A Predictive and Mitigative Modelling Framework for Intercell Interference in Heterogeneous Networks Using Enhanced Feedforward Neural Networks

Abstract

The rapid densification of cellular systems into heterogeneous networks (HetNets), comprising macrocells, picocells, and other small cells, has improved spectrum efficiency but also intensifies intercell interference (ICI), particularly at cell boundaries. This interference degrades signal quality, reduces throughput, and limits quality of service (QoS), making its management a critical research challenge. To address this issue, this study develops an enhanced Feedforward Neural Network (eFFNN) framework for predictive and mitigative ICI control in LTE HetNets. Empirical drive-test data including RSRP, SINR, throughput, latency, and packet loss were collected, preprocessed, and used to train the model. The eFFNN consists of three hidden layers with dropout and batch normalization, ReLU activation, the Adam optimizer, and a dynamic weighted binary cross-entropy loss function that prioritizes false-negative reduction in interference detection. Simulation results show significant improvements over the conventional FFNN: packet loss was reduced to 0.16% (compared to 0.75% in the baseline) at –102.67 dBm interference, and decision confidence increased to 0.91 (versus 0.8). Under severe interference (–103.78 dBm), the eFFNN maintained packet loss at 2.31%, outperforming the ordinary FFNN at 3.15%. These findings highlight the effectiveness of the proposed predictive–mitigative approach in enhancing QoS and interference resilience in heterogeneous LTE networks.