Skip to content

Latest commit

 

History

History
39 lines (25 loc) · 2.66 KB

README.md

File metadata and controls

39 lines (25 loc) · 2.66 KB

Overview

The Synchronous Data Stream (SDS) Framework manages data streams and provides tools for developing and optimizing embedded applications using DSP, ML, or Edge AI algorithms. It enables real-time capture of multiple data streams—such as sensor, audio, and video inputs—alongside algorithm outputs directly on target hardware. Using the MDK-Middleware, these streams may be stored in files on a host computer or on memory cards in the embedded system.

Data capturing and playback in Target System

The captured data streams are useful in various steps of the development cycle, for example to:

  • Validate physical input signals from sensors or output of algorithms.
  • Provide input data to Digital Signal Processing (DSP) development tools (such as filter designers) or MLOps systems (for AI model training).
  • Provide input data for simulation using Arm Virtual Hardware (AVH-FVP) models for testing and validation, for example in CI systems.

CI Workflow with Simulation

With integration into MLOps systems, the SDS Framework can be used to provide input data to ML/AI development systems for model classification, training, and performance optimization.

MLOps Integration

Features

  • Implements a flexible data stream management for sensor, audio, and video interfaces that process data streams.
  • Supports data streams from multiple interfaces (i.e. for sensor fusion) including provisions for time drifts.
  • Record real-world data with original data sources of the target hardware for analysis and development.
  • Playback real-world data for algorithm validation using target hardware or Arm Virtual Hardware - FVP.

The captured data streams are stored in SDS data files. A YAML metadata file can be used to describe the content. The SDS data files have several use cases such as:

  • Input to Digital Signal Processing (DSP) development tools such as filter designers
  • Input to Machine Learning (ML) model classification, training, and performance optimization
  • Verify execution of DSP algorithm on Cortex-M targets with off-line tools

Python-based utilities are provided for recording, playback, visualization, and data conversion

Links