局部迭代NLMS算法研究

针对回波抵消器中大量抽头系数需要更新的问题，分析比较各种局部迭代NLMS算法的性能和计算复杂度。根据权系数局部迭代的简化原理，将滤波器系数分成多个系数子集，通过每次迭代仅更新权系数部分子集的方法，减少算法的计算量。并对不同的子集生成策略进行复杂度和滤波性能分析，分析结果表明，基于权系数瞬时梯度估计的Mmax系数局部迭代方法，以及基于最小化干扰原理的选择性权系数局部迭代算法，其性能与全系数更新算法相当，而计算量与连续局部迭代算法相比仅附加少量比较运算。

Research on the Partial-Update NLMS Algorithms

Aims at the large number of filter coefficients diminishes the usefulness of the adaptive filtering algorithm owing to increased complexity, summarizes some partial-update adaptive signal processing algorithms. Computational complexity, provides convergence and stability properties of the class of Sequential Partial-Update algorithm and Stochastic Partial-Update algorithms. In addition, based on the magni- tude of the corresponding gradient estimate in the direction of every individual coefficient at a given iteration and the principle of mini- mum disturbance, derives two simple and direct criterions ranking the regressor vector blocks according to their largest values or squared Euclidean norms for the selection of coefficients to be updated. Examples of such applications which in acoustic echo cancellation show that the Mmax tap and selective-partial-update algorithms have good convergence performance and allow significant savings to be achieved on computational requirements.