PDCCH Resource Allocation Using ML for CCE and Power Trade-Offs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current PDCCH resource allocation methods in 5G NR networks face challenges in optimizing CCE-AL assignment and power allocation due to computational complexity, leading to suboptimal performance in terms of PDCCH capacity and coverage, as they often treat scheduling entities individually without a full view of the system.
Innovation Solution
A two-step approach involving an optimization framework to maximize the number of accommodated scheduling entities per slot while minimizing CCE consumption, followed by a machine learning technique trained on a labeled training set to predict CCE assignments and power allocation, using integer linear programming to solve offline optimization problems and leveraging UE location information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If higher CCE aggregation levels are used, then PDCCH coverage is improved, but PDCCH capacity is reduced due to earlier exhaustion of CCE resources
Solution Approach 1:
The system dynamically changes the CCE aggregation level parameter based on UE location and channel conditions. Machine learning models predict the optimal aggregation level by analyzing historical data and current network state, allowing the system to adapt the CCE-AL parameter to maximize both coverage and capacity under varying conditions.
Solution Approach 2:
The CCE aggregation level assignment transitions from static to dynamic. The system continuously adjusts aggregation levels based on real-time feedback from machine learning models that consider current network load, UE positions, and channel quality, enabling flexible optimization of the trade-off between coverage and capacity.
2Reliability
If higher transmit power per CCE is used, then PDCCH coverage is improved by improving channel estimation accuracy, but total power resources are exhausted earlier
Solution Approach 1:
The system dynamically adjusts the transmit power parameter for PDCCH based on predicted channel conditions and UE locations. Machine learning models optimize power allocation by analyzing patterns in channel quality indicators and network state, enabling the system to use higher power only when and where necessary to maintain coverage while conserving total power resources.
3Reliability
If joint optimization of CCE-AL assignment and power allocation is performed, then both PDCCH capacity and coverage are maximized, but computational complexity increases
Solution Approach 1:
The system performs preliminary optimization offline by collecting historical network data and training machine learning models before deployment. This pre-computation of optimal policies based on integer linear programming formulations allows the system to make fast online decisions without real-time computational burden, resolving the contradiction between optimization performance and computational complexity.
Solution Approach 2:
Machine learning models serve as intermediaries between the complex joint optimization problem and real-time decision making. The models translate intricate optimization calculations into simple prediction rules that can be executed quickly in online operations, enabling joint optimization of CCE-AL and power allocation without direct computational burden during runtime.
4Reliability
If a full view of all scheduling entities is obtained, then global optimal solution can be found, but implementation complexity in user plane control design increases
Solution Approach 1:
The system creates a simplified virtual representation (copy) of the network state including all scheduling entities, their locations, and channel conditions. Machine learning models operate on this virtual copy to determine global optimization policies, which are then applied to the actual network without requiring complex real-time coordination of all entities.
Data Source
AI summary
According to some embodiments, a method is performed by a network node for physical downlink control channel (PDCCH) resource allocation. The method includes obtaining a data set representing a plurality of scheduling entities (SEs). Each of the SEs is associated with a signal quality, a priority, and/or a downlink control information (DCI) size. The method further includes determining a number of control channel elements (CCEs) and power allocation for the CCEs for each of the SEs based on the signal quality, the priority, and/or the DCI size associated with each of the SEs and a total power available, a power boosting threshold, and/or a total number of CCEs available. The method further includes generating a machine learning training set for online CCE and power allocation based on the determined number of CCEs and power allocation for the CCEs for each of the SEs.