Near-fieldamplitude and phase measurements are routinely used to characterize microwave and millimeter-wave antennas. At frequencies above about 100GHz the near-field phase become more and more dificult to measure accurately. Phase retrieval research that began in the mid 1980´s has strived for over two decades to demonstrate its potential in antenna near-field measurements. We believe that the advent of GPUs is about to change how phase-retrieval is regarded in terms of accuracy and verification.
More details can be found in this paper: Junkin, G. "Planar Near-Field Phase Retrieval Using GPUs for Accurate THz Far-Field Prediction", IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 61, NO. 4, APRIL 2013, pp 1763
The following gives a small example of what is now possible.
This is an example of an unblocked 0.5m diameter antenna at 186GHz that has been engraved with text.
The aperture has been rotated by 2 degrees in azimuth. The aperture also contains some low-order surface deformations, including astigmatism and defocus.
The video above shows the screen output from an nVidia GTX480 as it retrieves the phase of the aperture wave distribution.
The algorithm is based on Anderson's PTP http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=04642828
combined with Successive Over-Relaxation (SOR) . http://en.wikipedia.org/wiki/Successive_over-relaxation
The first plane of intensity data is located at a distance of one aperture radius. The separation between planes is only 93 mm, or about 0.0025 times the Rayleigh Range.
As can be appreciated from the following images, the near-field intensity data is only slightly different at each plane. Because of this, receiver drift must be controlled pretty precisely and numerical simulations suggest that about 0.5% gain variation is acceptable.