Theses supervised by Prof. Dr. Tuna Tuğcu
10 theses · Boğaziçi University
Urban cellular wireless network planning with 3D geographical grid structures
The largest traffic on the Internet is generated by multimedia content, and the share of wireless mobile access to the multimedia content is ever increasing. The delivery process in the content depends on the type of the content, and whether it requires a strictly time sensitive live session, or it can be delivered on demand, irrespective of time of request. As a general solution in the IP networks, the delivery is taking place as multiple individual transmissions from the source to all receivers, replicating the same content at different times, or at the same time if the content is streaming live. In this thesis, we propose an advanced collaborative multicast routing model for delivering bandwidth hungry streaming content such as IPTV to multiple users having low quality wireless connection, with the help of other nearby users having a better connection quality utilizing the wireless mesh network topology. The model reduces the amount of data replication, which causes significant overhead in the network transmissions and intermediate computations, improving the overall throughput of the wireless network. An essential issue in wireless mobile networks is where to position the base stations on the network coverage area. We also propose an alternative adaptive mixed path signal propagation estimation model for planning the base station locations in new deployment areas, based on the signal characteristics of existing networks. In doing so, we utilize digital 3D maps of urban areas and signal measurements to train the adaptive model, and to exploit local similarities in cities in order to estimate the signal channel characteristics and calculate the potential base station locations for covering the new area with a specific wireless communication technology.
Improving the performance of LoRaWAN IoT System
One of the technologies developed in LPWAN according to the IoT concept is the LoRaWAN protocol. This protocol works together with LoRa modulation. So far, studies on this protocol included only uplink communication. Moreover, the performance of the pure ALOHA type MAC protocol used in this protocol was not fully observed. Recently, investigations, including downlink communication, have revealed that the pure ALOHA type MAC protocol and downlink traffic have a negative impact on network scaling. A solution is provided to minimize these negative effects in the LoRaWAN protocol. This solution is a scheduling method for synchronizing the gateway with the end devices. In addition, the proposed solution provides two different methods of sending aggregated acknowledgment messages, which are sent to the last devices from the gateway. This proposed solution has been observed to reduce downlink traffic when it is run with various test scenarios and to increase the capacity of the last device to which a gateway can serve. In other words, the proposed solution had a positive contribution to LoRaWAN scaling.
Handover with network slicing in 5G networks
5G suggests many advantages but these advantages bring some problems to be solved. Network Slicing is one of the key concepts that $5G$ introduces. Slicing enables having multiple isolated virtual networks on top of the same physical infrastructure. Thus, each slice can provide different services with diverse Quality of Service (QoS) requirements. In 5G, Software Defined Networking (SDN) and Network Function Virtualization (NFV) are critical to support network slicing. In the literature, several problems of network slicing are studied. Two outstanding areas of focus are admission control and resource allocation. Most of the studies are on the Core Network resources although it is essential to investigate radio resource allocation in order to maintain an end-to-end isolation for slices. While there are considerable contributions around Radio Access Network (RAN) resource allocation, optimizing the throughput of the network is not fully achievable via admission control. In this thesis, we mainly focus on handover to maintain the usability and high utilization of the radio resources of the networks. When the number of users within one cell suddenly increases up to the limit of the base station, all of the incoming requests may not be handled and most of the users may suffer from not being able to use the offerings of the slices that they demand. We develop an optimization problem to optimize the radio resources and propose an heuristic to reach similar results by leveraging handover in considerably low computing times and with the simulation results we show that our heuristic can present solutions close to the optimal within a short time frame.
Reinforcement learning based handover mechanism for next generation mobile communication systems
Next-generation mobile communication networks have been established on critical enabling technologies such as millimeter-wave usage, cloud-native architectures, and new intelligent algorithms to meet the increasing demands of new services and requirements. One important research area for the new generation of networks is Radio Resource Management (RRM) applications. In this thesis, a reinforcement learning-based handover (HO) mechanism is designed by the concept of Contextual Multi-Armed Bandit (CMAB) algorithm named CHARM (CMAB-Based Handover Algorithm in Reinforcement Mechanism) and considering Open-Radio Access Network (O-RAN) architecture. The speed of user equipment (UE) and Signal-to-Interference-plus-Noise Ratio (SINR) of the serving Base Station (BS) parameters are evaluated as the context information for the algorithm. The proposed algorithm is compared with the traditional algorithm of 3rd Generation Partnership Project (3GPP) and a rival reinforcement algorithm in the literature under different channel conditions such as Urban Macro (UMa), Urban Micro (UMi) propagation, and different intensities of BS and obstacles on the map. The results show that our algorithm outperforms the traditional 3GPP HO algorithm and the rival algorithm for average information rate under every channel condition. According to the simulations, it is also highly competitive for average HO numbers.
