V. R ELATED W ORK
The increasing interest in wireless technologies for indus- trial applications has motivated the development of algorithms and solutions to support this type of applications . These solutions, implemented at the network and lower layers, aim at improving communication reliability, security, and delay. The work in this paper proposes an application-level solution for reliability, that can be integrated with lower level solutions, to further improve the reliability and enforce the consistence of the system. Similarly, Feng et al.  propose an application- level solution that relies on packet loss knowledge to keep controlling the system even when data is lost. Nevertheless, given the distributed nature of WSANs, guaranteeing a con- sistent state of the global system in this case is challenging.
Let us say that node M1 and the AP both have packages to exchange. If AP gains the channel after the contention, it starts transmitting in the HD mode to M1. When M1 receives AP’s package successfully, it sends an acknowledgment packet (ACK) back to the AP with a bit called “head of line” (HOL) set to 1 in HD mode. HOL = 1 means that the first packet in the M1 buffer is destined for AP. At this point, both nodes know that they should switch to FD mode since both have packages to exchange. After the first FD transmission, two nodes may still have more data packets to handshake. In this case, AP and M1 pick a back-off number randomly from [0 : CW AP max ] and [0 : CW M 1 max ] independently, where CW max is the maximum contention window of each node. Then, they transmit these numbers in their ACK packages as the shared random back-off (SRB). M ax(SRB AP , SRB M 1 ) will be chosen at both nodes as the shared back- off number and both nodes start to countdown until the back-off timers expire. The only difference between SRB and a general back-off is that the countdown process for SRB doesn’t pause once the channel gets busy. Instead, after the expiry of the back-off the nodes wait for DIFS, check the channel and start their next FD transmissions if the channel is not occupied. In case another node wins the contention before SRB expires, AP and M1 should go through the regular contention to access the channel for the remaining data packets.
allows a node to act as a relay for others, but, beyond one hop, a protocol is required for routing packets through the multi- hop network. Note that the constrained devices with limited memory and processing resources can be interconnected by a variety of links, such as IEEE Std 802.15.4, Low Power WiFi or Power Line Communication (PLC) links. LLNs are transitioning to an end-to-end IP-based solution to allow inter- operability across the networks. To enable and manage such an interconnection, we use the IPv6 Routing Protocol for Low- Power and Lossy Networks (RPL), one of the most adopted routing protocols for the IoT. RPL has been standardized in the Internet Engineering Task Force (IETF) Routing Over Low power and Lossy networks (ROLL) Working Group (WG) and is adapted to noisy wireless environments and also takes into consideration the computational, memory, and energy constraints of battery-operated network nodes.
1.2. MOTIVATION AND OBJECTIVES
kinds of applications. However, this solution will face the same problem as ZigBee. In fact, it will not be easy to integrate it with Internet, because it would require the deployment of a proxy between the wireless and the wired network. The proxy translates headers in the border router. A second solution is to provide a new transport layer over UDP. This solution should offer a reliable data transfer that is not offered by UDP, which consists of data loss detection and retransmission, and congestion control and avoidance. However, in order to design that protocol with all these requirements, the transport protocol header should include a sequence number, an acknowledgment number, and control flag and other fields that are already included in the TCP header. A third proposition, which is currently under development in the Constrained Restful Environments (CoRE) WG in the IETF, is to leave the congestion control and the loss recovery in the application layer. However, this solution provides a framework for a limited class of applications. At the moment of writing this thesis, the WG provides only a solution for HTTP over UDP. For all these reasons, in this thesis, we choose the fourth and last proposition, which consists of keeping TCP over the low power networks. The choice of TCP allows us to keep all mechanisms provided by TCP for loss recovery and congestion control. In this work, we distinguish most limitations of the TCP deployment over low power networks, and we propose a solution for each one.
