Desk presence detection using an HLK-LD2420 24GHz radar sensor and LaskaKit ESP32-LPKit.

• HLK-LD2420 datasheet (PDF)

Python 3 and PlatformIO:

python3 -m venv venv
source venv/bin/activate       # fish: source venv/bin/activate.fish
pip install platformio

On macOS you also need the CH9102 USB driver for the LaskaKit programmer. Install from https://www.wch.cn/downloads/CH34XSER_MAC_ZIP.html if /dev/cu.wchusbserial* doesn't appear when the board is plugged in.

The LD2420 module has 5 pins. Some modules don't expose a TX pin — OT1 acts as serial data output instead.

  LD2420 sensor          ESP32-LPKit
  +-----------+          +-----------+
  | VCC ------+------>---+ 3.3V      |
  | GND ------+------>---+ GND       |
  | OT1 ------+------>---+ GPIO17    |  (sensor data out -> ESP32 receives)
  | RX  ------+----<-----+ GPIO16    |  (ESP32 sends -> sensor data in)
  | OT2       |          |           |  (not connected)
  +-----------+          +-----------+

pio run -t upload && pio device monitor

The LD2420 protocol driver lives in lib/LD2420/ and can be copied into other PlatformIO projects. Usage:

#include <LD2420.h>

LD2420 radar(Serial2, 17, 16);   // UART, RX pin, TX pin

void setup() {
    radar.setMaxDistance(2.1);    // 2.1m = gate 3
    radar.setTimeout(10);
    radar.setDebug(&Serial);     // optional: verbose config output
    radar.begin();
}

void loop() {
    radar.update();
    if (radar.isPresent()) {
        // someone is at the desk
    }
}

For full gate threshold control, build an LD2420Config struct and pass it via radar.setConfig(cfg) before begin().

• MCU: LaskaKit ESP32-LPKit (ESP32-WROOM) with CH9102 USB-C programmer
• Sensor: HLK-LD2420 24GHz radar presence sensor (3.3V, powered directly from ESP32)
• Firmware version: v1.6.1
• Serial port: /dev/cu.wchusbserial*

• UART2 pins: RX=GPIO17, TX=GPIO16 (swapped from what pinout suggests — verified working)
• I2C power gate: GPIO4 must be HIGH to enable I2C (SDA=21, SCL=22) — not used in this project
• Programmer shows as /dev/cu.wchusbserial* (CH9102 driver)

All frames are little-endian.

Command frames:

FD FC FB FA | length(2) | command(2) | payload(N) | 04 03 02 01

• Length field = 2 (command bytes) + payload size

Energy data frames (output at ~10 Hz when in Energy mode):

F4 F3 F2 F1 | length(2) | presence(1) | distance(2) | gate_energy(2) × 16 | F8 F7 F6 F5

Simple mode output (factory default): ASCII text: ON\r\n or OFF\r\n followed by Range XX\r\n (distance in cm)

Read gate params response: [value(4)] × N starting at byte 10.

The sensor infers the register count from the frame length. Do NOT prepend a register count — the sensor interprets those bytes as a register address, silently writing to the wrong location.

Correct single register write (cmd 0x0007):

FD FC FB FA  08 00  07 00  [reg_addr 2B] [value 4B]  04 03 02 01

Wrong (what we had initially — caused all writes to be garbled):

FD FC FB FA  0A 00  07 00  01 00  [reg_addr 2B] [value 4B]  04 03 02 01
                           ^^^^^ sensor reads this as reg 0x0001

Payload is always 6 bytes: param_id(2) + mode(2) + padding(2)

1. Enter config mode (0x00FF with protocol version 0x0002)
2. Write min_gate + max_gate + timeout in ONE 0x0007 frame (3 register pairs, 18 bytes payload)
3. Loop gates 0-15: write move+still thresholds per gate (2 register pairs per 0x0007 frame, 12 bytes payload, 125μs delay between gates)
4. Set system mode (0x0012)
5. Exit config mode (0x00FE — saves to NVM)

Factory defaults:

• Gate 0: move=60000, still=40000
• Gate 1: move=30000, still=20000
• Gate 2: move=400, still=200
• Gates 3-6: move=250, still=200
• Gate 7+: move=250, still=100-150

Thresholds are energy values — higher = less sensitive. Set to 65535 to effectively disable a gate.

Tuning approach:

1. Run in Energy mode with nobody present, note idle energy per gate
2. Set thresholds ~50% above idle energy
3. Walk through detection zone, verify presence triggers
4. For through-wall blocking: set gates beyond the room to 65535 AND reduce MAX_GATE as a hard cutoff

Response command byte = sent command | 0x0100. Status at bytes 4-5: 0 = success.

• MIN_GATE=0, MAX_GATE=3 (0-2.1m, covers person sitting at desk)
• TIMEOUT=10s
• Gates 0-3: factory-default thresholds
• Gates 4-15: disabled (65535)
• Energy mode active — per-gate energy output for monitoring
• Through-wall detection to sleeping room (behind 15cm brick wall) confirmed blocked

pio run -t upload && pio device monitor

• #define RAW_DUMP — raw hex byte dump for debugging wiring/baud issues
• Comment it out — full config + monitoring (Energy or Simple mode auto-detected)

1. Always do a raw byte dump first — confirms wiring before touching protocol
2. The LD2420 has no TX pin on some modules — OT1 serves as serial output
3. Protocol has no num_regs field — this was the main bug; sensor infers count from length
4. MAX_GATE is a hard cutoff — most effective way to prevent through-wall detection
5. Config is saved to NVM on exit config — persists across power cycles
6. ESPHome source is the authoritative protocol reference — manufacturer docs are incomplete
7. Loose wires = silent failure — sensor just stops responding, solder connections for production