Achieving ultra-reliable low-latency communication (URLLC) in next-generation cellular networks with programmable data planes
Recent advancements in wireless technologies towards the next-generation cellular networks have brought a new era that made it possible to apply cellular technology on traditionally-wired networks with tighter requirements, such as industrial networks. The next-generation cellular technologies (e.g., 5G and Beyond) introduce the concept of ultra-reliable low-latency communications (URLLC). This thesis presents a Software-Defined Networking (SDN) architecture with programmable data planes for the next-generation cellular networks to achieve URLLC. Our design deploys programmable switches between the cellular core and Radio Access Networks (RAN) to monitor and modify data traffic at the line speed. We introduce the concept of intra-cellular optimization, a relaxation in cellular networks to allow pre-authorized in-network devices to communicate without being required to signal the core network. We also present a control structure, Unified Control Plane (UCP), containing a novel Ethernet Layer control protocol and an adapted version of link-state routing information distribution among the programmable switches. Our implementation uses P4 with an 5G implementation (Open5Gs) and a UE/RAN simulator. We implement a Python simulator to evaluate the performance of our system on multi-switch topologies by simulating the switch behavior. Our evaluation indicates latency reduction up to 2x with intra-cellular optimization compared to the conventional architecture. We show that our design has a ten-millisecond level of control latency, and achieves fine-grained network security and monitoring.
Weighted adaptive load and handoff equilibrium: a novel resource allocation framework for cloud radio access networks with load prediction and handoff optimization
The effective assignment of Remote Radio Units (RRUs) to Baseband Units (BBUs) in Cloud Radio Access Networks (C-RAN) remains a critical challenge, particularly in large-scale deployments where dynamic load patterns and frequent handoffs necessitate rapid adaptation. This thesis addresses this challenge through an innovative optimization approach that balances computational efficiency with performance in RRU-BBU assignments. The thesis presents a novel approach through the development of the WeigHted Adaptive Load and handoff Equilibrium (WHALE) framework, addressing three critical challenges: RRU load prediction, handoff optimization, and load balancing. The prediction framework combines Long Short-Term Memory (LSTM) and Generalized Autoregressive Conditional Heteroskedasticity (GARCH) models. In moderate-scale tests, WHALE achieves most possible handoff transitions with nearly half the reduction in execution time. For large-scale scenarios, it significantly reduces execution time while maintaining performance comparable to Genetic Algorithm (GA). The prediction-based variant (WHALE-P) achieves nearly all of GA's performance in handoff optimization and outperforms GA in load balancing with notably lower error rates. Keywords: Cloud Radio Access Networks, RRU Load Prediction, Handoff Optimization, LSTM-GARCH
Designing a simulatable environment for genetic circuitbased receivers in molecular communications
Inspired by nature, Molecular Communications (MC) has emerged over the past decade as a field that utilizes molecules as information carriers. This approach offers enhanced biocompatibility compared to traditional electromagnetic methods. The interdisciplinary nature of MC and the extensive research conducted in this area underscore its potential, particularly for medical applications. Yet, challenges such as intersymbol interference (ISI) and lower data rates, stemming from the inherent stochastic nature of MC, persist. While theoretical studies abound, bridging these concepts to practical applications remains a significant challenge, emphasizing the gap between theory and real-world implementation. Adopting an interdisciplinary approach, this thesis explores the integration of synthetic biology and molecular communications to address practical challenges in the field generating a realistic network environment. To combat the challenge of ISI, the pre-equalizer method is employed. This approach uses two interacting molecule types to mitigate the signal's heavy-tail effect. The approach is fine-tuned by a novel framework Parameter Finder to adjust the parameters of the communication scenario and MOL-Eye diagram is also integrated to enhance the precision of the molecular communication system. The findings reveal that integrating the system model with the pre-equalizer approach, alongside the use of the Parameter Finder tool, is highly beneficial for fine-tuning simulation parameters in the biological implementation of the pre-equalizer method. Additionally, the MOL-Eye diagram proves to be an effective tool for enhancing the selection of optimal parameters, providing a robust assessment of the performance of molecular signals.
