数字调制识别SVM分类器参数优化及仿真

对于基于SVM数字信号调制识别分类器,参数选取过程中如何优化惩罚因子和径向基核函数参数问题,提出了一种改进算法。该算法将自适应惯性权重粒子群算法和k折交叉验证法结合,利用交叉验证法计算粒子适应度值,通过粒子群算法实现最优参数值搜索,最终得到分类器惩罚因子和径向基核函数参数最优值。仿真结果表明,该算法性能明显优于网格搜索法和遗传算法。

Research and Simulation on the Parameter Optimization of Signal Modulation Recognition Classifier Based on SVM

For digital signal modulation recognition classifier based on SVM, an improved algo- rithm is proposed in the process of parameter selection of how to optimize the penalty factor and radial basis kernel function parameter. Combining adaptive weight particle swarm optimization and cross validation, the algorithm is proposed for the parameter selection of modulation recogni- tion classifier, which uses cross validation to calculate particle fitness value and uses particle swarm algorithm to finish parameter search. The performance in parameter selection than the grid simulation results show the method has better search method and the genetic algorithm.