Pilot contamination is a problem is comes into picture when channel coefficients are estimated in one cell but this channel estimation is not properly done with the intervention of UE of an other close cell . This problem is addressed in the

https://uts.cc.utexas.edu/~jjose/files/Pilot_Contamination_ISIT_2009.pdf

A method which is normally used is to reduce the effect of pilot contamination is uplink training of non orthogonal sequences.

The simplest solution is to introduce a reuse pattern for the pilot resources, such that adjacent cells are using orthogonal pilot sequences. The data transmission can still be performed with full frequency reuse, but the pilot transmission can be protected in this way. This is for example done in the following paper:

http://dx.doi.org/10.1109/OnlineGreenCom.2013.6731024

Instead of avoiding pilot contamination, you can try to reduce it by scheduling and signal processing.

One way is to exploit statistical information (allocate the same pilot sequence to users that are very different from a statistically point of view so you can still tell them apart). This is done in the following paper:

http://arxiv.org/abs/1203.5924

Another way is to use both data and pilot sequences when estimating the channels, where the data sequences basically make the pilot sequences longer and different between the users. This is done in the following paper:

http://arxiv.org/abs/1309.6806