I²C Address Converter

Why one device has two addresses

An I²C address is seven bits, but nothing on the bus is seven bits wide. Every transfer begins with a whole byte, so the address is shifted up one place and the spare low bit carries the direction: 0 to write, 1 to read.

A device with the 7-bit address 0x48 therefore appears on the bus as 0x90 when written to and 0x91 when read from. All three numbers describe the same device. Datasheets print whichever the author preferred, sometimes without saying which, and that is the entire source of the confusion.

The quickest sanity check: a 7-bit address never exceeds 0x7F, and an 8-bit write byte is always even. If a number is above 0x7F it is already shifted; if it is odd it is a read byte.

Reserved addresses

The specification holds back sixteen of the 128 addresses, which leaves 112 for devices. Both reserved blocks sit at the ends of the range:

If a calculation lands a device in one of these blocks, the most likely explanation is that an 8-bit value was treated as a 7-bit one. Halving it usually produces a sensible address.

Reading a bus scan

A scanner walks every address from 0x08 to 0x77, sends the write byte, and records which ones acknowledge. The grid above uses the same sixteen-column layout as i2cdetect, so a scan result can be compared cell for cell, and selecting a cell shows the parts that commonly answer there.

Two failure patterns are worth recognising immediately. A scan that finds nothing at all is almost always electrical: no pull-up resistors, SDA and SCL swapped, or the device unpowered. A scan that findsevery address means SDA is being held low, which the scanner cannot distinguish from an acknowledgement. Neither is an addressing problem, and neither is fixed by changing the address.

Choosing an address in a block

Many parts expose address pins so several can share a bus. A PCF8574 has three, giving eight addresses from 0x20 to 0x27; a 24Cxx EEPROM does the same from 0x50 to 0x57; most sensors have one pin and two choices, such as the BME280 at 0x76 or 0x77.

Tie those pins deliberately rather than leaving them floating. A floating address pin produces a device that appears at an address that drifts, which presents as a peripheral that works until something nearby changes.

When two parts genuinely cannot be separated, an I²C multiplexer such as the TCA9548A gives each one its own downstream segment, and the multiplexer itself occupies a single address.

10-bit addressing

The 10-bit scheme exists to extend the address space, and it works by borrowing the reserved0x78–0x7B block. The first byte is 11110 followed by the top two address bits and the direction bit; the second byte carries the remaining eight.

In practice it is rare. Seven-bit devices ignore the reserved prefix so the two schemes coexist on one bus, but very few parts implement 10-bit addressing and many controllers support it only partially. If you are running out of addresses, a multiplexer is usually the more practical answer.

When the address is right and it still does not answer

Addressing is only the first byte. If a scan sees the device but transfers fail afterwards, look at the pull-up value against the bus capacitance, at clock stretching if the part uses it, and at whether the register access pattern matches what the datasheet specifies — many devices need a repeated START rather than a STOP between writing the register pointer and reading it back.

For decoding the bytes that come back once addressing works, thehex converter handles width and byte order, and theCRC calculator covers the checksums some sensors append.