Méthodes et outils mathématiques pour la modélisation des réseaux IoT
Thèmes scientifiques :
- D - Télécommunications : compression, protection, transmission
Nous vous rappelons que, afin de garantir l'accès de tous les inscrits aux salles de réunion, l'inscription aux réunions est gratuite mais obligatoire.
36 personnes membres du GdR ISIS, et 36 personnes non membres du GdR, sont inscrits à cette réunion.
Capacité de la salle : 73 personnes.
IMPORTANT: L'inscription à la journée est gratuite mais obligatoire! Tous les participants doivent être inscrits pour pouvoir accéder à l'ISEP et une pièce d'identité leur sera demandée à l'entrée dans les locaux.
L'internet des objets connectés, mythe ou réalité, est en phase de développement intensif, tant sur le plan technique que commercial et marketing. Si des solutions techniques (Sigfox, Lora, 5G) se développent très rapidement, elles reposent soit sur des techniques originales dont il apparaît difficile d'évaluer les performances, soit sur l'adaptation de solutions dédiées initialement aux réseaux cellulaires.
D'un point de vue théorique, les réseaux IoT reposent sur un paradigme différent de celui des réseaux mobiles usuels. La question essentielle n'est plus d'atteindre des hauts débits, mais plutôt une grande fiabilité et des délais très faibles, pour la transmission sans protocoles de petites quantités d'informations. Le problème de transmission d'information dans l'IoT devient donc un problème de partage de ressource opportuniste ou un problème de transmission multi-utilisateur avec des flux impulsifs. Des modèles d'interférences, intégrant l'impulsivité des communications, doivent être considérés, impactant ainsi la dérivation des bornes atteignables car la traditionnelle contrainte de puissance lors de la maximisation de l'information mutuelle en régime asymptotique, ou non, doit être adaptée.
Objectifs de cette journée
Identifier les méthodes et formalismes théoriques permettant d'appréhender correctement la problématique du développement de réseaux d'accès radio optimisés pour l'IoT. Les présentations de cette journée s'articuleront autour des sujets ci-dessous :
- Théorie de l'information, en particulier en régime non asymptotique (paquets courts),
- Géométrie stochastique et outils associés permettant de modéliser des réseaux denses,
- Outils de modélisation de canal non gaussien, pour prendre en compte les interférences impulsives,
- Performances des codes correcteurs d'erreurs et construction de code pour IoT
- Méthodes d'estimation ou de codage distribuées.
- Traitement du signal en émission/réception optimisées pour IoT.
Comité d'organisation :
- Claire Goursaud, INSA Lyon, laboratoire CITI (firstname.lastname@example.org)
- Anne Savard, IMT Lille Douai, laboratoire IRCICA (email@example.com)
- Philippe Mary, INSA Rennes, laboratoire IETR (firstname.lastname@example.org)
Cette journée du GdR ISIS est également soutenue par la chaire IoT INSA-SPIE ICS et le projet ANR ARburst. Elle sera ponctuée par des interventions d'orateurs invités du monde de la recherche académique internationale et industrielle.
