# Artificial Intelligence  
## Radio Transceiver (AIR-T)  
### AIR-T Embedded Series Product Line  
#### Overview  
Key Specifications  
Deepwave’s AIR-T is the first software defined radio with embedded high performance computing. It contains three unique digital processors for any application:  
- **FPGA** for strict real-time operations  
- **GPU** for highly parallelized processing  
- **CPU** for control, I/O, and software applications  
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.  
- **Dual Channel MIMO Transceiver**  
  - 300 MHz to 6 GHz  
  - 100 MHz bandwidth Rx (per channel)  
  - 100 MHz bandwidth Tx (per channel)  
- **Digital Signal / Deep Learning Processors**  
The AIR-T allows users to easily incorporate artificial intelligence into their radio frequency and wireless technologies.  
- **High-speed digital I/O (GPIO/UART)**  
  - 1 Gbps Ethernet  
    - **Dual Power Mode:**  
      - ARM Cortex-A57 CPU (4 core)  
      - NVIDIA Denver2 CPU (2 core)  
      - NVIDIA Pascal GPU (256 core)  
- **Dual Power Mode:**  
  - 22 / 14 Watts  
#### Software Support  
- K Keras  
- HPC with CUDA toolkit using C/C++ or Python interfaces  
- GNURadio  
  **THE FREE & OPEN SOFTWARE RADIO ECOSYSTEM**  
#### Signal Processing Support for  
Deepwave Digital, Inc.  
1429 Walnut St, Suite 1000, Philadelphia, PA 19102  
[www.deepwavedigital.com](/content/site-root.html)  
[salesteam@deepwavedigital.com](mailto:salesteam@deepwavedigital.com)  
#### Mechanical  
- **Size** - 6.7 x 6.7 x 1.4 inches  
- **Weight** - 0.35 kg (0.8 lbs)  
Support for industry leading SDR development environment  
### Deep Learning  
---  
### DEEPWAVE DIGITAL  
#### Performance  
The AIR-T uses 256 GPU cores to create a highly parallel compute environment making wideband processing for software defined radio (SDR) applications obtainable.  
Using the embedded NVIDIA  
The AIR-T uses zero copy memory access to overcome the data transfer overhead typically associated with GPU processing.  
#### Real-time DSP Measurements  
### AIR-T Hardware Abstraction  
#### Fourier Transform Size  
### AIR-T Hardware  
- **Pre-trained AI Cores**  
- **AI Frameworks**  
- **User Developed Applications**  
### Embedded Series Models  
##### Performance Table  
|  | Air107A | Air107B | Air107C | Air107D |  
| --- | --- | --- | --- | --- |  
| GPU Cores | 256 | 256 | 256 | 256 |  
| CPU Cores | 6 | 6 | 6 | 6 |  
| Shared Memory | 8GB | 8GB | 8GB | 8GB |  
| FGPA Model | XC7A75T | XC7A75T | XC7A200T | XC7A200T |  
| Logic Cells | 75,520 | 75,520 | 215,360 | 215,360 |  
| DSP Slices | 180 | 180 | 740 | 740 |  
| Memory | 3,780 | 3,780 | 12,140 | 12,140 |  
| Enclosure | No | Yes | No | Yes |  
Data Rate (Gbits/s)