Andreas König

Methodology, Algorithms, and Emerging Tool for Automated Design of Intelligent Integrated Multi-Sensor Systems

The emergence of novel sensing elements, computing nodes, wireless communication and integration technology provides unprecedented possibilities for the design and application of intelligent systems. Each new application system must be designed from scratch, employing sophisticated methods ranging from conventional signal processing to computational intelligence. Currently, a significant part of this overall algorithmic chain of the computational system model still has to be assembled manually by experienced designers in a time and labor consuming process. In this research work, this challenge is picked up and a methodology and algorithms for automated design of intelligent integrated and resource-aware multi-sensor systems employing multi-objective evolutionary computation are introduced. The proposed methodology tackles the challenge of rapid-prototyping of such systems under realization constraints and, additionally, includes features of system instance specific self-correction for sustained operation of a large volume and in a dynamically changing environment. The extension of these concepts to the reconfigurable hardware platform renders so called self-x sensor systems, which stands, e.g., for self-monitoring, -calibrating, -trimming, and -repairing/-healing systems. Selected experimental results prove the applicability and effectiveness of our proposed methodology and emerging tool. By our approach, competitive results were achieved with regard to classification accuracy, flexibility, and design speed under additional design constraints.

Minimizing power consumption in wireless sensor networks by duty-cycled reconfigurable sensor electronics

In the last decade, wireless sensor networks (WSN) have gained considerable importance and momentum. They are used with a growing diversity of sensors in a plethora of applications, e.g., from Ambient Intelligence (AmI), Assisted Living (AAL) to agriculture. WSN based measurement and instrumentation systems commonly have to operate under tight power consumption constraints. In particular, WSN benefiting from micro-electro-mechanical-system (MEMS) miniaturization have a very limited power budget due to poor energy density of storage elements. While highly optimized microcontrollers are available by now, todays sensors and sensor electronics are the predominant cause of power consumption in WSN. The standard approach tries to minimize standby currents by low-power sensor, bridge, and amplifier design. In contrast, this paper investigates an approach of duty-cycled, reconfigurable sensor electronics for resistive bridge sensors, that seamlessly integrates into microcontroller sleep modes. The circuit is optimized for the particular case of anisotropic magneto-resistive (AMR) sensor based localization in autonomous WSN. In a systematic analysis amplifier currents, slew-rate, and read-out or on-time Ton are optimized for minimum power consumption of a tri-axial AMR sensor. At the required read-out rates, e.g., one measurement per minute, energy consumption can be reduced to a factor 1.5·10−6 of the continuous operation. A reconfigurable chip in a standard 0.35 µm bulk CMOS technology is under preparation adding flexibility for different sensor types and calibration needs.

Mixtrinsic Multi-Objective Reconfiguration of Evolvable Sensor Electronics

Recently, evolvable hardware has been developed with the objective to deal with sensor electronics problems such as static and dynamic deviations. However the industrial specifications and requirements are not considered in the hardware-learning loop during the intrinsic optimization. Indeed, it minimizes the error between the required output and the real output generated by a given test signal. In our work, we optimize the standard specifications of the hardware to obtain predictable behavior hardware. However, some of the industrial specifications need expensive equipments and some others are time consuming. In this paper we introduce a new approach, which simulates a set of the specifications that is hard to be measured due to the cost/time requirements, e.g., A0, Phi, etc. On the other hand, the set of specifications that is more sensitive to the instance deviations is measured intrinsically, e.g., offset, CMR, etc. We employ the programmable operational amplifier from L. Lakshmanan et al., (2005) as a case study. Our approach succeeds to optimize the amplifier to meet the industrial specifications at low cost setup.

Cost Effective 3D Localization for Low-Power WSN Based on Modified Sammon’s Stress Function with Incomplete Distance Information

A crucial aspect of mobile 3D wireless sensor network application is cost effective localization of individual nodes. GPS based localization is costly and not feasible in many applications. Among GPS free localization methods, Multi-Dimensional Scaling (MDS) based localization methods has been well accepted solution for localization problem, due to low localization error. MDS based localization uses the connectivity information among the nodes to find the positions of nodes in the network. MDS needs a full distance matrix as input. Various methods, e.g., Dijkstra’s algorithm, have been used to generate full distance matrix from incomplete distance matrix obtained from the connectivity information of nodes at the price of O(n2) added computational complexity. This paper presents a novel idea of trying to generate the position information at lowered computational cost and correspondingly reduced power requirement without need for full distance matrix. We will present a modified cost function and demonstrate Sammon’s Mapping with standard gradient descent techniques as well as Genetic algorithms and Particle Swarm Optimization. Our experimental results show, that the proposed localization approach can lead to savings in numerous practical cases. With regard to mapping error reduction, standard GA and PSO so far did not offer a improvement. In future work variations of GA and PSO will be investigated.

Intelligent magnetic sensing system for low power WSN localization immersed in liquid-filled industrial containers

Wireless sensor networks (WSN) have become an important research domain and have been deployed in many applications, e.g., military, ambient intelligence, medical, and industrial tasks. The location of wireless sensor nodes is a crucial aspect to understand the context of the measured values in industrial processes. Numerous existing technologies, e.g., based on radio frequency (RF), light, and acoustic waves, have been developed and adapted for requirements of localization in WSN. However, physical constraints of the application environment and each localization technology lead to different aptness. In particular, for liquid media in industrial containers, determining the location of every sensor nodes becomes very challenging. In this paper, a localization concept based on intelligent magnetic sensing system using triaxial anisotropic magnetoresistive (AMR) sensor with appropriate switched coils combined with a centralized localization algorithm based on iterative nonlinear mapping (NLM) is presented. Here, our system is extended by low power and fast localization based on triangulation for feasible local position computation. The experimental results, both in the air as well as in liquid filled stainless steel container, delivered in the average an absolute localization error in the order of 6 cm for both NLM and triangulation. In future work, we will scale our approach to industrial container size required for beer brewing industry and increase the accuracy and speed by timely electronics and calibration.

