"The full name of air-t is artistic intelligence radio - transceiver, i.e. artificial intelligence radio transceiver.
After listening to the application of artificial intelligence in voice, image and data, I believe most people have not heard of the combination of artificial intelligence and radio. What kind of technology is this? What changes will it bring to the radio field?
Air-t characteristics
Air-t is a software defined radio development board with embedded high-performance computing function launched by deepwave, which supports frequencies ranging from 300 MHz to 6 GHz. By allowing the AI engine to fully control the hardware, air-t greatly reduces the price and performance bottlenecks required to realize autonomous signal recognition, interference suppression and other functions in the radio, so as to realize fully autonomous SDR.
The air-t development board is 17.0 * 17.0 * 3.5cm in size and contains three unique digital processors. To exaggerate, it can even provide any desired function for signal processing applications.
FPGA: for strict real-time operating environment
GPU: for high performance parallel processing
CPU: used for control, I / O port, DSP and software application
Air-t parameters
Air-t development board allows users to easily integrate AI into their RF and wireless technologies. This unique combination not only brings high-performance computing, artificial intelligence and deep learning, but also enables advanced graphics processing and rendering on an embedded platform. Air-t is multifunctional and can be used as a high-performance parallel SDR, data recorder or reasoning engine for deep learning algorithm. The embedded GPU allows SDR applications to process bandwidth greater than 200 MHz in real time.
Let's briefly understand the parameters of the air-t development board:
Dual channel MIMO transceiver analog devices 9371 (specification)
-300 MHz to 6 GHz
-100 MHz bandwidth RX * 2
-100 MHz bandwidth TX * 2
Digital signal / deep learning processor
-Xilinx artix 7 FPGA (specification)
-NVIDIA Jetson TX2 [arm cortex-a57 CPU (4 cores); NVIDIA denver2 CPU (2 cores); NVIDIA Pascal GPU (256 cores); 8 GB memory] (specification)
Connectivity
-GPS synchronization via 1 PPS and 10 MHz
- USB 3.0，USB 2.0 / 3.0，SATA
-High speed digital I / O (GPIO / UART)
-1 Gbps Ethernet
Dual power mode
-22 / 14 watts
NVIDIA Jetson TX2
The principle block diagram of the whole air-t development board is as follows:
performance
Air-t uses 256 GPU cores (NVIDIA Jetson TX2) to create a high-performance parallel computing environment, which can realize broadband processing of Software Defined Radio (SDR) applications. As shown in the figure below, in the real-time SDR application, using NVIDIA Jetson TX2, the bandwidth processing of air-t is 250% higher than that of Intel 7500u CPU and 1350% higher than that of arm cortex-a57 (4-core). Air-t uses zero copy memory access to overcome the data transmission overhead usually associated with GPU processing.
Software support
The air-t development board supports Ubuntu 16.4 system. You can transplant existing GNU radio applications in a few minutes through soapysdr, or easily deploy neural networks or high-performance applications using custom GNU radio modules;
At the same time, air-t supports the use of official hardware drivers to build its own customized software. Air-t can easily program GPU using Python or C + +. Of course, if you like, you can program FPGA using high-level language VHDL / Verilog, which of course requires a certain knowledge reserve.
Development process
Air-t provides customized and open Ubuntu system and customized FPGA module connected with GNU radio, so the development can be realized without changing the existing code. With 256 GPU cores, it is quite easy to develop and deploy AI applications on hardware without writing CUDA or VHDL code.
Of course, the simplest way is to communicate directly with the driver by using Python or C + +. However, for powerful users, air-t is an open platform, which can directly program FPGA and GPU.
From the board itself, air-t is a brand-new radio development solution for engineers and researchers who develop low-cost AI, deep learning and high-performance wireless systems. However, with its excellent hardware conditions, air-t can even process any type of data, For example, you can combine the traditional uses of image and video processing with RF. With the help of USB 3.0 and 2.0, Gigabit Ethernet and high-speed IO, there are many ways to input and output data on air-t, or you can connect other sensors and fuse the data through air-t.
In addition, air-t can demodulate signals in an integrated platform and apply deep learning to the generated image, video or audio data. For example, you may once need multiple devices to directly receive signals containing audio and then perform speech recognition. However, air-t integrates this hardware into an easy-to-use software package. From speech recognition to digital signal processing, the integrated NVIDIA GPU can provide you with the performance required by these edge computing applications.
Summary
In a word, high-performance SDR is seamlessly integrated with the most advanced in-depth learning hardware. Air-t has opened a new chapter of SDR, which may also be the most attentive and popular AI development board in 2018! Of course, it may also be one of the most expensive AI development boards at present, with a price of $4995.
Original address: https://www.eeboard.com/evaluation/air-t/
Original statement: This article is original by Aiban. Reprint is declined“
Our other product:
