I just released a new paper onto the arXiv entitled "Improving the fidelity of single photon preparation from conditional down-conversion via asymmetric multiport detection". Quite a mouthful you say. Indeed. In this paper I look at the conditional preparation of single photons via parametric down-conversion. This is a very common technique for preparing single photons for use in quantum optics experiments. The idea is that the down-converter probabilistically spits out pairs of photons, where each photon in the pair goes a different direction. Then, if we measure one of the outputs and detect a photon, we know with highly probability that there was a photon in the other output, due to the fact that they always pair up.
To operate well this requires number resolving photo-detectors. That is, detectors that can tell you exactly how many photons were detected. The reason for this is that the down-converter can spit out multiple pairs simultaneously and we want to filter out the cases where there was exactly one pair. Unfortunately present day detectors can't do this. They can only tell you if you have no photons or some photons, but they can't discriminate between different numbers of photons. To remedy this several authors have suggested approaches for improvement. The most elementary is the 2-port device. This consists of a simple beamsplitter, a device that sends half your photons in one direction, and the other half in the other. Both of the two outputs are then measured using normal non-number resolving detectors. Now, suppose two photons were incident. Then there is a 50% probability that the photons will go to different outputs. Then the sum of the two measurement results (i.e. a total of two detection events) gives you the correct result. Note that without the 2-port, 100% of the time you'll get the wrong answer. So we've reduced the error probability. This idea generalizes to an N-port. Then the probability of getting a wrong answer goes as 1/N, which approaches zero for large N. So this trick effectively allows us to approximate a number resolving detector using a sequence of non-number resolving detectors and an N-port.
In this paper I describe a slight variation on the simple 2-port idea and show that by using an asymmetric beamsplitter instead of a half-half beamsplitter, you can significantly improve the operation of this scheme.