一种基于奇异值分解的改进信噪比盲估计算法

Improved blind SNR estimation algorithm based on singular value decomposition

针对空间分解类信噪比(SNR)估计算法中子空间维数估计复杂度较高,低信噪比下估计偏差较大的问题,提出了一种改进的子空间维数估计算法。该算法首先利用样本自相关矩阵的奇异值序列进行后向差分得到梯度序列,对梯度序列每一项与后5项之和的比值进行搜索,最大比值所对应的奇异值序号作为信号子空间维数,最后计算信噪比。合适数据长度下的仿真结果表明:在信噪比-5 d B^20 d B范围内,常规通信信号的信噪比估计平均偏差小于0.5 d B,标准差小于1 d B;该算法提升了低信噪比下的估计性能,运算量较小,无需知道调制方式、载波频率、符号率等先验信息,在低信噪比时对信噪比时变的跟踪估计更为准确,且对复杂高阶调制信号同样适用。

Signal Noise Ratio（SNR） estimation algorithms adopting subspace decomposition exhibit some disadvantages such as high complexity of estimating dimension of subspace and large deviation under low SNR region. An improved algorithm to estimate the dimension of subspace is proposed. Firstly, autocorrelation matrix of receiving signals is constructed to decompose the singular values. Then the gradient array is obtained from singular values through backward deviation. The ratio of each element to the sum of next five elements in gradient array is searched to find the max ratio. The sequence number corresponding to the max ratio is the dimension of signal subspace. Finally, SNR estimation value is calculated. Simulations under appropriate length of data indicate that the mean bias of SNR estimation is below 0.5dB and the standard deviation is below 1dB for normal modulated signals with SNR from -5dB to 20dB. This algorithm improves estimated performance in low SNR region and reduces the amount of calculation without knowing the parameters such as modulation mode, carrier wave frequency and symbol frequency beforehand. It has better performance of SNR tracking estimation in low SNR region and is also suited to complex high order modulation signals.