May 19, 2026 Embedded Development

MEMS Sensor Selection Guide: Accelerometer/Gyroscope Comparison

When doing embedded development, sensor selection is an unavoidable hurdle. Especially for MEMS sensors - accelerometers, gyroscopes, IMUs - there are so many models, prices ranging from a few yuan to hundreds, making it easy for beginners to fall into pitfalls.

Today we’ll discuss how to choose MEMS sensors, and compare several popular models to help you avoid detours.

What Exactly Are MEMS Sensors?

MEMS (Micro-Electro-Mechanical Systems) are miniature electromechanical systems. Simply put, they integrate mechanical structures (like cantilever beams, proof masses) and circuits on a single chip.

There are three common types of MEMS sensors:

Accelerometer: Measures linear acceleration, can sense tilt, vibration, shock
Gyroscope: Measures angular velocity, can sense rotation, turning
IMU (Inertial Measurement Unit): Combination of accelerometer + gyroscope, some also include magnetometer

First step in selection: Understand what you need to measure.

Want to make a pedometer, tilt detection? → Accelerometer is enough
Want gesture recognition, attitude control? → Need gyroscope
Want to make drones, balance cars, VR devices? → Go directly for IMU

Popular Model Comparison

There are quite a few common MEMS sensor models on the market. I’ve picked 5 commonly used ones to compare:

Model	Type	Accelerometer Range	Gyroscope Range	Interface	Price	Application Scenario
MPU6050	IMU	±2/4/8/16g	±250/500/1000/2000°/s	I2C	¥8-15	Best for beginners, balance cars, drones
ICM20948	IMU	±2/4/8/16g	±250/500/1000/2000°/s	I2C/SPI	¥25-40	High precision, VR/AR, gesture recognition
BMI088	IMU	±3/6/12/24g	±125/250/500/1000/2000°/s	I2C/SPI	¥30-50	Industrial grade, robots, vibration analysis
ADXL345	Accelerometer	±2/4/8/16g	None	I2C/SPI	¥10-20	Pure acceleration measurement, tilt detection
L3GD20H	Gyroscope	None	±245/500/2000°/s	I2C/SPI	¥12-25	Pure angular velocity measurement, turning detection

Selection Recommendations:

Beginner learning: MPU6050, cheap, lots of documentation, rich libraries
High precision needs: ICM20948, low noise, small temperature drift
Industrial applications: BMI088, shock resistant, wide temperature range
Cost sensitive: ADXL345 or L3GD20H, single function is sufficient

Hardware Connection Example

Taking MPU6050 as an example, wiring is very simple:

MPU6050	ESP32	Description
VCC	3.3V	Power (some modules support 5V)
GND	GND	Ground
SCL	GPIO21	I2C clock
SDA	GPIO22	I2C data
INT	GPIO15	Interrupt (optional)
ADO	GND	I2C address selection (0x68 or 0x69)

Note: MPU6050 is a 3.3V device. If using a 5V microcontroller (like Arduino Uno), you need level shifting.

Code Practice: Reading Sensor Data

Below we use ESP32 + MPU6050 to demonstrate how to read acceleration and gyroscope data.

1. Install Library

# PlatformIO
pio lib install "MPU6050 by Electronic Cats"

# Arduino IDE
# Search "MPU6050" in library manager and install

2. Basic Reading Code

#include <Wire.h>
#include <Adafruit_MPU6050.h>
#include <Adafruit_Sensor.h>

MPU6050 mpu;

void setup() {
  Serial.begin(115200);
  Wire.begin();

  // Initialize MPU6050
  if (!mpu.begin()) {
    Serial.println("MPU6050 initialization failed, check wiring!");
    while (1);
  }

  // Configure range
  mpu.setAccelerometerRange(MPU6050_RANGE_8_G);
  mpu.setGyroRange(MPU6050_RANGE_500_DEG);
  mpu.setFilterBandwidth(MPU6050_BAND_21_HZ);

  Serial.println("MPU6050 initialized successfully!");
}

void loop() {
  sensors_event_t a, g, temp;
  mpu.getEvent(&a, &g, &temp);

  // Print acceleration (unit: m/s²)
  Serial.print("Accel X: "); Serial.print(a.acceleration.x);
  Serial.print(" Y: "); Serial.print(a.acceleration.y);
  Serial.print(" Z: "); Serial.println(a.acceleration.z);

  // Print angular velocity (unit: rad/s)
  Serial.print("Gyro X: "); Serial.print(g.gyro.x);
  Serial.print(" Y: "); Serial.print(g.gyro.y);
  Serial.print(" Z: "); Serial.println(g.gyro.z);

  delay(100);
}

3. Calculate Tilt Angle

Accelerometer can calculate static tilt angle:

float getPitch() {
  sensors_event_t a, g, temp;
  mpu.getEvent(&a, &g, &temp);

  // pitch = atan2(-accX, sqrt(accY² + accZ²))
  float pitch = atan2(-a.acceleration.x, 
                      sqrt(a.acceleration.y * a.acceleration.y + 
                           a.acceleration.z * a.acceleration.z));
  return pitch * 180 / PI;  // Convert to degrees
}

float getRoll() {
  sensors_event_t a, g, temp;
  mpu.getEvent(&a, &g, &temp);

  // roll = atan2(accY, accZ)
  float roll = atan2(a.acceleration.y, a.acceleration.z);
  return roll * 180 / PI;
}

Note: Tilt angles calculated by accelerometer are inaccurate in dynamic scenarios (because motion acceleration interferes). In this case, you need to fuse gyroscope data using Kalman filter or complementary filter.

