Quantization: 6 dB Noise Increase per Bit Lost

How does quantization impact sound quality? Try it by yourself.

16 bits
1 bit 16 bits
Comments

If \(x(n)\) is the original discrete signal, and \(x_q(n)\) is its quantized version, the error \(e(n)=x(n)-x_q(n)\) is white noise.

In this example, \(x(n)\) is a musical signal quantized on 16 bits, which is considered as noise-free. By changing the number of bits, one can see and hear the noise level coming \(6\,\mathrm{dB}\) up for every lost bit.

Notice the quantized signal does not \(seem\) to change much until b<8. This is only due to the fact that the display area is itslef quantized on a fixed number of pixels.