Tensor Streaming Protocol for Real-Time Data Collection
Data collection is widely recognized as one of the most significant bottlenecks in machine learning. Sourcing data from data vendors, open-source datasets, or surveys can be expensive, and the collected data often requires extensive cleaning and formatting before use. To simplify data collection, DEEPCRAFT™ Studio provides the Tensor Streaming Protocol, which allows you to stream data from any sensor or development kit directly into DEEPCRAFT™ Studio. Because the streamed data already conforms to the format required by DEEPCRAFT™ Studio, no additional data formatting is needed, saving you valuable time.
This topic explains what the Tensor Streaming Protocol is, how to implement your own streaming firmware, and which ready-to-use streaming firmware is available for Infineon development kits.
What is the Tensor Streaming Protocol?
The Tensor Streaming Protocol is a set of rules that defines how data is streamed from a sensor, development board, or kit into DEEPCRAFT™ Studio. Using the protocol, you can build firmware that streams data for collecting, training datasets and for evaluating machine learning models in real time.
The protocol defines the communication mechanism between a client (DEEPCRAFT™ Studio) and a board. Key characteristics include:
- Multiple transport options: The protocol is designed to work over TCP, UDP, serial port, and Bluetooth.
- Multiple data streams: The protocol handles concurrent data streams from sensors, models, and playback devices, enabling efficient data transfer and processing on embedded systems.
- Based on Protocol Buffers: The protocol is built on protobuf3, and a DotNetCli test client is provided to help you validate your implementation.
To learn more about the protocol, refer to the tensor-streaming-protocol GitHub repository.
Streaming Firmware for Infineon development kits
Infineon provides ready-to-use streaming firmware for selected development kits. The firmware is built on the Tensor Streaming Protocol and enables the supported kits to stream sensor data directly into DEEPCRAFT™ Studio.
Infineon development kits can be used as rapid prototyping platforms for edge machine learning workflows. With the provided streaming firmware, you can collect real-time sensor data from supported onboard sensors, use the data to create training datasets, generate device-ready code, and deploy models on the hardware.
Each supported kit has a dedicated page with kit-specific details, including supported sensors, firmware download links, flashing prerequisites, and step-by-step flashing instructions. Select your kit from the table below and follow the Streaming Firmware section on the corresponding kit page.
| Development kit | Supported sensors |
|---|---|
| PSOC™ Edge E84 HMI Kit | Microphone, accelerometer, gyroscope, humidity-temperature sensor, radar |
| PSOC™ Edge E84 AI Kit | Microphone, accelerometer, gyroscope, magnetometer, pressure-temperature sensor, humidity-temperature sensor, radar |
| PSOC™ Edge E84 Evaluation Kit | Microphone, accelerometer, gyroscope, magnetometer |
| PSOC™ 6 AI Evaluation Kit | Microphone, accelerometer, gyroscope, magnetometer, pressure sensor, radar |
You can flash the firmware onto the kits through the Web-based OOB webpage or ModusToolbox™ Programmer . Refer to the linked kit page for the flashing instructions.
Implement your own streaming firmware
To stream data from a sensor or an Infineon development board for which no streaming firmware is provided, implement custom firmware using the Tensor Streaming Protocol. Custom firmware allows you to add support for additional sensors and to collect data at different sampling rates.
Use the reference implementation , available in the example-firmware folder of the Tensor Streaming Protocol GitHub repository, as a starting point for your custom firmware.
For step-by-step instructions on registering one or more sensors using Tensor Streaming Protocol, refer to Registering sensors for data streaming.
After building your firmware, refer to Flashing custom firmware using ModusToolbox™ to flash it onto your kit.