Beam Selection via UE Trajectory Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current beam selection methods in 5G and IoT networks face high latency and poor quality of experience due to the high computational burden of traditional single beam selection and GPS-enhanced beam selection approaches, which are ineffective in multi-gNB systems and indoor settings, especially with hundreds of beams and multiple UEs.
Innovation Solution
A beam selection system that proactively selects a beam pattern based on the spatial trajectories of UEs using a policy unit and a beam prediction application deployed at a near Real-Time RIC, leveraging AI and machine learning to predict optimal beam patterns and reduce latency and interference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional single beam selection is used, then beam selection can be implemented, but high computational burden and latency occur
Solution Approach 1:
The system performs preliminary actions by predicting future UE positions and pre-determining beam patterns before actual beam selection is needed. The AI model forecasts UE trajectories and pre-calculates optimal beam patterns based on predicted positions, allowing the system to prepare beam selections in advance rather than computing them in real-time when needed.
Solution Approach 2:
The system creates simplified copies or representations of the complex beam selection problem by using AI models to generate predicted beam patterns that approximate the optimal solution. Instead of performing exhaustive computational searches for the perfect beam, the system uses trained models that produce sufficiently accurate beam predictions much faster than traditional methods.
2Measurement precision
If GPS-enhanced beam selection is used, then beam selection accuracy improves, but it becomes ineffective in indoor settings and multi-gNB systems
Solution Approach 1:
The system enables UEs to self-determine their positions and trajectories using local sensor data (accelerometers, gyroscopes, magnetometers) without relying on external GPS infrastructure. The AI model processes these self-collected measurements to predict UE motion and determine optimal beam patterns, making the system autonomous and independent of external positioning systems.
Solution Approach 2:
The system changes the input parameters from GPS coordinates to local sensor measurements (acceleration, orientation, magnetic field data). By transforming the positioning approach from absolute GPS-based coordinates to relative sensor-based measurements, the system becomes applicable in environments where GPS is unavailable or ineffective.
3Reliability
If beam selection is performed for every supported UE, then service quality improves, but computational burden increases preventing support for multiple UEs
Solution Approach 1:
The system performs preliminary trajectory predictions and beam pattern calculations for multiple UEs simultaneously using the AI model. By forecasting the future states of multiple UEs and pre-determining their beam patterns in advance, the system avoids the need to perform separate complex computations for each UE at the moment of beam selection.
Solution Approach 2:
The AI model creates simplified predictions of UE trajectories and beam patterns that serve as approximate solutions for multiple UEs. These predicted beam patterns act as copies or surrogates for the actual optimal beams, allowing the system to handle multiple UEs with reduced computational overhead while maintaining acceptable service quality.
Data Source
AI summary
A beam selection system that includes a memory operably connected to processing circuitry. The memory configured to store instructions that request call control information from a device; receive a plurality of user equipment (UE) trajectories and parameters from the device; predict with a first artificial intelligence (AI) engine a plurality of beam patterns within a beam index at a plurality of nodes corresponding with a predicted UE location based on the plurality of UE trajectories and parameters of each UE; transmit the plurality of predicted beam patterns to the plurality of nodes corresponding with a corresponding UE of a plurality of UEs; receive a plurality of UE selected beam patterns from the plurality of nodes; train a second AI engine based the plurality of UE selected beam patterns. The processing circuitry is further configured to update the first AI engine based on learned results from the second AI engine.