研究目的
To propose a new DMA solution, CubeDMA, tailored for hyperspectral images to handle memory access patterns efficiently and overcome limitations of existing DMA cores.
研究成果
The CubeDMA core efficiently handles HSI data access patterns, outperforming existing AXI DMA cores in terms of throughput and resource efficiency, particularly for block-wise and BSQ orderings. It eliminates overheads from block descriptors and supports flexible configurations. Future work includes automated verification, adding new access patterns, and optimizing throughput further.
研究不足
The CubeDMA core requires re-synthesis if generic parameters (input data width, component bit-width, number of parallel output components) are modified. Performance in BSQ ordering is limited by DataMover's idle cycles, and further improvements are needed for higher throughput. The core is specific to FPGA implementations and may not be directly applicable to other hardware platforms.
1：Experimental Design and Method Selection:
The study evaluates existing FPGA-related DMA solutions and proposes a custom DMA core (CubeDMA) for hyperspectral imaging. It involves designing the core using VHDL, synthesizing and implementing it with Xilinx Vivado tool, and testing on a Zedboard development board with a Zynq-7020 FPGA. Theoretical models include address computation and data restructuring for various access patterns.
2：Sample Selection and Data Sources:
A real HSI image of size 512×2000×128 collected by HICO imager is used for testing. Generic parameters such as [N_comp, BPC] pairs are varied.
3：List of Experimental Equipment and Materials:
Zedboard development board, Zynq-7020 FPGA, Xilinx Vivado tool, DDR memory, custom processing cores (e.g., FIFO module, CCSDS-123 compression core).
4：Experimental Procedures and Operational Workflow:
The CubeDMA core is synthesized and implemented. Functional testing involves streaming data from memory to processing cores, comparing results with reference software (Emporda), and measuring performance metrics like throughput and resource utilization.
5：Data Analysis Methods:
Performance is analyzed in terms of throughput (MB/s), resource utilization (LUTs, registers, block RAM, DSP), and comparison with existing AXI DMA cores. Statistical analysis includes speed-up calculations and pattern efficiency.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
