Anytime a message is transferred over a physical medium, the possibility exists that it may be corrupted by noise. Accordingly, since the earliest days of data communication, various mechanisms have been devised to detect when the data received was not the same as the data sent.One of the simplest error detection schemes is parity checking. Each data byte is sent with an extra bit, which is called the parity bit. The value of the parity bit depends on the number of 1 bits in the byte, and also on the type of parity checking used. When Odd Parity is employed, the parity bit is a 1 when the number of 1 bits in the byte is odd. Otherwise, the parity bit is a 0. When Even Parity is used, a parity bit of 1 indicates an even number of 1 bits. Parity checking is easily implemented in hardware and is a feature found on most data comm chips. Its speed and ease of use make it an attractive and popular error detection mechanism.
Yet, parity checking is rather inefficient. In asynchronous communications, each eight-bit data byte is "framed" by a start bit and a stop bit, for a total of 10 bits. Adding a parity bit to the data byte increases the character size 10 percent. Furthermore, parity checking can only detect an odd number of errors per byte. If two bit errors occur in a single byte, they cancel each other. The parity is unchanged, and the error goes undetected. In an extreme case, all eight data bits in the byte could be reversed, but the parity check would not detect the error.
A more efficient method of detecting errors is the checksum. A checksum is calculated by adding together the values of all of the data bytes in the message. Checksums can be eight, sixteen, or thiry-two bits wide (overflow from the addition is ignored). In a typical application, the checksum is appended to the end of the message. The receiver verifies the message by re-calculating the checksum on the data and comparing its result to the checksum that was sent. )
Simple checksums are easy to implement in software and do not bog the processor down. When checksums are employed, data can be transmitted without the overhead of parity bits. This is a consideration that becomes more important as the size of the message increases. However, checksums fall prey to an entire class of errors that can be termed "transposition errors." Imagine that a message is sent containing the sequence 0x31 0x33. With just two bit errors, the sequence could be incorrectly received as 0x33 0x31, and yet still produce a "correct" checksum.