Optimizing path planning for reduced congestion using reinforcement learning
Overpopulation in urbanized areas has been causing traffic density to increase and push the boundaries of the existing infrastructure units. Especially during peak hours, many people suffer from the congestion they encounter in their everyday commute. Aside from the time being spent on the road, traffic also causes air pollution, increased gas consumption, and even negative psychological impacts. Expanding existing roads with new lines, or promoting public transportation do not balance out the increasing congestion, and there is a need for an alternative approach to path planning. Autonomous driving technology, driving assistance software, increased IoT installations in road infrastructure, and the accessibility of the internet make modern vehicles smarter each day and revolutionize the transportation industry. Today, many drivers make their route planning based on path suggestion applications using GPS. With the help of these applications, drivers can see the least congested and fastest routes possible based on their destination locations. However, these applications provide individual-optimum solutions for the driver and may fail to estimate congestion correctly due to the lack of other vehicles' path information. Following the literature in the multi-agent path-finding domain, a new solution is proposed for the congestion problem. The solution is based on prioritizing system optimality and increasing the utilization of existing road networks. By using a reinforcement learning algorithm and a scalable orchestration architecture it is shown that overall congestion is reduced and less time is spent on the roads compared with individual-oriented solutions.
Network slicing management for IoT devices at home using machine learning algorithms
5G increased the speed and quality after 4G and enabled the development of new areas with the latest technology and advantages it brought. One of the groundbreaking technologies of 5g is network slicing. Network slicing simultaneously creates multiple virtual slices over the physical layer, enabling a single base station to customize those slices and improve its service according to different features and requirements. Thanks to network slicing, virtual slices with separate customizations have been created for the Internet of Things (IoT) or Ultra Reliable Low Latency Communications (URLLC), one of today's most rapidly developing technologies, enabling these areas to be integrated even faster. Resource allocation is one of the most significant areas that need improvement for network slicing. Resource allocation is the bandwidth that the base station should distribute to slices. It is critical to set this width correctly because if it is not set correctly, there may be disruptions in the users' service. Different solutions to the resource allocation problem have been presented in the literature. Among these, some studies aim to predict resource allocation with machine learning, and some papers also provide optimization with deep learning. In the thesis, the objective is to solve the problem with more than one algorithm. To solve this problem, it is attempted to get accurate results with Gurobi, a Mixed-Integer Linear Programming (MILP) solver. Afterward, Gurobi's results are compared with heuristic, Random, Balancing, Score-Based, and Reinforcement Learning solutions in different situations and scenarios. Resource allocation is the bandwidth that the base station should distribute to slices. It is critical to set this width correctly because if it is not set correctly, there may be disruptions in the users' service. Different solutions to the resource allocation problem have been presented in the literature. Among these, some studies aim to predict resource allocation with machine learning, and some papers also provide optimization with deep learning. In the thesis, the objective is to solve the problem with more than one algorithm. This thesis seeks to optimize and maximize the use of base stations and serve the maximum number of users. To solve this problem, it is attempted to get accurate results with Gurobi, a Mixed-Integer Linear Programming (MILP) solver. Afterward, Gurobi's results are compared with heuristic, Random, Balancing, Score-Based, and Reinforcement Learning solutions in different situations and scenarios.
Graph neural network based handover optimization framework
In the evolving landscape of mobile networks, an innovative handover optimization framework for next-generation networks in Open Radio Access Network settings is presented in this thesis. The research, focusing on embedding mobile networks including user equipments and base stations to better capture the network dynamics in handover decisioning, is facilitated through the utilization of Graph Neural Network (GNN) based framework. The core objective is to optimize critical aspects such as load balancing, handover cost, throughput gain, and coverage gain. This framework, called GNN-HOF (Graph Neural Network Based Handover Optimization Framework), is a significant departure from traditional proximity-based methods, leveraging advanced machine learning techniques to better understand and predict network dynamics. The efficacy of the approach is validated through extensive testing in simulated environments as well as real-world urban scenarios in Stuttgart and Monaco using the Simulation of Urban Mobility (SUMO) tool. The results are compelling, demonstrating that the proposed framework consistently outperforms the baseline method across all key metrics.