The Research On The Improved Algorithm Of Minimum Variance Adaptive Beamforming For Ultrasound Imaging

Ultrasound imaging is widely applied in the fields of medical diagnosis,industry and power nondestructive testing because of its superiority of safety,non-invasive,convenience,low-cost and real-time imaging.Ultrasound imaging testing system includes transmitting,receiving,beamforming and imaging modules.The beamforming module is the core of the whole system,which directly determines the ultrasound imaging quality.The adaptive beamforming algorithm,taking the minimum variance(MV)algorithm as an example,has certain improvement in resolution and contrast compared to the delay and sum(DAS)algorithm.However,there are still two problems that limit the imaging quality and application of the adaptive beamforming algorithm.(1)MV beamformer involves the inversion of high-dimensional covariance matrix in the process of dynamic weighting,especially when fused with eigenspace method(ESBMV)to suppress noise interference,which involves the eigenvalue decomposition and ranking operations of the covariance matrix.Therefore,the algorithm has high complexity,poor robustness,and significantly reduced imaging efficiency,which makes it difficult to meet the real-time requirements of ultrasound imaging.(2)The narrowband assumption of MV beamformer contradicts the broadband characteristics of ultrasound signals.Although the subband minimum variance(MVS)algorithm transforms ultrasound broadband signals into narrowband subband signals through Short-time Fourier transform(STFT),which can break through the resolution limit of the MV algorithm.However,due to the spectrum leakage and uncertain time-frequency concentration by using STFT,artifacts are caused,which are not suitable for continuous target imaging.In view of the above problems,the MV algorithm from the perspective of beam-domain conversion and frequency-domain conversion are conducted as follows:(1)In order to solve the problems of high complexity and poor robustness of the MV and ESBMV algorithms,a low-complexity eigenspace-based minimum variance(LCMV)beamforming algorithm with power method is proposed.The LCMV algorithm first uses discrete cosine transform to convert the ultrasound echo signal to low-dimensional beam domain,and determines the optimal dimension reduction coefficient based on the minimum distortion principle and beam-domain conversion efficiency.Then,the power method is used to get the maximum eigenvalue and the corresponding eigenvector of the covariance matrix.Finally,the low-energy signals corresponding to other eigenvalues are ignored,and the covariance matrix inversion is simplified to vector multiplication and projection.Simulation and experimental results show that the LCMV algorithm has better resolution and efficiency than MV and ESBMV algorithms,and is more robust to changes of ultrasound velocity,which means it can be applied to different ultrasonic non-destructive testing occasions.(2)For the problems of the time-frequency uncertainty and spectrum inaccuracy by using STFT in the MVS algorithm,a minimum variance beamforming(STFTMV)algorithm based on optimal frequency-domain segmentation is proposed.Firstly,an improved time-frequency concentration criterion based on logarithmic window energy is proposed,and the optimal window of STFT is selected to achieve the optimal segmentation of the frequency narrowband sub-signals.Then,the computation of frequency-domain beamforming is halved by using the conjugate symmetry of STFT.Finally,the frequency-domain subband signals of each array element are reconstructed by using the non-overlapping characteristic of STFT window.Simulation and experimental results show that the STFTMV algorithm has better resolution,robustness and efficiency than MV and ESBMV algorithms.Moreover,combined the improved frequency-domain coherence factor,the STFTMV algorithm can achieve better contrast.(3)For the problems of spectrum leakage and continuous imaging blur in the MVS algorithm,a subband minimum variance(ABMVS)algorithm based on spectral pursuit model which combined with the alternating multipliers iterations is proposed.Firstly,by establishing a spectral pursuit model of ultrasound signal,the problem of solving frequency-domain subband signals can be transformed into the problem of solving the frequency coefficient of each time window after STFT of ultrasound signals.Then,the penalty function to verify the frequency sparsity of ultrasound signal is introduced to obtain the maximum posterior probability solution of the frequency coefficient estimation problem.Finally,the alternating direction method of multipliers method is used to iteratively obtain the ultrasound signal spectrum to find the exact solution of the spectral pursuit model.Simulation and experimental results show that the ABMVS algorithm has better resolution and contrast than MV and ESBMV algorithms,and has the strongest ability to suppress noise.