I think the more precise concern is not so much wide dynamic range, but rather logarithmic sensing in humans. That is, in order to achieve the 110-120 dB of dynamic range, the human ear/brain system cleverly uses a logarithmic scale. Very faint sounds can be resolved very well, while the difference in loudness of extremely loud sounds cannot. People have a hard time discerning the difference between a loudness of, say, 107 dBA vs 110 dBA, even though the latter is twice as powerful as the former.
So, as Daniel explains, if you quantize the signal on a linear scale, you are wasting a whole lot of values for very, very loud sounds, while not providing enough values to the very faint sounds that we can discern much better.
In short, for sound or vision, we only prefer non linear quantization because our own sensors are non linear. If we sensed loudness or light intensity on a linear scale, then we'd probably prefer linear quantization.
Non-uniform quantisation (such as A-law or mu-law companding) is useful for signals such as speech with a very wide dynamic range, as a linear quantisation scheme would result in a high signal-to-quantisation-noise ratio for very small-amplitude speech signals, causing severe distortion. Companding the signal (non-linearly compressing prior to quantisation by a linear ADC and expanding post-DAC) overcomes this by effectively using a larger part of a linear ADC's dynamic range for small amplitudes.
I think the more precise concern is not so much wide dynamic range, but rather logarithmic sensing in humans. That is, in order to achieve the 110-120 dB of dynamic range, the human ear/brain system cleverly uses a logarithmic scale. Very faint sounds can be resolved very well, while the difference in loudness of extremely loud sounds cannot. People have a hard time discerning the difference between a loudness of, say, 107 dBA vs 110 dBA, even though the latter is twice as powerful as the former.
So, as Daniel explains, if you quantize the signal on a linear scale, you are wasting a whole lot of values for very, very loud sounds, while not providing enough values to the very faint sounds that we can discern much better.
In short, for sound or vision, we only prefer non linear quantization because our own sensors are non linear. If we sensed loudness or light intensity on a linear scale, then we'd probably prefer linear quantization.
increment number of S/H at A/D result difficult building circuit it,because build the like circuit's almost is not applicable\.
Companding is essential in every quantization process to reduce the noise in dealing with any audio or video analogue signal for their digital conversion.
In logarithmic quantizers the quantization step d(x) is proportional to the level x : d(x)=c*x. So, it keeps the signal to quantization noise ratio d(x)/(2x) constant over (almost) the whole dynamic range. I write almost, because for low level signals the quantization is linear, otherwise it would require infinite many levels, since log(x) tends to minus infinite if x tends to zero. In linear quantizers the signal to noise ratio is either unnecessarily high in case of high level signals or poor for low level ones.
to overcome large dynamic amplitude signal quantification
ie : log of the signal amplitude before quantization is used to implement best quantification of small amplitude
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing