The random function in the deformable mirror is not functioning properly

Oswald522 commented 1 year ago

When using dm.random(rms), it does not return a consistent surface with the specified RMS value. The source code is located in deformable_mirror.py and needs the following modifications:

# deformable_mirror.py
 def random(self, rms):
        '''Set the dm actuators with random white noise of a certain rms.

        Parameters
        ----------
        rms : scalar
            The dm surface rms.
        '''
        self._actuators = np.random.randn(self._actuators.size) 
     # add this line
        self._actuators *= (rms)/np.std(self.surface))
    # or this line
        self._actuators *= (rms)/np.sqrt(np.mean(self.surface**2))

ehpor commented 1 year ago

While the code does exactly what we intended it to do, I agree that your version is way more useful and I'm totally for changing this. I think the way to do is, is to use a different name for the new function and leave the old function as is (or even deprecate it?). I want to add a few more features to the new function.

Ability to supply weights for the std(). We usually want to take the RMS surface only within the (projected) pupil of the telescope.
Ability to add a relative RMS distribution over actuators. That way, this function is also useful for DMs with global actuator modes (ie. Zernike/sinusoidal modes, which are non-physical but very useful for theoretical simulation work).
Supply a Numpy RNG or seed.

We can also add the second and third points to the existing function, which would be a reason to not deprecate it. So the new function signatures would be:

def random(self, rms, seed=None):
    '''Set the DM actuators with random white noise of a certain RMS.

    Parameters
    -----------
    rms : scalar or ndarray
        The RMS per actuator. If this is a scalar, the same RMS will be used for all actuators.
    seed : None, int, ndarray of ints, SeedSequence, BitGenerator or Generator
        A seed to use for the random noise. If this is a noise generator, this will be used as the source of the noise.
    '''

def random_surface(self, surface_rms, rms_distribution=None, std_weights=None, seed=None):
    '''Set the DM to a random surface with a certain RMS.

    Parameters
    -----------
    surface_rms : scalar
        The requested RMS of the DM surface.
    rms_distribution : ndarray or None

    std_weights : Field or None
        The weights to use for the surface RMS. If this is None (default), then the surface will be weighted equally.
    seed : None, int, ndarray of ints, SeedSequence, BitGenerator or Generator
        A seed to use for the random noise. If this is a noise generator, this will be used as the source of the noise.
    '''

BTW, I don't like normalizing the surface after the fact. This results in a surface that is always exactly the requested RMS, but that wouldn't happen naturally. Naturally, there is a bit of variation in the RMS. So you'd have to precompute the normalization factor in advance, and not post-facto.

ehpor commented 1 year ago

@Oswald522 Do you want to take a first stab at this? You can create a PR and put me as reviewer.

ehpor / hcipy

The random function in the deformable mirror is not functioning properly #203