Method for the rapid estimation of figures of merit for multiple devices based on nonlinear modeling

• US 6,892,155 B2
• Filed: 11/19/2002
• Issued: 05/10/2005
• Est. Priority Date: 11/19/2002
• Status: Expired due to Fees

• Current Assignee: Agilent Technologies, Inc.
• Original Assignee: Agilent Technologies, Inc.
• Inventors: Gee, Kevin, Tufillaro, Nicholas B.
• Primary Examiner(s): Barlow, John
• Assistant Examiner(s): Pretlow, Demetrius
• Application Number:
  US10/299,169
• Publication Number:
  US 20040098215A1
• Time in Patent Office: 903 Days
• Field of Search: 702/108, 702/117, 702/109, 702/118, 702/127, 703/2, 703 13- 18
• US Class Current: 702/108
• CPC Class Codes: G06F 30/367 Design verification, e.g. using simulation, simulation program with integrated circuit emphasis [SPICE], direct methods or relaxation methods

Abstract

A system and a general method estimate figures of merit based on nonlinear modeling and nonlinear time series analysis. Terms in a nonlinear behavioral model that depend on nonlinear combinations of a fixed input signal value are precomputed, optimizing the behavioral model such that figures of merit are evaluated from a single short stimulus vector. The optimized nonlinear behavioral model can then be applied to evaluate figures of merit for multiple devices under test (DUTs) in a manufacturing line. A process continually verifies and adjusts the nonlinear behavioral model based on sub-sampling multiple DUTs in a manufacturing line and comparing their figures of merit based on nonlinear modeling with those based on conventional measurement procedures.

Claims

1. A method of estimating a figure of merit for a nonlinear device under test, comprising:

developing from a first sub-sampling of input signals and corresponding output signals for said device a nonlinear behavioral model fitted mathematical function for said device containing terms depending on linear and nonlinear combinations of said input signals and linearly on a set of parameter coefficients;

deriving from said nonlinear behavioral model fitted mathematical function a figure of merit having terms depending on linear and nonlinear combinations of said input signals and linearly on a set of parameter coefficients;

before said nonlinear behavioral model is applied, pre-computing in advance for a substantially fixed value of said input signal all terms in said nonlinear model that depend on nonlinear combinations of said input signals; and

evaluating said parameter coefficients, such that said nonlinear behavioral model is optimized.

2. The method of claim 1 wherein said nonlinear behavioral model fitted mathematical function contains terms depending on linear and nonlinear combinations of said input signals and/or time derivatives of said input signals.

3. The method of claim 2 further comprising applying said optimized nonlinear behavioral model to estimate a figure of merit for a plurality of said nonlinear devices using a second sub-sampling of said input signal having said substantially fixed value for all of said plurality of said nonlinear devices.

4. The method of claim 2 further comprising before said nonlinear behavioral model is applied verifying said nonlinear behavioral model using a third sub-sampling of said input signals and corresponding output signals.

5. The method of claim 3 wherein said optimized nonlinear behavioral model is continually verified and adjusted using a fourth sub-sampling of said input signals and corresponding output signals.

6. The method of claim 2 wherein said nonlinear behavioral model fitted mathematical function is fitted to an embedded data set.

7. The method of claim 6 wherein said embedded data set is embedded using a method selected from the group consisting of lagged embedding, singular value decomposition, and wavelet decomposition individually and in combination.

8. The method of claim 6 wherein said behavioral model fitted mathematical function is fitted to said embedded data set using a method selected from the group consisting of least squares fit, least squares fit with multivariate polynomials, radial basis functions, neural networks, universal function approximation, and local linear modeling.

9. The method of claim 2 wherein said parameter coefficients are evaluated using operations selected from the group consisting of Kalman filtering, variants of Kalman filtering, and matrix algebra operations.

10. The method of claim 9 wherein said matrix algebra operations comprise pseudoinverse matrices and/or singular value decomposition.

11. The method of claim 2 wherein said figure of merit is selected from the group consisting of correlation coefficient (ρ

), error vector magnitude, adjacent channel power ratio, third-order intermodulation distortion, and power spectral density.

12. A system for estimating a figure of merit for a nonlinear device under test, comprising:

an excitation signal generating module operable to apply a plurality of input signals to an input port of said nonlinear device;

a data acquisition module operable to sample said plurality of input signals and to measure a plurality of output signals generated by said nonlinear device in response to said plurality of input signals; and

a signal processing computer operable to create from a sub-sampling of said plurality of input and output signals an embedded data set and further operable to fit a nonlinear behavioral model function to said embedded data set, and operable to compute at least one figure of merit using said nonlinear behavioral model function.

13. The system of claim 12 wherein said signal processing computer is operable to create said embedded data set using a method selected from the group consisting of lagged embedding, singular value decomposition, and wavelet decomposition individually and in combination.

14. The system of claim 12 wherein said signal processing computer is operable to fit said nonlinear behavioral model function using a method selected from the group consisting of least squares fit, least squares fit with multivariate polynomials, radial basis functions, neural networks, universal function approximation, and local linear modeling.

15. The system of claim 12 wherein said nonlinear behavioral model function contains terms depending on parameter coefficients and on linear and nonlinear combinations of said input signals and/or time derivatives of said input signals.

16. The system of claim 15 wherein said signal processing computer is operable to evaluate said parameter coefficients using operations selected from the group consisting of Kalman filtering, variants of Kalman filtering, and matrix algebra operations.

17. The system of claim 16 wherein said matrix algebra operations comprise pseudoinverse matrices and/or singular value decomposition.

18. The system of claim 12 wherein said figure of merit is selected from the group consisting of correlation coefficient (ρ

), error vector magnitude, adjacent channel power ratio, third-order intermodulation distortion, and power spectral density.

19. A system for estimating a figure of merit for a nonlinear device under test, comprising:

an excitation signal generating module operable to apply a plurality of input signals to an input port of said nonlinear device;

a data acquisition module operable to sample said plurality of input signals and to measure a plurality of output signals generated by said nonlinear device in response to said plurality of input signals; and

a signal processing computer operable to create from a sub-sampling of said plurality of input and output signals an embedded data set and further operable to fit a nonlinear behavioral model function to said embedded data set, wherein said function contains terms depending on parameter coefficients and on linear and nonlinear combinations of said input signals and/or time derivatives of said input signals.

20. The system of claim 19 wherein said signal processing computer is operable to evaluate said parameter coefficients using operations selected from the group consisting of Kalman filtering, variants of Kalman filtering, and matrix algebra operations.