Cell Cluster Power Control Using ML-Based DL Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for optimizing downlink (DL) transmit powers in wireless communication networks face challenges due to high complexity, uncertainty, and costs associated with 3PP tools and KPI analysis, which fail to consider effects on neighboring cells and uplink coverage, leading to inefficient power management.
Innovation Solution
A computer-implemented method utilizing a machine learning (ML) model, specifically deep reinforcement learning, to emulate and optimize DL transmit powers based on deployment and measurement data, enabling efficient training and offline optimization of DL powers across a cluster of cells.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3PP tools are used to estimate the impact of changes in DL transmit powers, then propagation models can assess the effects, but the costs and complexity increase considerably
Solution Approach 1:
The patent uses ray tracing simulations to create virtual copies of radio wave propagation paths through the environment. Instead of relying on complex 3PP tools with extensive propagation model tuning, the system generates synthetic measurement data by simulating signal paths, reflections, and blockages in a virtual model of the network, thereby obtaining accurate propagation characteristics without the high costs and complexity of traditional 3PP tools
Solution Approach 2:
The patent replaces the mechanical/mathematical propagation models with machine learning models trained on ray tracing simulation data. The ML models learn propagation patterns from the simulated data and can predict signal characteristics without requiring complex mathematical modeling, thereby substituting the traditional propagation model approach with a data-driven alternative that is both accurate and computationally efficient
2Reliability
If Monte Carlo simulations are used to model user distribution and traffic profiles, then performance assessment is possible, but significant uncertainty is introduced
Solution Approach 1:
The patent performs ray tracing simulations in advance to generate comprehensive training data that captures various propagation scenarios, user distributions, and traffic patterns. This preliminary generation of realistic simulation data allows the subsequent ML models to be trained on accurate representations of network behavior, eliminating the need for uncertain Monte Carlo assumptions during actual performance assessment
Solution Approach 2:
The system uses itself to generate the training data it needs through automated ray tracing simulations. Rather than relying on external propagation models or assumptions about user distribution, the system autonomously generates realistic propagation scenarios and uses these to train its own ML models, creating a self-sufficient framework that avoids external sources of uncertainty
3Productivity
If KPI analysis with dynamic thresholds is used to optimize DL transmit powers, then performance optimization is possible, but the complexity and manual intervention requirements increase
Solution Approach 1:
The patent implements a closed-loop feedback system where ML models continuously predict the impact of power changes on network performance based on current state information. The system monitors actual performance metrics, compares them with predictions, and uses this feedback to iteratively optimize power settings. This automated feedback mechanism replaces complex manual KPI analysis with an intelligent system that learns from ongoing network operations and automatically adapts to changing conditions
Solution Approach 2:
The system transforms the complex multi-parameter optimization problem into a learning problem by changing the approach from rule-based threshold analysis to ML-based prediction. The ML models learn optimal power settings by analyzing patterns in the data rather than relying on predefined thresholds, thereby simplifying the optimization process while maintaining or improving performance
4Loss of energy
If DL transmit powers are optimized to reduce interference and save energy, then overall network performance improves, but the complexity of coordinating multiple cells increases
Solution Approach 1:
The patent merges the optimization problems of multiple individual cells into a single unified ML model that considers the entire network simultaneously. Rather than having each cell optimized independently or through complex iterative coordination protocols, the system trains a comprehensive model on network-wide ray tracing data that captures inter-cell interference and cooperation effects, allowing centralized optimization that naturally coordinates all cells while reducing overall energy consumption
Data Source
AI summary
A computer-implemented method for optimization of downlink (DL) transmit powers in a wireless communication network includes acquiring deployment data describing a deployment of a cluster of cells of the wireless communication network. Further, the method includes acquiring measurement data representing measurements performed with respect to a plurality of connections established between wireless devices and the wireless communication network. Further, the method includes, based on the deployment data and the measurement data, emulating effects of applying different DL transmit powers in at least some cells on the plurality of connections. Further, the method includes estimating state information for each of the cells based on the emulated effects. Further, the method includes, based on the estimated state information, training a machine learning (ML) model for optimization of the DL transmit powers of the cells.