Pipeline többprocesszoros jelfeldolgozó rendszer magas szintű szintézise

OData támogatás
Konzulens:
Dr. Arató Péter
Irányítástechnika és Informatika Tanszék

Voice Source Localization processing needs specific response(performance) time due to the efficiency. Then improving this performance would not be efficient by applying a uniform parallel processing architecture or using a high performance single processing unit. So like this case, the pipeline signal processing is required and this way would be more efficient. To get performance efficeint pipeline structure, High level synthesis(HLS) of a pipeline signal processing can be applied.

For applying this HLS, First of all, based on the literature provided by the supervisor, summarize the main properties of the high level synthesis methods by focusing to the pipeline systems. (Task1)

After understanding the main properties, based on the paper provided by the supervisor, construct the initial data flow graph representation from the algorithm of the voice source localization.(Task2)

Then by applying the Chaco graph decomposition software, generate some proper segmentations of the initial data flow graph and evaluate the segmentations regarding the effect of the applied decomposing algorithms and the number of segments.(Task3)

Next based on the evaluation, select at least two beneficial segmentations and apply them as input data flow graphs in high level synthesis tool PIPE. (Task4)

And generate the cost versus restart time (initialization interval) diagram for each selected segmentation. (Task5)

And then define at least two advantageous restart times for each selected segmentation and determine the multiprocessing structure in each case.(Task6)

Furthermore evaluate the resulted multiprocessing structures and compare them with the structure in the paper provided by supervisor. And I will suggest proper implementations (hardware or software) for each processing unit in each structure.(Task7)

Finally apply the initial data flow graph determined in second step directly without segmentation as input in high level synthesis tool PIPE and repeat Tasks 5, 6, 7 by interpreting nodes instead of segments. (Task8)

Letölthető fájlok

A témához tartozó fájlokat csak bejelentkezett felhasználók tölthetik le.