DDPG Power Control for Wireless Interference Coordination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current power control and interference coordination methods in wireless communication systems, such as the weighted minimum mean square error (WMMSE) algorithm, are too complex for real-world implementation, necessitating solutions with lower latency and reduced computation power.
Innovation Solution
A user-centric power control and interference coordination scheme based on deep deterministic policy gradient (DDPG) is employed, which does not require training datasets and uses four neural networks to precisely determine power allocation, thereby addressing the complexity and latency issues of existing methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the weighted minimum mean square error (WMMSE) algorithm is used for power control and interference coordination, then the optimal power allocation is achieved, but the computational complexity and latency increase significantly
Solution Approach 1:
The patent replaces the traditional iterative mathematical optimization algorithm (WMMSE) with a deep neural network-based reinforcement learning approach. The neural network learns the optimal power allocation policy through training, substituting the complex iterative computation with a trained model that provides near-optimal solutions with significantly reduced computational complexity and latency.
Solution Approach 2:
The patent employs a preliminary training phase where the neural network is trained offline using simulated network data to learn optimal power allocation strategies. This preliminary action allows the system to store learned knowledge in the neural network weights, enabling fast real-time power control without requiring complex online computation, thus resolving the contradiction between accuracy and computational complexity.
2Reliability
If the WMMSE algorithm is used for power control, then the interference coordination is optimized, but the latency increases making it unsuitable for real-time networks
Solution Approach 1:
The patent substitutes the iterative WMMSE algorithm with a trained neural network that provides near-real-time power allocation decisions. The neural network, after offline training, can rapidly infer optimal power control parameters from current network state inputs, dramatically reducing latency while maintaining interference coordination performance close to WMMSE.
Solution Approach 2:
The patent performs preliminary training of the neural network offline using extensive simulated network scenarios. This preliminary action pre-computes the optimal power allocation strategies for various network conditions, allowing the system to quickly adapt to real-time changes without performing complex iterative calculations, thus achieving low latency while maintaining reliability.
3Measurement precision
If the WMMSE algorithm is used for power allocation, then the optimal solution is obtained, but the computation power required is too high for practical deployment
Solution Approach 1:
The patent replaces the computationally intensive WMMSE algorithm with a neural network-based approach. The neural network, trained offline using computational resources, shifts the computation burden from real-time operation to the training phase. During actual power control operations, the trained network requires minimal computation power, achieving near-optimal power allocation with significantly reduced real-time computational requirements.
Solution Approach 2:
The patent performs the computationally heavy work of learning optimal power allocation strategies during an offline training phase using simulated data. This preliminary action allows the system to store the learned knowledge in the neural network parameters, enabling practical deployment with limited real-time computation power while maintaining power allocation optimality close to WMMSE.
Data Source
AI summary
A method performed by a UE may include: receiving a pilot signal from a first number of first BSs; generating a serving BS matrix, wherein the serving BS matrix indicates that the UE accesses a second number of first BSs among the first number of first BSs; measuring CSI between the UE and each of the first number of first BSs; generating a CSI matrix based on the measured CSI between the UE and the first number of first BSs; encoding the serving BS matrix and the CSI matrix; and transmitting the encoded serving BS matrix and the encoded CSI matrix to one of the second number of first BSs.