OmniSIG® Studio

With DeepSig’s OmniSIG® Studio software tools, customers can curate their own RF datasets, train state-of-the-art deep learning inference models for custom wireless sensing applications, and deploy them to edge-sensing devices. OmniSIG® Studio contains DeepSig’s industry-leading baseline RF dataset for machine learning, including many consumer wireless signals. Customers can also incorporate their own custom data, signals, and signatures to train the AI sensor.

OmniSIG® Studio Tools

The OmniSIG® Studio contains a full suite of tools that allow a user, without machine learning experience, to create powerful signal classification systems within hours without writing a single line of code.  The Studio includes the following tools:

• File management tools for organizing your RF datasets
• Tools for labeling RF datasets including very bursty or hopping signals, wideband and narrowband signals, or even unknown signals
• Automated labeling tools that give the user and quick way to label thousands of signals in seconds
• The core training mechanism that allows users to select the files they want to include in the new model and begin training
• Knobs to tweak RF augmentation phases and certain neural network architecture parameters for expert users
• Model evaluation page that allows users to immediately measure the performance of the trained model using RF data test sets
• Convenient output model format that easily drops into the OmniSIG® sensor software for immediately deployment of new capabilities

This tool suite is a market-first, enabling customers to custom-tune DeepSig’s deep learning models for signal detection and classification for their specific RF signatures and applications.

OmniSIG® Studio was designed by engineers with decades of industry experience to enable signal processing on complex-valued RF sample data. It contains specialized features to assist in working with large RF datasets that don’t exist elsewhere.

RF Data Management

AI workflows are driven by data. DeepSig has built a comprehensive labeling tool designed for working with signal recordings. The OmniSIG® Studio supports a variety of different signal recording formats and provides an easy way to visualize signal captures and label them for use in AI systems, using automated, semi-automated, and hand-tuning methods, all while using a convenient web-based UI that can be accessed remotely by multiple users.

The OmniSIG® Sensor and OmniSIG® Studio combine to enable a data engine that can improve signal detection and classification accuracy by capturing spurious data. Labeling the spurious data, retraining the network, and redeploying automatically improves the sensor’s ability to detect signals and signatures of interest. Leveraging the OmniSIG® Studio lets your team focus on identifying and triaging the rare cases with anomalies or interference to improve system performance without dealing with the complexities of deep learning techniques.

Training, Validating and Deploying

The world of artificial intelligence is moving fast, and DeepSig’s machine learning research scientists are at the top of the field. A primary goal of the OmniSIG® Studio is making the latest advances in AI available to customers without requiring machine learning expertise. We accomplish this by incorporating these advances into the OmniSIG® Sensor and providing a simple-to-use web interface to configure model training.

Using either datasets provided by DeepSig (included with an OmniSIG® Studio license) or custom signal captures, customers train the OmniSIG® deep learning model until it reaches the desired level of performance.  The Studio provides a running measure of training accuracy, and the user can save off the current model without stopping training.  This provides for a quick way for users to test and evaluate the current model without sacrificing time due to stopping training.  This workflow has been shown to reduce signal classification development time 10 fold compared to traditional DSP development workflows.  All of these features can be done using a single desktop-class GPU, training an OmniSIG® model takes only a few hours, depending on the required level of performance.