SHT31, HTU21D and AM2320 on One Arduino I²C Bus: Power, Pull-Ups and Valid Readings

How three sensors share one I²C bus

The Arduino host connects its SDA and SCL lines to all three modules and shares a common ground with them. Each device needs its own supply connection at a voltage its installed board supports. Before applying power, verify the host logic voltage and the rail feeding the SDA/SCL pull-up resistors; a shared data bus does not imply that every module can share an unchecked 5 V supply.

• SHT31: tie ADDR LOW for 7-bit address 0x44 or HIGH for 0x45; do not leave it floating.
• HTU21D: use 7-bit address 0x40. Check the actual breakout schematic because its regulator, if present, does not necessarily protect the SDA and SCL pins.
• AM2320: use 7-bit address 0x5C. Its wake behavior and slower sampling interval need their own firmware handling.

Those distinct addresses let the devices coexist, but a scan only establishes communication at an address. The electrical checks and complete reading procedure below determine whether the arrangement is safe and useful.

First check the voltage domains

Sensirion specifies SHT31 operation from 2.15 to 5.5 V, while TE lists the HTU21D sensor's normal supply as 1.5 to 3.6 V and a 3.8 V peak. Aosong specifies 3.1 to 5.5 V for AM2320. These ranges describe components, not the entire assortment of third-party modules sold under those names. A 5 V supply or a 5 V SDA/SCL pull-up can exceed the HTU21D chip's limit unless that specific breakout provides a suitable regulator and level translation. A regulator alone does not prove that its signal pins are shifted. SHT3x datasheet; TE sensor range table; AM2320 datasheet.

Practical choice: document every board's power path and logic levels, then choose a common bus rail or proper bidirectional level shifting. A nominal 3.3 V rail is within all three listed supply ranges, but AM2320 has only a 0.2 V margin above its 3.1 V minimum; verify the rail at the sensor under worst-case conditions. Check the Arduino host's input-high requirement too. Do not assume a 3.3 V pull-up is always read as HIGH by a 5 V host.

Count the pull-ups already on the bus

I²C SDA and SCL are open-drain lines. Several breakout boards may each carry pull-up resistors, while the host or another module adds more. Their parallel combination changes the effective resistance. Three 10 kΩ resistors in parallel produce roughly 3.33 kΩ; two 4.7 kΩ resistors produce about 2.35 kΩ. These are arithmetic examples, not universal recommended values.

Inspect or measure the installed network and record where it pulls: 3.3 V, 5 V, or another rail. Then verify rise time at the intended clock rate and LOW-level sink current using the weakest connected device's limits. A working short cable on a bench may fail after cable length, capacitance or module count changes. Sensirion's datasheet provides the SHT3x I²C electrical constraints; the other devices and the host must also pass.

Scan, wake, read, verify: keep four states separate

Begin at a conservative bus speed supported by all connected devices. Log the 7-bit address and error state rather than treating a scan as the measurement test. SHT31's ADDR input selects 0x44 when tied LOW or 0x45 when tied HIGH; Sensirion says not to leave it floating. AM2320's datasheet also expresses its bus address as the 8-bit write byte 0xB8, corresponding to 7-bit 0x5C. Its wake sequence can make a simplistic scanner or first transaction look like a missing sensor. Follow the documented wake/read timing and avoid polling it faster than its specified measurement interval.

This state model is more useful than a single sensor.begin() == true flag. It tells a technician whether a fault is electrical, protocol-related, data-integrity related or a genuine environmental mismatch. It also prevents a last-known-good value from masquerading as a new observation.

Decision order: first confirm the device responds at its expected address, then obtain a complete and valid sample with the correct driver, and only then compare its temperature and RH against an independent reference. A scan result alone should never be presented as a sensor accuracy result.

Arduino diagnostic code for all three sensors

This original example initializes each sensor separately, reads temperature and humidity, and prints a clear error when initialization or acquisition fails. It uses a 100 kHz I²C clock for AM2320 and waits two seconds between readings. Verify breakout voltage and logic levels before running it.

The example uses the documented APIs of the AM232X library, Adafruit HTU21DF and Adafruit SHT31. Keep device status and NaN checks visible so a failed read is not mistaken for a valid measurement.

