Section: New Results

Wireless Networks

Participants : Osama Arouk, Btissam Er-Rahmadi, Adlen Ksentini, Yassine Hadjadj-Aoul, Quang Pham Tran Anh, Hyunhee Park, César Viho.

We continue our activities around wireless and mobile networks, where we focus particularly on 4G/5G networks as well as on a new mobile architecture known as mobile cloud.

LTE improvements. In [35] , we investigated, at both the core network (EPC) and Radio Access Network (RAN), the impact of caching the shared content among users. We reviewed the different locations were data could be cached and their impacts on user QoS/QoE. In [33] , we proposed several new mechanisms to handle the gateway relocation in the context of highly decentralized mobile network. To evaluate these mechanisms, we proposed an analytical model based on Markov Chains, whereby we captured the randomness of user mobility and its impact on the user QoS in terms of the probability to be connected to the optimal gateway, the drop rate, etc. In [32] , we devised an agile admission control mechanism that anticipates QoS/QoE degradation and proactively defines policies for admitting UEs handing-in from the macro network to the small cell network. It also enables IP flow mobility between small cells and macro networks. We provided an analytical model to the admission control mechanism based on Markov Decision Processes (MDP). The ultimate objective of the proposed model is to derive the optimal policy (i.e., reject or accept flows in the macro or the small cell) which maximizes users' QoE under different load scenarios (low and high load user traffic). Another work regarding small cells in LTE was proposed in [76] , where we used the small cell principle to extend the mobile network coverage in emerging countries that not include a wired infrastructure. The proposed framework aims to backhaul the small cell with the less costly connection, while ensuring minimal QoS to users. In this vein, we formulated this problem through an Integer Linear Program (ILP), and solve it for small network sizes. For large instances of the network size, we proposed two new heuristics. In [30] , we investigated network decentralization in conjunction with the Selective IP Traffic Offload (SIPTO) approaches to handle the mobile increased data traffic. We first devised different approaches based on a per destination domain name basis, which offer operators a fine-grained control to determine whether a new IP connection should be offloaded or accommodated via the core network. Two of our solutions are based on Network Address Translation (NAT) named simple-NATing and twice-NATing, while a third one employs simple tunneling and a forth proposal adopts multiple Access Point Names (APNs). We also proposed methods enabling User Equipment (UEs), both in idle and active modes and while being on the move, to always have efficient Packet Data Network (PDN) connections. A qualitative analysis and a simulation study compared the different approaches with respect to cost, complexity, service continuity and network performance, demonstrating the significance of the proposed schemes for multimedia applications.

M2M. We addressed another type of traffic that appeared these last years, namely Machine to Machine (M2M) communication or Machine Type Communication (MTC). Such traffic is known by its intensity and its impact on increasing congestion in both parts of 4G networks, the Radio Access Network (RAN) and the core network. The main spirit of the proposed solutions is to proactively anticipate system overload by reducing the amount of MTC signaling messages exchanged in normal network operations. In [49] we introduced a solution that operates at the core network. We proposed that the Mobility Management Entity (MME), or an alike core network node, computes the device trigger rate that alleviates congestion, and communicates this value to the MTC-Interworking Function (MTC-IWF) element that enforces MTC traffic control, via admission control or data aggregation, on the device trigger request rate received from the different MTC servers.

As mentioned earlier, the MTC would impact not only the EPC part, but also the RAN. Group paging is currently considered as one of the most efficient mechanisms proposed to alleviate the problem of the RAN overload. In [42] we introduced a new solution to improve the performance of the current group paging method and overcome its disadvantages. The proposed solution is intended for MTC devices in connected mode state, in which they have an RRC context without being synchronized with the network. In [41] we devised a novel algorithm which estimates the network status (the number of active devices), thus better controlling the RAN access. Unlike most existing methods that consider only one channel, the proposed solution uses the statistics of all the channels in order to estimate the number of arrivals (UE and MTC devices) in each RA (Random Access) slot.

Most of the above-proposed solutions are basically incremental ones. In [31] , we devised a complete new architectural vision to support MTC in mobile networks. This vision relies on the marriage of mobile networks and the cloud, specifically based on Network Function Virtualization (NFV). The proposed solution simplifies the network attach procedure for MTC devices by creating only one NFV MTC function that groups all the usual procedures. By doing so, the proposed solution is able to create and scale instances of NFV MTC functions on demand and in an elastic manner to cope with any sudden increase in traffic generated by MTC devices.

