Core Concepts
A deep and transfer learning-based algorithm that accurately predicts the optimal target cell for handover in 6G and beyond networks, enabling dynamic optimization of handover decisions and seamless integration of new network elements like UAV base stations.
Abstract
The content discusses a deep and transfer learning-based algorithm for predicting the optimal target cell for handover in 6G and beyond cellular networks. The key highlights are:
The algorithm is designed to address the challenges of increased handovers due to smaller cell coverage and higher signal attenuation in next-generation networks. It utilizes sequential user equipment (UE) measurements, such as Reference Signal Received Power (RSRP) and Signal-to-Interference-plus-Noise Ratio (SINR), to predict the future serving cell.
The algorithm is compliant with the Open Radio Access Network (O-RAN) architecture and can be deployed as an xApp in a Near-Real-Time RAN Intelligent Controller (near-RT RIC) for real-time handover management decisions.
The algorithm incorporates two types of dynamic events that can occur in the network: (1) dynamic changes in handover decision parameters, such as incorporating load balancing and energy efficiency considerations, and (2) the dynamic introduction of new Unmanned Aerial Vehicle (UAV) base stations for enhanced coverage and capacity.
To address these dynamic events, the algorithm employs a deep learning-based architecture with an encoder, stacked Long Short-Term Memory (S-LSTM) layers, and a decoder. It also utilizes transfer learning techniques to quickly adapt to new decision rules or the introduction of new UAV base stations, reducing the retraining time by up to 91% and 77%, respectively.
The evaluation results demonstrate that the proposed algorithm achieves a 92% accuracy in predicting the future serving cell, outperforming classical supervised machine learning algorithms like Random Forest, Multi-Layer Perceptron, and Gated Recurrent Unit.
Stats
The algorithm was evaluated using a simulated cellular network with the following key parameters:
Cellular network configuration: Urban Macrocell (UMa) with 7 macrocells in a hexagonal layout and an intersite distance of 1000 m.
Transmission specifications: Carrier frequency of 4 GHz (Macro) and 5 GHz (UAV), bandwidth of 20 MHz, and transmit power of 35 dBm (Macro) and 20 dBm (UAV).
Handover parameters: Handover Margin (HOM) of 3 dB and Time-To-Trigger (TTT) of 100 ms.
Path loss models: NLOS for macrocells and LOS for UAVs.
UE parameters: Distributed UEs with a full buffer traffic model, height of 1.5 m, velocity of 60 km/h, and sensitivity of -110 dBm.
Quotes
"Our algorithm enables network operators to dynamically adjust handover triggering events or incorporate UAV base stations for enhanced coverage and capacity, optimizing network objectives like load balancing and energy efficiency through transfer learning techniques."
"We demonstrate that our framework can predict the target cell with 92% accuracy using a time window W of 70 measurements. Finally, we show that it reduces the retraining time by 91% when dynamic load balancing handover decision rules are incorporated and by 77% when new UAV base stations are introduced, respectively."