Coping with Network Dynamics Using Reinforcement Learning Based Network Optimization in Wireless Sensor Networks

Due to the constant increase of in density of wireless network devices, inter-network cooperation is increasingly important. To avoid conflicts, advanced algorithms for inter-network optimization utilize cognition processes such as reinforcement learning to enable networks to become capable of solving complex optimization problems on their own with minimal outside intervention. This paper investigates the inherent trade-offs that occur when using reinforcement learning techniques in dynamic networks: the need to keep the network running optimally whilst, at the same time, different (suboptimal) network settings need to be continuously investigated to cope with changing network conditions. To cope with these network dynamics, two existing algorithms, “epsilon greedy” and Softmax, are compared to a novel approach, based on a logarithmic probability distribution function. It is shown that, depending on the expected level of dynamics, the new algorithm outperforms existing solutions.     

IDRA: A flexible system architecture for next generation wireless sensor networks

Wireless sensor networks consist of embedded devices (sensor nodes), equipped with a low-power radio. They are used for many applications: from wireless building automation to e-health applications. However, due to the limited capabilities of sensor nodes, designing network protocols for these constrained devices is currently very challenging. Therefore, this paper presents the IDRA platform: an information driven architecture designed to support next-generation applications on resource constrained networked objects. IDRA supports simple but useful optimizations at an architectural level. These include support for cross-protocol interactions, energy efficiency optimizations, QoS optimizations (packet priorities, dynamic protocol selection), mobility support and heterogeneous network support. The paper shows how the development of protocols is improved by using an architecture which delegates specific tasks to a central system, decreasing the memory requirements of associated network protocols. A thorough experimental performance analysis demonstrates that IDRA is much more scalable in terms of memory requirements, energy requirements and processing overhead than traditional system architectures. Finally, the paper discusses how the optimizations presented in this paper can be used for the clean-slate design of architectures for other wireless or wired network types.     

All-optical 2R regeneration based on integrated asymmetricMach-Zehnder interferometer incorporating MMI-SOA

2R regeneration is proposed with an integrated Mach-Zehnder interferometer containing a monomodal semiconductor optical amplifier (SOA) and an SOA with a 1 × 1 multimode interference (MMI) coupler-SOA in the arms. Simulations and measurement of the static transfer function show a digital shape with a high extinction ratio improvement     

A negotiation-based networking methodology to enable cooperation across heterogeneous co-located networks

In a future internet of things, an increasing number of every-day objects becomes interconnected with each other. Current network solutions are not designed to connect a large number of co-located devices with different characteristics and network requirements. To cope with increasingly large and heterogeneous networks, this paper presents an ‘incentive driven’ networking approach that optimizes the network performance by taking into account the network goals (‘incentives’) of all individual devices. Incentive driven networking consists of the following steps. First, devices dynamically search for co-located devices with similar network preferences and hardware and/or software capabilities. Next, if such devices are found, communities consisting of interconnected objects with similar network expectations are formed on an ad hoc basis. Due to the similarities between the involved devices, it is easier to optimize the network performance of each individual community. Finally, different communities can cooperate with each other by activating and sharing (software or hardware) network resources. The paper describes which (future) research is needed to realize this vision and illustrates the concepts with a number of simple algorithms. Through an experimental proof-of-concept implementation with two networks of resource-constrained embedded devices, it is shown that even these simple algorithms already result in improved network performance. Finally, the paper describes a large number of example use cases that can potentially benefit from our innovative networking methodology.     

QoS Challenges in Wireless Sensor Networked Robotics

Wireless sensor networks and mobile robotics are two hot research topics. Integrating them leads to a wide range of new applications in many different environments such as terrestrial, underwater, underground and aerial. Where sensor networks are mainly used for large-scale monitoring and control, mobile robotics are used for performing fine-scale actions and automation. Network heterogeneity together with stringent Quality of Service (QoS) demands from applications such as voice and video make QoS support very challenging. Therefore, this paper investigates the QoS challenges in wireless sensor networked robotics and presents a novel QoS framework as solution to cope with these challenges.     

