When an analog audio signal is converted to a digital signal, a sample is taken from the analog signal 44100 times per second for CD resolution and 96000 times or more per second for High Definition (HD) recordings. For each sample taken, the signal voltage is measured and converted to an integer number. This number – the sample value – represents the voltage at one time instance with some precision, referred to as the resolution and depends on the number of bits in the integer word.
The CD resolution is 16 bits according to the Red Book standard and because of this, the allowable input voltage range for the AD-converter (typically ±12.3 V for a professional AD-converter) is divided into 65 536 (2^16) voltages that can be represented by the integer sample values. The discrepancy between the true analog voltage and the voltage represented by the sample is an error, which is sometimes referred to as noise. Under some assumptions, a maximum signal-to-noise ratio (SNR) can be calculated and is typically given roughly as 6 dB per bit, resulting in 96 dB for a Red Book CD.
Nowadays, most AD-conversions are made with 24-bit converters with 16 777 216 discrete voltages and a maximum signal-to-noise ratio of 144 dB. This is actually a much larger signal-to-noise ratio that what we can achieve with analog circuits, so in effect the (SNR) is determined by the analog circuits (typically microphones and microphone amplifiers) preceding the AD-converter.
A PCM-encoded signal is thus a stream of digital numbers represented as 16, 24 or 32 bit integers or 32-bit floating point numbers. The floating point numbers typically have a 24-bit mantissa (determining the SNR) and an 8-bit exponent (determining the dynamic range). This is the format used in e.g. WAV files. PCM stands for Pulse Code Modulation.