Advances in GPU have facilitated design and execution of complex and computation-intensive deep learning models. As the model complexity increases, the risk of encountering problems due to very large model size, individual tensor size, Not a Number (NaN) value, and memory leak increases as well. When untreated, these problems lead to substantial increase of execution time, generating unpredictable results, and memory leak exceptions. In this paper, we address these problems and particularly large tensor support, C++ kernel changes, and recompilation of the TensorFlow framework. In addition, issues related to NaN value debugging with existing debugging toolkits and solutions to alleviate memory leaks will be explored. Based on experience gained from our analysis, we propose solutions related to better tensor dimension sanity checks, alternative tensor loop procedures, different ways of applying kernels to tensors, a debug trace file filter method, and ways how memory leak exceptions can be resolved. While these problems and solutions may be applicable to running any complex and computation-intensive deep learning model, we described how we encountered them in a use case, in which we designed a deep learning model for activity and gesture recognition using radio data aiming to mitigate domain shift problem.

Internet of Things (IoT) devices that are wirelessly connected in mesh networks often need mutual clock time synchronization, to enable chronological ordering of sensor events, coordination of asynchronous processes across devices, or network-wide coordination of actuators. Accurate time synchronization is also critical for IoT device event logging and debugging. The widely-adopted Network Time Protocol (NTP) is less suitable for constrained mesh networks due to its high network load and sensitivity to asymmetric network delays. In this paper, Mesh Time-synchronization Protocol (MTP) is proposed for IP-based IoT wireless mesh networks and shown to outperform NTP. MTP exploits one resource-rich node on the mesh, the gateway or border router, to be the time reference or to synchronize with a (global/UTC) time reference. This reference time is efficiently disseminated throughout the mesh, using radio broadcasts. Our MTP implementation achieves a median accuracy of 0.48 ms, 1.01 ms, 1.50 ms, 2.06 ms on an event-driven embedded OS platform and 0.74 ms, 1.60 ms, 2.43 ms, 3.28 ms on a multi-threaded embedded platform respectively for mesh nodes 1, 2, 3 and 4 hops away from the gateway. Compared to NTP, the accuracy error variability over repeated measurements is an order of magnitude lower for MTP.

Dynamic Street Lighting Systems (DSLS) have been advertised as a means to reduce cost, reduce light pollution and increase feelings of safety. The dynamism is based on sensing and actuation, and uses communication between the light poles. While there have been many proposals for dynamic street lighting there is actually little data available about their operation. In this paper we present results from the monitoring of an installed DSLS in Eindhoven, The Netherlands for a period of more than a year. Besides questions of energy saving, we examined commissioning problems, effects of message loss, effects of the weather and general performance characteristics. The contribution includes a novel model for describing the outcomes of metrics as a function of sensor information. The results demonstrate that energy saving can be a strong driver for the introduction of DSLS but economic savings much less; the results also show a versatile usage of the collected data.

Many smart city lighting applications require information about the location of light poles, in particular about which light poles are neighbors along the street with respect to passing traffic. This paper addresses the problem of deriving the topology of light poles in a smart city, relying only on the data gathered from Passive Infrared Sensors attached to the light poles. A statistical algorithm is presented and evaluated based on a data set gathered from a real deployment in a residential area.

Simulations are often used to improve the development and test time of industrial embedded systems in different domains, such as healthcare, railway, automobile and aerospace. This paper introduces a general framework capable of simulating real components simultaneously with simulated ones, in real-time, and with a low impact on the performance of the overall system. Our framework is loosely dependent on the application and the domain where the application is developed in. More importantly, it enhances interoperability and reuse, because it paves the way to reuse existing simulating tools and models and to integrate separately developed models from earlier stages in the design into a single coherent simulation.

