Colored Noise and Regularization Parameter Selection for Waveform Metrology

We study six algorithms to select a regularization parameter for deconvolution problems appearing in high-speed communication measurement applications. We investigate these algorithms in the presence of unspecified noise correlation analyzing their performance as components of a multivariate random variable and study their joint distribution by Monte Carlo. We find that several parameter selection algorithms, despite their widespread use, are not robust to unspecified noise correlations. Specifically, the discrepancy principle fails to return adequate regularizations for rough noise, while the generalized cross validation (GCV), unbiased predictive risk, and information complexity selectors can fail for smooth noise. For some experimental configurations, GCV failed completely, returning zero successful inversions out of 500 noise instantiations. These parameter selection algorithms share in the characteristic that they do not contain mechanisms to monitor quantities derived from the parameter-dependent solution vector. By contrast, the L-curve and quasi-optimality criteria do contain such mechanisms, and furthermore exhibited significantly fewer failures and correlated highly with the optimal inversion across all noise levels and correlations.

[1]  Behavior near zero of the distribution of GCV smoothing parameter estimates , 1995 .

[2]  A. Bakushinskii.  Remarks on choosing a regularization parameter using the quasioptimality and ratio criterion , 1985 .

[3]  C. Vogel.  Non-convergence of the L-curve regularization parameter selection method , 1996 .

[4]  Paul D. Hale,et al.  Waveform metrology and a quantitative study of regularized deconvolution , 2010, 2010 IEEE Instrumentation & Measurement Technology Conference Proceedings.

[5]  H. Engl.  Discrepancy principles for Tikhonov regularization of ill-posed problems leading to optimal convergence rates , 1987 .

[6]  Mark A. Lukas,et al.  Comparing parameter choice methods for regularization of ill-posed problems , 2011, Math. Comput. Simul..

[7]  Peter M. Harris,et al.  The evaluation of uncertainties in the analysis of calibration data , 1999, IMTC/99. Proceedings of the 16th IEEE Instrumentation and Measurement Technology Conference (Cat. No.99CH36309).

[8]  C. Vogel.  Computational Methods for Inverse Problems , 1987 .

[9]  A. N. Tikhonov,et al.  REGULARIZATION OF INCORRECTLY POSED PROBLEMS , 1963 .

[10]  S.M. Riad,et al.  The deconvolution problem: An overview , 1986, Proceedings of the IEEE.

[11]  Nicholas G. Paulter,et al.  The IEEE standard on transitions, pulses, and related waveforms, Std-181-2003 , 2004, IEEE Transactions on Instrumentation and Measurement.

[12]  Dylan F. Williams,et al.  A Statistical Study of De-Embedding Applied to Eye Diagram Analysis , 2012, IEEE Transactions on Instrumentation and Measurement.

[13]  Naomi Altman,et al.  Kernel Smoothing of Data with Correlated Errors , 1990 .

[14]  Dominique Schreurs,et al.  Applications of vector non-linear microwave measurements , 2010 .

[15]  C. Dorrer,et al.  High-speed measurements for optical telecommunication systems , 2006, IEEE Journal of Selected Topics in Quantum Electronics.

[16]  Santhosh George,et al.  Parameter choice by discrepancy principles for ill-posed problems leading to optimal convergence rates , 1994 .

[17]  J. R. M. Hosking,et al.  FRACTIONAL DIFFERENCING MODELING IN HYDROLOGY , 1985 .

[18]  Markus Reiss,et al.  Regularization independent of the noise level: an analysis of quasi-optimality , 2007, 0710.1045.

[19]  Thomas Huckle,et al.  Tikhonov–Phillips regularization with operator dependent seminorms , 2012, Numerical Algorithms.

[20]  Gene H. Golub,et al.  Matrix computations , 1983 .

[21]  Gene H. Golub,et al.  Generalized cross-validation as a method for choosing a good ridge parameter , 1979, Milestones in Matrix Computation.

[22]  An autocorrelation approach for parameter estimation of fractional order equal-root autoregressive models using hypergeometric functions , 1980 .

[23]  M. Hanke.  Limitations of the L-curve method in ill-posed problems , 1996 .

[24]  Dylan F. Williams,et al.  Traceable Waveform Calibration With a Covariance-Based Uncertainty Analysis , 2009, IEEE Transactions on Instrumentation and Measurement.

[25]  M. A. Lukas.  Robust generalized cross-validation for choosing the regularization parameter , 2006 .

[26]  Wen-ke Chen.  - Passive and Active Filters , 2018, Signals, Systems, Transforms, and Digital Signal Processing with MATLAB.

[27]  Mark A. Lukas,et al.  Robust GCV choice of the regularization parameter for correlated data , 2010 .

[28]  Thomas Bonesky.  Morozov's discrepancy principle and Tikhonov-type functionals , 2008 .

[29]  Paul D. Hale,et al.  Calculation of Pulse Parameters and Propagation of Uncertainty , 2009, IEEE Transactions on Instrumentation and Measurement.

[30]  E. Somersalo,et al.  Statistical inverse problems: discretization, model reduction and inverse crimes , 2007 .

[31]  T. Sarkar,et al.  Some mathematical considerations in dealing with the inverse problem , 1981 .

[32]  Ronny Ramlau,et al.  Convergence rates for Morozov's discrepancy principle using variational inequalities , 2011 .