基于FPGA和GPU混合架构 实时相关器的设计进展II 裴鑫 2014.07.24
N_FFT=1024 A.A* B.B* Amp(A.B*) Phase(A.B*)
改进 1、镜像处理 2、滤波系数调整 3、发现采样时钟问题 4、量化处理
51.2MHz,-10dBm,N_FFT=1024 A.A* B.B* Amp(A.B*) Phase(A.B*)
51.2MHz,-10dBm,N_FFT=4096 A.A* B.B* Amp(A.B*) Phase(A.B*)
51.2MHz,-10dBm,N_FFT=32768 A.A* B.B* Amp(A.B*) Phase(A.B*)
51.2MHz,-10dBm,N_FFT=65536 A.A* B.B* Amp(A.B*) Phase(A.B*)
增加数字混频模块 3/8 混频
333.6MHz,-6dBm,N_FFT=1024,3/8mixing,50MHz LPF A.A* B.B* Amp(A.B*) Phase(A.B*)
333.6MHz,-6dBm,N_FFT=4096,3/8mixing,50MHz LPF A.A* B.B* Amp(A.B*) Phase(A.B*)
333.6MHz,-6dBm,N_FFT=32768,3/8mixing,50MHz LPF A.A* B.B* Amp(A.B*) Phase(A.B*)
FPGA逻辑资源占用 Used Availiable Utilization 采用ROACH进行混合架构相关器开发的资源占用情况 Logic Unit Used Availiable Utilization Slices 8641(11047) 14720 58%(75%) Slice Registers 25346(34715) 58880 43%(58%) LUTs 20907(29832) 35%(50%) LUT-Flip Flop pairs 30026(39386) 50%(66%) RAMB18×2s 15(24) 244 6%(9%) RAMB36_EXPs 9(220) 3%(90%) DSP48Es 252(322) 640 39%(50%) 采用ROACH进行混合架构相关器开发的资源占用情况 括号内为2048点FFT相关机资源占用情况
成本估算(VEGAS) FPGA+CPU: FPGA+GPU+CPU: (5108+8770+1300)×16+10000=$252848 ( 4000+2900+1300)×8+20000=$83584
应用实例 SETI BURST LEDA-OVRO HERA PRESTO …
HERA: Hydrogen Epoch of Reionization Array
System Overview
LEDA: Large-Aperture experiment to Detect the Dark Ages OVRO: Owens Valley Radio Observatory NIC: Network Interface Controller HERA: Hydrogen Epoch of Reionization Array PRESTO: PulsaR Exploration and Search TOolkit