Orateurs invités :
- Petar Popovski (Aalborg University)
- Marios Kountouris (Huawei)
- Marco Di Renzo (L2S)
- Michele Wigger (Telecom ParisTech)
- Christophe Fourtet (Sigfox)
9h30 - 9h45: Accueil et introduction
9h45 - 10h15: Petar Popovski, Aalbord University, "On access protocols in ultra-reliable wireless communications"
10h15 - 10h30: Lilian Besson, CentraleSupélec Rennes, "Multi-armed bandit learning in IoT networks: learning helps even in non-stationary settings"
10h30 - 10h45: Laurent Clavier, IMT Lille-Douai, IRCICA, "Modeling interference with alpha-stable distributions and copula for receiver design in wireless communications"
10h45 - 11h00: Pause
11h00 - 11h30: Michèle Wigger, Telecom ParisTech, "Distributed hypothesis testing over networks"
11h30 - 11h45: Yasser Mestrah, CReSTIC, "System robust communication in impulsive environment"
11h45 - 12h00: Ingmar Land, Huawei, "Polar codes for short lengths and low rates"
12h00 - 13h30: Pause déjeuner
13h30 - 14h00: Christophe Fourtet, Sigfox, "New approaches for massive IoT"
14h00 - 14h15: Alex The Phuong Nguyen, IMT Atlantique, "State of the art in short packet communications: a physical layer comparison for block fading channels"
14h15 - 14h30: Emmanuel Boutillon, Université Bretagne Sud, LabSTICC, "Construction de code NB-LDPC de très bas rendement pour l'IoT"
14h30 - 14h45: Pause
14h45 - 15h15: Marios Kountouris, Huawei, "Delay performance in mission-critical MIMO communications"
15h15 - 15h30: Romain Chevillon, Polytech Nantes, IETR, "Stochastic geometry approaches for D2D communications in dense heterogeneous networks"
15h30 - 15h45: Apostolos Avranas, Telecom ParisTech, Huawei, "Energy-latency tradeoff in ultra-reliable-low latency communication with short packets"
15h45 - 16h00: Pause
16h00 - 16h30: Marco Di Renzo, L2S, CNRS, "System-level modeling and optimization of the energy-efficiency in cellular networks - A stochastic geometry framework"
16h30 - 16h45: Malcolm Egan, INSA Lyon, CITI, "Capacity sensitivity and non-Gaussian interference models in IoT"
16h45 - 17h00: Lina Mroueh, ISEP, "Statistical dimensioning of massive and dense NB-IoT network"
17h00 - 17h15: Clôture de la journée
Résumés des contributions
Titre : System-Level Modeling and Optimization of the Energy Efficiency in Cellular Networks - A Stochastic Geometry Framework
Orateur : Marco Di Renzo CNRS, L2S Paris-Saclay
Résumé : In this talk, we analyze and optimize the energy efficiency of downlink cellular networks. With the aid of tools from stochastic geometry, we introduce a new closed-form mathematical expression of the potential spectral efficiency (bit/sec/m2) in the interference-limited regime. Unlike currently available mathematical frameworks, the proposed analytical formulation explicitly depends on the transmit power and density of the base stations. This is obtained by generalizing the definition of coverage probability and by accounting for the sensitivity of the receiver not only during the detection of information data, but during the cell association phase as well. Based on the new analytical representation of the potential spectral efficiency, the energy efficiency (bit/Joule) is formulated in a tractable closed-form expression. The resulting optimization problem is comprehensively studied and it is mathematically proved that the energy efficiency is a unimodal and strictly pseudo-concave function in the transmit power, given the density of the base stations, and in the density of the base stations, given the transmit power. Under these assumptions, therefore, a unique transmit power and density of the basestations exist, which maximize the spectral efficiency per consumed power. Numerical results are illustrated in order to confirm the obtained mathematical findings and to prove the usefulness of the proposed framework for optimizing the network planning and deployment of cellular networks from the energy efficiency standpoint. This is a joint research activity with Alessio Zappone, Thanh Tu Lam and Merouane Debbah.
Titre : Delay Performance of Mission-Critical MIMO Communications
Orateur : Marios Kountouris, Huawei R&D Paris.
Résumé : Ultra-reliable, low latency communications (URLLC) are currently attracting significant attention due to the emergence of mission-critical IoT applications and device-centric communication. URLLC will entail a fundamental paradigm shift from throughput-oriented system design towards holistic designs for guaranteed and reliable end-to-end latency. A deep understanding of the delay performance of wireless networks is essential for efficient URLLC systems. In this talk, we present a framework for analyzing the network layer performance of wireless systems under statistical delay constraints. Using tools from stochastic network calculus, we are able to derive probabilistic delay bounds for various service processes. We employ this framework to analyze the delay performance of multi-antenna (MIMO) diversity schemes with perfect and imperfect channel knowledge. The effect of transmit power, number of antennas, and finite blocklength channel coding on the delay distribution is also investigated. Our higher layer performance results reveal key insights of MIMO channels and provide useful guidelines for the design of mission-critical IoT systems that can guarantee stringent latency requirements.
