Data-Driven Radio Environment Map Estimation Using Graph Neural Networks

Ali Shibli,Tahar Zanouda
2024-06-09
Abstract:Radio Environment Maps (REMs) are crucial for numerous applications in Telecom. The construction of accurate Radio Environment Maps (REMs) has become an important and challenging topic in recent decades. In this paper, we present a method to estimate REMs using Graph Neural Networks. This approach utilizes both physical cell information and sparse geo-located signal strength measurements to estimate REMs. The method first divides and encodes mobile network coverage areas into a graph. Then, it inputs sparse geo-located signal strength measurements, characterized by Reference Signal Received Power (RSRP) and Reference Signal Received Quality (RSRQ) metrics, into a Graph Neural Network Model to estimate REMs. The proposed architecture inherits the advantages of a Graph Neural Network to capture the spatial dependencies of network-wide coverage in contrast with network Radio Access Network node locations and spatial proximity of known measurements.
Networking and Internet Architecture,Machine Learning,Signal Processing
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to construct accurate Radio Environment Maps (REMs). REMs are crucial for many applications in the telecommunications field, such as network configuration and parameter optimization, spectrum estimation, coverage optimization, finding the optimal location for new base stations, and proactive resource management. However, traditional methods face challenges in dealing with data sparsity and spatial dependence, and usually require domain knowledge (such as calculating path loss), which limits their use in practical application scenarios. To address these issues, this paper proposes a method based on Graph Neural Networks (GNNs) to estimate REMs. Specifically, this method utilizes physical cell information and sparse geolocated signal strength measurements (characterized by the Reference Signal Received Power (RSRP) and Reference Signal Received Quality (RSRQ) indicators) to estimate REMs. The following are the main contributions of this method: 1. **Model Architecture**: A Graph Convolution Networks (GCNs) architecture suitable for REM estimation is proposed, and this architecture can compete with other solutions in terms of performance. 2. **Comparative Analysis**: An exhaustive comparative evaluation of the proposed GNN model with current machine - learning models (such as XGBoost, Fully - Connected Networks (FCN), and TabNet) is carried out. 3. **Data Fusion**: Multiple types of data are integrated, including physical cell information, geolocated signal strength measurements, and geographical information. Through these improvements, this method can not only capture the spatial dependencies within the network coverage more accurately, but also has good generalization ability between different geographical regions. ### Formula Summary - **Accuracy in Classification Tasks**: \[ \text{Accuracy}=\frac{\text{TP}+\text{TN}}{\text{TP}+\text{TN}+\text{FP}+\text{FN}} \] where \(\text{TP}\) represents True Positives, \(\text{TN}\) represents True Negatives, \(\text{FP}\) represents False Positives, and \(\text{FN}\) represents False Negatives. - **\(R^{2}\) Score in Regression Tasks**: \[ R^{2}=1 - \frac{\sum_{i = 1}^{n}(y_{i}-\hat{y}_{i})^{2}}{\sum_{i = 1}^{n}(y_{i}-\bar{y})^{2}} \] where \(y_{i}\) is the true value, \(\hat{y}_{i}\) is the predicted value, and \(\bar{y}\) is the average of the true values. These formulas are used to evaluate the performance of the model in classification and regression tasks.