Interference Prediction in Partially Loaded Cellular Networks Using Asymmetric Cost Functions

The traffic patterns of users in a base station coupled with the sub-frame blanking due to HetNets leads to partially loaded base stations and dynamic interference patterns. Hence, the channel quality indicator information obtained through pilots at a particular time instant would not be reflective of the interference that a user might experience in the immediate future. In this letter, we consider interference prediction in a slot using previous interference values. We observe that conventional Wiener filter based linear prediction that minimizes mean square error performs very badly and hence propose a variant where the weights used for prediction are biased. We also use an asymmetric cost function called Linex and observe through simulations that these two methods perform significantly better than conventional linear prediction. In some cases, the ergodic rate obtained due to the predicted interference is nearly 90% of the actual rate achievable.