BLE GATT sensor bridge for Edge Impulse - Edge Impulse Documentation

This project runs in two modes from the same codebase. In local sensor mode a board with an on-board IMU (Nesso N1, Arduino Nano 33 BLE Sense) reads sensor data, optionally runs Edge Impulse inference, and advertises results via BLE GATT to an Android phone. In relay mode a Nordic Thingy:53 connects to an EI-Golioth peripheral and forwards its inference notifications to Android.
Reference code: https://github.com/edgeimpulse/ei-zephyr-ble-gatt-client

Overview

  LOCAL SENSOR MODE                       RELAY MODE
  ─────────────────                       ──────────
  ┌────────────────┐  BLE GATT            ┌────────────────┐  BLE GATT (client)  ┌───────────────┐
  │  Nesso N1      │ ──────────────────►  │  Thingy:53     │ ◄────────────────── │  EI-Golioth   │
  │  (ESP32-C6)    │  sensor + inference  │                │   inference notify  │  (nRF52840)   │
  │                │                      │                │                     └───────────────┘
  │  BMI270 IMU    │                      │  Relay + serve │
  └────────────────┘                      └───────┬────────┘
                                                  │ BLE GATT (server)
  ┌────────────────┐  BLE GATT                    │
  │  Nano 33 BLE   │ ──────────────────►          ▼
  │  (nRF52840)    │  sensor + inference   ┌──────────────┐
  │                │                       │  Android app │
  │  LSM9DS1 IMU   │                       └──────────────┘
  └────────────────┘

Both modes expose the same GATT service so the same Android app connects to either without changes.

Modes at a Glance

	Local Sensor mode	Relay mode
Board	Nesso N1 (ESP32-C6) or Nano 33 BLE Sense	Nordic Thingy:53
BLE role	Peripheral (server)	Central (client) + Peripheral (server)
Data source	On-board BMI270 / LSM9DS1	EI-Golioth peripheral over BLE
Edge Impulse inference	Optional (drop in a `model/` dir)	On the peripheral, not here
Kconfig	`CONFIG_EI_SENSOR_LOCAL=y` (auto via board conf)	default (`CONFIG_EI_SENSOR_LOCAL=n`)

Hardware

Sensor-side boards (local sensor mode)

Board	SoC	IMU	Notes
Arduino Nesso N1	ESP32-C6 (RISC-V)	Bosch BMI270 (6-axis)	Requires `west blobs fetch hal_espressif` before building; needs Zephyr ≥ v4.1.0
Arduino Nano 33 BLE Sense	nRF52840	ST LSM9DS1 (6-axis) + HTS221 + LPS22HB + APDS9960	All sensors on-board, no extra config needed
Any board with a supported IMU	—	LSM9DS1, BMI270, or FXOS8700	Add a `boards/<board>.overlay` and a matching `.conf`

Monitor board (relay mode)

Board	SoC	Notes
Nordic Thingy:53	nRF5340	BLE 5.3, 1350 mAh Li-Po; no on-board sensors used

Prerequisites

Zephyr SDK 1.0+ and West 1.5.0+
For Nesso N1 only: the riscv64-zephyr-elf toolchain (install with ./setup.sh -t riscv64-zephyr-elf from the SDK directory) and the Espressif HAL blobs (see Step 1b)
For relay mode: a running EI-Golioth peripheral (see example-edge-impulse)

1. Initialize the Repository

west init -m https://github.com/edgeimpulse/ei-zephyr-ble-gatt-client.git
cd ei-zephyr-ble-gatt-client
west update

west update fetches Zephyr RTOS main (required for the Nesso N1 board), the Edge Impulse Zephyr SDK module, and all dependencies.

