基于向量量化的简化DFT滤波信道估计  

DFT-based Channel Estimation Simplified with Vector Quantization

作　　者：张逸帆 姜明 还冬锐 巩帅聪 ZHANG Yifan;JIANG Ming;HUAN Dongrui;GONG Shuaicong(School of Information Science and Engineering,Southeast University,Nanjing 210089,China)

机构地区：[1]东南大学信息科学与工程学院,江苏南京210089

出　　处：《无线电工程》2024年第10期2296-2304,共9页Radio Engineering

基　　金：国家自然科学基金(62271137)。

摘　　要：在5G通信系统中,正交频分复用(Orthogonal Frequency Division Multiplexing, OFDM)技术被广泛应用于物理层,而OFDM信道估计算法通常涉及大量的矩阵运算,其中矩阵乘法计算对整个系统的实现复杂度有着重要影响。基于离散傅里叶变换(Discrete Fourier Transform, DFT)滤波的OFDM信道估计算法利用一对DFT/离散傅里叶反变换(Inverse Discrete Fourier Transform, IDFT),在时域上消除最小二乘法(Least Squares, LS)、最小均方误差(Minimum Mean Square Error, MMSE)等基础估计算法结果中的噪声,算法简单有效,但引入的DFT/IDFT导致其实现复杂度显著上升。从DFT滤波中的IDFT入手,提出一种简化的OFDM信道估计实现方法。该方法借助MADDNESS算法,将IDFT的计算过程转化为向量量化(Vector Quantization, VQ)的查表过程,去除了大量的乘法和加法操作,大幅降低了IDFT过程的计算复杂度。针对MADDNESS算法中质心优化过程的内存开销问题,引入了子采样方法,在降低岭回归算法所需内存开销的同时,保证了系统整体性能几乎无损。仿真结果表明,与采用精确矩阵乘法的IDFT运算相比,所提出的简化方法在信道估计准确度上的性能损失可控制在0.5 dB以内,同时计算复杂度仅为精确算法的8%。In 5G communication systems,Orthogonal Frequency Division Multiplexing(OFDM)technology is widely used in the physical layer,and OFDM channel estimation algorithms usually involve a lot of matrix operations,among which the matrix multiplication has a significant impact on the implementation complexity of the entire system.The OFDM channel estimation algorithm,utilizing Discrete Fourier Transform(DFT)filtering and employing a pair of DFT/Inverse Discrete Fourier Transform(IDFT)operations,effectively eliminates noise from basic estimation methods such as Least Squares(LS)and Minimum Mean Square Error(MMSE)in the time-domain.Although this approach is straightforward and effective,the introduction of DFT/IDFT transformations significantly increases its implementation complexity.A simplified channel estimation method for OFDM system based on IDFT in the DFT filtering is proposed.By using MADDNESS to transform the calculation process of IDFT into the table lookup process of Vector Quantization(VQ),a large number of multiplication and addition operations are removed,so the computational complexity of IDFT is greatly reduced.To address the memory overhead problem in the centroid optimization process of MADDNESS,the subsampling method is introduced to reduce the memory overhead required by the ridge regression algorithm while ensuring that the overall system performance is almost lossless.The simulation results show that compared with the IDFT operation using the precise matrix multiplication,the performance loss of the simplified method in channel estimation accuracy can be controlled within 0.5 dB,while the computational complexity is only 8%of the precise algorithm.

关 键 词：正交频分复用 信道估计 向量量化 MADDNESS 岭回归 

分 类 号：TN929.5[电子电信—通信与信息系统]