Active queue management for LTE uplink in eNodeB

摘要

Long-Term Evolution (LTE) is an evolved radio access technology of the 3rd generation mobile communication. It provides high peak bit rates and good end-to-end Quality of Service (QoS). Nevertheless, the wireless link is still likely to be the bottleneck of an end-to-end connection. Thus, having a sophisticated method to manage the queues of the mobile terminal is important. For Wideband Code Division Multiple Access (WCDMA), an Active Queue Management (AQM) algorithm managing the buffer based on the queue size was proposed. In LTE, due to its largely varying bit rates, the queue-size-based approaches are not suitable anymore. Thus, earlier studies have proposed a delay-based AQM to provide a better performance in LTE. For LTE uplink, the existing algorithm is supposed to be implemented in the User Equipment (UE). On the other hand, the implementation of an AQM in the UE is not mandatory. Until now, only a quite simple delay-.based queue management method called Packet Data Convergence Protocol (PDCP) discard is standardized by 3GPP. However, this method is not adaptive and cannot thus guarantee a good throughput. The purpose of this thesis is to develop an AQM method for LTE uplink to enhance the performance of TCP traffic. In order to have a better control of the LTE uplink traffic from the network side, the AQM algorithm is proposed to be implemented in the eNodeB. It retains the delay-based approach; to achieve it, a method is developed to estimate the queuing delays of the UE from the eNodeB side. The delay estimation is based on the changes in Buffer Status Reports (BSRs) and the amount of data delivered in the eNodeB. In LTE, BSRs are created and transmitted by the UE to report the queue length waiting for uplink transmission. A number of simulations are done to study the performance of the delay estimation and the resulting AQM algorithm. The new AQM algorithm is also compared with other algorithms, i.e. delay-based AQM implemented in the UE, PDCP discard and drop-from-front. The results show that the delay-based algorithm implemented in the eNodeB performs almost as well as when implemented in the UE. The results also show that the advantaged of delay-based algorithms comparing to the drop-from-front and PDCP discard are evident; they maintain a high throughput and the low end-to-end delay in most of the scenarios.