Browsing by Author "Vienamo, Aapo"

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    Reaaliaikainen pistepilven tiedonkäsittely
    (2016-12-23) Vienamo, Aapo; Hyyti, Heikki; Sähkötekniikan korkeakoulu; Forsman, Pekka
  • Thumbnail Image
    Item
    Resource and performance modeling of 5G beamforming and channel estimation ASIC block
    (2021-01-25) Vienamo, Aapo; Karjalainen, Janne; Sähkötekniikan korkeakoulu; Ryynänen, Jussi
    Beamforming and channel estimation are techniques for improving a 5G mobile networks spectral efficiency and robustness. The computation process required in beamforming and channel estimation is data-intensive, which calls for employing a bespoke accelerator core to implement the computation in hardware. The accelerator core is designed in a hardware/software co-design process which involves concurrent model-based development of both hardware and software. The modeling task ad- dressed in the thesis work is performance and resource modeling of the accelerator core. The goal is to create a model that would act as a tool that allows system architects to determine the effects of design decisions in terms of processing performance and buffer memory requirements. In the thesis work, the performance and resource model of the accelerator is implemented as a transaction-level simulation model. The model is a non-functional model, meaning that it does not implement the functionality of the accelerator, but only the performance characteristics. The simulation model is augmented with an instrumentation system that allows for extracting measurements from the simulation to enable determining performance and resource usage attributes, such as end-to-end latency, throughput, utilization, and buffer usage levels. The simulation model was evaluated in terms of its ability to recreate a wide range of behavior in response to varying design parameters and was found to be able to exhibit the expected performance phenomena. In terms of facilitating rapid iteration modeling workflow, the simulation model implementation was determined to be of sufficiently high performance to facilitate iterative and interactive workflow.