基于GPP平台的LTE上行信道估计高效实现

摘要

Real-time processing is the key technique in the LTE base station system on GPP platforms.In the uplink channel estimation,Wiener interpolation is adopted for good performance,however,it causes a high computation complexity,which becomes the bottleneck of the real-time processing in LTE uplink channels and requires the optimization based on the features of GPP platforms.Using partial linear interpolation and pre-storing the coefficients of Wiener interpolation,the computation complexity of Wiener interpolation is reduced significantly.Then,SSE instruction sets and pipelining techniques are used for optimization.Maintaining the estimation performance and using a single core,the optimized module could process the uplink channel estimation of 100 physical resource blocks in 6.36 microseconds,which is only 1.5％ of the initial method.Therefore,the uplink channel estimation is no longer the bottleneck of the real-time processing in GPP software defined radio systems.%处理的实时性是基于GPP(general purpose processor)平台的LTE基站系统的关键指标.在上行信道估计环节,往往需要维纳插值来获得良好性能；但其过高的计算复杂度使信道估计成为上行物理信道实时性处理的计算瓶颈,需要结合GPP特点进行算法实现的优化.通过引入部分线性插值和预存维纳插值系数的方式,显著降低维纳插值的运算复杂度.再使用SSE指令集和“流水线”处理的优化,在保持性能基本不变的情况下,在单核处理器上处理100个物理资源块的上行信道估计仅需6.36 μs,为优化前的1.5％,使得高性能上行信道估计不再成为GPP软件无线电实时处理的瓶颈.