Cross-layer AI Power Control for mMIMO SINR Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In wireless network environments, massive multiple-input multiple-output (mMIMO) systems face challenges in power allocation to ensure guaranteed quality of service (QoS) and maximize throughput, particularly due to the complexity and time constraints of optimizing signal-to-noise-plus-interference ratio (SINR), which current centralized methods struggle to address efficiently.
Innovation Solution
A distributed cross-layer power control engine is implemented across radio access network (RAN) layers, utilizing artificial intelligence/machine learning (AI/ML) applications in non-real time and real time RIC layers to optimize spectral and energy efficiency, with a framework that includes policy power control, refined SINR power control guidance, and real-time power adjustment, enabling dynamic SINR management and power allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If centralized power allocation optimization is performed at the local edge, then power allocation can adjust to channel state changes, but computational capacity limitations prevent timely optimization
Solution Approach 1:
The patent segments the power allocation optimization process into two distinct parts: an offline training phase that computes optimal power allocation policies using historical channel state data, and an online execution phase that applies pre-computed policies in real-time. This segmentation transfers the heavy computational burden from the resource-constrained edge device to a more powerful training environment, enabling timely power allocation at the edge without overwhelming computational capacity.
Solution Approach 2:
The patent performs preliminary computation of power allocation policies during an offline training phase using historical channel state information. These pre-computed policies are stored and later applied during online operation when channel states are observed. This preliminary action allows the system to prepare optimal solutions in advance, eliminating the need for real-time optimization at the edge and ensuring timely power allocation decisions.
2Productivity
If more users are paired to increase network capacity, then throughput increases, but inter-user interference increases and degrades customer experience
Solution Approach 1:
The patent implements dynamic user pairing and power allocation that adapts to varying channel conditions. By using historical channel state data to train power allocation policies, the system dynamically adjusts which users are paired together and how much power is allocated to each user based on current conditions. This dynamic approach maximizes network capacity while minimizing inter-user interference, as the system can avoid pairing users when channel conditions would lead to excessive interference.
Solution Approach 2:
The patent changes the parameter being optimized from raw power values to power allocation policies that are functions of channel state information. By representing power allocation as a policy that maps channel states to optimal power values, the system can systematically balance throughput maximization with interference minimization across different user pairing scenarios, adjusting parameters based on observed channel conditions rather than using fixed allocations.
3Reliability
If power allocation optimization is performed in real-time, then QoS can be guaranteed, but computational complexity prevents timely completion
Solution Approach 1:
The patent segments the optimization process into offline policy training and online policy application. The offline phase uses historical data to compute optimal power allocation policies that guarantee QoS, while the online phase simply applies these pre-computed policies when channel states are observed. This segmentation ensures QoS guarantees are built into the policies themselves, eliminating the need for time-consuming real-time optimization while maintaining reliability.
Solution Approach 2:
The patent performs preliminary computation of power allocation policies that embed QoS guarantees during an offline training phase. These pre-computed policies incorporate QoS constraints and are validated beforehand. During online operation, the system only needs to apply these pre-validated policies, ensuring QoS is guaranteed without requiring time-consuming real-time optimization computations.
Data Source
AI summary
The technology described herein is directed towards a distributed cross-layer intelligent power control engine in a communications network architecture that determines power control per user equipment (UE) per access point, using an AI/ML model at each layer. One application (e.g., in a non-real time controller) outputs candidate minimum required signal-to-noise-plus-interference ratio (SINR) policy, and another application (e.g., in a near-real time controller) adjusts the candidate SINR data to provide an environment-aware refined SINR threshold. A third, real time application (e.g., in a real time controller) determines the real time power allocation coefficients per UE per access point based on current conditions such as channel coefficients/parameters and/or UE enrichment information. The distributed cross-layer intelligent power control engine can optimize spectral efficiency and energy efficiency within SINR constraints for a group of UEs based on policy data, and adjust as the network environment changes.