Andersen et al. propose a rather elegant method to sidestep these difficulties: perform the processing optically, before the wavefront is even recorded by a computer. Their solution uses a hologram to project an array of focused spots onto an array of pinholes. At first, this might seem to resemble the well-known Shack-Hartmann wavefront sensor that reconstructs the shape of a wavefront by measuring its tilt at a number of different points. However, their sensor effectively measures the curvature of the wavefront at each point.
The first advantage of this approach is that it is easier to read out---whereas a Shack-Hartmann sensor requires us to determine the shift in position of each spot, the readout of this holographic sensor is simply comparing the intensity at two points. Secondly, by aligning the points at which the curvature is measured with the actuators on the mirror, the relationship between detector signal and applied correction is simplified enormously. In a closed-loop configuration (i.e. the wavefront that is sensed is the one that has been corrected), one simply needs to determine whether each actuator ought to be moved forward or back and by approximately how much---any small deviations from linearity in the sensor will be cancelled out by the closed-loop control.
Implementing this system requires an array of photodiodes behind the pinholes, and a microcontroller to read out the photodiodes, feed the measurement into the control loop and output a drive voltage to each actuator on the mirror. With no complicated mirror response functions to deal with, and no images to analyze, the control loop has a latency of 10μs. Consequently, the system's bandwidth is now limited by the intertia of the deformable mirror's membrane to an impressive 10kHz---a figure which the team reckon could be improved at least tenfold with a faster microprocessor and mirror.
While the implementation using a 32-actuator mirror is very promising, one of the major advantages of the holographic adaptive optic system here is that it scales well to more sophisticated corrective elements. While most algorithms would take twice or four times as long to compute drive voltages for a 64-actuator mirror, the processing here can all be performed in parallel to keep the speed exactly the same.
Another intriguing possibility for this technique is the replacement of their optically-recorded hologram with a liquid crystal Spatial Light Modulator. By replacing optical holograms with computer-generated ones, we gain extra control over the system, potentially allowing even more of what would usually be computationally intensive work to be done by the light itself.
Overall, this paper has proved an exciting concept; we can use holograms to simplify a complicated control algorithm into something that can be performed in real time by simple hardware. I hope we can look forward not only to faster adaptive optics systems, but to similar approaches being applied elsewhere to speed up normally computationally intensive image processing tasks.
You must log in to add comments.