One-wire sensor timing

Step through the exchange. Every step is the same wire, changing hands.

One wire, two owners

The line idles high, held there by a pull-up resistor — usually already on the module. Your pin is an input, doing nothing.

To ask for a reading you make the pin an output and pull the line low for a millisecond or more. Then you let go, switch back to input, and from that moment the sensor is driving the same wire. It answers with a low, then a high, then forty bits, then releases the line and goes quiet for at least two seconds.

Nothing in that sequence is negotiated. Both ends simply agree, in advance, what the timings mean. That is what makes it cheap and what makes it brittle.

The bit is the pulse width

Every data bit starts with a 50 µs low that carries no information at all. What follows is either a short high (26–28 µs) or a long one (70 µs). Your code times that high pulse and compares it with about 50 µs. That comparison is the entire decoder.

What actually goes wrong

Reading too often. A DHT22 needs two seconds between reads, a DHT11 one. Ask sooner and it returns the previous sample or nothing at all.

No pull-up. Bare sensors need a 4.7 kΩ or 10 kΩ resistor from data to 3V3. Most modules already have one; a three-pin board almost certainly does, a four-pin bare sensor does not.

Long wires. Beyond about 20 cm the edges soften and the pulse widths blur. This protocol has no error correction, only detection — you find out by getting nothing.

Sharing the pin. One sensor per pin. There is no address, so two DHTs on one wire both answer at once and neither can be understood.

Use the library

#include <DHT.h>
#define DHTPIN  4

DHT dht(DHTPIN, DHT22);

void setup() {
  Serial.begin(115200);
  dht.begin();
}

void loop() {
  float t = dht.readTemperature();
  float h = dht.readHumidity();
  if (isnan(t) || isnan(h)) Serial.println("bad read");   // expect a few
  else Serial.printf("%.1f C  %.1f %%\n", t, h);
  delay(2000);                                            // the minimum
}

Timing this by hand is a good exercise and a bad plan: the library already disables interrupts around the critical section and retries. Checking for nan is not optional — a few failed reads per hundred is normal, not a fault.

The name collision

This is not the same thing as 1-Wire with a capital W. Dallas 1-Wire is a real bus with 64-bit addresses and many devices on one line, and it is the next page. The DHT just happens to use one wire, which is a description, not a protocol.