orbital and planetary exploration missions. Furthermore, she will present how she succeeded in securing new research opportunities through spinning-out technology from Space Robotics to Medical robotics. She will conclude by sharing her long-term vision on advancing the capabilities of autonomous space robots and its application to other user-driven application domains.
Turning materials into evolving soft robots
Robots have been successful in the applications which require speed, power and precision, but very limited where soft and delicate interactions are necessary. In this context the use of soft functional materials opened the door to a range of new types of machines, soft robots, that can not only interact with soft objects in the environment but also make robots themselves into soft structures for better adaptability in different situations. We have been exploring a set of alternative technologies to design and construct complex soft robots, such as multi-material 3D printing, electrically conductive elastomers, and model-free design automation processes. With the recent rapid progress of these technologies, we are able to develop new kinds of robots that we can characterised as “morphologically computing machines”. We are exploring how such a new paradigm of design processes can be realised, though there are a number of known challenges such as design of complex mechanical structures, sensing of physical interactions, and modelling, simulation and control in general. In this talk, I would like to introduce some of our recent soft robotics projects in our laboratory and their extension to model free design automation.
Hardware-Based Security Solutions for the Internet of Things
The internet of Things technology is expected to generate tremendous economic benefits, this promise is undermined by major security threats. First of all the vast majority of these devices are expected to communicate wirelessly, and will be connected to the Internet, which makes them especially susceptible to confidentiality threats from attackers snooping for messages contents. Second, most IoT devices are expected to be deployed in remote locations with little or no protection; therefore they can be vulnerable to both invasive and side channel attacks, malicious adversaries can potentially gain access to a device and apply well know power or timing analyses to extract sensitive data that might be stored on the IoT node, such as encryption keys, digital identifiers, and recorded measurements. Furthermore, with ubiquitous systems, it can no longer be assumed that the attacker is remote. Indeed, the attack could even come from within the system itself, from rogue embedded hardware (e.g. Trojans). A large proportion of IoT devices operate in an energy-constrained environment with very limited computing resources, this makes the use of typical defence mechanisms such as classic cryptography algorithms prohibitively expensive. The challenges for building secure IoT are threefold:
1) How to develop hardware which is inherently resilient to physical attacks
2) How to implement complex security protocols with very limited resources
3) How to detect/diagnose anomalous behaviour of an IoT device
This talk addresses the above three questions, as follows.
The first part of this talk addresses the first question, it presents two novel approaches for enhancing the security and reliability of physically unclonable functions, one of the enabling technologies designing Tamper resistant Hardware
The first technique proposes a physically unclonable function using instruction cache, typically found in all embedded processors. The design is optimised to improve resilience to ageing effects. The second approach aims to enhance the security of physically unclonable functions against modelling attacks by combining these with low cost cryptographic primitives such as permutation and substitution. The proposed techniques make it affordable, secure and reliable to use physically unclonable technology in resources constrained systems.
The second part of this talk addresses the second question, it presents a new authentication protocol based on PUF technology, Then power consumption and memory utilization of the proposed protocol were estimated and compared with the existing solutions, namely: DTLS (datagram transport layer security) handshake protocol and UDP (user datagram protocol). Our results indicate that the proposed PUF based authentication saves up to 45% power and uses 12% less memory compared to DTLS handshake authentication.
The third part of this talk addresses the final question, it presents a new detection technique for malicious/abnormal behaviour of embedded using data from Hardware Performance Counters (HPCs).
Finally the talk concludes with a summary of outstanding challenges
Joint Automatic Gain Control and Receiver Design for Large-Scale Multiuser MIMO Systems with Coarsely Quantized Signals
This seminar will present recent work on joint design of the automatic gain control (AGC) and a linear minimum mean square error (MMSE) receive filter for large-scale multiuser multiple input multiple output (MU-MIMO) systems with coarsely quantized signals. The optimization of the AGC is based on the minimization of the mean square error (MSE) and the proposed receive filter takes into account the presence of the AGC and the effects due to quantization. Moreover, we will also provide a lower bound on the capacity of the MU-MIMO system by deriving an expression for the achievable rate. Numerical results will illustrate the performance of the proposed approach against existing techniques.
Biometrics and engineering: what is still missing?
Biometrics-based technology is a name we hear everywhere these days. It is basically related to any authorisation or authentication processthat uses a physiologic or behavioural information that is collected from us to make sure we are who we say we are. Despite the fact that we think this is a closed issue, biometrics data holds quite a lot of open problems that still need attention, epecially now, with the "Smart-anything" revolution. In fact, some of the "closed" problems are not so closed as we think!
In this talk, we will explore the main biometrics open problems that are related with engineering and discuss possible opportunities of research in this area.