In beacon-enabled mode, IEEE 802.15.4 introduces the
concept of superframes (Figure 1). Each coordinator sends periodically – every Beacon Interval (BI) – a beacon, piggybacking the control information. Then, transmissions from its children take place using a slotted CSMA-CA so- lution during the first part of the superframe (CAP) and with dedicated timeslots (GTS) in the second part. A GTS (Guaranteed Time Slot) has to be reserved a priori by a child
Conclusion and Future Work
In this thesis, we considered queue-length-based transmission scheduling for simul- taneously exploring channel conditions and servicing the traffic demand in wirelessnetworks. We considered two general models for the behavior of the channel condi- tions. In the first model, packets failed according to a Bernoulli process with unknown mean, and the problem facing our network was to learn the channel statistics through receiver feedback. In the second model, packet transmissions failed because of colli- sions with an adjacent, uncooperative network that generated packets according to a Bernoulli process. In this model, the adjacent network would attempt retransmission of its packets until its queue was empty, and thus the interference experienced by our network was correlated over time through the queueing dynamics of the adjacent net- work. We began by studying networks consisting of a single user that was under our control and used the resulting insights to extend our analysis to multi-user networks. In Chapter 2, we developed a set of policies that learn channel rates to minimize queue length regret. We proved there exist policies that have regret that is O(1), order optimal, when the best available channel has a success rate that dominates the packet arrival rate to the system. Likewise, we showed that any traditional bandit algorithm may have a regret that grows Ω(log T ). Therefore, to be order optimal, a policy cannot just maximize offered service to the queue and must take the queue backlog into account when making transmission decisions.
the cost of the sub-optimality: resources may not be used optimally (e.g., a better path exists);
the reconfiguration cost: the controller needs to trigger some reconfigurations. For this transient period, resources may be twice reserved, and control packets have to be transmitted.
Since RAW expects to support real-time flows, we have to support soft-reconfiguration, where the novel ressources are reserved before the ancient ones are released. Some mechanisms have to be proposed so that packets are forwarded through the novel track only when the resources are ready to be used, while maintaining the global state consistent (no packet re-ordering, replication, etc.)
For a periodic networking pattern such as an automation control loop, this number is proportional to the Mean Time Between Failures (MTBF). When a single fault can have dramatic consequences, the MTBF
expresses the chances that the unwanted fault event occurs. In data networks, this is rarely the case. Packet loss cannot never be fully avoided and the systems are built to resist to one loss, e.g., using redundancy with Retries (HARQ) or Packet Replication and Elimination (PRE), or, in a typical control loop, by linear interpolation from the previous measuremnents.
Chen et al. (2010) identify the MAC layer responsibilities of IEEE 802.15.4: generating network beacons, synchronizing to network beacons, supporting MAC association and disassociation, supporting MAC encryption, employing unslotted/slotted CSMA/CA (Carrier Sense Multiple Access/Collision Avoidance) mechanism for channel access, and handling guaranteed time slot (GTS) allocation and management. The authors also mention some concerns about Zigbee: WBANs operate at 2.4 GHz and suffer from significant and highly variable path loss near the human body causing Zigbee to yield unsatisfactory performance; the maximum supported data rate is only 250 kbps which is inadequate to support real-time and large-scale WBANs; and there are other issues such as power, data rate, and frequency. ZigBee Health Care public application profile provides a flexible framework to meet Continua Health Alliance requirements for remote health and fitness monitoring. ZigBee may have a better chance to be adopted in the area of home automation and industrial automation and control. Hughes et al. (2012) pose that although 802.15.4 has been proven as a suitable standard for multi-hop WSNs providing QoS and adequate data rates, it is not the best solution for low power communications in WBANs.
I. I NTRODUCTION
The literature on sensor networks has extensively dealt with the extraction and communication of information ob- tained through the sensors. Communicating the obtained information with a high data rate or reliability, while consum- ing as little energy or power as possible, is the primary objec- tive of works in this field, see e.g., , . The approaches used to reduce the power consumption are based on the communication requirements and wireless aspects like access protocols, scheduling and transmission power control . On the other hand, the literature on networked control systems (NCS) concentrates on applications in which actuation plays a major role, see e.g., , , , . These are systems in which the plant and the controller communicate with each other via a communication network. In this context, the energy expenditure incurred is often assumed to be directly related to the frequency of communication through the network, meaning that the less we transmit, the less the energy expenditure due to communication.
IEEE 802.15.4-2015 TSCH implements a channel hopping approach to combat external interference and signal fading and, thus, to achieve high reliability. At the beginning of a timeslot, a device verifies in the schedule if it has to stay awake. If the cell is allocated to the device, the physical frequency to use for transmission/reception is derived from the ASN of the timeslot and the channel offset assigned to this cell. Nodes regularly broadcast Enhanced Beacons (EB) to disseminate control information for unattached devices. The EBs may also be used as a mean for nodes already part of the network to re-synchronize their clocks. To this aim, an EB frame may contain synchronization of ASN and Join Metric, channel hopping sequence identifier and TSCH slotframe.
