A LTE UPCUL architecture design combining Multi-Blackboards and Pipes & Filters architectures

Abstract: Context. The single blackboard architecture is widely used in the LTE application area. Despite its several benefits, this architecture limits synchronization possibilities of the developed systems and increases the signal operational latency. As a result the DSP (Digital Signal Processing) utilization is suboptimal. Objectives. In this thesis, we design a new architecture, which combines concepts of Multi-Blackboards and Pipes & Filters architectures, as a replacement for the current single blackboard architecture at Ericsson. The implementation of the new architecture makes the environment asynchronous. We evaluate the new architecture at simulated environment of Ericsson with 222225 connection items from 9000 base stations all over the world. Each connection item has a complete UE session and one of possible connection statuses, e.g. active, inactive, connected, DRX sleeping, postponed. These connection items can be from any country in the world. Methods. We design a new architecture for UPCUL component of LTE network based on analysis of real network data from Ericsson. We perform a case study to develop and evaluate the new architecture at Ericsson. Results. We evaluate the new architecture by performing a case study at Ericsson. The results of case study show that the new architecture not only increases DSP utilization by 35%, but also decreases signal operational latency by 53%, FO operation time by 20% and FO operation cycles by 20%. Also, the new architecture increases correctness performance. Conclusions.  We conclude that the new architecture increases DSP utilization and decreases the signal operational latency, therefore, improves performances of UPCUL component of LTE.  Due to time constraints, we only considered four LTE FOs (Function Objects) and relative signals. Future work should focus mainly on the other FOs and signals. We also analyze unconsidered FOs, and make an integration solution table which contains solutions to integrate these unconsidered FOs into the new architecture.  The second avenue for future work is to re-size the size of the two blackboard storages. We find out that the maximum memory size of needed UE sessions per sub-frame is only 1.305% of the memory size of all UE sessions (31650 bytes). So the memory size of blackboard storage should be adjusted on the basis of needed UE sessions instead of all UE sessions.