// Original NYFEA three-sensor diagnostic example.
// Libraries: Adafruit HTU21DF, Rob Tillaart AM232X, Adafruit SHT31.
// Check each breakout's supply and I2C logic levels before applying power.

#include <Wire.h>
#include <Adafruit_HTU21DF.h>
#include <AM232X.h>
#include <Adafruit_SHT31.h>

Adafruit_HTU21DF htu;
AM2320 am2320;
Adafruit_SHT31 sht;

bool htuReady = false;
bool amReady = false;
bool shtReady = false;

void printPair(const char *name, float tempC, float humidity, bool valid) {
  Serial.print(name);
  Serial.print(F(": "));
  if (!valid || isnan(tempC) || isnan(humidity)) {
    Serial.println(F("read error"));
    return;
  }
  Serial.print(tempC, 2);
  Serial.print(F(" C, "));
  Serial.print(humidity, 2);
  Serial.println(F(" %RH"));
}

void setup() {
  Serial.begin(9600);
  Wire.begin();
  Wire.setClock(100000);  // AM2320 is specified for a 100 kHz bus.
  delay(1000);

  htuReady = htu.begin();
  shtReady = sht.begin(0x44);  // Use 0x45 only when ADDR is tied HIGH.
  amReady = am2320.begin();  // Required by current Rob Tillaart AM232X.

  Serial.println(F("Three-sensor I2C check"));
  Serial.print(F("HTU21D init: "));
  Serial.println(htuReady ? F("OK") : F("FAILED"));
  Serial.print(F("SHT31 init: "));
  Serial.println(shtReady ? F("OK") : F("FAILED"));
  Serial.print(F("AM2320 init: "));
  Serial.println(amReady ? F("OK") : F("FAILED"));
}

void loop() {
  if (htuReady) {
    const float t = htu.readTemperature();
    const float rh = htu.readHumidity();
    printPair("HTU21D", t, rh, true);
  } else {
    Serial.println(F("HTU21D: not initialized"));
  }

  if (amReady) {
    const int status = am2320.read();
    if (status == AM232X_OK) {
      printPair("AM2320", am2320.getTemperature(),
                am2320.getHumidity(), true);
    } else {
      Serial.print(F("AM2320: read error, status "));
      Serial.println(status);
    }
  } else {
    Serial.println(F("AM2320: not initialized"));
  }

  if (shtReady) {
    const float t = sht.readTemperature();
    const float rh = sht.readHumidity();
    printPair("SHT31", t, rh, true);
  } else {
    Serial.println(F("SHT31: not initialized"));
  }

  Serial.println(F("------------------------------"));
  delay(2000);  // Avoid polling AM2320 faster than its sample interval.
}

Keep the AM2320 status code and the NaN checks in the output. A returned number without a valid status should not enter the comparison.

Use the failure pattern to choose the next test

Do not treat every bad value as a calibration problem. Record the first gate that fails, then run the corresponding check.

CRC is device-specific: SHT31 places CRC-8 after each temperature and RH word; AM2320 uses a different CRC-16 frame. A library's successful begin() result does not replace checking the actual read status.

When three sensors disagree, compare air before comparing brands

Place the sensing openings in the same representative airflow, away from the Uno regulator, display backlight, warm wiring and direct breath. Log temperature and RH together, with timestamps and valid-data flags. Let the assemblies equilibrate after handling, power-up and any enclosure change. A warmer local sensor often reports lower RH even when all devices share the same moisture content; a humidity-only ranking can therefore misidentify a thermal placement issue.

How much can a small temperature difference matter? At unchanged water-vapor partial pressure, air that is 50% RH at 25 °C would be about 44.4% RH at 27 °C, using the saturation-vapor-pressure relation. The roughly 5.6 percentage-point drop is a calculation, not an NYFEA test result or a correction to apply blindly. It shows why a board warmed by a nearby regulator can look “dry” without any change in moisture content.

Use an independent reference if the outcome will support a design decision. Compare repeated stable windows, not a single screenshot of three numbers. Record the reference's uncertainty and the product's allowable error before naming a “winner.” If a device differs, swap its physical position with another device while keeping its electrical channel and code traceable. A position-dependent difference suggests airflow or thermal bias; a device-dependent difference merits calibration, contamination and protocol investigation.

Use this comparison method to separate sensor differences from placement, temperature and firmware effects. Do not infer accuracy from a single set of readings.

FAQ