Titre : Distributed Hypothesis Testing over Networks
Orateur : Michèle Wigger, Telecom ParisTech, Paris.
Résumé : As part of the internet of things (IoT), the number of sensor nodes that wish to communicate with each other has exploded and is expected to further increase dramatically. Such an increase of communication devices inherently leads to involved communication and hypothesis testing scenarios, and thus calls for new coding and testing strategies. The talk presents new strategies and corresponding error exponents for different network scenarios, and it proves information-theoretic optimality of the proposed strategies in some cases. Special attention is given to scenarios where information collected at a sensor is desired at multiple decision centres and where communication is multi-hop involving sensor nodes as relays. In these networks, sensors generally compete for network resources, and relay sensors can process received information with sensed information or forward intermediate decisions to other nodes. Depending on the studied error exponents, some of these intermediate decisions require special protection mechanisms when sent over the network. The talk is based on joint work with Sadaf Salehkalaibar, Roy Timo, and Ligong Wang.
Titre : On Access Protocols in Ultra-Reliable Wireless Communications
Orateur : Petar Popovski, Aalborg University.
Résumé : Ultra-reliable low latency communication (URLLC) is an important new feature brought by 5G, with a potential to support a vast set of applications that rely on mission-critical links. Establishing and transmitting a short data packet is associated with multiple steps of the access protocols that involve auxiliary procedures (e.g. channel estimation) and exchange of metadata. The requirement for very high reliability is propagated and further augmented to each of those steps. This talk elaborates on the models used for access protocols and, specifically, provides a critical look on the assumptions used in random access protocols. The talk presents models for grant-free access and contrasts them to the traditional resource reservation procedures used in cellular systems. Specifically, the talk highlights the design alternatives for the authentication procedures, usually treated as an "afterthought" in the theoretical treatment of access protocols. Finally, a discussion is provided on techniques based on coded random access and successive interference cancellation with latency/reliability constraints.
Titre : Modeling interference with alpha-stable distributions and copula for receiver design in Wireless Communications
Orateur : Laurent Clavier, IMT Lille-Douai, IRCICA.
Résumé : Interference (signals coming from undesired transmitters) is a limiting factor in many wireless communication situations and especially for Internet of Things. It is often considered as a Gaussian random variable with independent components but these assumptions (Gaussian and independent) are no longer available in many cases. A better model of interference can be the alpha-stable distribution with copulae to take the dependence structure into account. The main aim of this presentation is to obtain an evaluation (explicit or numeric) of the Log- Likelihood Ratio (LLR) in this interference model. We show that the copula framework allows us to split the LLR into a part depending only on the copula and the marginal distribution of the interference and another which is exactly the LLR for independent noise (that only depends on the marginal distributions). To have closed-form solutions, we will focus on the case where the noise follows a Cauchy distribution, which is an alpha-stable distribution with alpha=1. We also consider the Archimedean copula family which also allows to have tractable solutions. We can then observe the impact of impusive noise and dependence on the decision regions.
Titre : Statistical dimensioning of massive and dense NB-IoT network
Orateur : Lina Mroueh, ISEP
Résumé : One of the main challenges of Radio Resource Management (RRM) in the uplink of NB-IoT is to optimize the tradeoff between the energy consumption of the sensors, the allocated bandwidth at the collector and the density of the collectors in the network. Given the density of aggregators and simultaneously active sensors, dimensioning the LPWA network consists to statistically determine the bandwidth required to serve all sensors with a given confidence margin. The event that arises when the collector has no more radio resources to attribute to active sensors will be denoted by radio resource outage event. This talk will focus on the statistical dimensioning problem that aims to determine the adequate bandwidth that minimizes the occurrence of radio resource outage event. The average statistical behavior of the network is considered: active sensors and collector are randomly distributed in a given area according to a random Poisson Point Process marked by the randomness of the wireless channel. The main goal of this talk is to provide statistical dimensioning tools to estimate the required bandwidth in function of the network load and the collector density. The impact of the receiver diversity on the network dimensioning will be also addressed.
