Network Bulls
www.networkbulls.com
Best Institute for CCNA CCNP CCSP CCIP CCIE Training in India
M-44, Old Dlf, Sector-14 Gurgaon, Haryana, India
Call: +91-9654672192
analog circuit can handle only one call at a time, whereas a digital circuit can handle many.
There are two main types of digital circuits: Common Channel Signaling (CCS) and Channel Associated Signaling
(CAS). CAS circuits are available in two speeds: Tl at 1.544Mbs supports 24 calls, and El at 2.048Mbs supports 30
calls. (For these values, we are assuming the calls are not compressed; more on this later). CCS circuits are designated as
PRI T l , PRI E l , and BRI. A PRI Tl can support 23 calls, a PRI El 30, and a BRI only 2.
The use of a digital circuit by definition implies that the voice signal must be digitized; the conversion from analog to
digital is performed by a codec. The following sections discuss the conversion of analog to digital.
Digitizing Analog Signals
There are four steps in the process of digitizing analog sound:
1. Sample the analog sound at regular intervals
2. Quantize the sample
3. Encode the value into a binary expression
4. Optionally compress the sample
Sampling could be done any number of times per second; the more samples taken per second, the higher the audio
quality, but the amount of digital data produced is much larger. Nyquist's theorem states that the sampling interval should
be 2x the highest frequency of the sample to produce acceptable audio quality during playback. Because the highest
frequency in human speech that we want to reproduce in telephony is around 4000 Hz, the sampling rate for standard tollquality
digital voice is 8000 intervals per second. By contrast, CD music audio, which must encode both much higher and
much lower frequencies, samples at about 192,000 times per second.
Quantizing refers to making a digital approximation of an analog waveform. Imagine drawing an arc on a chessboard; if
you had to define the arc using only the square it was in for each row (segment) and column (interval), you would end up
with a stepped pattern that was sort of close to the original arc but not exact. This is exactly the process that happens with
quantization: the codec chooses a segment value that is as close as possible to the analog value at the interval it was
sampled, but it cannot be exact. To make the quantization more accurate, each sample is divided into 16 intervals that are
adjusted to more closely match the sampled wave. Furthermore, the segments are actually more fine-grained at the origin
than at the high and low ranges. This is because most of the human speech we are trying to capture accurately is in this
center range of the scale; there are fewer sounds at the very highest and lowest values.
© 2008 Cisco Systems Inc. All rights reserved. This publication is protected by copyright. Please see page 147 for more details.
FIGURE 11
Quantizing the Digital
Sample
Encoding the signal is a simple process. We have a single 8-bit code word to identify whether the analog signal was a
positive or negative voltage, what value the signal was quantized to (which segment), and finally, which interval is represented
by the code word. The first bit identifies either positive voltage (1) or negative (0). The next three bits represent
the segment. There are eight segments in the positive range and eight segments on the negative range, so three bits
provide the necessary encoding for the quantization. The last four bits identify the interval. A code word example is
shown next:
In this case, the first 1 indicates a positive voltage; the next digits of 001 indicate this is the first segment (on the positive
side), and 1100 indicates the twelfth interval.
The code word is 8 bits; we generate a code word 8000 times per second (the sample rate). This gives us a bitrate output
of 8 x 8000 = 64,000 bps (64 kbps). The process we just described is known as Pulse Code Modulation (PCM) and is the
standard for uncompressed digital voice in telephony. One voice stream thus requires 64k of bandwidth for transport.
1 0 0 1 1 1 0 0
© 2008 Cisco Systems Inc. All rights reserved. This publication is protected by copyright. Please see page 147 for more details.
NOTE
The determination of
voice quality is based on
the Mean Opinion Score
(MOS). This is a subjective
measurement,
created by gathering the
opinion of live human
listeners. A sample
recording is played, and
the listeners give it a
score out of 5, where 5 is
best. The same sample is
played using different
compression or processing
methods and scored
again. Because MOS is
so subjective, other
quality measurements
exist that are more
empirical and more accurate.
For reference, standard
PCM encoding
(G.711) scores 4.1, and
G.729 scores 3.92.
Compression is not a required step, but it is often done to save bandwidth in VoIP environments. The two main types of
compression we are concerned with are the following:
• Adaptive Differential PCM (ADPCM): This method does not send entire code words, but instead sends a smaller
code that represents the difference between this word and the last one sent. This is not commonly used today,
because it produces lower voice quality and compresses down only to about 16 kbps.
• Conjugate Structure Algebraic Code Excited Linear Prediction (CS_ACELP): As the name suggests, this is
more complex compression. Based on a dictionary or codebook of known sounds made by a standardized American
male voice, the digital sample is analyzed and compared to the dictionary. The dictionary code that is the closest to
the sample is sent. The codebook is constantly learning. The output of this compression is typically 8 kbps—with
very little degradation of voice quality. This compression is widely used in VoIP.
No comments:
Post a Comment