JPEG2000中小波滤波器的定点分析及其VLSI实现

Finite-precision Analysis and VLSI Architecture of Wavelet Filters in JPEG2000

对JPEG2 0 0 0中推荐的 5 /3整数滤波器和 9/7实数滤波器进行了硬件实现时所需要的有限精度分析 ;确定了小波变换过程中各个参数的最佳数据宽度 ,还确定了整个变换系统的数据通路的数据宽度。基于lifting的小波变换的特点结合嵌入式延拓算法提出了两种小波变换———折叠结构和长流水线结构 ;对两种结构进行了分析比较。最后 ,对折叠结构和相关的其它结构在所需存储单元的数量、存储单元的访问次数、处理能力以及功耗等方面进行了分析比较 ,可以看出文中提出的结构在性能上有明显优点。

The wavelet transformation has become the core algorithm of JPEG2000 and there are two wavelet filters recommended in standard. The finite-precision of these two wavelet filters has been performed. The optimized bit-width of every parameter during the transformation has been decided. The bit-width of data-path of the transformation system has been validated. Moreover, two kinds of VLSI architectures-folded architecture and pipelining architecture-were proposed. The embedded extension algorithm was employed in thses two architectures. Then, the comparisons between the folded architecture and the relative architecture were performed in the memory requirement, number of memory access, throughput, and energy dissipation etc. The result shows that the performance is better than the relative architecture. These two architectures are suitable for the hardware implementation of JPEG2000 and various real-time image/video applications.