Signalling minimizing handover parameter optimization algorithm for LTE networks

In currently deployed cellular networks, a lot of time and effort is put by operators into the optimization of radio resource management functions. The use of self-optimization algorithms will eliminate the need for manual monitoring and offline analysis, as these algorithms constantly strive to improve network performance in an automated fashion. However, expectations have to be kept realistic as in some cases it may not be possible for network performance to be significantly improved. A trade-off has to be made between the performance enhancement that these algorithms can achieve and the resources they need for doing so. This article proposes a mechanism which will minimize signalling determined by the use of a handover (HO) self-optimization algorithm. This algorithm derives the proper values of the main HO control parameters based on the observed network performance, while the signalling minimizing mechanism will be used as a stop condition, as, due to shadow fading or coverage holes, performance can only be improved up to a point. After this point, performance will only slightly vary with different control parameter settings but the observed gain will not compensate for the possible instability and signalling load introduced by these changes. By using the signalling minimizing mechanism in combination with the HO self-optimization algorithm, network performance will be maintained while the signalling load will be significantly diminished.     

Symbiotic Networks: Towards a New Level of Cooperation Between Wireless Networks

In the future, many wireless networks, serving diverse applications, will co-exist in the same environment. Today, wireless networks are mostly optimized in a rather opportunistic and/or selfish way: optimizations methods only use a local view of the network and environment, as they try to achieve the best performance within its own network. The optimizations are very often limited to a single layer and cooperation between networks is only happening through the use of gateways. In this paper, we suggest an alternative paradigm for supporting cooperation between otherwise independent networks, called ‘symbiotic networking’. This new paradigm can take many forms, such as sharing of network resources, sharing of nodes for communal routing purposes and sharing of (networking) services. Instead of optimizing network parameters within the individual networks, symbiotic networking solutions operate across network boundaries. Parameters are optimized between the networks and communal protocols are developed, leading to a more global optimization of the scarce network resources. In this paper, we describe several scenarios which can profit from symbiotic networking and illustrate a strategy for supporting networking protocols which can operate across network boundaries. Ultimately, through the disappearance of network boundaries and the introduction of cross-layer/cross-node/cross-network cooperation, symbiotic networks takes the notion of cooperation to a new level, paving the way for a true network symbiosis.

Material optimisation for AlGaN/GaN HFET applications

An optimisation of some growth parameters for the epitaxy of AlGaN–GaN based heterostructure field effect transistors (HFET) at low pressure in a new 3 _* 2″ MOVPE reactor is presented. Some possible processes for the growth of semi-insulating buffers have been identified and are described. TEM analysis shows that the insulating character is not due to a high density of dislocations, whereas SIMS analysis shows that classical impurity (Si, O and C) concentrations are in the same range as in conductive undoped layers. Further studies are needed to identify the traps responsible for the compensation of the GaN layers. The properties of the two-dimensional electron gas (2DEG) located at the AlGaN–GaN interface can be tuned by modifying the characteristics of the AlGaN layer and of the insulating buffer. The best mobility (1500 cm² V⁻¹ s⁻¹ for n6×10¹² cm⁻²) is obtained when using a thick buffer layer, whereas the sheet carrier density is found to increase with the Al content in the undoped supply layer and reaches 1.1×10¹³ cm⁻² for a composition of 24%.     

Realization of a short integrated optic passive polarizationconverter

For the first time, passive polarization converters with lengths of less than 1 mm and high polarization conversions are realized on InGaAsP-InP. Angled facets in a periodic coupler are used to obtain efficient conversion. Polarization conversions of more than 70%, respectively more than 90%, are measured for two different realizations of the angled facets. The most promising technology used in the realization results in a propagation loss of less than 1 dB/mm for both polarizations     

Differences between super-centrifuged and membrane-filtrated soil solutions obtained from bulked and non-bulked topsoils by soil centrifugation

The composition of the soil solution must be known to estimate the mobility and availability of plant nutrients and contaminants. When soil centrifugation is used to obtain the soil water for analysis, bulked soil samples are normally used in order to reduce random variations. In the present study, large systematic differences between the centrifugates from bulked and core samples are reported. For most of the soil solution components, no, or only small differences could be detected between super-centrifuged and membrane-filtrated solutions.