Estimation of Vector Autoregressive Parameters and Granger Causality From Noisy Multichannel Data

Area of Science:

• Signal Processing
• Biomedical Engineering
• Time Series Analysis

Background:

• Multichannel recordings are crucial in biomedical applications but often suffer from measurement noise.
• Accurate estimation of multivariate autoregressive (MAR) parameters is essential for analyzing such data.
• Existing methods may struggle with noise and complex MAR processes.

Purpose of the Study:

• To propose a novel method for estimating MAR parameters from noisy multichannel data.
• To introduce the nonlinear Cadzow method as a robust alternative to existing techniques.
• To demonstrate the method's effectiveness in denoising and parameter estimation.

Main Methods:

• A multivariate generalization of the Cadzow method was employed.
• The proposed method is termed the nonlinear Cadzow method.
• Performance was evaluated using simulated and experimental local field potential data.

Main Results:

• The nonlinear Cadzow method demonstrated superior performance compared to higher order Yule-Walker and Kalman EM methods on simulated data.
• The method achieved better results in estimating Granger causality from noisy data.
• Application to monkey cortical data yielded results consistent with known cortical physiology.

Conclusions:

• The nonlinear Cadzow method provides denoised estimates of MAR parameters, outperforming current approaches.
• This method offers significant potential for impact in biomedical applications involving noisy multichannel data.
• The technique is valuable for analyzing functional connectivity and other complex signal processing tasks.