identification of parallel wiener
play

Identification of parallel Wiener- Hammerstein systems with a - PowerPoint PPT Presentation

Identification of parallel Wiener- Hammerstein systems with a decoupled static nonlinearity M. Schoukens, K. Tiels, M. Ishteva, J. Schoukens 1 Parallel Wiener-Hammerstein Flexible Simple LTI SNL LTI 2 Identifiability Full rank linear


  1. Identification of parallel Wiener- Hammerstein systems with a decoupled static nonlinearity M. Schoukens, K. Tiels, M. Ishteva, J. Schoukens 1

  2. Parallel Wiener-Hammerstein Flexible Simple LTI SNL LTI 2

  3. Identifiability Full rank linear transform 3

  4. Identifiability Full rank linear transform  coupled nonlinearity 4

  5. Best Linear Approximation Input signals 5

  6. Best Linear Approximation Input signals Combination of dynamics! 6

  7. Best Linear Approximation Input signals Combination of Fixed poles dynamics! Moving zeros 7

  8. Best Linear Approximation Common denominator – Fixed poles – Moving zeros 8

  9. Decomposing the dynamics numerator coefficients D U  V 9

  10. Partition the dynamics 10

  11. Partition the dynamics 11

  12. Partition the dynamics 12

  13. Partition the dynamics 13

  14. Decoupling 14

  15. Decoupling a Static Nonlinearity 15

  16. Decoupling a Static Nonlinearity Homogeneous 2 nd degree  Decoupling = matrix diagonalization 16

  17. Decoupling a Static Nonlinearity Homogeneous 3 rd degree 17

  18. Decoupling a Static Nonlinearity Homogeneous 3 rd degree  Decoupling = tensor diagonalization 18

  19. Decoupling a Static Nonlinearity Polynomial of degree n – Combine Homogeneous Tensors of different degrees – Add constant input  Decoupling = tensor diagonalization 19

  20. Measurement Example Multisine input: System: 5 amplitudes Custom built circuit 12 th order dynamics 20 realizations 2 periods Diode-resistor NL 16384 samples 20

  21. Measurement Example Validation Error (mV) rms(e) Coupled, 3 rd Degree 11.92 Decoupled, 3 rd Degree 31.05 Decoupled, 15 th Degree 0.66 21

  22. Measurement Example 22

  23. Measurement Example 23

  24. Conclusion • Different LTI models  parallel Wiener-Hammerstein model • Tensor Diagonalization  Decoupling polynomial • Low complexity • High flexibility • Good performance It works! 24

  25. Identification of parallel Wiener- Hammerstein systems with a decoupled static nonlinearity M. Schoukens, K. Tiels, M. Ishteva, J. Schoukens 25

Recommend


More recommend