Description
Date depot: 6 octobre 2023
Titre: Data-Driven Wireless Spectrum Supervision
Directrice de thèse:
Lina MROUEH (LISITE)
Encadrant :
Idowu AJAYI (LISITE)
Domaine scientifique: Sciences pour l'ingénieur
Thématique CNRS : Signal et communications
Resumé: Connecting industry 4.0 requires integrating various wireless communication technologies to provide a connected ecosystem for secure and reliable data exchange. Different wireless technologies are used including short range unlicensed IoT communication such as BLE, WiFi, Zigbee or long-range licensed and unlicensed communication such as low power wide area network (LPWAN) like LoRa or cellular network as NB-IoT or 5G network. The selection of a technology depends on the purpose and the target use case. While 5G provides high-speed, low-latency communication for a broad range of applications, LoRa focuses on low-power, wide-area connectivity for IoT devices deployed in hard-to-reach areas. Monitoring and regulating the wireless spectrum remain a key asset to enhance the spectrum usage specifically in a context where the radio-frequency bands become scarce resources. In this research work, we will focus on the supervision of two unlicensed bands: 868 MHz band and 5 GHz band.
Wireless network Identification and classification using artificial intelligence (AI) architectures such as convolutional neural networks (CNN), generative adversarial networks (GAN), etc. have received a lot of interest in recent times [1][2][3][4]. In this thesis, we will consider the unlicensed 868 MHz, used in Europe for various wireless communication technologies, such as LoRa, Sigfox, wireless Meter-Bus (used for remote metering reading), RFID, and smart home devices technologies. The first objective of this thesis is to develop AI-based algorithms that can automatically detect and classify different types of interference or signal anomalies (transmission power, duty-cycles) in the 868 MHz band. The designed AI-based algorithm is expected to recognize the chirp spread spectrum (CSS) modulation used in a LoRa network and its spreading factor, and to detect the potential anomalies (duty-cycles and transmission power), and the interference pattern generated by other technologies. Interference generated by the remote utility meters (water, gas, electricity) will be considered.
The second objective of the thesis is to study the co-existence of 5G with the other 802.11 technologies (802.11n, 802.11p, 802.11ac) that operate in this band. Although the primary focus of 5G network remains on licensed spectrum, the 5G is expected to leverage unlicensed using 5G-License Assisted Access (LAA) network in this band [6]. We will consider indoor scenarios in which Wi-Fi networks have primary access to the 5 GHz band, and the LAA devices are considered as secondary users. Protocols and mechanisms are currently considered in the standards to ensure fair spectrum sharing between Wi-Fi devices and LAA sharing. In this part of the thesis, we will extend the AI algorithms proposed for the 868 MHz band to the case of 5 GHz band to identify the interference pattern induced by OFDM technologies operating in this band as well as the aggregated carrier. Based on that, we will propose a radio-resource allocation algorithm to distribute the radio-resources on the uplink and downlink of a 5G-LAA network with massive MIMO antenna. A multi-agent reinforcement learning-algorithms will be proposed to optimize the transmission power of WiFi and 5G and to select the convenient numerology depending on the use-case. The reward function of the algorithm will be related to the positions of the sensing devices to reflect the impact of beam direction on the agent action on the environment. Indeed, this AI algorithm will select the adequate Wi-Fi channels to enable co-existence between 5G and WiFi.
In the third part of the thesis, we will analytically study how the co-existence strategy between 5G and Wi-Fi [8]and how this will affect the 5G network stability, and more specifically the buffer states in the downlink. For this, mathematical tools based on spatial Point Poisson Process and the Discrete Time Markov Chain will be respectively used to model the nodes spatial distribution and their traffics in cellular networks. Infinite and finite size of buffers at the base station will be considered. Based on a predefined scheduling policies and finite or infinite buffer assumptions at the base station, we will study the stability of the queue and the resulting delay depending on the wireless environment and the data-driven scheduling policy.
Doctorant.e: Ebo Ife Olalekan