6.5. Connectivity Devices
TI’s connectivity devices combine wireless communication with edge ML capabilities for IoT applications.
6.5.1. Overview
Connectivity devices enable:
Wireless sensor networks with edge AI
Smart home and building automation
Industrial IoT with local inference
Bluetooth and Sub-GHz sensing applications
These devices perform ML inference locally, reducing latency and network bandwidth while maintaining wireless connectivity.
6.5.2. Supported Devices
SimpleLink CC27xx Family
Device |
Core |
Features |
|---|---|---|
CC2755 |
Cortex-M33 (96 MHz) |
Bluetooth 5.4 LE, Thread, Zigbee, Matter |
SimpleLink CC13xx Family
Device |
Core |
Features |
|---|---|---|
CC1352 |
Cortex-M4 (48 MHz) |
Sub-GHz, Bluetooth 5.0 LE, multi-protocol |
6.5.3. CC2755
The CC2755 is a modern wireless MCU with strong ML capabilities.
Key Features:
96 MHz Arm Cortex-M33 core
Bluetooth 5.4 LE Audio
Thread and Zigbee support
Matter protocol ready
1 MB Flash
256 KB SRAM
TrustZone security
ML Capabilities:
CPU-based inference (no NPU)
Support for models up to ~10k parameters
Suitable for classification and anomaly detection
Configuration:
common:
target_device: 'CC2755'
training:
model_name: 'CLS_4k' # Standard models
6.5.4. CC1352
Multi-band wireless MCU for Sub-GHz and Bluetooth applications.
Key Features:
48 MHz Arm Cortex-M4F core
Sub-GHz radio (for long range)
Bluetooth 5.0 LE
352 KB Flash
80 KB SRAM
Ultra-low power
ML Capabilities:
More constrained than CC2755
Best for small models (<4k parameters)
Ideal for simple classification tasks
Configuration:
common:
target_device: 'CC1352'
training:
model_name: 'CLS_1k' # Small models
6.5.5. AM26x Family
The AM26x family uses Arm Cortex-R5F cores for real-time industrial applications with Ethernet connectivity.
Device |
Cores |
Features |
|---|---|---|
AM263 |
Quad R5F (400 MHz) |
Industrial Ethernet, high performance |
AM263P |
Quad R5F (400 MHz) |
Enhanced security features |
AM261 |
Single R5F (400 MHz) |
Cost-optimized industrial |
Key Features:
Real-time capable Cortex-R5F cores
Industrial Ethernet (EtherCAT, PROFINET, EtherNet/IP)
High-speed ADCs and PWMs
Suitable for larger ML models
Configuration:
common:
target_device: 'AM263' # or AM263P, AM261
training:
model_name: 'CLS_6k' # Can handle larger models
6.5.6. Typical Applications
Wireless Sensor Networks
Deploy ML-enabled sensors with wireless backhaul:
# Vibration sensor with Bluetooth reporting
common:
task_type: 'generic_timeseries_anomalydetection'
target_device: 'CC2755'
training:
model_name: 'AD_2k'
Use cases:
Structural health monitoring
Environmental sensing
Asset tracking with condition monitoring
Smart Home/Building
Local inference for privacy and responsiveness:
# Occupancy detection
common:
task_type: 'generic_timeseries_classification'
target_device: 'CC2755'
training:
model_name: 'CLS_2k'
Use cases:
Occupancy sensing
HVAC optimization
Security systems
Industrial IoT
Edge inference with industrial protocols:
# Motor monitoring with Ethernet reporting
common:
task_type: 'generic_timeseries_classification'
target_device: 'AM263'
dataset:
dataset_name: 'motor_fault_classification_dsk'
training:
model_name: 'CLS_6k'
Use cases:
Predictive maintenance
Quality inspection
Process anomaly detection
Long-Range IoT (Sub-GHz)
Remote sensing with minimal power:
# Remote vibration sensor
common:
task_type: 'generic_timeseries_anomalydetection'
target_device: 'CC1352'
training:
model_name: 'AD_500' # Minimal model
6.5.7. Power Optimization
Connectivity devices often run on batteries:
Duty Cycling
Run inference periodically, sleep between:
Wake on timer or sensor threshold
Perform inference
Transmit only if anomaly detected
Return to sleep
Model Size vs Battery Life
Smaller models use less energy per inference:
Model Size |
Inference Energy |
Battery Impact |
|---|---|---|
500 params |
~10 µJ |
Minimal |
2k params |
~50 µJ |
Low |
4k params |
~150 µJ |
Moderate |
Transmission Optimization
Send only alerts, not raw data
Batch non-urgent communications
Use lowest sufficient TX power
6.5.8. Memory Constraints
Connectivity devices have varying memory:
Device |
Flash |
RAM |
Recommended Model |
|---|---|---|---|
CC2755 |
1 MB |
256 KB |
Up to CLS_6k |
CC1352 |
352 KB |
80 KB |
Up to CLS_2k |
AM263 |
2 MB |
512 KB |
Up to CLS_13k |
Note: Wireless stack consumes significant memory. Plan model size accordingly.
6.5.9. Development Tools
Code Composer Studio (CCS)
Install SimpleLink SDK for CC27xx/CC13xx
Install AM26x SDK for AM26x devices
Use SysConfig for wireless stack configuration
SimpleLink SDK
Wireless protocol stacks
Example applications
Power management
TI 15.4-Stack
For Sub-GHz mesh networks with CC13xx.
Industrial Communications SDK
For AM26x industrial Ethernet protocols.
6.5.10. Wireless Protocol Considerations
Choose protocol based on application needs:
Protocol |
Device |
Best For |
|---|---|---|
Bluetooth LE |
CC2755, CC1352 |
Short range, smartphone integration |
Thread/Zigbee |
CC2755 |
Home automation mesh |
Sub-GHz |
CC1352 |
Long range, building penetration |
Industrial Ethernet |
AM26x |
Factory automation, real-time |
6.5.11. Getting Started
Choose device based on connectivity needs
Install appropriate SDK
Account for wireless stack memory usage
Select model size within remaining memory
Test inference + communication together
# Example: BLE sensor node
common:
task_type: 'generic_timeseries_anomalydetection'
target_device: 'CC2755'
dataset:
dataset_name: 'your_sensor_dataset'
training:
model_name: 'AD_2k'
training_epochs: 30
compilation:
enable: True
6.5.12. Next Steps
See Device Overview for complete device list
Read Non-NPU Deployment for CPU inference
Explore Anomaly Detection for sensor monitoring