1b. Nesso N1: Fetch Espressif Blobs

ESP32 targets require closed-source binary blobs for the BLE and Wi-Fi stack. Fetch them once after west update:

west blobs fetch hal_espressif

The Arduino Nesso N1 board (arduino_nesso_n1/esp32c6/hpcore) is provided by Zephyr main. Make sure west.yml pins zephyr to revision: main and that you have built the riscv64-zephyr-elf toolchain in your Zephyr SDK install. Export ZEPHYR_SDK_INSTALL_DIR so CMake picks up the right SDK:

export ZEPHYR_SDK_INSTALL_DIR=$HOME/zephyr-sdk-1.0.1

2. (Optional) Add an Edge Impulse Model

If you want the sensor board to run local inference and notify the result label over BLE:

In Edge Impulse Studio go to Deployment → Zephyr library, click Build, and download the .zip
Extract the archive into a model/ directory next to the project root:

ei-zephyr-ble-gatt-client/
model/
  CMakeLists.txt
  edge-impulse-sdk/
  model-parameters/
  tflite-model/

CMakeLists.txt auto-detects the model/ directory and links the SDK at build time. Without it the firmware streams raw sensor data only.

3. Build

Nesso N1 (local sensor mode)

west build --pristine -b arduino_nesso_n1/esp32c6/hpcore

The board qualifier /esp32c6/hpcore selects the RISC-V HP application core. prj.conf enables CONFIG_EI_SENSOR_LOCAL=y, CONFIG_BMI270=y, CONFIG_I2C=y, and CONFIG_SENSOR=y so no manual Kconfig changes are needed.

Arduino Nano 33 BLE Sense (local sensor mode)

west build --pristine -b arduino_nano_33_ble

Nordic Thingy:53 (relay mode)

west build --pristine -b thingy53/nrf5340/cpuapp

Or set your board once in .west/config:

[build]
board = arduino_nesso_n1/esp32c6/hpcore

Then just run west build --pristine.

4. Flash

west flash

For Nordic boards you can specify a runner:

west flash --runner jlink
west flash --runner nrfjprog

For Nesso N1 / ESP32-C6, west flash uses the built-in ESP-IDF flasher via the ESP32-C6’s native USB-Serial-JTAG port — no external programmer or BOOT/RESET dance required. Just plug in via USB-C and run west flash.

5. Monitor Serial Output

# Nordic boards
ls /dev/tty.usbmodem*
minicom -D /dev/tty.usbmodem14401 -b 115200

# Nesso N1 (ESP32-C6) — use the Espressif monitor tool
west espressif monitor
# or
ls /dev/tty.usbserial*
minicom -D /dev/tty.usbserial-0001 -b 115200

Local sensor mode (Nesso N1 / Nano 33 BLE Sense)

Boot log on a Nesso N1 (ESP32-C6) with west espressif monitor:

*** Booting Zephyr OS build v4.4.0-3391-gc2961410b98f ***
========================================
  Edge Impulse BLE GATT Client
  Mode: Local Sensor Collection
  Board: arduino_nesso_n1/esp32c6/hpcore
========================================
[00:00:00.137,000] <inf> bt_hci_core: HW Platform: Espressif Systems (0x0004)
[00:00:00.137,000] <inf> bt_hci_core: HW Variant: ESP32-C6 (0x0005)
[00:00:00.137,000] <inf> bt_hci_core: Identity: 58:8C:81:50:37:92 (public)
[00:00:00.137,000] <inf> gatt_client: Bluetooth initialized
[00:00:00.139,000] <inf> gatt_server: EI GATT server advertising as "EI-Monitor"
[00:00:00.139,000] <inf> ei_sensor: BMI270 initialised
Starting local sensor collection (sampling every 10 ms)...
[00:00:00.139,000] <inf> ei_sensor: Local sensor loop started (10 ms interval)

Without a model the firmware streams raw sensor samples:

Sensor data: ax=-0.12 ay=9.81 az=0.04 gx=0.01 gy=-0.02 gz=0.00
Sensor data: ax=-0.11 ay=9.80 az=0.05 gx=0.00 gy=-0.01 gz=0.01

With a model (model/ present at build time):

=== Inference Result ===
Label: idle
Confidence: 91.2%
DSP Time: 52 ms
Classification Time: 8 ms
========================

Relay mode (Thingy:53)

========================================
  Edge Impulse BLE GATT Client
  Mode: BLE Relay (EI-Golioth Monitor)
========================================

Scanning for EI-Golioth devices...

Device found: EI-Golioth (RSSI: -45 dBm)
Connecting...

>>> Connected to EI-Golioth device!

