B. P. Lathi, Zhi Ding - Modern Digital and Analog Communication Systems-Oxford University Press (2009)

304 SAMPLING AND ANALOG-TO-DIGITAL CONVERSION
order the chef's recipe. All the ingredients are purchased locally and the cooking is also done
locally. The Alaskan family can satisfy their gourmet craving without receiving a single food
item form New York! Clearly, the last scenario captures the idea of model-based vocoders.
LPC vocoders essentially deliver the recipe (i.e., the LPC parameters) for voice synthesis at
the receiver end.
Practical High-Quality LP Vocoders
The simple dual-state LPC synthesis of Fig. 6.35 describes no more than the basic idea behind
model-based voice codecs. The quality of LP vocoders has been greatly improved by a number
of more elaborate codecs in practice. By adding a few bits, these LP-based vocoders attempt
to improve the speech quality in two ways: by encoding the residual prediction error and by
enhancing the excitation signal.
The most successful methods belong to the class known as code-excited linear prediction
(CELP) vocoders. CELP vocoders use a codebook, a table of typical LP error (or residue)
signals, which is set up a priori by designers. At the transmitter, the analyzer compares the
actual prediction residue to all the entries in the codebook, chooses the entry that is the closest
match, and just adds the address (code) for that entry to the bits for transmission. The synthesizer
receives this code, retrieves the corresponding residue from the codebook, and uses it to modify
the synthesizing output. For CELP to work well, the codebook must be big enough, requiring
more transmission bits. The FS-1016 vocoder is an improvement over FS-1015 and provides
good quality, natural-sounding speech at 4.8 kbit/s. 25 More modern variants include the RPE­
LTP (regular pulse excitation, long-term prediction) LPC codec used in GSM cellular systems,
the algebraic CELP (ACELP), the relaxed CELP (RCELP), the Qualcomm CELP (QCELP)
in CDMA cellular phones, and vector-sum excited linear prediction (VSELP). Their data rates
range from as low as 1.2 kbit/s to 13 kbit/s (full-rate GSM). These vocoders form the basis of
many modern cellular vocoders, voice over Internet Protocol (VoIP), and other ITU-T G-series
standards.
Video Compression
For video and television to go digital we face a tremendous the challenge. Because of the high
video bandwidth (approximately 4.2 MHz), use of direct sampling and quantization leads to
an uncompressed digital video signal of roughly 150 Mbit/s. Thus, the modest compression
afforded by techniques such as ADPCM and subband coding 26 • 27 is insufficient. The key to
video compression, as it turns out, has to do with human visual perception.
A great deal of research and development has resulted in methods to drastically reduce the
digital bandwidth required for video transmission. Early compression techniques compressed
video signals to approximately 45 Mbit/s (DS3). For the emerging video delivery technologies
of HFC, ADSL, HDTV, and so on, however, much greater compression was required.
MPEG approached this problem and developed new compression techniques, which provide
network or VCR quality video at much greater levels of compression. MPEG is a joint effort
of the International Standards Organizations (ISO), the International Electrotechnical Committee
(IEC), and the American National Standards Institute (ANSI) X3L3 Committee. 28 • 29
MPEG has a very informative website that provides extensive information on MPEG and JPEG
technologies and standards (http://www.mpeg.org/index.html/). MPEG also has an industrial
forum promoting the organization's products (http://www.m4if.org/).
The concept of digital video compression is based on the fact that, on the average, a
relatively small number of pixels change from frame to frame. Hence, if only the changes
are transmitted, the transmission bandwidth can be reduced significantly. Digitizing allows
the noise-free recovery of analog signals and improves the picture quality at the receiver.