Getting Data Off the Radar
Two Main Methods
There are two main ways to collect data from TI mmWave radars:
1. Raw ADC Streaming
- Uses a DCA1000 capture board.
- Streams unprocessed ADC samples directly to a computer over Ethernet.
- All signal processing is done later in MATLAB or Python.
- Offers maximum flexibility to reprocess data in different ways.
Drawbacks:
- Very large files
- Requires extra hardware
- Higher bandwidth needs
2. Feature Vector Streaming
- The radar processes data on-board using its built-in hardware accelerator and MCU.
- Only compressed feature vectors (e.g 6 numbers per frame) are sent out via UART or SPI.
- Saves bandwidth and storage, ideal for embedded systems.
Drawbacks:
- Less flexibility - if processing algorithms change, the stored features may no longer match.
Data Format: TLV (Type-Length-Value)
TI’s radar SDK uses a TLV structure for UART data:
| Field | Description |
|---|---|
| Type | Identifies what kind of data (e.g., point cloud, features) |
| Length | Number of bytes in this data block |
| Value | The actual data (e.g., feature vector array) |
Each packet corresponds to one radar frame and includes padding so the total packet size aligns with 32 bytes (for efficient hardware transfers).
Summary of Trade-offs
| Method | Data Type | Processing Location | Advantages | Disadvantages |
|---|---|---|---|---|
| Raw ADC Streaming | Raw complex samples | Off-chip (PC) | Full control, tunable processing | High data rate, large files |
| Feature Vector Streaming | Compact feature vectors | On-chip (radar) | Low bandwidth, real-time capable | No offline reprocessing |