Wireless Sensor Networks (WSN). WSNs are complex systems that are mainly limited by the battery life of the nodes in order to have an adequate performance. In most cases, it is possible to have a re-deployment of new nodes in order to prolong the systems lifetime. This leads to a situation where some nodes have a low energy level while other nodes (the majority of nodes a few instants after the re-deployment procedure) have high energy levels. In these environments, it is clear that ancient nodes, those with low energy levels, have to contend for the shared medium against the majority of high energy nodes. As such, the remaining battery life of low energy nodes would be rapidly consumed. In [64] , we propose to extend the battery life of low energy nodes by means of assigning prioritized access to the shared channel to those nodes. The goal is to content among a low population of such nodes, while delaying the contention access of high energy nodes which can support higher number of collisions before energy depletion. This is done by studying two different transmission strategies referred to as “hard” and “soft” transmission probabilities. Results show that a soft transmission strategy achieves better results in terms of reduced energy consumption than both the conventional protocol or a hard transmission assignment.

The communication between nodes is the greedy factor to the energy consumption. One important mechanism to reduce the energy consumption is the in-network data aggregation. This mechanism removes repeated and unnecessary data readings and thus cuts on the energy used in communications. In [14] we reviewed the state of art on this topic. Then, we proposed a classification of the available solutions according to the way the aggregation is done. In [15] , we addressed the reliable minimum data aggregation scheduling problem in wireless sensor networks under multi-channel frequency use. The proposed solution ensures the collection reliability and reduces the latency in disseminating aggregated data to the base-station over multi-frequency radio links. Another mechanism to improve energy efficiency is to optimize link scheduling when using TDMA-based techniques and data fragmentation when using slotted CSMA/CA access methods. In this line, we proposed a protocol, named DLSP, with the objective of achieving both low energy consumption and low latency in Wireless Sensor Networks. DLSP takes advantage of the spatial reuse of interference-free time slots by means of conflicts graphs. Unlike the previous studies that often consider saturated nodes, we propose to relax the saturation assumption in order to maintain good performance when some of the nodes have no data to send  citemouloud:hal-01101396. In [55] , we noticed that the standardized slotted CSMA/CA may lead to a wastage of the bandwidth utilization and an additional transmission delay. This drawback is mainly caused by Deferred Transmission in the CSMA/CA algorithm at the end of the superframe, when there is not sufficient time to complete the frame transmission. Thus, we proposed to fragment a data frame into a short frame and attempt its transmission in the current frame and transmit the remaining frame in the next superframe. The data fragmentation mechanism was modeled using a Markov chain. A non-saturated traffic and acknowledgement transmission are considered in our analysis.

High data rate WiFi networks. The IEEE 802.11ac Task Group (TGac) is actively working on an amendment that allows WLAN to reach a maximum aggregate network throughput up to 7 Gbps on bands below 6 GHz. In particular, the standard envisions a maximum Medium Access Control (MAC) throughput of at least 500 Mbps for a single user, and at least 1 Gbps in case of multiple users. In [36] we proposed an analysis of the IEEE 802.11ac TXOP Sharing mechanism, which was recently introduced by the 802.11ac group, by providing a Markov chain-based model. Based on the proposed Markov chain, we provided an analytical model of the achievable throughput for each AC. Accordingly, we can analyze the impact of the TXOP Sharing on the throughput of each AC, hence highlighting the improvement achieved in terms of bandwidth utilization and channel access fairness among the different ACs.

Mobile cloud. One of the 5G-architecture visions considers the usage of clouds to build mobile networks and help in decentralizing mobile networks on demand, elastically, and in the most cost-efficient way. This concept of carrier cloud becomes of vital importance knowing that several cloud providers are distributing their cloud/network, globally deploying more regional data centers, to meet their ever-increasing business demands. As an important enabler of the carrier cloud concept, network function virtualization (NFV) is gaining great momentum among industries. NFV aims for decoupling the software part from the hardware part of a carrier network node, traditionally referring to a dedicated hardware, single service and single-tenant box, that is using virtual hardware abstraction. Network functions become thus a mere code, runnable on a particular, preferably any, operating system and on top of a dedicated hardware platform. The ultimate objective is to run network functions as software in standard virtual machines (VMs) on top of a virtualization platform in a general-purpose multi-service multi-tenant node (e.g., Carrier Grade Blade Server) put into the cloud. In [26] , we presented a LISP-based implementation of the Follow Me Cloud (FMC) concept, whereby mobile services hosted in federated clouds follow mobile users as they move and according to their needs. This implementation clearly demonstrates the feasibility of the FMC concept. On the other hand, service migration in FMC may be an expensive operation given the incurred cost in terms of signaling messages and data transferred between DCs. Indeed, decision on service migration defines therefore a tradeoff between cost and user perceived quality. In [48] we addressed this tradeoff by modeling the service migration procedure using a Markov Decision Process (MDP). The aim was to formulate a decision policy that determines whether to migrate a service or not when the concerned User Equipment (UE) is at a certain distance from the source DC.

