Extreme Wideband Signal Processing via Zero-Copy Shared Memory
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computing equipment is limited to processing about 20 MHz instantaneous bandwidth of electromagnetic spectrum, leading to system bottlenecks and failures when attempting to process larger bandwidths, and existing digital signal processing frameworks exert back pressure and require data copying, which is computationally burdensome and inefficient.
Innovation Solution
The system employs a shared memory data structure in a SIMD processor to store digitized sensor samples, allowing zero-copy access for multiple applications, bypassing CPU processing and enabling parallel processing of up to 100 MHz or more instantaneous bandwidth without data copying, using direct memory access and circular buffers to maintain time coherency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If data copying is used to share signal data with processing applications, then data sharing is enabled, but computational resources are tied up and system bottlenecks occur
Solution Approach 1:
The patent applies the opposite of copying - zero-copy sharing - where data is shared between applications without creating copies. The shared memory buffer allows multiple applications to access the same signal data simultaneously without duplication, eliminating the computational burden of copying while maintaining data sharing capability.
Solution Approach 2:
The patent introduces a shared memory buffer as an intermediary mechanism between the signal collection system and processing applications. This buffer acts as a mediator that enables efficient data sharing without requiring applications to copy data, thus resolving the contradiction between data sharing and computational efficiency.
2Productivity
If bandwidth processing is increased beyond 20 MHz, then spectrum processing capability is improved, but system bottlenecks and failures occur
Solution Approach 1:
The patent segments the signal processing workflow into distinct phases: data collection in shared memory buffers, then processing by multiple applications. This segmentation allows the system to handle high bandwidth data (500 MHz+) by dividing the processing tasks across multiple applications that access the same data through zero-copy sharing, thereby improving reliability while increasing productivity.
Solution Approach 2:
The patent transitions from traditional CPU-based sequential processing to GPU-based parallel processing with zero-copy memory access. This dimensional change in the processing architecture enables the system to handle bandwidths exceeding 500 MHz without system failures, as the GPU can process large datasets in parallel while maintaining system stability through efficient memory management.
3Adaptability or versatility
If multiple applications access signal data simultaneously, then processing versatility is improved, but data copying becomes computationally burdensome
Solution Approach 1:
The patent creates a universal shared memory buffer that serves multiple applications simultaneously. The buffer is designed to be accessed by any number of applications through zero-copy sharing, making the system highly versatile while avoiding the computational overhead of copying data for each application. This resolves the contradiction by enabling multi-application processing without sacrificing computational efficiency.
Data Source
AI summary
A method for digital signal processing of sensor data includes receiving digitized samples of sensor signals via a network connection; converting the digitized samples into a standardized format; storing the converted digitized samples in a shared memory data structure in memory of a single instruction multiple data (SIMD) processor; and providing zero-copy read access to the converted digitized samples stored in the shared memory data structure to a plurality of applications.