The Software Suite Powering RF AI Workflows
Radio frequency data is ubiquitous and contains valuable information to inform operations in aerospace, autonomous vehicles, communications networks, and other applications. However, RF high data rates and signal complexity make analysis and usage a challenging task.
Deepwave’s mission is focused on providing easier access to RF intelligence, and it created two product lines to address these needs: AIR-T hardware and AirStack software.
AirStack software manages AI and signal processing applications on Deepwave’s edge platform, as well as an expanding set of platforms from the RF technology ecosystem.
How It Works
Deepwave’s software and hardware platform offers a new approach to analyzing RF data. Instead of recording RF signals and transmitting raw IQ data across IT networks for historical analysis, Deepwave’s patented AIR-T hardware architecture and AirStack software allow AI models to analyze RF signals on the edge. This yields two sets of benefits.
The first is that low-latency ML model results can inform critical actions at the edge without requiring communication with an external network. Supporting navigation on autonomous vehicles is one example of how this capability can be applied.
The second benefit of AI-generated intelligence at the edge is data reduction. Instead of attempting the technical feat of sending GBs per second of raw RF data across a network, the Deepwave platform only delivers ML classification results. This reduces data backhaul by a factor of more than 10 million. Converting RF data into RF intelligence not only makes data transfers across networks more manageable, but it also can serve as timely intelligence to inform multi-sensor inputs to operations and command platforms.
AirStack in Action
Consider, for example, the need to convert FM radio signal content into text and then share the results with an external system in real-time. The high-level steps for this scenario are:
1. Initialize
The SDR must be initialized with the correct signal processing application and AI models and be waiting to be commanded.
2. Capture
The SDR must be tuned to the specific frequency of interest and commanded to receive RF signals with nanosecond precision, then converted into digital baseband data.
3. Interpret
The digital signal data is sent via low-latency protocols to powerful general-purpose processors (CPU/GPU). The CPU handles baseband-to-audio conversion, and the GPU executes the ASR AI model.
4. Publish
The resulting output text is combined with the SDR’s metadata and securely published to a larger operations platform where it undergoes further interpretation and data fusion.
Following is a closer look at the FM radio signal-to-text workflow, and how the full AirStack software portfolio powers this use case:
1. The AIR-T Sensor is running AirStack Edge to provide a remote command and control interface with a FIPS-grade encrypted datalink.
2. Using the AirStack Edge API, the Operations Platform makes a capabilities request on an AI generated textual output from a specified RF channel.
3. AirStack Edge returns the onboard capabilities library, including FM demodulation with automatic speech recognition (ASR).
4. The Operations Platform then communicates with AirStack Edge to initialize the edge sensor to perform the capture, demodulation, ASR, and the database location for the publishing of the results.
5. Next, the AIR-T Sensor receives and analyzes FM signals using desired RF AI application running on the AIR-T’s integrated FPGA, CPU, and GPU hardware. The onboard hardware drivers and high-performance processing libraries are provided by the AirStack Core operating system.
6. When a signal is received, it is processed by the AIR-T Sensor and passed to AirStack Edge for the AI Inference. Once completed, the results are published to the preconfigured database that is accessible by the Operations Platform.
7. Based upon the interpretation of the results, the Operations Platform may use the AirStack Edge API to reconfigure the AIR-T Sensor for a new task.
Product Overviews
AirStack software plays a central role in developing and managing low-latency RF intelligence. The main products cover remote management, custom edge application development, and high-performance computing operating systems.
AirStack Edge
Remote Application Management. Deploys and manages RF AI application suites at the edge, which include integrating multiple applications and models to execute larger RF intelligence tasks.
Explore Edge
AirStack Core
Operating System. Runtime and APIs to power high-performance RF AI computing. Bundled APIs provide direct AIR-T tuning and control of SDR and AI models, and custom application development capabilities.
Explore Core
AirStack BitStream
FPGA Toolkit. An FPGA application development framework for the AIR-T platform. This boosts analysis by enabling custom apps to pre-process RF data on the FPGA before it routes to the integrated CPU and GPU
Explore BitStream
Learn More
Frequently Asked Questions
How does the AirStack suite support the creation of a scalable RF System of Action for enterprise deployment?
AirStack delivers a cohesive System of Action by assigning clear roles to each product: AirStack Core provides the high-performance operating system, enabling on-device AI analysis. AirStack BitStream allows hardware customization for optimal data processing. Crucially, AirStack Edge serves as the scalable orchestration layer, enabling secure command and control for entire fleets of sensors. This layered approach ensures that high-value RF intelligence can be deployed, managed, and updated consistently across thousands of distributed edge nodes.
What are the distinct functional boundaries between Core, Edge, and BitStream in an RF Intelligence workflow?
AirStack’s boundaries reflect the computing stack: BitStream is the FPGA toolkit, handling custom hardware logic and ultra-low latency pre-processing. Core is the runtime, providing the operating system, drivers, and APIs for AI model execution on the CPU/GPU. Edge is the application management layer, offering the API for remote initialization, result publishing, and reconfiguring the entire RF AI application suite. This separation ensures developers can specialize without complex full-stack integration burden.
How does AirStack’s architecture address the challenge of remote lifecycle management and model updates at the edge?
AirStack solves remote management through its Edge API-centric design. All updates, deployment tasks, and model versioning are handled through AirStack Edge. A central operations platform communicates with Edge to upload new AI models, reconfigure signal channels, or initialize a new mission profile. This capability guarantees agility, allowing RF Intelligence applications to adapt rapidly to changing Electromagnetic Spectrum Monitoring environments without requiring costly, manual intervention at the sensor level.
How does the AirStack architecture achieve the massive data reduction necessary for operational command platforms?
Data reduction is a joint benefit achieved by converting high-volume raw RF data into low-bandwidth intelligence directly on the AIR-T. AirStack Core and BitStream perform the computationally intensive signal analysis and AI inference at the edge. AirStack Edge then securely publishes only the resulting low-bandwidth ML classification or textual output, which can be 10 million times smaller than the raw data stream. This process serves as timely intelligence to directly inform multi-sensor inputs to external command platforms.