改进变步长LMS在组合学习结构预失真的研究

针对预失真间接学习结构易受加性噪声、模数转换（Analog to Digital Converter，ADC）量化噪声等影响，提出一种通过设置判别门限自适应切换直接学习结构和间接学习结构的组合学习结构数字预失真方案。该方案在直接学习结构中采用最小二乘法（Recursive Least Square，RLS）算法对参数进行快速粗估计，切换至间接学习结构时采用改进变步长最小均方（Least Mean Square，LMS）算法进一步提取参数。分析仿真表明，组合学习结构的预失真方案其线性化性能较间接学习结构有很大提升，且在算法收敛速度基本持平的情况下有效抑制了间接学习结构中的非相关噪声。

Study on a improved step size LMS algorithm in composite learning architecture pre-distortion

To overcome disadvantages of additive noise and quantization noise of analog to digital converter (ADC) in the indirect learning architecture, a new composite learning architecture scheme in digital pre-distortion is proposed, which switching direct and indirect learning structure of learning structure depending on the setting appropriate threshold adaptively. In direct learning structure, the recursive least square (RLS) algorithm is used to fast estimate the parameters, and the minimum mean square (LMS) algorithm is adopted to extract the parameters when it is switched to the indirect learning structure. The simulation and analyses show that the linearization performance of the proposed scheme is improved greatly than which in the indirect learning structure. Furthermore, with two convergence rates of the algorithm equal practically it can restrain the non-correlative noises in indirect learning structure effectively.