Source Coding Theory
Source coding (i.e. data compression) is the process of creating binary rep-
resentations of the data from sources such as speech, images, audio, video
or English text. This course gives a broad introduction to the theory and
practice of lossy coding (i.e. quantization), where perfect reproductions are
not possible or require too many bits (e.g. speech, images, audio, video)
and also of lossless coding, where perfect reproductions are required (e. g.
text). The lossy codes include scalar, vector, transform (e.g. JPEG,
MPEG), subband (e.g. wavelet), predictive and adaptive quantizers (e.g.
CELP); and the theory is mainly high resolution quantization theory. The
lossless techniques include Huffman, run-length, Lempel-Ziv, and arithme-
tic codes; and the theory is entropy theory. Particular attention is paid to
coding speech and images. Students will gain experience in source coding
through a term project. The course is oriented toward first and second
year graduate students. It is a kernel course for both communications and
signal processing. No previous introduction to source coding is presumed.
Detailed Syllabus (attached)
Time and Place
MWF 1:30-2:30, 3150 Dow (but it may change)
Prof. David L. Neuhoff
4215 EECS Building
- Regular hours:
MWF 2:30-3 and beyond if needed. I'll at least see everyone who's
here by 3.
Tues. 1-2:30 or Thurs. 1-2:30, whichever is the last one before the homework
- By appointment at some other time: Contact me
by email, phone or stopping by my office.
- Drop in: I'll try to see you, but can't make guarantees.
There is no required course textbook. You should, however,
obtain a copy of the following paper:
R.M. Gray and D.L. Neuhoff, "Quantization," IEEE
Trans. Inform. Theory, pp. 2325-2383, Oct. 1998. You may download it
from IEEE through the library web page:
or directly by clicking
See also the list of books on reserve for the class in the Media
Union. Those in bold are likely to be the most useful references.