http://www.ssicentral.com/lisrel/techdocs/IPUG.pdf

If you focus a beam you destroy the coherence of the beam, because of the angular distribution in the illuminated area. Hence the coherence in a hologram is always worse if a beam is focused. If you make a parallel beam the illuminated area becomes big and with a limited brightness the exposure time becomes to big. How to overcome the problem is to make a elliptic illumination. In an elliptic illumination you can increase the intensity by focusing in one direction.(short direction of the elliptic illumination)  You are using the one perpendicular to the fringe spacing of the biprism in an TEM. If you align the long axis of the elliptic illumination perfectly, the coherence in this direction is not destroyed since the beam in this direction is still parallel. Hence you have best of both worlds, good intensity on the detector and good coherence in the direction where it counts for the application.

Thank you Dr. Bert

In case it is still relevant, this paper might t be helpful:

http://www.sciencedirect.com/science/article/pii/030439919390130P

This could help too (page 11 and ref.s): http://www.ncbi.nlm.nih.gov/pubmed/12533207