Contribución a la mejora del control de flujo en redes de acceso inalámbrico

Résumé

The framework of this thesis is the backhaul of radio access networks (RANs). We refer to backhaul as the infrastructure connecting the base stations of a cellular networks to either the radio network controller (RNCs) nodes or the core network nodes. In particular, this thesis addresses the issues associated to the congestion of the backhaul and the control algorithms that manage the resources of this infrastructure. The potential problems caused in the RAN by the backhaul congestion are of different nature depending on the RAN technology. We will provide a historical overview of the evolution of the RAN technologies over the last two decades, focusing on the functionalities relying on the backhaul, and how each generation imposes different and somewhat more stringent demands on the backhaul infrastructure. Along this work we focus chronologically on a specific RAN generation, starting with 3G, in particular Universal Mobile Telecommunication System (UMTS), then 3.5, High Speed Data Access (HSPA) and 4G Long Term Evolution (LTE). On each one, we study the impact of backhaul congestion on the RAN performance and propose a strategy to optimally control and manage the resources of the system. The objectives of this thesis can be formulated as follows: 1. To develop and evaluate effective control schemes to improve the performance of the transport channel synchronization functionality under backhaul congestion. The technological framework of this first objective is 3G (UMTS). 2. To develop and evaluate own control mechanisms that jointly consider the radio interface and the backhaul. The technological framework of this second objective is 3.5G (HSPA). 3. To study the impact of backhaul congestion on downlink scheduling over the radio interface. The technological framework of this second objective is 3.5G (HSPA). 4. To develop and evaluate mechanisms for coordinate resource allocation at both the radio interface and the backhaul. 5. To address previous objective with proposals that are compliant with the technical specifications of the involved protocols. With regard to this last objective, we consider that it is not fully realistic to come up with algorithms implying changes in systems that are already standardized and widely deployed. However, 3GPP standards do not pretend to define every algorithm involved, since the 3GPP's main objective is to clearly specify the interfaces to allow the interoperability between operators and deferent vendors' devices. The internal mechanisms contained in deferent layers of the protocol stack to accomplish each task are generally left open to operator or vendor choice. Especially resource management functions (such as scheduling or congestion control). This approach leverages innovation and research within the industry and the academia, and allows that mechanisms as the ones presented in this thesis be feasible within existing standardized cellular networks. The main objective of transport channel synchronization in UMTS is to make sure that the frames sent by the RNC arrive on time to the Nodes B (NBs) to be transmitted over the radio interface. For this task, the 3GPP specifies an algorithm known as timing adjustment (TA) that controls the delay suffered by the frames over the interface (Iub) connecting each NB with its corresponding RNC. The TA can add or subtract a certain quantity to the transmission delay. We show that the typical mechanism reacts too slowly in situations where the Iub delay increases abruptly, e.g. under transient congestion of the Iub. Besides, this classic algorithm shows potential instability issues in scenarios of very high delay. We address this problem using the tools of discrete-time control theory, which allows us to propose a new scheme that assures stability under any circumstances and improves the classical mechanism. The performance evaluation is carried out by means of simulation and considering realistic traffic scenarios for the Iub. With the introduction of HSPA (3.5G) in UMTS, the scheduling function was moved from the RNC to the Node B, imposing the inclusion of new data buffers at the NBs. Additionally, this redistribution of the data storage function between the RNCs and the NBs created the need of a flow control mechanism regulating the data transfers on the Iub. We model and analyze this mechanism as a quadratic optimization problem, and exploit this approach to propose a new flow control scheme that minimizes the end-to-end delay over the RAN by considering no only the situation of the NB buffer, but also of the RNC buffers. Our approach is a backhaul-aware optimal flow control system for RAN. Finally, in LTE (4G) systems, the peak transmission rates achievable by a user over the radio interface have boosted compared to previous generations. For the first time operators, vendors and the academic community agree on the need to optimize the backhaul resources, not only the radio resources. In fact, our point is that backhaul congestion impacts the performance of radio resource allocation and this function should be redesigned taking into consideration the capacity limitation of the backhaul. We use network utility maximization (NUM) techniques to address the problem of joint radio-backhaul scheduling. We use a dual decomposition approach to propose a low-complexity, distributed mechanism, that can make resource allocation decisions in a subframe basis. Finally, we incorporate the optimal control policy for tande queues in our mechanism, improving even more the performance compared to non-coordinated schedulers.