HetGNN Routing Link Scheduling for Heterogeneous IoT Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing routing technologies are inadequate for heterogeneous wireless IoT networks, which consist of both current-generation and next-generation devices with varying communication capabilities, leading to complex interference and inefficiencies in data transmission.
Innovation Solution
A machine learning-based routing link scheduling scheme using a heterogeneous graph neural network (HetGNN) to optimize data transmission by formulating the routing link scheduling problem as a combinatorial optimization problem, solved using a primal-dual approach with zero duality gap, to avoid interference and maximize network throughput.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If routing link scheduling is implemented in heterogeneous wireless IoT networks, then network efficiency and throughput are improved, but device complexity and algorithm complexity increase
Solution Approach 1:
The patent transforms the routing link scheduling problem from a complex combinatorial optimization problem into a parameterized form suitable for machine learning. By parameterizing the scheduling decisions and using neural network parameters to represent scheduling policies, the system achieves optimal scheduling without solving complex optimization problems in real-time, thus improving throughput while managing complexity.
Solution Approach 2:
The patent replaces traditional mechanical optimization algorithms (primal-dual methods) with a machine learning-based neural network system. Instead of solving optimization problems through iterative mathematical computations, the system uses trained neural networks to directly predict optimal scheduling decisions, significantly reducing computational complexity while maintaining high throughput.
2Loss of time
If machine learning based routing link scheduling is used, then interference and channel access delay are reduced, but computational resources and training time increase
Solution Approach 1:
The patent performs scheduling policy optimization in advance through offline training of the neural network. The system pre-computes optimal scheduling decisions for various network scenarios during the training phase, storing these as learned parameters. During actual network operation, the system only needs to infer decisions from pre-trained models, eliminating real-time computational overhead and reducing channel access delay.
3Adaptability or versatility
If heterogeneous network support is implemented, then network versatility and functionality are improved, but routing problem complexity increases
Solution Approach 1:
The patent designs a universal neural network-based scheduling framework that handles multiple types of heterogeneous devices (single-link and multi-link nodes) through a single unified model. The neural network is trained to recognize and appropriately schedule different device types using the same architecture, eliminating the need for separate routing algorithms for each device category and simplifying the overall system.
Data Source
AI summary
A heterogeneous multi-hop wireless network is provided. The network consists of single-link data nodes, multi-link data nodes and data centers. The single-link data node supports one communication interface, the multi-link data node supports two communication interfaces and the data center is considered as multi-link node. The routes from all data nodes to data centers are given. The network is configured to schedule data transmissions from all data nodes to data centers without transmission interference and channel access delay. The routing link scheduling problem is formulated as an optimization problem with constraints. Due to the NP-Hard complexity of the formulated optimization problem, the scheduling policies are then parameterized for the application of graph neural network techniques. The parameterized optimization problem is solved using primal-dual approach with zero duality gap. A heterogeneous graph neural network (HetGNN) algorithm is provided to train the primal-dual problems.