Deep Reinforcement Learning for Wireless Network Cell Parameter Optimization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional wireless network optimization methods face challenges in achieving lower optimization costs, shorter optimization times, and higher accuracy as networks become more complex, requiring significant resources and time, and are not adaptable for one-shot optimization across different networks.

Innovation Solution

The implementation of deep reinforcement learning (DRL) techniques to train neural networks for selecting actions to adjust cell parameters in wireless networks, using experience tuples and expert experience to improve training efficiency and accuracy, allowing for one-shot optimization without the need for extensive site visits or accurate engineer parameters.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional wireless network optimization methods are used, then optimization can be performed with existing techniques, but optimization time and cost increase significantly as networks become more complex

Engineering Contradiction:
Improveoptimization accuracyVSAvoidoptimization time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent replaces conventional mechanical optimization methods with deep reinforcement learning (DRL) techniques. The DRL agent learns optimal network configurations through trial and error in a simulated environment, substituting traditional manual or algorithmic optimization approaches with intelligent agents that can adapt to network complexity without proportionally increasing optimization time

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent performs preliminary training of DRL agents in simulated environments before deploying them to real networks. Experience tuples are generated and stored in advance, allowing the system to have pre-learning optimization knowledge ready before actual network optimization is needed, thus reducing real-world optimization time

Inventive Principle:
Principle #10Preliminary action

2Reliability

If conventional optimization methods are used, then existing tools can be utilized, but optimization cost increases with network complexity

Engineering Contradiction:
Improveoptimization accuracyVSAvoidoptimization cost
Core Design Contradiction:
ReliabilityVSLoss of energy

Solution Approach 1:

The patent substitutes expensive conventional optimization tools and processes with DRL-based optimization. The DRL agent requires minimal human intervention and can operate autonomously, reducing labor costs and tool licensing fees associated with traditional optimization methods while maintaining or improving accuracy

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent creates virtual copies of network environments for training DRL agents. By using simulated network copies instead of real networks during the training phase, the system avoids the high costs of trial-and-error optimization in production environments, while the learned policies can be applied to real networks

Inventive Principle:
Principle #26Copying

3Adaptability or versatility

If conventional optimization methods are used, then standard procedures can be followed, but adaptability to different networks is limited

Engineering Contradiction:
Improvenetwork adaptabilityVSAvoidmethod complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent implements dynamic adaptation through DRL agents that continuously learn from experience tuples generated during training and operation. The agents can adapt to different network configurations, topologies, and conditions by learning new policies, making the optimization method highly versatile across various network types without requiring manual reconfiguration

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent develops universal DRL agents that can optimize multiple types of wireless networks (LTE, 5G, etc.) and handle various optimization scenarios (coverage, capacity, energy efficiency) using the same underlying framework. The experience tuple structure and training methodology are generalizable across different network contexts

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Loss of time

If one-shot optimization is implemented using DRL, then optimization time is reduced, but training data requirements increase

Engineering Contradiction:
Improveoptimization timeVSAvoidtraining data volume
Core Design Contradiction:
Loss of timeVSQuantity of substance

Solution Approach 1:

The patent performs preliminary generation of experience tuples in simulated environments before actual optimization is needed. By pre-training DRL agents with synthetic training data that mimics real network conditions, the system reduces the need for extensive real-world data collection while enabling rapid one-shot optimization deployment

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses virtual copies of network environments to generate training data. Instead of collecting massive amounts of real network data, the system creates synthetic experience tuples from simulated networks, reducing data acquisition costs and enabling efficient training without proportionally increasing data volume requirements

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3635505B1System and method for deep learning and wireless network optimization using deep learning
Publication Date: 2023.09.06 HUAWEI TECH CO LTD
  • EP3635505B1 patent drawingFigure 1~2
  • EP3635505B1 patent drawingFigure 3~5
  • EP3635505B1 patent drawingFigure 6

AI summary

A neural network is trained using deep reinforcement learning (DRL) techniques for adjusting cell parameters of a wireless network by generating a plurality of experience tuples, and updating the neural network based on the generated experience tuples. The trained neural network may be used to select actions to adjust the cell parameters. Each experience tuple includes a cell identifier, a first state, a second state, an action applied to the cell that moves the cell from the first state to the second state, a local reward, and a global reward. The neural network is updated based on whether or not each action is acceptable, which is determined based on the global reward and the local reward associated with each action.