In orthogonal frequency division multiple access systems, inter-cell interference (ICI) can be considered as a collision between resource blocks (RBs), which can be reduced by employing a power control strategy at colliding RBs. This paper presents a random neural network (RNN) and a genetic algorithm-based hybrid cognitive engine (CE) architecture to reduce the ICI and achieve the coverage and capacity optimization in a long-term evolution uplink system. The embedded CE within eNodeB learns from the local environment about the effect of ICI on the reliability of communications. Consequently, the CE dynamically selects the optimal transmission power for serving users based on an experienced signal-to-interference-plus-noise ratio and an ICI on a scheduled RB in the subsequent transmission time intervals. The CE also suggests acceptable transmit power to users operating on the same scheduled RB in adjacent cells through the X2 interface (a communication interface between eNodeBs). The RNN features help the CE to acquire long-term learning, fast decision making, and less computational complexity, which are essential for the development and practical deployment of any real-time cognitive communication system. In six different test cases, the simulation results have shown improvements up to 87% in long-term learning and a quick convergence of the RNN as compared with artificial neural network models. Moreover, the gains of 7% in average cell capacity and 118% in system coverage have been achieved as compared with a fractional power control method.
- random neural networks
- neural network models