Epiq Solutions Paves the Way for AI/ML Integration in RF Spectrum Analysis

The world of AI and ML is evolving at a rapid pace. While much of the fanfare is focused on analyzing the surprisingly human-like output of ChatGPT, the usage of AI/ML in defense applications is also getting significant attention.  This is especially true for use cases where there is an ever-increasing amount of digital data to process in order to understand the RF spectrum and wireless threat landscape. In these applications, AI/ML techniques can bring a unique perspective to provide insights where classical processing may fall short for latency, data throughput, or processing SWaP-C.  

The Challenge of RF Spectrum Data

Of all the applications that can generate a firehose of data to process, digitized RF spectrum coming from RF sensors is near the top of the list.  Processing these large streams of digital data becomes even more challenging when thinking about size, weight, and power-constrained deployment scenarios.  Distributed processing “at the edge” refers to a system that has all of the data processing happening in the same integrated platform where the data is being collected.  For an RF system, this translates to a system where the antenna, software-defined radio (SDR), and CPU/GPU for executing an AI/ML application would all be self-contained in a single system. An example system configuration for AI/ML edge processing could be a payload bay of a small unmanned air system (sUAS) where an SDR is creating raw digitized data from the RF spectrum and is processing that digitized RF in a neural net running on a local CPU or GPU to perform signal classification, channel estimation, direction finding, RF anomaly detection, and more.

Epiq's Vision: Enabling AI/ML at the RF Edge with Flexible Hardware

With these applications in mind, Epiq is excited to provide a sneak peek at a family of newly announced standalone SDR platforms focused on enabling our customers to tackle AI/ML applications at the edge.  The first platform is Matchstiq™ X40, which combines a wide bandwidth multi-channel SDR transceiver supporting operation up to 18 GHz with an NVIDIA Jetson Orin NX 16GB module to provide a powerhouse of computational signal processing.

The second platform is our Matchstiq™ G40, which combines two of our Sidekiq™ NV100 M.2 2280 cards on a carrier card with the same NVIDIA Jetson Orin NX 16GB module to handle advanced signal processing requirements.  For customers interested in RF recording, analysis, and playback, the Matchstiq™ G20 integrates a single Sidekiq™ NV100 SDR card and an M.2 SSD for local non-volatile data storage.

Each platform is intended to run standalone, with high-rate data transport and computational signal processing executing locally.  Leveraging the continued advances that NVIDIA has brought forward over the last 5+ years, the latest NVIDIA Jetson Orin NX 16GB modules are groundbreaking devices in terms of their processing capability: at a power consumption that scales between 10W and 20W, the Orin NX can achieve up to 70 trillion operations per second (TOPS) of processing to support the most demanding neural net models.  

Similar to our existing portfolio of COTS SDRs, our standard libsidekiq software API is supported on the Matchstiq™ X40 and the G20/G40.  This API provides a consistent interface across our SDR portfolio to simplify configuration, control, and streaming with each flexible RF transceiver. This AI/ML product family is architected to give customers a range of capabilities, trading off RF coverage, bandwidth, and channel count against SWaP-C. See the product table below.

*Power consumption is application-dependent

If running AI/ML-enabled applications on a standalone SDR is of interest to you, our Matchstiq™ X40, G20, and G40 will all be commercially available starting in 2024.

Want to get any additional details on these platforms or any other existing SDR products? Contact our team —we're always happy to answer any questions or tell you more about our tech!