ADPCM (Adaptive Differential Pulse Code Modulation) is an algorithm that is used to compress the data rate of audio communication in telephony while retaining acceptable audio quality. The most common use of this algorithm is to reduce the normal telephone audio rate of 64kbit/s (known as DS0) to 32kbit/s. This doubles the capacity of the telephone network, as it allows for twice as many phone calls to be multiplexed on higher bandwidth links. This is particularly important on international trunks on the phone network. Other bit rates supported are 16, 24 and 40kbit/s. The 16 and 24kbits/s rates are used for overloaded telephone systems, and 40kbit/s are used for phone line modems that are faster than 4.8kbit/s.  ADPCM is used as an encoder/decoder algorithm for DECT wireless phone systems.

 

The ADPCM algorithm is defined by telecommunications standard G.726. The algorithm operates on a pulse code modulated (PCM) signal. PCM is a way to digitally represent analog audio signals and is used in CD players, computers, and many other common devices. The instantaneous amplitude of the audio is sampled with a special logarithmic weighting called μ-law or a slight variant called A-law. This allows for the audio signals to be encoded with acceptable noise and distortion across the whole amplitude variation (dynamic range). The ADPCM algorithm transcodes from these PCM audio signals to and from ADPCM encoding.

 

Audio is compressed using two methods. The first method encodes the audio samples as differential values rather than absolute values. Because audio varies amplitude at frequencies slower than the sample rate, the data can typically be encoded with higher efficiency in this format. The second method is a predicative algorithm where the receiver is modeled in the transmitter and the result of how the receiver is predicted to decode the signal is subtracted from the input signal. This creates the differential signal in such a way that the receiver can take this input and work backward to reconstruct the original audio samples. Because the system is encoded differentially, initial conditions need to be assumed and managed by a method of synchronization and adjustment.

 

ADPCM voice compression trades off noise in a special way that also pertains to modern higher quality and more efficient audio coding algorithms used in VoIP communication. The algorithm compresses noise and distortion so to be relatively the same across the wide dynamic range of hearing that tends to be masked, by the way, the ear's acoustic mechanisms work. This is called lossy compression. How well this is done is measured by a mean opinion score (MOS) of people listening to encoded samples. Today's algorithms used in VoIP and cellphone communication can compress to over four times the ADPCM data rates for the same MOS.