

The lesson learned from the uniform quantization experiment is that uniform quantization produces poor quality for small sounds. In this approach, non-uniform quantization should overcome this shortcomming. Logarithmic compressing and expanding (companding) improves the Signal-to-quantization-noise ratio for signals with a large dynamic range like speech. Logarithmic curves like the European A-law are troublesome to implement but can be approximated using piecewise linear segments. In this online simulation app, 13 straight line segments are used - 13-Segment-Kennlinie.
Figure out these questions using the online simulation:
- How much can the Signal-to-quantization-noise ratio improve for low amplitude signals compared to linear / uniform quantization?
- Compare the minimum quantization interval size with uniform quantization!
Listen to signals using the audio play measure (right click on wire):
- Check the sound quality for low signal amplitudes.
External links
Companding: Logarithmic Laws, Implementation, and Consequences
Click on source symbol ![]() |
Select and adjust input signal ![]() |
Click on Quantizer ![]() |
Vary the bit per sample resolution ![]() |
Right click on wire ![]() ![]() |
Open any measure for a signal ![]() ![]() |
![]() |
![]() |
A sine signal is processed by the compressor to amplify the parts with small amplitudes. |
The same signal after a 6-bit quantization. |
![]() |
![]() |
Finally the expanded signal. You can see a much finer resolution at low signal levels. |
Probability density of the total error. |
![]() |
![]() |
For comparison here is again a uniform quantized signal. |
And here is the same signal processed through a non uniform quantization. |
![]() Signal-to-quantization-noise ratios of uniform and non-uniform quantization.
|