=== Inference Result ===
Label: wave
Confidence: 97.0%
DSP Time: 47 ms
Classification Time: 6 ms
Total Time: 53 ms
========================

How It Works

Local sensor mode

BLE stack + GATT server start advertising as EI-Monitor
ei_sensor_init() binds to the on-board IMU via the Zephyr Sensor API (device resolved at compile time from devicetree)
ei_sensor_run_loop() samples the IMU every CONFIG_EI_SENSOR_SAMPLE_INTERVAL_MS milliseconds (default 10 ms = 100 Hz)
Each sample ([accel_x, accel_y, accel_z, gyro_x, gyro_y, gyro_z] in m/s² and rad/s) is notified to connected Android centrals via gatt_server_notify_sensor_data()
If a model is present and the feature buffer is full, run_classifier() fires and the label/confidence is notified via gatt_server_notify_inference()

Relay mode

BLE stack starts in both Central and Peripheral roles (CONFIG_BT_MAX_CONN=2)
GATT client scans for EI-Golioth by name, connects, discovers the EI service, and subscribes to the inference characteristic
GATT server simultaneously advertises as EI-Monitor so Android can connect
Inference notifications from the peripheral are forwarded to Android via gatt_server_notify_inference()

Sensor driver (`ei_sensor.cpp`)

The IMU is selected at compile time via DT_HAS_COMPAT_STATUS_OKAY():

Devicetree compatible	Board	Channels
`st,lsm9ds1`	Nano 33 BLE Sense	`ACCEL_XYZ`, `GYRO_XYZ`
`bosch,bmi270`	Nesso N1	`ACCEL_XYZ`, `GYRO_XYZ`
`nxp,fxos8700`	NXP boards	`ACCEL_XYZ`, `GYRO_XYZ`

GATT service layout

Characteristic	UUID suffix	Properties	Payload
Inference result	`...def1`	READ, NOTIFY	`inference_result_t` (label, confidence, timing)
Sensor data	`...def2`	READ, NOTIFY	`float[]` — raw IMU axes
Device state	`...def3`	READ, WRITE	Status flags

Project Structure

ei-zephyr-ble-gatt-client/
├── CMakeLists.txt                          # Auto-detects model/ for optional inference
├── Kconfig                                 # EI_SENSOR_LOCAL, EI_SENSOR_SAMPLE_INTERVAL_MS
├── prj.conf                                # BLE central + peripheral, C++17, logging
├── west.yml                                # Zephyr v4.0.0 + EI SDK Zephyr module
├── boards/
│   ├── arduino_nesso_n1.conf              # Sets EI_SENSOR_LOCAL=y, CONFIG_BMI270=y
│   ├── arduino_nesso_n1.overlay           # Placeholder (BMI270 in upstream DTS)
│   ├── arduino_nano_33_ble.conf           # Sets EI_SENSOR_LOCAL=y, all sensor drivers
│   ├── arduino_nano_33_ble.overlay        # Enables all on-board sensors
│   └── thingy53_nrf5340_cpuapp.overlay    # Routes UART0 for serial console
└── src/
    ├── main.cpp                            # Mode switch, startup, callbacks
    ├── ble/
    │   ├── gatt_client.cpp / .h           # BLE Central: scan, connect, subscribe
    │   └── gatt_server.cpp / .h           # BLE Peripheral: advertise, notify Android
    └── sensors/
        ├── ei_sensor.cpp                   # IMU read loop, optional EI inference
        └── ei_sensor.h                     # Public API + EI_SENSOR_IMU_AXES constant

Customising the Example

Adjust the sensor sampling rate

In prj.conf or a board-specific .conf:

CONFIG_EI_SENSOR_SAMPLE_INTERVAL_MS=20   # 50 Hz instead of 100 Hz

Port to a different board

Add boards/<your_board>.overlay and boards/<your_board>.conf
In .conf, set CONFIG_EI_SENSOR_LOCAL=y and enable the IMU driver (e.g. CONFIG_ICM42688_P=y)
In .overlay, enable the IMU node if it is not already on in the upstream board DTS
Add a DT_HAS_COMPAT_STATUS_OKAY(your_compat) branch in src/sensors/ei_sensor.cpp

Change the target peripheral name (relay mode)

In src/ble/gatt_client.cpp:

static const char *target_device_name = "MY-DEVICE";

Increase memory for larger EI models

CONFIG_HEAP_MEM_POOL_SIZE=65536
CONFIG_MAIN_STACK_SIZE=32768

Troubleshooting

Nesso N1 build error: board not found

The Nesso N1 was added after Zephyr v4.0.0. Update west.yml to revision: v4.1.0 (or main) and run west update again.