Optimal power control in wirelessnetworks is studied in ; it shows that maximizing concave rate utility functions can be formulated as geometric programming (GP) problems in the high-SINR regime, which are convex and hence their solutions can be computed efficiently. In the medium- to low- SINR regime,  describes an iterative approximation method to maximize a concave rate utility function by solving a series of GPs, and  proposes an algorithm to maximize a linear function of the rates by approximating the feasible SINR region by a series of polyblocks. In , sufficient conditions are presented for the convexity of the feasible QoS region, with optional constraints on the allocation of user power. Utility maximization through joint optimization of adaptive modula- tion, rate allocation, and power control is investigated in . In  power control is studied in frequency-selective Gaussian interference channels. Outage probabilities corresponding to different fading distributions for network users and interferers are derived in . For interference-limited wirelessnetworks, optimal power control is considered in  under Rayleigh fading subject to outage probability constraints.
Innovative Internet of Things (IoT) applications with strict performance and energy consumption requirements and where the agile collection of data is paramount are rousing. Wireless sensor networks (WSN) represent a promising solution as they can be easily deployed to sense, process, and forward data. The large number of Sensor Nodes (SNs) composing a WSN are expected to be autonomous, with a node's lifetime dictated by the battery's size. As the form factor of the SN is critical in various use cases such as industrial and building automation, minimizing energy consumption while ensuring availability becomes a priority. Moreover, energy harvesting techniques are increasingly considered as a viable solution for building an entirely green SN and prolonging its lifetime. In the process of building a SN and in the absence of a clear and well-rounded methodology, the designer can easily make unfounded decisions about the right hardware components, their configuration and data reliable data communication techniques such as automatic repeat request (ARQ) and forward error correction (FEC). In this thesis, a methodology to better optimize the design, configuration and deployment of reliable ultra-low power WSNs is proposed. Comprehensive and realistic energy and path-loss (PL) models of the sensor node are also established. Through estimations and measurements, it is shown that following the proposed methodology, the designer can thoroughly explore the design space and make most favorable decisions when choosing commercial off-the-shelf (COTS) components, configuring the node, and deploying a reliable and energy-efficient WSN.
Fig. 1. Evaluation of distorted videos by real humans
III. A DMISSION C ONTROL M ECHANISM
We propose an admission control mechanism based on PSQA tool previously described. We place our context in wire- less access environments such as IEEE 802.11 standards with infrastructure mode, meaning that all traffic passes through an access point (AP). This choice has been made because we want the access point to act as controller equipped with PSQA tool. The idea is to have access points monitor MOS for each connection in order to have knowledge of the perceived quality level of the service and then decide whether to accept a new connection or not accordingly.
Fig. 1. The topology structure of the distributed cooperative MHWN.
II. SYSTEM MODEL
The system consists of a set of N nodes with same computation and transmission capabilities, commu- nicating through bidirectional wireless links between each other. There are wireless gateways providing access to the other networks (i.e.,Internet), and the system architecture is composed of such two kinds of communications as Node-to-Gateway and Node-to-Node, as shown in Fig.1. The characteristics of the wireless cooperative architecture and its differences with the traditional always-connected model motivate the need to revisit the design of the protocols designed for wired infrastructure. In this paper, we employ the cooperative protocol introduced in , which designs a network-wide broadcasting protocol that exploits cooperative diversity and addresses the challenges of: (a) enabling cooperation and (b) exploiting the diversity benefits due to cooperation.
