Quality After A-D Conversion

An analog waveform is characterized by a smooth and continuous change in its amplitude. The quality of the converted (digitized) waveform is determined by how closely it resembles the original (analog) waveform. The first major difference arises from the sampling interval. The word "interval" implies that the resultant waveform is NOT continuous. A digitized waveform can never be continuous, but it can approach that state by reducing the interval between samples. In other words, the digitized output will approach the continuity of the analog waveform if the sampling rate is increased sufficiently.

The second difference arises from using a limited set of integers to describe the samples. To elaborate: Suppose we use a set of 2000 integers to describe a dynamic range of 2000 millivolts (mv). That works out to 1 number for 1 mv (2000mv/2000 numbers). In other words we cannot describe (resolve) differences smaller than 1 mv! We can improve the resolution by using larger numbers (numbers with more digits) to describe the amplitude of the signal waveform. The digitized output will approach the smoothness of the analog waveform if we increase the number size.

These two parameters (sampling rate and sampling resolution) define the quality of the digitized signal. The higher the value of these parameters, the better is the quality. At the same time it must be noted that the higher the value of these parameters, the larger is the size of the resultant file! It must also be noted that high resolution, high speed DACs are (comparatively) expensive devices.

So it's basically a tug-of-war (or compromise) between quality on the one hand and a combination of file size and device cost on the other.

A stereo file has 2 channels of sound. This means that a stereo file is double the size of a monophonic (single channel) file.

Compression adds yet another factor to the question of quality. The most popular compression methodology nowadays is called MP3. At its best compression rate of 20kbps (Windows ACM codec) the quality is noticeably poor, and at 32kbps (Lame encoding) it is downright intolerable!

As a test we ripped a track from an up-market audio CD at 44100Hz, 16bit, stereo (a). The file size was about 28MB.

Then the file was resampled at 11025Hz, 16bit, stereo (b). The file size was about 7MB.

Next, the (original) file was MP3-compressed at 20kbps using the Windows ACM codec (c). The file size was 400 kB!

Finally, the (original) file was MP3-compressed at 128kbps using the Windows ACM codec (d). The file size was 2.5MB.

File size CompressionFormat Quality

(a) 28000kB -----Original 100%

(b) 7000kB 4:1Wave, 11kHz Poor

(c)400kB 70:1MP3, 20kbps Poor

(d) 2500kB 11:1MP3, 128kbps 99%

The quality of (b) and (c) was almost the same, poor. Whereas (b) exhibited a slight raspiness, a form of amplitude distortion, (c) exhibited frequency distortion.

And (d) was PRACTICALLY the same as the original, with a 11:1 disk economy to boot!

The secret appears to be in the fact that MP3 advances the sampling rate in slabs as the bitrate increases. At a bitrate of 32kbps the sampling rate is 11kHz, at 64kbps the SR is 22kHz, and at 96kbps the SR is 44kHz.

So the best deal is to compress using MP3 with a bitrate of 96kbps or higher.