Nesso N1 flash error or blank serial output

ESP32-C6 requires binary blobs. Run west blobs fetch hal_espressif once after west update. The Nesso N1 uses native USB-Serial-JTAG so no BOOT/RESET sequence is needed — flashing is fully automatic over USB-C. If the serial console is blank, use west espressif monitor instead of minicom to open the port.

IMU not detected at runtime (local sensor mode)

Check the serial log for the device not ready error — it prints the compatible string it searched for
Verify the IMU node has status = "okay" in the devicetree

Enable I²C and sensor debug logs:

CONFIG_I2C_LOG_LEVEL_DBG=y
CONFIG_SENSOR_LOG_LEVEL_DBG=y

Android cannot connect (local sensor mode)

Confirm the board is advertising; a BLE scan on the phone should show EI-Monitor
Check that gatt_server_init() completed without error in the serial log
Only one Android central can connect at a time (CONFIG_BT_MAX_CONN=2 reserves one slot for the EI-Golioth peripheral in relay mode)

Relay mode: no EI-Golioth device found during scan

Confirm the peripheral is powered on and advertising as EI-Golioth
Enable BLE scan debug output:
```
CONFIG_BT_LOG_LEVEL_DBG=y
```
The scan filter matches by exact advertised name — verify the peripheral’s CONFIG_BT_DEVICE_NAME matches exactly

No serial output on Thingy:53

Set your terminal to 115200 baud, 8N1. On macOS use ls /dev/tty.usbmodem* to find the port. The boards/thingy53_nrf5340_cpuapp.overlay already routes zephyr,console to uart0.

Next Steps

Zephyr series overview
IMU inference on Zephyr — build a standalone inference node
ei-zephyr-golioth-integration — the relay-mode peripheral with Golioth cloud sync

Documentation Index

​Overview

​Modes at a Glance

​Hardware

​Sensor-side boards (local sensor mode)

​Monitor board (relay mode)

​Prerequisites

​1. Initialize the Repository

​1b. Nesso N1: Fetch Espressif Blobs

​2. (Optional) Add an Edge Impulse Model

​3. Build

​Nesso N1 (local sensor mode)

​Arduino Nano 33 BLE Sense (local sensor mode)

​Nordic Thingy:53 (relay mode)

​4. Flash

​5. Monitor Serial Output

​Local sensor mode (Nesso N1 / Nano 33 BLE Sense)

​Relay mode (Thingy:53)

​How It Works

​Local sensor mode

​Relay mode

​Sensor driver (ei_sensor.cpp)

​GATT service layout

​Project Structure

​Customising the Example

​Adjust the sensor sampling rate

​Port to a different board

​Change the target peripheral name (relay mode)

​Increase memory for larger EI models

​Troubleshooting

​Next Steps

​Resources

Overview

Modes at a Glance

Hardware

Sensor-side boards (local sensor mode)

Monitor board (relay mode)

Prerequisites

1. Initialize the Repository

1b. Nesso N1: Fetch Espressif Blobs

2. (Optional) Add an Edge Impulse Model

3. Build

Nesso N1 (local sensor mode)

Arduino Nano 33 BLE Sense (local sensor mode)

Nordic Thingy:53 (relay mode)

4. Flash

5. Monitor Serial Output

Local sensor mode (Nesso N1 / Nano 33 BLE Sense)

Relay mode (Thingy:53)

How It Works

Local sensor mode

Relay mode

Sensor driver (`ei_sensor.cpp`)

GATT service layout

Project Structure

Customising the Example

Adjust the sensor sampling rate

Port to a different board

Change the target peripheral name (relay mode)

Increase memory for larger EI models

Troubleshooting

Next Steps

Resources