Wireless Split Inference Point Adjustment for Channel-Latency Tradeoffs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wireless communication systems face challenges in optimizing split points for split inference in AI/ML models, particularly in terms of channel quality and latency, which affect inference performance.
Innovation Solution
A method and device for adjusting a split point in a wireless communication system by considering channel quality and computing power, allowing for dynamic adjustment of split points based on uplink channel quality information and latency constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If the split point is set closer to the terminal for local inference, then inference latency is reduced, but channel quality requirements increase and may not be satisfied in poor radio conditions
Solution Approach 1:
The patent implements dynamic adjustment of the split point between terminal and network based on real-time channel quality feedback. When channel quality is good, the split point moves toward the terminal for lower latency; when channel quality deteriorates, the split point moves toward the network to ensure reliable transmission. This dynamic reconfiguration resolves the contradiction by adapting the inference architecture to current radio conditions.
Solution Approach 2:
The patent changes the parameter of split point location based on channel quality metrics. By monitoring uplink channel quality information and adjusting the split point position accordingly, the system optimizes the balance between latency and reliability. The split point can be moved between different layers of the neural network or between terminal and network based on channel conditions.
2Reliability
If the split point is set closer to the network for centralized inference, then channel quality impact is reduced, but inference latency increases due to additional transmission time
Solution Approach 1:
The system dynamically adjusts the split point position based on channel quality feedback. When channel conditions improve, the split point moves closer to the terminal to reduce latency; when conditions worsen, it moves toward the network for better reliability. This resolves the contradiction by making the architecture adaptive rather than static.
Solution Approach 2:
The patent implements a feedback mechanism where the terminal reports uplink channel quality information to the network, and the network adjusts the split point accordingly. This closed-loop control enables the system to respond to changing channel conditions and optimize the balance between latency and reliability in real-time.
3Productivity
If the split point is adjusted frequently to optimize performance, then inference performance is maximized, but system complexity and signaling overhead increase
Solution Approach 1:
Instead of continuous adjustment, the patent employs periodic adjustment of the split point based on threshold-based triggers. The terminal monitors channel quality and only requests split point adjustment when quality crosses certain thresholds. This reduces signaling overhead and system complexity while maintaining inference performance optimization.
4Loss of time
If more computing power is allocated to the terminal for local inference, then latency is reduced, but the terminal's computing power consumption increases
Solution Approach 1:
The patent dynamically adjusts the computing power allocation between terminal and network by changing the split point position. When channel quality is good, more computation is moved to the terminal for lower latency; when channel quality is poor, computation is moved to the network to conserve terminal energy. This resolves the contradiction by making computing power allocation adaptive to channel conditions.
Data Source
AI summary
In order to provide device and method for adjusting split point in wireless communication system, a method of operating a terminal in a wireless communication system may comprise requesting a base station for uplink channel quality information, receiving, from the base station, the uplink channel quality information, determining a first split point based on the channel quality information, performing first split inference based on the determined first split point and transmitting, to the base station, first split inference result information.