info america

Video Over Wireless
As explained in the title of this book, our focus is on the video processing and communications
techniques for the next generation of wireless communication system (4G
and beyond). As the system infrastructure has evolved so as to provide better QoS support
and higher data bandwidth, more exciting video applications and more innovations
in video processing are expected. In this section, we describe the big picture of video
over wireless from video compression and error resilience to video delivery over wireless
channels
Video Compression Basics
Clearly the purpose of video compression is to save the bandwidth of the communication
channel. As video is a specific form of data, the history of video compression can be
traced back to Shannon’s seminal work [1] on data compression, where the quantitative
measure of information called self-information is defined. The amount of self-information
is associated with the probability of an event to occur; that is, an unlikely event (with
lower probability) contains more information than a high probability event. When a set of
independent outcome of events is considered, a quantity called entropy is used to count
the average self-information associated with the random experiments, such as:
H =p(A
i
)i(A
i
) = −p(A
i
) log p(A
i
)
where p(Ai ) is the probability of the event Ai , and i (Ai ) the associated self-information.
Entropy concept constitutes the basis of data compression, where we consider a source
containing a set of symbols (or its alphabet), and model the source output as a discrete
random variable. Shannon’s first theorem (or source coding theorem) claims that no matter
how one assigns a binary codeword to represent each symbol, in order to make the source
code uniquely decodable, the average number of bits per source symbol used in the source
coding process is lower bounded by the entropy of the source. In other words, entropy
represents a fundamental limit on the average number of bits per source symbol. This
boundary is very important in evaluating the efficiency of a lossless coding algorithm.
Modeling is an important stage in data compression. Generally, it is impossible to
know the entropy of a physical source. The estimation of the entropy of a physical
source is highly dependent on the assumptions about the structure of the source sequence.
These assumptions are called the model for the sequence. Having a good model for
the data can be useful in estimating the entropy of the source and thus achieving more
efficient compression algorithms. You can either construct a physical model based on the
understanding of the physics of data generation, or build a probability model based on
empirical observation of the statistics of the data, or build another model based on a set
of assumptions, for example, aMarkov model, which assumes that the knowledge of the
past k symbols is equivalent to the knowledge of the entire past history of the process
(for the kth order Markov models).
After the theoretical lower bound on the information source coding bitrate is introduced,
we can consider the means for data compression. Huffman coding [2] and arithmetic
coding [3] are two of the most popular lossless data compression approaches. The Huffman
code is a source code whose average word length approaches the fundamental limit set
by the entropy of a source. It is optimum in the sense that no other uniquely decodable set
of code words has a smaller average code word length for a given discrete memory less
source. It is based on two basic observations of an optimum code: 1) symbols that occur
more frequently will have shorter code words than symbols that occur less frequently;
2) the two symbols that occur least frequently will have the same length [4]. Therefore,
Huffman codes are variable length code words (VLC), which are built as follows: the
symbols constitute initially a set of leaf nodes of a binary tree; a new node is created as
the father of the two nodes with the smallest probability, and it is assigned the sum of
its offspring’s probabilities; this new node adding procedure is repeated until the root of
the tree is reached. The Huffman code for any symbol can be obtained by traversing the
tree from the root node to the leaf corresponding to the symbol, adding a 0 to the code