Titre : Stochastic geometry approaches for D2D communications in dense heterogeneous networks
Orateur : Romain Chevillon, Polytech Nantes, IETR
Résumé : Stochastic geometry (SG) is a useful tool to estimate the interferences and the data rates in wireless networks. 5G is being defined and includes numerous types of heterogeneous links: LTE-based communications, Device-to-Device (D2D) links, etc. Despite the adaptability of SG, its use remains complex, and various approaches are necessary, in particular for D2D communications, depending on the studied network environment. D2D links can be implemented either in the cellular band, sharing the LTE spectrum (Inband Underlay) or using a dedicated portion of this spectrum (Inband Overlay), or in unlicensed bands (Outband D2D). In the case of Uplink Inband communications, SG easily allows to quantify the SINR by means of Point Processes. However, if D2D communications use the Downlink, it is much more complicated to quantify this ratio, due to the joint distributions of the users and the base stations. Besides, the choice of hexagonal or Voronoï cells influences the associated calculations. Furthermore, the propagation environment involves different small-scale fading models(Rayleigh or Nakagami-m fading). In this work, we propose to present and analyze the various approaches of SG for D2D communications in dense heterogeneous networks, and depict some analytical results for these perspectives.
Titre : Energy-Latency Tradeoff in Ultra-Reliable-Low-Latency Communication with Short Packets
Orateur : Apostolos Avranas, Telecom ParisTech, Huawei
Résumé : Emerging mobile networks (5G new radio) are envisioned to support mission-critical Internet-of-Things (IoT) applications (e.g., industrial control, automated transportation and robotics) and ultra-reliable low-latency communication (URLLC) scenarios with strict requirements in terms of latency (ranging from 1 ms and below to few milliseconds) and reliability (higher than 99.999%). This entails a fundamental paradigm shift from throughput-oriented system design towards holistic designs for guaranteed and reliable end-to-end latency. In this talk, we consider a URLLC system with short packets employing hybrid automatic repeat request (HARQ). We study the fundamental energy-latency tradeoff and analyze the energy consumption for incremental redundancy (IR) HARQ and compare it to the no HARQ case. Moreover, we derive closed-form expressions for the outage probability of IR-HARQ with both variable finite-length codewords and power. Our results show that when the feedback delay is more than half the latency constraint, it is beneficial in terms of energy to use one-shot transmission (no HARQ). This is joint work with Marios Kountouris (Huawei France) and Philippe Ciblat (Telecom ParisTech).
Titre : Capacity sensitivity and non-Gaussian interference models in IoT
Orateur : Malcolm Egan, INSA Lyon, CITI
Résumé : A feature of the internet of things is short transmissions. A consequence is that in large-scale networks, the active interferer set can change rapidly--leading to dynamic interference. In this setting, the interference is impulsive and not well-modeled as Gaussian. As such, fundamental capacity limits are not known in closed-form, which poses challenges for network design and resource allocation. In this presentation, we introduce the capacity sensitivity framework, which provides a means of characterizing the effect of changing network parameters on the channel capacity in the presence of non-Gaussian interference. As an illustration, we study the case of alpha-stable interference, which is a popular model for impulsive interference.
Titre : Polar codes for short lengths and low rates
Orateur : Ingmar Land, Huawei
Résumé : IoT has specific requirements on channel codes that differ from other wireless scenarios. In particular, the focus will be on short codes due to the small message sizes and low rates due to the required reliability. In this presentation we will present a new polar code construction with inner edge-permutations. This construction allows for larger minimum distance and thus superior error-rate performance under successive-cancellation list decoding with moderate list lengths. These kind of codes are therefore interesting candidates for IoT.