Robustness investigation of a SVM-based knock detection method

Knock detection on the basis of structure-borne noise signals that are processed through a sequence of filtering, rectifying, and integrating has been performed in series for many years. It is implemented in Bosch engine management systems that are used by automotive manufactures all over the world. The trend towards higher power density, lower fuel consumption and legal limitation of emissions calls for increasingly complex engines, thereby diversifying the requirements for reliable knock detection.

Comparison of Effective Assessment Functions for Optimized Sensor System Design

Currently, the design of the signal processing and recognition architecture for intelligent sensor systems is still a tedious and labor-intensive task. Optimization techniques, e.g., from gradient descent, stochastic search or evolutionary computation are available to accelerate and automate this procedure. However, appropriate assessment or cost functions are required. This paper presents effective state-of-the-art techniques and compares them to two novel, salient modifications. These are a normalized compactness measure, which has salient properties as it is non-parametric, easy-to-use, fine-grained, competitive performing, and the sum-volumetric k-NN classifier. The aims of this paper are to compare the computation complexities, discriminant properties or capabilities, and sensitivity to control parameter. The methods will be described and compared for the task of dimension reduction by feature selection. Achieved results underpin their saliency for general optimized sensor system design.

Towards Generic On-the-Fly Reconfigurable Sensor Electronics for Embedded System- First Measurement Results of Reconfigurable Folded

The adaptation and robust sensing capabilities of living organisms remain envy to engineers. Several research efforts have been started to mimic these capabilities and exploit them in technical devices, and systems. Embedded systems for sensor applications, comprise of irreplaceable analog and mixed signal components. The considered electronics and sensors themselves are prone to numerous static and dynamic influences and mismatches. Precise design methodology, trimming/calibration is mandatory to restore the functionality. Recent block level granular approaches using field programmable analog array and the more recent approaches from evolutionary electronics providing transistor level granularity using field programmable transistor array offers considerable extensions. In our work, we started on a new medium level granular approach called field programmable medium granular mixed signal array (FPMA) providing basic building blocks of heterogeneous array of active and passive devices to build established circuit structure which are adaptive, fault-tolerant, bio-inspired, and dynamically reconfigurable i.e., trimmable. FPMA also supports rapid prototyping. Our design objective is to create cells of clear compatibility to that of industrial standards having predictable behavior and maintaining quality along with the incorporation of design knowledge. In this paper, measurement results of our first dynamic reconfigurable operational amplifier in an extrinsic fashion are presented. Specific generic configurable cells under the control of optimization techniques are considered. The aspired embedded system architecture will be illustrated and finally the summary of results will be furnished

True Front-to-Back Analogue IC Designers' Training

Analogue and mixed-signal sensor electronics comprise a small yet essential fraction of a large variety of embedded and integrated systems in industrial application systems. While the number of devices is considerably smaller than in massively parallel digital systems, the design of each device requires much more care, time and experience; moreover, learning the skills required for analogue and mixed-signal sensor electronics can be tedious, and these challenges are not taken up by a sufficient number of students to meet industry demands, in particular of the booming sensor and microelectromechanical systems industries. To attract more students to this topic, we offer design courses that combine theoretical and practical skills based on state-of-the-art tools and technologies. Students undertake (individually or in small groups) analogue/mixed signal design projects, but, due to the time constraints common in engineering courses, they are often unable to test to their designs. Drawing from research work and results on dynamically reconfigurable sensor electronics, we have tackled this problem by giving students an opportunity to download their dimensioning solution into certain reconfigurable devices and to carry out testing and measurement to characterise their designs. To this end, we introduced the true front-to-back (TFB) analogue IC designers training kit, used in the CAD laboratory at the Institute of Integrated Sensor Systems (ISE) in TU-Kaiserslautern for various engineering curricula. This paper gives the details of the teaching concepts and state of implementation, the reconfigurable hardware platform and typical results for student design and measurement activities. Finally, we give an outlook on upcoming reconfigurable chips.

Automated and Holistic Design of Intelligent and Distributed Integrated Sensor Systems with Self-x Properties for Applications in Vision, Robotics, Smart Environments, and Culinary Assistance Systems

The ongoing advance in micro technologies gives rise to increasingly versatile and capable sensors as well as unprecedented computational power and communication options in diminishing scale. The notion of smart dust summarizes ubiquitous computing and sensing application systems, which can serve for local as well as global information acquisition and decision making. For off-the-shelf-nodes, and even more for dedicated physical designs, the system design process becomes increasingly challenging and potentially intractable. Automated design methods emerging for intelligent systems are introduced as a remedy. These considerations will be extended to variations, that multiple system instances have to face in real-world applications and potential compensation by incorporation of self-x properties. These concepts are elucidated for the case of reconfigurable and evolvable sensor electronics. Finally, an application perspective of the presented approach for integrated distributed sensing in home automation, assisted living, and in particular, smart kitchen applications, denoted as culinary assistance systems will be presented.