Parameter estimation of chirp signal under low SNR

Abstract

Aiming at solving the problem of great difficulty and low accuracy existing in parameter estimation of chirp signal under low signal-to-noise ratio (SNR) condition, an algorithm of accurate parameter estimation of chirp signal is proposed. First, the algorithm extracts ridge frequency of chirp signal based on Short Time Fourier Transform. Second, the protruding glitch frequencies are eliminated through median filter with proper size and the smoothing frequencies are obtained corresponding to the time. Third, the frequency time frequency-modulated (FM) line is fitted coarsely by the least-square linear fitting method, and some frequency points are removed, which are far away from the FM line. Repeat the process several times until the sample correlation coefficient of the fitted line is in high degree when the optimum chirp line is fitted out, so chirp rate and initial frequency are obtained. Simulation shows that the correct rate of parameter estimation is very high. When SNR is not less than −17 dB, the correct rate is 100%, and when the SNR is −18 dB, the correct rate is still capable of up to 99%. This algorithm has a lower SNR threshold of about −17 dB. When the SNR is greater than threshold, parameter estimation accuracy is close to Cramer Rao low bound (CRLB), higher than parameter estimation based on the phase field method. Measured data experiments show that the algorithm can reasonably fit the chirp line of measured chirp signal, which better characterizes the FM features of chirp signal.

摘要

创新点

针对低信噪比下线性调频(Linear Frequency Modulation, LFM)信号参数估计难度大以及精度不高的问题, 本文提出一种基于短时傅立叶变换(Short Time Fourier Transform, STFT)的最小二乘调频直线拟合的LFM信号参数估计算法。 该算法首先通过LFM信号的STFT提取出脊线频率并通过中值滤波滤除凸出的毛刺频率点, 然后通过最小二乘法粗拟合出频率-时间调频直线并剔除偏离调频直线距离较大的频率点, 迭代进行该过程多次, 直到拟合直线的样本相关程度很高, 最终拟合出最优的频率-时间直线, 从而得到调频斜率和起始频率。 仿真表明, 本算法调频斜率估计正确率较高, 信噪比不小于−17dB时, 正确率达到100%, 信噪比为−18dB时, 正确率仍能高达99%。 本算法具有较低的信噪比门限, 约为−17dB; 高于信噪比门限时, 参数估计精度接近克拉美-罗限(Cramer Rao Low Bound, CRLB), 精度高于基于相位域的参数估计方法。 实测数据实验表明, 本文算法能够合理地拟合出实测LFM信号的调频直线, 较好地表征了LFM信号的调频特征。