AI Channel Estimation for Varying PRB Set Sizes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional AI/ML-based channel estimation methods are limited by fixed input and output layer sizes that match specific channel estimation problem sizes, making them ineffective for varying PRB or mini-slot sizes, as they require retraining for each new size.
Innovation Solution
A method using a single neural network (NN) or a bank of NNs to split and combine channel estimation inputs and outputs across varying PRB or mini-slot sizes, employing pre-processing and post-processing units to adapt to different dimensions, allowing a single NN architecture to handle multiple sizes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a single NN architecture with fixed input and output layer sizes is used, then the NN can be trained efficiently for a specific channel estimation problem size, but it cannot address channel estimation problems with different sizes (varying PRB set sizes, number of antennas)
Solution Approach 1:
The patent applies segmentation by dividing the channel estimation problem into multiple independent sub-problems, each corresponding to a specific PRB set size. Multiple NN architectures are trained separately for different problem sizes, allowing each NN to be efficiently trained for its specific size while the system as a whole can handle varying sizes by selecting the appropriate NN.
Solution Approach 2:
The patent implements universality through a framework that manages multiple NN architectures, where each NN is specialized for a particular problem size. The system becomes universal by accommodating different channel estimation problem sizes through the collection of specialized NNs, allowing the same system to serve multiple problem size requirements.
2Adaptability or versatility
If multiple NN architectures are trained for different channel estimation problem sizes, then the system can handle varying PRB set sizes and antenna configurations, but the system complexity and training resources increase
Solution Approach 1:
The patent applies dynamics by making the NN architecture selection dynamic based on the input problem size. The system dynamically chooses which NN to use depending on the specific channel estimation problem parameters (PRB set size, number of antennas), allowing the system to adapt its complexity to match the requirements of each specific task.
Solution Approach 2:
The patent implements parameter changes by varying the NN architecture parameters (input layer size, output layer size) to match the specific problem size requirements. Each NN is configured with parameters that correspond to its designated problem size, allowing the system to handle different sizes by changing which NN instance is activated rather than using a single fixed architecture.
Data Source
AI summary
One example method includes receiving, by a pre-processing unit, a rough channel estimated array, splitting, by the pre-processing unit, the rough channel estimated array into a group of smaller arrays that each have a size that is smaller than a size of the rough channel estimated array, providing, by the pre-processing unit, the smaller arrays to a single NN (neural network), processing, by the NN, the smaller arrays to generate respective refined channel estimation outputs for each of the smaller arrays, and combining, by a post-processing unit, the smaller arrays having the respective refined channel estimation outputs to generate an output array with a size that is the same as the size of the rough channel estimated array.