Research On Recognition Of Digital Signal Modulation Via Ensemble Learning

During the development of radio,the key step to communication system is to confirm the signal modulation scheme whether is traditional radio or cognitive radio.Therefore,there exits great influence for signal modulation scheme recognition.Facing with the low and unknown signal-noise ratio(SNR),the common classification algorithms can not have satisfied performance.To boost the accuracy of communication modulation scheme recognition,the ensemble learning is utilized for combination decision tress in this paper.Firstly,some entropy-based features are extracted from various kinds perspective to embody signal characteristics.Meanwhile,the analysis for decision tree properties is made under single SNR which exposes some drawbacks.And the result for models is presented: the more complex of model construction,the higher accuracy is.As a trade-off between the complexity and performance,the state-of-art method,boosting,occurs in the application of digital signal modulation schemes.It is proved by research that the boosting based algorithms,i.e.,Adaboost,gradient boosting,Xgboost,are effective.A series of experiments under single and unknown SNRs are taken to distinct the superiority of boosting algorithms form widely used machine learning with recognition and efficiency respectively.In the end,based on Bayesian hypothesis and Bayesian theory,this paper presents the confidence machine learning.This method does not only predict the raw datasets,but also appraise the recognition quality.Hense,our method can offer more credible information.