Full duplex radio increases the frequency efficiency but its performance is limited by the self-interference （SI）. We first analyze the multiple noises in the full duplex radio system and model such noises as an α ...Full duplex radio increases the frequency efficiency but its performance is limited by the self-interference （SI）. We first analyze the multiple noises in the full duplex radio system and model such noises as an α - stable distribution. Then we formulate a novel non-Gaussian SI problem. Under the maximum correntropy criterion （MCC）, a robust digital non-linear self-interference cancellation algorithm is proposed for the SI channel estimation. A gradient descent based algorithm is derived to search the optimal solution. Simulation results show that the proposed algorithm can achieve a smaller estimation error and a higher pseudo signal to interference plus noise ratio （PSINR） than the well-known least mean square （LMS） algorithm and least square （LS） algorithm.展开更多
高斯混合模型采用固定混合数结构的建模方法并不符合说话人语音特征分布的多样性,从而出现过拟合或者欠拟合的情况并影响系统的识别性能。提出一种混合数可变的自适应高斯混合模型并将其应用于说话人识别。模型训练中根据说话人语音特...高斯混合模型采用固定混合数结构的建模方法并不符合说话人语音特征分布的多样性,从而出现过拟合或者欠拟合的情况并影响系统的识别性能。提出一种混合数可变的自适应高斯混合模型并将其应用于说话人识别。模型训练中根据说话人语音特征参数分布的聚类特性,采用吸收合并与分裂机制动态调整混合数以获得更加精确的拟合性能,提高系统识别率。实验结果显示,在特征参数MFCC和BFCC(Bilinear Frequency Cepstrum Coefficients)下相对误识率分别下降了41.41%和22.21%。展开更多
基金supported by the National Natural Science Foundation of China under Grants 61372092"863" Program under Grants 2014AA01A701
文摘Full duplex radio increases the frequency efficiency but its performance is limited by the self-interference （SI）. We first analyze the multiple noises in the full duplex radio system and model such noises as an α - stable distribution. Then we formulate a novel non-Gaussian SI problem. Under the maximum correntropy criterion （MCC）, a robust digital non-linear self-interference cancellation algorithm is proposed for the SI channel estimation. A gradient descent based algorithm is derived to search the optimal solution. Simulation results show that the proposed algorithm can achieve a smaller estimation error and a higher pseudo signal to interference plus noise ratio （PSINR） than the well-known least mean square （LMS） algorithm and least square （LS） algorithm.
文摘高斯混合模型采用固定混合数结构的建模方法并不符合说话人语音特征分布的多样性,从而出现过拟合或者欠拟合的情况并影响系统的识别性能。提出一种混合数可变的自适应高斯混合模型并将其应用于说话人识别。模型训练中根据说话人语音特征参数分布的聚类特性,采用吸收合并与分裂机制动态调整混合数以获得更加精确的拟合性能,提高系统识别率。实验结果显示,在特征参数MFCC和BFCC(Bilinear Frequency Cepstrum Coefficients)下相对误识率分别下降了41.41%和22.21%。
