On stream selection for interference alignment with limited feedback in heterogeneous networks

Abstract

This paper presents a stream selection based interference alignment approach with imperfect channel state information for heterogeneous networks. The proposed solution constructs stream sequences by selecting only the strongest stream of each user where the first stream of the constructed stream sequences is associated to a pico user. While selecting the streams, the channel matrices of the unselected streams are projected orthogonally to the virtual transmit and receive channels of the selected stream in order to align the interference in the null space of these virtual channels. In addition, the influence of imperfect channel state information on the proposed algorithm is analysed. A bit allocation scheme is given by deriving an upper bound on the rate loss because of quantisation. The simulation results are carried out by considering various scenarios with different locations of pico cells at the cell edge regions of the macro cell. The performance results show that the proposed algorithm with the imperfect channel state information achieves higher performance than the existing algorithms.

Deep learning based adaptive bit allocation for heterogeneous interference channels

“…The number of receive antennas is 2 at each UE. Simulation parameters can be found in the study of Aycan Beyazit et al (2016).…”

“…In order to find a local optimum, the optimization problems are handled for each cell. In the study of Aycan Beyazit et al (2016), a two-step solution is studied for the adaptive bit allocation scheme in heterogeneous networks. First the total number of feedback bits is shared to each cell considering the transmit powers and the interference levels, then the shared bits are adaptively and locally allocated to each channel in the considered cell.…”

“…Another property of the training data set of the handled problem is that both the input and output variables are continuous. Therefore, a regression algorithm is trained with a higher number of data obtained with the conventional bit allocation (CBA) scheme as in Aycan Beyazit et al (2016) for several different scenarios. In this study, a fully connected deep neural network (FC-DNN) is used as a regression method since there are multiple number of output variables due to the nature of the MIMO systems.…”