用于油画鉴别的自适应MPSoC中NoC仿真平台研究

Emulation platform dedicated to adaptive MPSoC-based NoC architecture for the art authentication application

实时光谱图像技术的应用是数据和通信的密集型算法的综合。因为片上系统(SoC)具有较大的灵活性及良好的性能,并可以直接采用IP核,基于片上网络(NoC)的SoC架构是实现这种实时系统最合适的解决方案。针对不同的应用算法,调整基于SoC的NoC架构是较为复杂的工作,因为设计空间探索和实验需要大量的时间。针对这些问题,提出了一种探索和评估SoC架构性能的仿真平台。该仿真平台基于NoC和参数化的评估模块,在FPGA上对SoC的通信结构进行仿真,并将时间性能和模拟结果进行对比。基于该仿真平台的系统评估主要用来考察用于油画鉴别应用的MPSoC的性能。

Real time spectral imaging applications are data and communication intensive algorithms. SoC architecture using NoC communication is the most appropriate solution as it brings flexibility, performances and immediate reuse of IPs cores. Tuning the SoC-based NoC architecture for a dedicated algorithm is a complex task as the design space is huge and experimentations take time. This paper proposes an emulation platform for the evaluation and exploration of performances of the architecture on FPGA. The emulation platform is based on the NoC and parameterized emulation blocks. Emulating the communication inside the FPGA gives precise timing performances compared to simulation. The evaluation is illustrated for a spectral imaging algorithm dedicated to art authentication on the architecture containing MIPS processors and the Hermes NoC.