利用分段时频峰值滤波法抑制磁共振全波信号随机噪声

Segmented time-frequency peak filtering for random noise reduction of MRS oscillating signal

磁共振信号极其微弱,容易受到周围环境中各种电磁噪声干扰.其中随机噪声,由于频带宽、不规则、无规律、与有效信号混叠,难以抑制.近年来,采用数量级为104~105 Hz采样频率收录的全波磁共振信号,以其携带丰富全面的信息量,为数据处理及解释提供了更多的潜能.然而,只要随机噪声的幅度大于信号幅度,拟合得到的信号特征参数准确度就会降低.目前普遍采用的数据叠加方法仅能抑制部分随机噪声,且需要多次采集信号,探测效率低.本文针对全波磁共振信号采样点数多和信号非线性强的特点,提出采用分段时频峰值滤波(STFPF)法消噪,将全波磁共振信号分成若干段,编码为解析信号的瞬时频率,采用短窗长PWVD计算解析信号的时频分布,利用时频分布沿瞬时频率集中的特性,通过提取时频分布的峰值获得信号的无偏估计,达到抑制全波磁共振信号随机噪声的目的.为了验证消噪效果,与传统叠加法进行对比分析,仿真结果表明,对于单次采集信号,信噪比低至-5dB时,STFPF方法依然能有效抑制信号中的随机噪声,消除随机噪声后信噪比提高23.19dB,信号的初始振幅拟合误差为3.03%,平均横向弛豫时间拟合误差为2.7%,消噪效果优于传统叠加法,且由于无需多次采集磁共振信号,可有效提高探测效率.模型数据的反演解释进一步验证了STFPF方法的有效性,本文研究结果为实际数据处理奠定了良好的基础.

Magnetic resonance sounding(MRS)signal is weak,and the measurement always suffers bad signal-to-noise ratio(SNR).For the random noise interference,it is difficult to remove,because the random noise has stochastic property and always mixing with desired signal.More information is provided to signal processing and inversion interpretation by MRS oscillating signal which recorded at 10 4-10 5 Hz sampling rate,but as long as the average magnitude of random noise is larger than that of signal,the accuracy of parameter estimation will be reduced.The common method for random noise reduction is stacking,but it requires multiple recordings and often limited to the slow measurement progress.Considering the MRS oscillating signal has the characters of large number of samples and high nonlinearity,the general objective of this study is to employ a segmented time-frequency peak filtering(STFPF)to recover the desired signal embedded in the noisy MRS oscillating data.Firstly,the noisy signal is divided into several segments.By transforming each segment of the signal into instantaneous frequency(IF)of the analytic signal,and calculating the time-frequency distribution of the analytic signal with a small windowed pseudo Wigner-Ville transform,then significant energy concentration is produced around the IF on the time-frequency plane.Subsequently,the unbiased estimation of underlying MRS signal is achieved by taking the peak of the time-frequency distribution of analytic signal.In order to verify the effectiveness of the proposed method,an analysis has been done which compared to the traditional stacking method.Numerical simulation shows that the desired MRS oscillating signal can be recovered from a single recording in noise level down to-5 dB by applying STFPF method.The SNR is increased 23.19 dB,the fitting error of initial amplitude is 3.03%and the fitting error of the transverse relaxation time is 2.7%after filtering.It performs better than the traditional stacking method.In addition,it is unnecessary to sample multi-recordings,and this will improve the measurement efficiency.Moreover,the effectiveness of STFPF method is further demonstrated by the retrieved model after inversion.The results of the proposed method in this study lay a good foundation for the real data processing.