Deepwave Digital’s AIR-T Brings Artificial Intelligence into a Radio Transceiver

Software-defined radio has been around for decades and fundamentally shifts most of the complex signal handling (receivers, transmitters, amplifiers, filters, etc.) over to a digital software platform implemented by PCs and embedded systems. With the rise of Wi-Fi, cellular telecommunications, and the IoT, we are increasingly bombarded with a myriad of different signals, and that’s on top of the RF spectrum. As you might imagine, these signals still rely on human analytics to sort those frequencies and reign in fraudulent others.

It turns out artificial intelligence can handle those signals better than humans, and although AI has been implemented into speech recognition, autonomous vehicles and gaming (among a host of other applications); it has not really been adapted for software defined radio or incorporated into an RF hardware solution. Hardware and software solutions provider Deepwave Digital stepped up to fill those voids with their AIR-T (Artificial Intelligence Radio-Transceiver) platform.

Deepwave’s AIR-T is essentially a high-powered SBC that utilizes a NVIDIA Jetson TX2 for signal processing and deep learning applications. It also sports FPGA for real-time operations and the integrated CPUs for hardware, I/O, and software applications.

According to Deepwave:

“This versatile system can function as a highly parallel SDR, data recorder, or inference engine for deep learning algorithms. The embedded GPU allows for SDR applications to process bandwidths greater than 200 MHz in real-time.”

Beyond the AIR-T’s Jetson TX2 GPU, the board is outfitted with an impressive list of hardware, including a Xilinx Artix 7 FPGA and an Analog Devices 9371 MIMO transceiver with a pair of RX channels (100MHz each) and two TX channels (250MHz each), along with Observation and Sniffer support. There is also a wealth of connectivity options as well, including GPS Sync (via 1 PPS and 10MHz), USB 2.0/3.0, SATA, Ethernet, Wi-Fi, Bluetooth, HDMI and high-speed digital GPIO/UART.

On the software side, the AIR-T uses Ubuntu 16.04 along with a custom GNURadio block for deploying neural networks. You can also program the GPU using Python or C++ to create your own custom GNURadio applications. Deepwave is expected to launch a crowdfunding campaign on Crowd Supply to get the AIR-T to the commercial market sometime soon, and there’s no word yet on pricing, but I expect it will be high in cost as the Jetson TX2 fetches over $500 alone.

UPDATE: AIR-T has gone live on Crowd Supply, and my initial estimate on pricing was correct as the AI radio transceiver fetches a hefty $5,699 pledge. But you do get what you pay for, as no other platform packs both high-end hardware and RF deep learning in a single package for SDR.