Common Problem Troubleshooting

Problem 1: All Readings Are 0

Possible causes:

Wiring error (SCL/SDA reversed)
I2C address incorrect (MPU6050 defaults to 0x68, ADO connected to VCC becomes 0x69)
Insufficient power voltage

Solution:

// Scan I2C devices
void scanI2C() {
  byte count = 0;
  for (byte addr = 1; addr < 127; addr++) {
    Wire.beginTransmission(addr);
    if (Wire.endTransmission() == 0) {
      Serial.print("Found device: 0x");
      Serial.println(addr, HEX);
      count++;
    }
  }
  if (count == 0) Serial.println("No I2C devices found");
}

Problem 2: Data Noise Is Large, Jumping Obvious

Possible causes:

Mechanical vibration interference
Large power supply ripple
Filter bandwidth set too high

Solution:

// Reduce filter bandwidth (trade response speed for stability)
mpu.setFilterBandwidth(MPU6050_BAND_5_HZ);

// Software filtering: moving average
#define SAMPLE_COUNT 10
float readAccelerometerX() {
  float sum = 0;
  for (int i = 0; i < SAMPLE_COUNT; i++) {
    sensors_event_t a, g, temp;
    mpu.getEvent(&a, &g, &temp);
    sum += a.acceleration.x;
    delay(2);
  }
  return sum / SAMPLE_COUNT;
}

Problem 3: Gyroscope Drift Is Severe

Symptom: Angular velocity doesn’t return to zero when stationary, angle continues to drift after integration.

Cause: Gyroscope has zero bias, needs calibration.

Solution:

// Calibrate zero bias (run when device is stationary)
float gyroBiasX = 0, gyroBiasY = 0, gyroBiasZ = 0;
void calibrateGyro() {
  Serial.println("Calibrating... Please keep device stationary");
  delay(1000);

  for (int i = 0; i < 100; i++) {
    sensors_event_t a, g, temp;
    mpu.getEvent(&a, &g, &temp);
    gyroBiasX += g.gyro.x;
    gyroBiasY += g.gyro.y;
    gyroBiasZ += g.gyro.z;
    delay(5);
  }

  gyroBiasX /= 100;
  gyroBiasY /= 100;
  gyroBiasZ /= 100;

  Serial.print("Bias: X="); Serial.print(gyroBiasX);
  Serial.print(" Y="); Serial.print(gyroBiasY);
  Serial.print(" Z="); Serial.println(gyroBiasZ);
}

// Subtract bias when using
float getGyroX() {
  sensors_event_t a, g, temp;
  mpu.getEvent(&a, &g, &temp);
  return g.gyro.x - gyroBiasX;
}

Advanced: Sensor Fusion

Single sensors have limitations:

Accelerometer: Accurate when static, but motion interferes when dynamic
Gyroscope: Accurate when dynamic, but long-term integration drifts

Solution: Use complementary filter or Kalman filter to fuse both.

Complementary filter is simple and effective:

float pitch = 0;
void updatePitch(float dt) {
  sensors_event_t a, g, temp;
  mpu.getEvent(&a, &g, &temp);

  // Pitch calculated from accelerometer
  float accPitch = atan2(-a.acceleration.x, 
                         sqrt(a.acceleration.y * a.acceleration.y + 
                              a.acceleration.z * a.acceleration.z)) * 180 / PI;

  // Pitch integrated from gyroscope
  pitch = pitch + g.gyro.y * dt;

  // Complementary filter: 98% trust gyroscope, 2% trust accelerometer
  pitch = 0.98 * pitch + 0.02 * accPitch;
}

Selection Summary

Finally, here’s a quick selection table:

Requirement	Recommended Model	Reason
Beginner learning, low cost	MPU6050	Cheap, lots of documentation, sufficient
High precision attitude calculation	ICM20948	Low noise, 9-axis (with magnetometer)
Industrial vibration monitoring	BMI088	High range, shock resistant, wide temperature
Pure tilt detection	ADXL345	Single function, low power
Pure rotation detection	L3GD20H	Single function, cost-effective

Procurement Suggestions:

Buy modules (with voltage regulation and level shifting) on Taobao/1688, more convenient than bare chips
Note the difference between “breakout boards” and “development boards” - the former is just sensor + minimal peripherals
For bulk procurement, you can get chips from distributors, reducing cost by 30-50%

Hope this blog post is helpful to you!