Flexible And Latency-optimized Passive Optical Networks For Mobile Fronthaul For Cloud-RAN Architecture

Wireless mobile network has been continuously evolving over the last two decades to meet the rapid growth on service demands and user requirements.The fifth generation(5G)mobile network is adopting the cloud radio access network architecture(Cloud-RAN or CRAN)to support enhanced capacity,massive connectivity and low-latency services.C-RAN is an architecture,where the baseband processing units(BBUs)of many cellular cells are pooled in a central location while the remote radio units(RRUs)are placed several kilometers away in cell sites.The interface connecting between BBUs and RRU is a major challenge for C-RAN deployment as it requires a huge number of high-capacity and lowlatency fiber-based connections which increases the cost of deployment of C-RAN.Several wireless and wireline technologies are currently being discussed as solutions for this interface.Passive optical networks(PONs)are outstanding solutions for such an interface due to their low cost because of their capability to share the fiber optic infrastructures and transmission equipment.However,PONs transport solutions have several challenges need to overcome in order to meet the requirement of such an interface.These challenges include;the latency,the traffic management,dynamic resource allocation,software define networking(SDN)and network function virtualization(NFV)support,etc.This thesis focus on addressing the latency and traffic variation management issue for single-wavelength PON-based C-RAN as well as the resource allocation issue for multi-wavelength-based CRAN.1)Traffic Variation Issue in Single-wavelength Virtualized C-RAN Architecture:In 5G virtualized C-RAN architecture,the processing chain of the traditional baseband unit of mobile network is decomposed and shared between the BBU and RRU.In such a CRAN architecture,the data traffic exchanged in the interface connecting between BBU and RRU(which is known as mobile fronthaul)is highly variable and bursty.To accommodate such a variable traffic using single-wavelength time division multiplexing PONs(TDMPONs)a flexible and low-latency bandwidth allocation mechanism is required.Taking ITUT standard-compliant TDM-PONs(e.g.XG-PON and XGS-PON)as a focus of our research.The most popular dynamic bandwidth allocation(DBA)proposals for ITU-T TDM-PONs are Giga-PON Access Network DBA(GIANT)and Round Robin DBA(RR-DBA).These proposals were developed to accommodate a semi-deterministic traffic profile of fixed lastmile broadband user.The major limitation of these proposal is using a pre-determined fixed limit for the optical network unit(ONU)upstream transmission during every allocation cycle which lead to inflexibly in allocating the bandwidth resources and poor utilization of XG-PON upstream frames especially when the ONU requests a data less than the predetermined fixed limit.To overcome this issue,we propose Optimized-RR DBA in which we introduce the idea of flexible adjusting of the ONU fixed limit according to the actual mobile fronthaul traffic to accommodate the traffic variation and tackle the problem of poor utilization of XG-PON upstream frames.The performance evaluation of Optimized-RR DBA over a simulated dynamic fronthaul traffic shows it superior network performance improvement compared to gGIANT and RR DBAs.2)Latency Issue in Single-wavelength Virtualized C-RAN ArchitectureTDM-PON(XG-PON)requires a DBA for scheduling the upstream transmissions for ONUS.Conventional DBA mechanisms for XG-PON have a high latency in the order of one millisecond which exceeds the requirement of mobile fronthaul(i.e.250~300?s).Although,in our first Optimized-RR DBA proposal we achieve a latency in order of 300?s.However,considering the quality of service(QoS)requirements for some 5G emerging usecases such as ultra-reliable low latency communication(URLLC),the fronthaul latency should be restricted to 250?s as maximum.There are many DBA proposals for low-latency mobile fronthaul over TDM-PONs.However,most of these proposals either consider modifying current TDM-PONs systems or modifying BBU processing and protocols.To achieve the mobile fronthaul latency over TDM-PON without modifying the current neither current PONs systems nor BBU processing and protocols,we propose a traffic estimation DBA proposal for XGS-PON mobile fronthaul.This method is based on estimating the amount of the mobile fronthaul data traffic reside in the ONU buffer using adaptive learning neural network function integrated into the DBA module of the optical line terminal(OLT).By utilizing this estimation method,we able to reduce the additional DBA processing delay and eliminate the delay required for sending the scheduling requests from ONU to OLT.As the result,the XGS-PON upstream latency is decreased as low as the latency required for mobile fronthaul in C-RAN architecture.3)Resources Management Issue in Multi-wavelength C-RAN ArchitectureA unified scheduling of the radio resource blocks(RBs)and the fronthaul transport resources(TBs)in C-RAN can offer several benefits such as facilitating the integration of the optical and radio access networking elements,and enabling the chance of decoupling the hardware from software in these elements.The idea of a unified scheduler for both radio and fronthaul resources has been discussed earlier for single-wavelength PON C-RAN under the name of Mobile-PON proposal.In Mobile-PON proposal,the authors assume allocating a fixed fronthaul TB to every radio RB.Such an assumption will lead to a reduction in fronthaul bandwidth efficiency and increase the delay because in practical mobile network the actual capacity of the RB varies depending on many factors such as the amount of data requested by the user,the channel quality condition and the modulation and coding schemes(MCS)used for user's uplink transmission.Moreover,single-wavelength PONs are incapable of supporting the huge capacity requirement of mobile fronthaul in 5G based C-RAN.To overcome these limitations,we formulate a joint resource allocation optimization problem for the radio and fronthaul resources considering a multi-wavelength C-RAN and taking these limitations into account.Due to the complexity of such an optimization problem,we propose a reinforcement learning scheduling algorithm that utilizes the actual user request,the channel quality condition and the MCS information to find the best allocations strategy of RB that maximizes the total system throughput,and the best size and scheduling policy of TB for each RB over the fronthaul wavelengths that minimizes the total uplink scheduling latency.The performance evaluation of the algorithm shows it potential in enhancing total system performance and outperforming the other heuristic methods proposed for the uplink scheduling.