利用记忆梯度法改进的变步长恒模盲均衡算法研究

针对传统恒模盲均衡算法收敛速度慢、固定步长条件下收敛速度和收敛精度之间存在矛盾的缺陷,提出了一种利用记忆梯度法改进的变步长恒模盲均衡算法。用记忆梯度算法替代最速梯度下降算法实现对恒模盲均衡中均衡器权值的调整,充分利用当前和前面迭代点的梯度信息,同时利用梯度信息变化率作为学习步长调整因子。新算法有效地提高了算法收敛速度,与共轭梯度法和拟牛顿法等改进算法比较,具有较低的计算复杂度和更好的均衡性能。计算机仿真证明了这一算法的有效性。

On a Variable-step Constant-modulus Blind Equalization Algorithm Modified by Memory Gradient Method

The traditional constant modulus blind equalization algrithm has defects including a slow convergence rate and,under a fixed step,conflicts between convergence rate and precision.Against those defects,this paper proposes a variable-step,constant-modulus,blind equalization algorithm modified by a memory gradient method.The equalizer weights are adjusted by the memory gradient method instead of the steepest descent algorithm.The new algorithm makes full use of gradient information on current and previous iteration points,while using gradient information changing rate as the learning step adjustment factor.It increases the convergence rate and,compared with other modified algorithms such as conjugate gradient and quasi-Newton methods,has lower computational complexity and better equalization performance.Computer emulation proved the effectiveness of this algorithm.