Hyperdimensional Code Words for Interference-Based Wireless Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data processing and communication systems face inefficiencies in transmitting and processing data, particularly in wireless communication contexts, where interference is typically viewed as an obstacle rather than a resource.
Innovation Solution
The use of hyperdimensional (HD) computing-based frameworks that leverage the robustness of HD representations to harness interference for over-the-air computation, enabling distributed training and inference of AI models by exploiting the superposition principle in wireless channels, allowing functions of HD code words to be computed without decoding individual code words.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional data transmission methods are used in wireless communication, then data can be transmitted, but interference is viewed as an obstacle reducing communication efficiency and reliability
Solution Approach 1:
The patent converts the harmful interference in wireless channels into a beneficial resource by using hyperdimensional code words where interference naturally performs distributed computation operations. Multiple transmitters can simultaneously transmit HD code words over the same wireless channel, and the interference at the receiver becomes the mechanism for computing functions (like summation) of the transmitted data, thereby improving communication efficiency and enabling over-the-air computation.
Solution Approach 2:
The patent transforms data into hyperdimensional representations with high dimensionality and specific mathematical properties. By changing the parameter of data representation from conventional low-dimensional formats to high-dimensional hyperdimensional code words, the system achieves robustness against interference and enables new computational capabilities in the wireless channel.
2Productivity
If distributed training and inference of AI models is performed using conventional methods, then computation can be distributed, but overhead and latency increase
Solution Approach 1:
The patent merges communication and computation into a single process by transmitting hyperdimensional code words that enable both data transfer and distributed computation simultaneously. The wireless channel itself performs computation through interference, eliminating separate communication and computation phases, thereby reducing overhead and latency in distributed AI training and inference.
Solution Approach 2:
The patent enables continuous useful action by allowing simultaneous transmission and computation. While conventional systems require sequential operations (transmit then compute), the hyperdimensional approach allows the wireless channel to continuously perform computation on transmitted data without interruption, improving productivity and reducing time loss.
3Loss of information
If HD code words are transmitted without compression, then full information is preserved, but communication resources are inefficiently used
Solution Approach 1:
The patent applies local quality by compressing HD code words selectively based on their information content and importance. Not all dimensions of hyperdimensional code words carry equal information value, so the system identifies and preserves only the critical components while compressing or discarding redundant parts, thereby maintaining essential information while reducing communication resource consumption.
Data Source
AI summary
According to various examples, a communication system is described comprising a receiver configured to receive, for each of a plurality of object classes, via a wireless communication channel shared among transmitters of a respective set of transmitters, a superposition of transmitted hyperdimensional code words, comprising, for each transmitter of the respective set of transmitters, a hyperdimensional code word transmitted via the wireless communication channel and encoding data of an object of the object class acquired by the transmitter, a memory configured to store, for each of the plurality of object classes, the received superposition in association with the class, a processor configured to classify a hyperdimensional code word representing an object to be classified by correlating the hyperdimensional code word with each stored superposition and to generate a classification result corresponding to the object class associated with a superposition fulfilling a predetermined criterion based on correlation results.