1 / 18

李雪梅 2014.03.21

学术报告. Low-area high-throughput Reed-Solomon decoder. 李雪梅 2014.03.21. 内容介绍. Part 1: 减少常数乘法器中异或门个数 Part 2 : 模块复用. Part 1: 减少常数乘法器中异或门个数. module mult_10(pro,a); output [4:0]pro; input [4:0]a; assign pro[4]= a[0]^a[3]^a[4]; assign pro[3]= a[2]^a[3]^a[4];

lars
Télécharger la présentation

李雪梅 2014.03.21

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. 学术报告 Low-areahigh-throughput Reed-Solomon decoder 李雪梅 2014.03.21

  2. 内容介绍 Part 1: 减少常数乘法器中异或门个数 Part 2 :模块复用

  3. Part 1: 减少常数乘法器中异或门个数 module mult_10(pro,a); output [4:0]pro; input [4:0]a; assign pro[4]= a[0]^a[3]^a[4]; assign pro[3]= a[2]^a[3]^a[4]; assign pro[2]= a[1]^a[2]^a[3]^a[4]; assign pro[1]= a[1]^a[2]; assign pro[0]= a[0]^a[1] ^a[4]; endmodule module mult_10(pro,a); output [4:0]pro; input [4:0]a; wire temp= a[3]^a[4]; assign pro[4]= a[0]^temp; assign pro[3]= a[2]^temp; assign pro[2]= a[1]^a[2]^temp; assign pro[1]= a[1]^a[2]; assign pro[0]= a[0]^a[1] ^a[4]; endmodule 10个XOR 8个XOR

  4. Part 1: 减少常数乘法器中异或门个数 搜寻公共项或匹配对或共享子式(common subexpression, CSE)的算法 Four steps: Step1:For any column i and j in matrix, find all possible pairs with bit-wise matches great than 1 Step 2:Choose the pair with the maximum bit-wise matches; If there are multiple such pairs, it is sufficient to randomly select one pair Step 3:Eliminate the redundancy from the best match; append an additional column at the right of the matrix to hold the redundancy; append an additional row to the bottom of the matrix to represent the selected CSE Step 4:Repeat steps 1-3 for all the columns in the matrix including the appended rows and columns until no subexpression appearing more than once

  5. Part 1: 减少常数乘法器中异或门个数 举例: 6个XOR 3个XOR

  6. Part 1: 减少常数乘法器中异或门个数 1D optimization 并行钱搜索电路或多项式选择电路

  7. Part 1: 减少常数乘法器中异或门个数 1D optimization 结果与分析 只能优化常数乘法器,不能优化加法器 6度并行的(31,25)码(t=3) 8度并行的(63,55)码(t=4) 16度并行的(255,239)码(t=8)

  8. Part 1: 减少常数乘法器中异或门个数 2D optimization

  9. Part 1: 减少常数乘法器中异或门个数 CSE算法的改进 多对匹配对均含有最大匹配数时,不采用随机挑选,选择重复性最小的那一对

  10. 不同方法的2D优化结果(均采用2t度并行) 采用repetitive方法的1D和2D结果(均采用2t度并行)

  11. Part 1: 减少常数乘法器中异或门个数 1D,2D优化结论: 由于搜索面积比1D大,所以比1D 效果好。2D不仅可以减少乘法器中异或门个数,并且可以移除加法器。 无论是1D还是2D优化,搜索面积越大效果越好。相同码长,并行度越高,纠错能力越大,效果越好

  12. Part 2: 模块复用 对syndrome、PS、CSFA进行模块复用,整个架构采用串行方式实现

  13. Part 2: 模块复用

  14. Part 2: 模块复用 并行syndrome结构

  15. Part 2: 模块复用 存在的问题: (1)CSFA都采用模块复用结构,需2t个周期用于钱搜索,还需2t个周期用于Forney计算,共需4t个周期求CSFA (2)钱搜索结束之后,需要用ram把钱搜索的结果存储起来,到求Forney时再把数取出来 (3)采用并行结构,每求一个error都需要一个inversion,共需p个inversion,占据较大面积 解决思路: 思路1:只把syndrome和PS进行模块复用,CSFA采用原始结构,需n个周期,避免ram或inversion带来面积的增加 思路2:把syndrome, PS, Chien Search进行模块复用,采用单独的结构求 ,就只需一个inversion,而不需要p个。但还是需要ram把对应的和存起来,Chien Search结束之后再求error值。

  16. The end!

  17. Part 2: 模块复用 存在的问题: (1)CSFA都采用模块复用结构,需2t个周期用于钱搜索,还需2t个周期用于Forney计算,共需4t个周期求CSFA (2)钱搜索结束之后,需要用ram把钱搜索的结果存储起来,到求Forney的时候再把数取出来,而ram占较大面积。 (3)采用并行结构,每求一个error都需要一个inversion,共需p个inversion,占据较大面积 解决思路: 思路1:只把syndrome和PS进行模块复用,CSFA采用原始结构,追求面积的减小 思路2:把syndrome, PS, Chien Search进行模块复用,采用单独的结构求, ,就只需一个inversion,而不需要p个。但还是需要ram把对应的和存起来,Chien Search结束之后再求error值。

More Related