vehicles with opposing driving directions). OBC does not require neither authentication nor association when exchanging data frames. To distinguish frames sent in OCB mode, 802.11p sets the value of Basic Service Set (BSS) identifier (BSSID) field in the data frame header to 0xFFFFFF, also known as wildcard value. IEEE 802.11p utilizes the Enhanced Distributed Channel Access (EDCA) mechanism to provide service differentiation. The basic mechanism of sharing the medium between vehicles relies on the Distributed Coordination Function (DCF) of CSMA/CA. IEEE 802.11p does not alter CSMA/CA rules in the 802.11  (the principles of "carrier sensing" and "collision avoidance"); carrier sensing is achieved through Clear Channel Assessment (CCA) and/or Network Allocation Vector (NAV). Collision avoidance is achieved using a back-off procedure. In a simplest communication scenario under CSMA/CA, if a vehicle has a frame to send, it first senses the wireless medium for Distributed Inter-frame Space (DIFS). If the medium is idle, the vehicle begins transmission of its frame. If the medium is busy, the vehicle performs a random back-off to wait before transmission. The countdown begins when the medium becomes idle. The above mechanism applies to both broadcast and unicast frames. Besides, EDCA enables 4 Quality of Service (QoS) classes by prioritizing data traffic within each node. Hence, each node maintains four queues. These queues have different Arbitrary Inter Frame Spacing (AIFS) and different back-off parameters; the higher the priority, the shorter AIFS. Each transmission queue of an Access Category (AC) operates as an independent DCF station (STA). Figure 1.4 shows the basic channel access procedure in DCF. Basically, in unicast communication, the sender transmits a packet and waits for an acknowledgment (ACK). If no ACK is received, a back-off procedure is invoked before a retransmission is allowed. For every attempt to send a packet, the size of the contention window (CW) is doubled from its initial value (CW min ) until a maximum value (CW max ) is reached. This enables to separate the nodes
In this work, a distributed control algorithm for energy management in a WSN has been proposed. The energy in the sensor nodes is modeled using a Hybrid Dynamical System representation. The WSN has to provide a given functionality (named “mission”) while taking into account the possibility for nodes to fall in an Unreachable condition. The optimality of our approach has been proven in the case of two nodes and an extension of the theorem is developed to encompass a more general case considering n nodes with harvesting systems. Simulation results on a realistic benchmark and comparison with an Model Predictive Control approach show the potential of the proposed control strategy, since the present distributed controller reduces the number of switches between two modes and the computational workload, besides making the WSN scalable and reliable. These benefits have been validated in experiments on a real test-bench.
Health assessment is a key step for Remaining Useful Life (RUL) estimation. Based on the analysis and the predeﬁned thresholds,
the machine/component's health state is identiﬁed. Sensory data is reported periodically to monitor critical components. This data corresponds to measurements of monitoring parameters and is useful to assess the machine/component's condition. Each monitoring parameter has a threshold; once reached, the system is considered to be in the corresponding state. Reliable health state estimations depend on accurate measurements and fast data processing. The information in question is often gathered by means of individual sensor nodes or via a wired network of sensors. Nevertheless, for some applications, the use of a Wireless Sensor Network (WSN) can be a requirement rather than a choice. For example, due to accessibility or extra weight issues, connecting the sensors through physical wires is not feasible. WSNs are designed for an efﬁcient event detection. They consist of a large number of sensor nodes deployed in a surveillance area to detect the occurrence of possible events. Such an activity necessitates efﬁciency, which is hard to achieve with the constraints of WSNs  .
The AP selection is not trivial, and we previously proposed a methodology called Common Ancestor (CA) , . More specifically, when a node n is choosing an AP node, it ensures that the selected AP has in its PS the node which is the PP of n. In order to support this selection mechanism, additional information is carried in the DIO control message that all nodes broadcast. The information carried is the list of IPv6 addresses corresponding to the m lowest ranking (and therefore most preferable) parents in node n’s parent set. We have extended the operating system used (Contiki OS) to include this information in the DIO message, in the DAG Metric Container (MC) field. 1 The interaction of RE with the TSCH schedule only concerns the required existence of cell(s) for transmitting packets from a node not only to its PP but also to the AP.
channels before hand-off is performed at time slots that are not dedicated for the current communications.
We thus consider the well-known CDMA over TDMA channel model at the MAC sub-layer proposed in (Lin et Liu, 1999) (Lin, 2001). That is, CDMA is overlaid on top of the TDMA infrastructure. Namely, multiple sessions can share a common TDMA slot via CDMA. The transmission and reception between two neighbours at the MAC layer are governed by the TDMA model. A node that wishes to transmit signals must use a free timeslot for transmission, and the node that wishes to receive the signals needs to listen to the transmitting node in the same timeslot. In this channel model, a radio station can only receive a single transmission at a time and cannot transmit and receive simultaneously. In addition, the channel is assumed to be time slotted. All nodes keep accurate common time (there exits a global clock or time synchronization mechanism). Each slot includes the number of redundant bits (e.g., for error control coding, retransmission) that must be sent when the channel has a low signal-to-noise ratio in order to get a successful transmission. Furthermore, the assumptions found in most other radio data link protocols (Baker et Ephremides, 1981; Chlamtac et Kutten, 1985; Chlamtac et Pinter, 1987a; Gerla et Tsai, 1995; Goodman et al., 1989; Lin et Gerla, 1997) are considered. That is, the physical layer can provide the service of the slotted channel. Only one data packet can be transmitted in each data slot.