NW-First AI Decoder Training With Raw CSI Feedback Sharing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current CSI compression techniques face challenges in achieving efficient model training and inference performance, particularly due to the coarse granularity of CSI feedback used in NW-first separate training schemes, which leads to performance degradation compared to UE-first and Joint E2E training methods.
Innovation Solution
The proposed NW-first separate training framework with raw dataset sharing addresses the performance degradation by providing the UE-side with raw, unquantized CSI feedback instead of its quantized version, enhancing the granularity of the input arguments for UE-side training without disclosing NW-side decoder model information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If quantized CSI feedback is used in NW-first separate training, then model training can proceed with protected proprietary information, but training performance degrades due to coarse granularity
Solution Approach 1:
The patent introduces an intermediate raw dataset sharing mechanism that mediates between the need for high-granularity training data and proprietary information protection. The raw CSI feedback data serves as an intermediary resource that can be shared for training purposes without revealing the proprietary decoder model, thus resolving the contradiction between information loss and performance.
Solution Approach 2:
The patent segments the training process into two distinct phases: first training the decoder with protected proprietary information, then training the encoder with shared raw data. This segmentation allows each component to be trained with appropriate data granularity while maintaining security boundaries, resolving the contradiction between information sharing and performance.
2Reliability
If raw unquantized CSI feedback is shared with UE-side, then training performance improves with finer granularity, but proprietary NW-side decoder model information may be disclosed
Solution Approach 1:
The raw CSI feedback dataset acts as an intermediary that enables high-performance training without direct exposure of proprietary decoder models. The dataset can be generated and shared through secure channels, allowing UE-side training to achieve fine granularity performance while the proprietary NW-side model remains protected.
Solution Approach 2:
The patent extracts the training value from raw CSI feedback data and separates it from the proprietary decoder model. By taking out only the necessary training data without extracting or exposing the proprietary model components, the system achieves high-performance training while preventing harmful information disclosure.
3Reliability
If Joint E2E training is used to achieve best performance, then model training and inference performance is optimized, but complexity increases and proprietary information protection becomes difficult
Solution Approach 1:
The patent segments the end-to-end training into separate NW-side decoder training and UE-side encoder training phases. This segmentation reduces system complexity by allowing independent training of each component while maintaining overall optimization, avoiding the need for complex joint training infrastructure.
Solution Approach 2:
The patent performs preliminary action by first training the decoder with protected proprietary information, then using the generated raw datasets for subsequent encoder training. This preliminary training step simplifies the overall process by preparing training data in advance, avoiding the need for complex real-time joint training systems.
Data Source
AI summary
Embodiments provide for an apparatus, method and computer program product at least for training an artificial intelligence decoder using at least an input to a hypothetical artificial intelligence encoder; determining a training dataset comprising the input to the hypothetical artificial intelligence encoder and at least one of: unquantized projected channel state information feedback, or dequantized projected channel state information feedback; transmitting the training dataset to a user equipment; and determining reconstructed channel state information, using the trained artificial intelligence decoder.