We present a solution for service discovery of resource constrained devices based on mDNS/DNS-SD. We extend the mDNS/DNS-SD service discovery protocol with support for proxy servers. Proxy servers temporarily store information about services offered on resource constrained devices and respond on their behalf while they are not available. We analyze two protocols for the delegation mechanism between a service provider and a proxy server: an active proxy protocol, as used in the mDNS/DNS-SD implementation by Apple, and a new, passive proxy protocol. We implement and simulate both approaches. Based on the delay and energy usage, we show that the second approach converges faster, thus saving more energy by allowing the resource constrained device to be turned off earlier.

Optimizing performance of wireless sensor networks leads to trade-offs between local computation and communication. With the usage of on-node processing, the lifetime of the sensor nodes can be extended as raw data does not need to be sent out and logged by an external device. In this paper we present our approach for on-node processing of ECG signals to signal abnormal heart rate conditions. We develop an algorithm that is significantly simpler than traditional approaches but has similar quality and that can run on sensors with very limited resources. We give results of our approach when applied to reference sequences. We also estimate the amount of information that gets transmitted over the wireless link in case of applying the on-node processing algorithm and compare it with the remote logging case. We propose fusing movement information with ECG data to provide a more accurate assessment of the patients current health.

Smart spaces typically consist of collaborating heterogeneous nodes with various resource capacities, e.g. processing, memory, storage, energy capabilities. Low capacity nodes can operate using very simple protocols, allowing them to provide and consume only simple services, as opposed to devices with higher capacity, which are not limited by such constraints. Providing seamless connection and collaboration between low and high capacity nodes, i.e. interoperability, remains a challenge. This paper presents an approach for interoperability among nodes of a smart lighting system via a gateway node considering nodes of two platforms: low capacity wireless light sensors and actuators on the one hand and high capacity Smart-M3 nodes on the other hand. The gateway node is able to communicate over heterogeneous networks using different communication protocols, transferring light model information between the smart space nodes.

In this paper we present a complete approach to programming sensor networks. The main contribution concerns the separation of three abstraction levels, viz., application level, network level and node level, leaving more room for standardization than with current practices. In addition we propose to program the network from an overall perspective rather than programming individual nodes. The proposed model is event- based, corresponding closely to the nature of wireless sensors and admits content-based addressing of nodes and groups of nodes. The paper describes how the three abstraction levels come together and gives examples of the approach at all three levels.

Service Oriented Architecture (SOA) is emerging as an enabling technology for sharing distributed heterogeneous resources on the network. Consequently, securing services is an increasing concern. Research issues include privacy protection for service providers, transparent access control for service consumers, secure service discovery and composition. In this paper, we present an access control approach which uses virtual communities to secure service discovery and access (VICSDA). Services grouped in virtual communities can only be discovered and accessed by authenticated community members. Meanwhile, services are autonomous to define their local access control policy. Moreover, behavior of these autonomous services is monitored in order to guarantee a better QoS provision. Using a virtual community overlay network on top of a SOA infrastructure, VICSDA can provide authentication, message confidentiality and integrity to secure service discovery and access. Better application performance can be achieved through VICSDA. We integrated VICSDA with a 3D video streaming application. This example provides us with some initial evidence that VICSDA is a viable solution to our target problems.

A redundant wire-driven parallel robot, the WARP manipulator, is a new robot mechanism which is suitable especially for high speed assembling of lightweight objects such as semiconductors. In this paper, a WARP manipulator was designed from the viewpoint of working space by considering conditions that the travelling plate must be abbe to generate arbitrary amounts of force and moment and the wires are not to be in contact. Simulation results of working space proved the validity of the design procedure described. A prototype of WARP manipulator was constructed. The kinematic parameters were calibrated for position control.

Essential points in robotic facilities for search and rescue at large-scale urban earthquakes were investigated. A parallel robot driven by multiple cables is proposed for transfer of debris. It is portable and is assembled rapidly in destroyed houses. An identification algorithm of mechanism parameters peculiar to this portable robot was studied. On the basis of identification error analysis, optimal number of measurement points and the size of an identification reference frame were obtained.