4.1 Quantization Techniques
4
Signals of widely varying amplitudes are present in radio receiver applications. Digitization of
these signals requires ADCs with large dynamic range. Most commercially available ADCs use
uniform quantization (for which quantization levels are evenly spaced) and therefore require high
resolution (a large number of bits) to achieve a large dynamic range. The problem for many RF
applications is that ADCs using uniform quantization cannot easily meet both the large dynamic
range and high sampling rates needed to digitize wide bandwidths at the RF or IF. In addition,
another problem is that uniform quantizers exhibit an SNR that varies with desired signal
amplitude; the SNR decreases as the input signal power to the quantizer decreases. (Ideally, a
large SNR would be maintained over a large variation in input signal power.) In an effort to
solve these problems, several quantization schemes other than uniform quantization have been
implemented.
These different quantization techniques include μ-law (and A-law) quantization, adaptive
quantization, and differential quantization [28]. While these techniques are typically associated
with speech signal processing applications, they may be useful for radio receiver applications
[27].
4
When this report was originally published, in this section the authors reproduced with permission some material
from L.R. Rabiner and R.W. Schafer, Digital Processing of Speech Signals, Englewood Cliffs, NJ: Prentice-Hall,
Inc., 1978, cited in this report as [27]. This textbook served as a fundamental reference text for both teaching the
fundamentals of digital speech processing and designing practical systems for over 30 years. In 2010, Rabiner and
Shafer published a new textbook on the same topic that incorporates advances in the technology of digital speech
processing that occurred in the interval and supersedes the 1978 textbook. The authors have therefore revised this
section to provide just an overview of the material. Readers who want more current in-depth information on this
topic should consult L.R. Rabiner and R.W. Schafer, Theory and Applications of Digital Speech Processing,
Englewood Cliffs, NJ: Prentice-Hall, Inc., 2010.
33
In μ-law quantization, a nonlinear compression (an approximation to logarithmic compression) is
used on the input signal before uniform quantization[28]. This increases the dynamic range of
the overall quantization. In addition, while for uniform quantizers the SNR varies considerably
with small changes in input signal amplitude, the SNR remains fairly constant over a wide range
of input signal amplitudes for μ-law quantizers. As an example of this, the SNR changes by 25
dB for input signal amplitude changes of 25 dB in a uniform quantizer. With appropriate scaling
of the input signal amplitude, the SNR changes by less than 2 dB in a μ-law quantizer (μ = 500)
for input signal amplitude changes of 25 dB.
Adaptive quantization is an approach by which the properties of the quantizer are varied
depending on the amplitude of the input signal. One example of adaptive quantization is when
the step size of the quantizer is adjusted based on the input signal amplitude. Note that while the
overall dynamic range is increased using adaptive quantization, the instantaneous dynamic range
is not increased.
5
. The SNR can be kept relatively constant for input signal amplitude changes of
40 dB, for example, by selecting the ratio of maximum to minimum step size of 100 [29].
Finally, the technique of differential quantization involves the quantization of the difference
between the actual input signal and an estimate of this input signal. Because this difference
signal is much smaller than the actual input signal if a good estimate is obtained, the dynamic
range is increased. The SNR can be maximized by optimizing the estimate of the input signal.
The potentially large processing load (number of computations within a given time interval)
required to implement adaptive or differential quantization in radio receiver applications must be
considered carefully. The speed of the processor must be able to handle the required processing
load adequately.
Chia sẻ với bạn bè của bạn: |