Titre : State of the art in short packet communications: a physical layer comparison for block fading channels
Orateur : Alex The Phuong Nguyen, IMT Atlantique
Résumé : Many wireless short packet systems have been proposed recently due to the new emerging use cases in wireless communications. These competing schemes exhibit quite different design choices at least from the physical layer perspective. An interesting question arising is then if there would be a particular waveform that could best fit the short length requirements in terms of transmission rate. Inspired by the fact that all these waveforms are different, a framework that is able to compare these modulation schemes is desired. To make an apples-to-apples comparison, two tools are needed: a metric model to quantify system performance and an equivalent channel for each modulation scheme. One of relevant metrics for short packet transmission is the maximal coding rate at a finite packet length and at a maximal packet error probability. No closed-form expression for this metric is available. Nevertheless, several theoretical frameworks for derivation of its converse (upper) and achievability (lower) bounds have been developed and tight numerically computable bounds have been recently obtained for MIMO Rayleigh fading channels. In this talk, we focus on four popular short packet modulation schemes that are Sigfox Ultra Narrowband (Sigfox), Direct Sequence Spread Spectrum (Ingenu), Chirp Spread Spectrum (LoRa) and CP-OFDM (NB-IoT). We propose a way to analyze their best achievable transmission rates at their typical conditions over multi-propagation block fading channels and then compare them for single and multiuser transmissions.
Titre : System Robust Communication "SyRoCo" in impulsive environment
Orateur : Yasser Mestrah, CReSTIC
Résumé : In a dense wireless network, interferences exhibit an impulsive behaviour especially when there is no central coordination and no power adaptation. Alpha-stable laws are often well-suited to imitate noise and interference noise. The receiver computes the log-likelihood ratio as input of the error-correcting decoder such as LDPC belief propagation. But, in order to perform this computation, the precise model of the noise is needed. We have developed a simple approximation of the LLR which showed to be robust enough for most case. Our solution depends on two parameters which can be learnt from a sequence of pilots or directly from the output of the channel. Accurate for long code words and/or training sequences, the results degrade significantly for shorter blocks. We will discuss solution to mitigate this loss and propose a robust estimation approach.
Titre : Multi-Armed Bandit Learning in IoT Networks: Learning helps even in non-stationary settings
Orateur : Lilian Besson, CentraleSupélec Rennes, IETR
Résumé : Setting up the future Internet of Things (IoT) networks will require to support more and more communicating devices. We prove that intelligent devices in unlicensed bands can use Multi-Armed Bandit (MAB) learning algorithms to improve resource exploitation. We evaluate the performance of two classical MAB learning algorithms, UCB1 and Thompson Sampling, to handle the decentralized decision-making of Spectrum Access, applied to IoT networks; as well as learning performance with a growing number of intelligent end-devices. We show that using learning algorithms does help to fit more devices in such networks, even when all end-devices are intelligent and are dynamically changing channel. In the studied scenario, stochastic MAB learning provides a up to 16% gain in term of successful transmission probabilities, and has near optimal performance even in non-stationary and non-i.i.d. settings with a majority of intelligent devices.
Titre : Construction de code NB-LDPC de très bas rendement pour l'IoT
Orateur : Emmanuel Boutillon, UBS, LabSTICC
Résumé : Dans cet exposé, nous présentons des résultats récents de code correcteurs d'erreurs très bas rendement (de 1/3 à 1/100 typiquement) permettant de décoder des messages très court (quelques dizaine de bits) à de très faible rapport signal à bruit. La structure de ces codes leur permet d'atteindre des performances très supérieures à celle du Narrow Band-LTE par exemple. Ces codes sont construits en association avec une modulation monoporteuse appelée CCSK (Cyclic Code Shift Keying). Au niveau de l'émetteur, la complexité est ainsi réduite au maximum.