Semantic Video Compression for Compute Offloading Over Impaired Links

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing cellular connectivity faces challenges such as bad coverage, low throughput, and dynamic network load, making it difficult to provide high-fidelity services to lightweight devices like XR glasses, which are compute-intensive and power-needy, especially for tasks like object detection and gesture recognition.

Innovation Solution

Implementing channel-aware semantic coding (CASC) that combines semantic source coding and channel coding to adapt to channel conditions, using multi-priority transmission (MPT) and hierarchical modulation simultaneous transmission (HMST) to optimize power consumption and bandwidth usage, and separating encoded data into high and low priority parts for efficient transmission.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Power

If compute-intensive tasks are performed locally on lightweight devices, then processing capability is improved, but power consumption increases

Engineering Contradiction:
Improveprocessing capabilityVSAvoidpower consumption
Core Design Contradiction:
PowerVSUse of energy by moving object

Solution Approach 1:

The patent segments video data into different priority levels (high priority and low priority portions) based on semantic importance. High priority portions containing critical semantic information are transmitted with higher reliability, while low priority portions are transmitted with lower reliability or compressed more aggressively. This segmentation allows the system to offload processing to remote devices while maintaining acceptable performance with reduced power consumption compared to full local processing.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent dynamically changes encoding parameters such as compression ratio, priority assignment, and transmission reliability based on channel conditions and device capabilities. When channel conditions are poor or device power is limited, the system adjusts parameters to transmit less data with higher compression. When conditions improve, it transmits more detailed data, thereby optimizing the trade-off between processing capability and power consumption.

Inventive Principle:
Principle #35Parameter changes

2Use of energy by moving object

If uncompressed or lightly compressed source data is offloaded to remote devices, then local power consumption is reduced, but bandwidth consumption and transmit power increase

Engineering Contradiction:
Improvelocal power consumptionVSAvoidbandwidth consumption
Core Design Contradiction:
Use of energy by moving objectVSQuantity of substance

Solution Approach 1:

The patent dynamically adjusts compression parameters based on channel conditions. When channel capacity is limited or transmit power is constrained, the system increases compression ratio to reduce bandwidth consumption. When channel conditions improve, it reduces compression to maintain higher fidelity, thereby optimizing the trade-off between local power savings and bandwidth consumption.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent divides video data into priority-based segments and applies different compression strategies to each segment. High priority portions are compressed less to preserve critical information, while low priority portions are compressed more aggressively. This selective segmentation reduces overall bandwidth consumption while maintaining essential visual quality, allowing lightweight devices to offload processing without excessive bandwidth usage.

Inventive Principle:
Principle #1Segmentation

3Quantity of substance

If high compression is applied to video data, then bandwidth usage is reduced, but inference fidelity at the decoder decreases

Engineering Contradiction:
Improvebandwidth usageVSAvoidinference fidelity
Core Design Contradiction:
Quantity of substanceVSMeasurement precision

Solution Approach 1:

The patent applies different quality levels to different portions of video data based on their semantic importance. High priority portions containing critical semantic information (such as key objects or actions) are encoded with higher quality and less compression, while low priority portions (such as background or less important details) are encoded with lower quality and higher compression. This local quality differentiation reduces overall bandwidth usage while preserving inference fidelity for the most important visual elements.

Inventive Principle:
Principle #3Local quality

4Productivity

If multiple priority levels are transmitted simultaneously, then transmission efficiency is improved, but device complexity increases

Engineering Contradiction:
Improvetransmission efficiencyVSAvoidencoding complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent segments video data into priority levels and transmits them through different channel coding schemes. High priority data uses more robust coding with higher redundancy, while low priority data uses less robust coding with lower redundancy. This segmentation enables simultaneous transmission of multiple priority levels with different reliability requirements, improving overall transmission efficiency by matching coding resources to data importance while managing encoding complexity through standardized priority categories.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentUS12501085B2Semantic compression for compute offloading
Publication Date: 2025.12.16 APPLE INC
  • US12501085B2 patent drawing
  • US12501085B2 patent drawing
  • US12501085B2 patent drawing

AI summary

Methods and systems for semantic encoding by a user equipment (UE) are configured for determining an expected power consumption for encoding video data that includes semantic features, the semantic features representing a meaning of information represented in video frames of the video data; encoding one or more video frames of the video data using a selected a semantic representation of one or more video frames of the video data, the semantic representation being selected based on the expected power consumption that is determined; and transmitting the encoded video data including the semantic representation.