In order to meet the general needs of mobile operators, efficient mobile cloud must give high importance to the placement/instantiation of mobile network functions (such as data anchor gateways) in the federated cloud. In [43] we argued the need of using service/application type and requirements as metrics for efficiently: (i) create virtual instance of the Packet Data Network Gateway (PDN-GW); (ii) select the virtual PDN-GW for UEs with specific application type. After modeling this procedure though a nonlinear Optimization Problem (OP) and proving it as a NP-hard problem, we proposed three solutions to solve this issue.

Wireless Local Area Networks. User-centric networking has emerged as a disruptive new communication paradigm. We particularly focused on its expressions in wireless networking and the challenges it brings about [23] . In this context, by means of testbed experiments and simple analytic models, we quantified the upper bounds on VoIP capacity of a purely user-centric secure VoIP communications scheme that we designed, identifying the major quality degradation factors. Our results have shown that typical user Wi-Fi equipment can sustain a satisfactory number of concurrent secure VoIP sessions with acceptable QoE and, at the same time, protection from malicious user activity can be offered to access providers, while a level of roaming privacy can be guaranteed [24] . We then studied the role of users in wireless network management tasks. In particular, we proposed a scheme where monitoring the topology of Wi-Fi deployments is crowdsourced to roaming users, who submit reports on wireless coverage in their vicinity [25] . Topology information can then be used as input to reconfiguration mechanisms, such as channel assignment schemes. Users cannot be assumed trustworthy, though. They can engage in fraudulent reporting, which, unless specific countermeasures are in place, can severely impact one's view of the network topology. To this end, we designed and implemented an architecture for accurate Wi-Fi topology discovery, devising a reputation-based mechanism to tackle realistic and simple to implement attacks. We have shown analytically and via simulation that, even in the presence of large numbers of attackers, our user-centric scheme significantly outperforms pure infrastructure-based approaches, where monitoring is carried out only by trusted Access Points.

In another line of research, we focused on efficiently integrating wireless users in an Information-Centric Network (ICN) architecture. In ICN, multicast content delivery is the norm. At the same time, wireless multicast is problematic. To address this issue, we took advantage of the content awareness inherent in ICN and proposed a relay-based approach for local wireless multicasting: ICN information scoping mechanisms assist in expressing content semantics and, in turn, encoding the heterogeneous performance requirements of different content/application types. Under this premise, we proposed a multiobjective optimization approach for relay selection and multicast transmission rate assignment which allows to optimize for reliability, delivery time, or energy cost on a per content basis [47] .

Energy saving. Another part of our activities in wireless network are related to energy saving. Indeed, one of the biggest problem today in the wireless world is that wireless devices are battery-driven, which reduce their operating lifetime. The experimental measurements we have achieved in [16] and [17] revealed that operating system overhead causes a drop in performance and energy consumption properties as compared to the GPP in case of certain low video qualities. We propose, thus, a new approach for energy-aware processor switching (GPP or DSP) which takes into consideration video quality. We show the pertinence of our solution in the context of adaptive video decoding and implement it on an embedded Linux operating system.

Adaptive Beam Scheduling for Scalable Video Multicast in Wireless Networks.  Design of efficient multicast for a scalable video coding (SVC) streaming combined with directional beamforming is a challenging issue. In [29] , we propose a QoE-aware directional beam scheduling (QBS) scheme which optimizes overall quality of experience (QoE) for multirate multicast of SVC, with beamforming in wireless networks. We optimally schedule different SVC layers to different beams and rate modulations. We provide a mixed integer linear programing (MILP) formulation of the problem, and then propose a heuristic algorithm. Extensive simulation results demonstrate that QBS can increase the overall QoE and can satisfy a minimum expected QoE for all users.