改进型 BP 神经网络自适应均衡器设计

针对传统的基于 BP 神经网络的自适应均衡器存在的缺陷,提出了一种改进的 BP 神经网络自适应均衡器的 FPGA 设计方法.从互连结构、训练法则及收敛算法等几个方面对 BP 网络进行改进,通过加入协调器提高 BP 网络收敛速度和容错性能,并根据 BP 网络的并行性能得到分块结构模型,同时采用流水线技术使计算速度明显提高.通过软件仿真,验证了这种结构简单、收敛速度较快的 BP 网络的有效性和可行性,并在ModelSim 工具上完成了自适应均衡器的仿真验证

The Improved Designing for Adaptive Equalizer Based on Optimize BP Network

To against the limitation of adaptive equalizer based on the old BP neural network,a new FPGA designing method of improved BP neural network is brought out.The method improved the BP network from the interconnect architecture,training algorithm and convergence algorithm by adding the coordinator to improve the BP′s convergence rates and error compatibility.Besides,the convergence rates is obviously accelerated by adopting the pipelining capability and concluding the separated bank structures according to BP network′s paralleled.Finally it was validated this simple-configuration,high-convergence rates BP network is validity and feasibility by software-simulation,and the simulation of adaptive equalizer is carried out by ModelSim.