Description

SebastienDeriaz

opened on Nov 10, 2020

Issue body actions

Given the transfer function
G = 1/s^3
We should see a constant negative phase of -270°, which is confirmed in matlab

However, both control.bode and control.matlab.bode give 90° instead of -270°. Although technically the same value, it does get confusing.

import control
import control.matlab
s = control.tf('s')

G = 1/s**3

control.bode(G);

control.matlab.bode(G);

What i've tried so far

• using different transfer functions (1/s^4 or 1/s^2 * 1/(s+10))
• specifying "omega" with a frequency range (as suggested here : https://ch.mathworks.com/matlabcentral/answers/99531-why-is-the-phase-that-is-returned-by-the-bode-function-wrapping-in-the-control-system-toolbox-5-1-r )
• extending the "omega" range

Is it possible to get the "absolute" phase of the transfer function ? If not, i would suggest that control.matlab.bode be corrected as to reflect the behavior of matlab's bode

Thanks for your help !

Activity

1. SebastienDeriaz commented on Nov 10, 2020

   

   SebastienDeriaz

   on Nov 10, 2020

   Author

   More actions

   My (limited) understanding of the function is that bode uses freqresp which uses evalfr and then numpy.angle. Since evalfr returns a complex number it isn't directly possible to get the "absolute" angle so the issue i'm reporting seems to require a dedicated code to determine a "reference phase" of some sort.
   My excuses for my current level of python (or gitHub) doesn't lead me much further than that
2. bnavigator commented on Nov 10, 2020

   

   bnavigator

   on Nov 10, 2020

   Contributor

   More actions

   Thanks, for bringing this up.

   Related issue: #388

   My excuses for my current level of python (or gitHub) doesn't lead me much further than that

   You are nevertheless very welcome to practice your Python and GitHub skills by trying to fix the issue and submitting a pull request 🧑‍🎓
3. ilayn commented on Nov 10, 2020

   

   ilayn

   on Nov 10, 2020

   More actions

   The phase correction is only possible when you have the pole-zero structure and the behavior w->0, so you have to compute the initial phase to get the right amount of shift which is typically done with a decision of adding a minus sign attached to the atan2 computation.

   When you supply a custom frequency grid then it is trickier to snap to the correct phase since then you have to evaluate more stuff to get to that grid from the pole-zero structure (if there is any, say FRD objects don't have it).

   Reacted by Juan Pablo Pierini
4. murrayrm

   added this to the 0.9.x milestone on Dec 26, 2020
5. murrayrm

   self-assigned this

   on Dec 29, 2020
6. murrayrm

   added

   PR pending

   on Jan 2, 2021
7. bnavigator commented on Jan 3, 2021

   

   bnavigator

   on Jan 3, 2021

   Contributor

   More actions

   Resolved in #494. If you think this is not solved, please reopen.
8. bnavigator

   closed this as completedon Jan 3, 2021
9. bnavigator

   removed

   PR pending

   on Jan 3, 2021
10. Juan-Gg commented on Aug 9, 2023

    

    Juan-Gg

    on Aug 9, 2023

    More actions

    It seems to work with poles, but it does not with zeros.

    For a single zero a constant phase of 90º would be expected, it gives -270º instead.

    from matplotlib import pyplot as plt 
    from control.matlab import *  # MATLAB-like functions
    
    %matplotlib widget
    s = tf('s')
    G = (s)
    plt.figure()
    bode(G, wrap_phase = False)
    plt.show()
    

    

Issue actions