Neural Network Coverage Prediction for 5G RF Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for coverage prediction in 5G wireless communication networks, such as statistical channel models and ray-tracing, are either inaccurate or excessively costly and time-consuming, particularly for mmWave communication, which requires precise antenna beam tuning and RF parameter optimization.
Innovation Solution
A neural network-based system that identifies regions of interest, determines system performance metrics using data samples including building height, terrain, foliage, clutter, and line-of-sight data, and generates accurate coverage predictions, optimizing RF parameters for improved network performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If statistical channel models are used for coverage prediction, then the prediction process is simple and fast, but the accuracy is poor with pathloss estimation errors as large as 20 dB or more
Solution Approach 1:
The patent introduces ray-tracing data as an intermediary between statistical channel models and ground truth measurements. The neural network learns to map simplified statistical model inputs to accurate coverage predictions by training on ray-tracing data, which captures detailed RF environment interactions without the computational cost of full ray-tracing. This intermediary training dataset enables the network to achieve ray-tracing-level accuracy with statistical model-speed prediction.
2Measurement precision
If ray-tracing technique is used for coverage prediction, then the accuracy is high with detailed azimuth angle, zenith angle, and power information, but the computational cost and time consumption are very expensive
Solution Approach 1:
The patent performs ray-tracing computations in advance to generate training data, then uses this pre-computed data to train a neural network. Once trained, the network can make rapid predictions without performing expensive real-time ray-tracing. The computationally intensive ray-tracing work is done preliminarily during the training phase, enabling fast inference afterward.
Solution Approach 2:
The patent creates a neural network model that copies the predictive capabilities of ray-tracing without replicating its computational process. The network learns the complex RF propagation patterns from ray-tracing data and reproduces accurate coverage predictions through learned weights and biases, avoiding the need to execute actual ray-tracing algorithms during operation.
3Reliability
If RF parameters are tuned during drive test stage, then the coverage data can be collected in real trial area, but the process takes several days and is expensive
Solution Approach 1:
The patent enables the network planning system to perform self-optimization using the trained neural network. Instead of requiring manual drive tests and expert tuning, the system automatically predicts coverage and identifies optimal RF parameters by leveraging the learned patterns from training data. The network serves its own optimization needs without external intervention.
Solution Approach 2:
The patent performs coverage prediction and parameter optimization preliminarily during the planning stage using the trained neural network, before actual deployment. This preliminary optimization identifies the best RF parameters in advance, eliminating or reducing the need for time-consuming post-deployment drive tests and manual tuning.
Data Source
AI summary
A server, method, and computer-readable storage medium for coverage prediction in wireless networks. The server includes a memory storing instructions and a processor operably connected to the memory, which is configured to execute the instructions to cause the server to identify a region of interest (RoI) for the coverage prediction; determine, using a neural network, a set of values for a system performance metric for areas in the RoI, respectively; and generate the coverage prediction for the RoI which associates the areas in the RoI with a determined value in the set of values. The set of values for the system performance metric is determined based on a plurality of data samples for a set of RoIs which include at least one of building height, terrain height, foliage height, clutter data that classifies land cover, line-of-sight indication data, antenna height, and ground truth data for the system performance metric.