Filetype PDF | Posted on 30 Jan 2023 | 3 years ago
Authentication
digital resources
262x Filetype PDF File size 0.19 MB Source: www.rle.mit.edu
Part III, Section 2, Chapter 1. Digital Signal Processing Research
Chapter 1. Digital Signal Processing Research Program
Academic and Research Staff
Professor Alan V. Oppenheim, Professor Arthur B. Baggeroer, Professor Anantha P. Chandrakasan, Professor
Jeffrey H. Lang, Professor Gregory W. Wornell, Giovanni Aliberti
Visiting Scientists and Research Affiliates
1 2 3
Dr. Bernard Gold, Dr. Hamid S. Nawab, Dr. James C. Preisig, Dr. Ehud Weinstein, Dr. Frank Kschischang
Graduate Students
Anthony J. Accardi, Rajeevan Amirtharajah, Richard J. Barron, Albert Chan, Brian Chen, Stark C. Draper,
Yonina C. Eldar, Christoforos N. Hadjicostis, Nicholas J. Laneman, Li Lee, Michael J. Lopez, Emin Martinian,
Scott E. Meninger, José O. Mur-Miranda, Haralabos C. Papadopoulos, Andrew I. Russell, Maya R. Said,
Matthew J. Secor, Alan J. Seefeldt, Charles K. Sestok, Wade P. Torres, Shawn M. Verbout, Kathleen E. Wage,
Huan Yao
Technical and Support Staff
Darla J. Chupp, Janice M. Zaganjori
1.1 Introduction nal design and analysis. Another research emphasis
is on structuring algorithms for approximate process-
The field of digital signal processing grew out of the ing and successive refinement.
flexibility afforded by the use of digital computers in
implementing signal processing algorithms and sys- In other research, we are investigating applications
tems. It has since broadened into the use of a variety of signal and array processing to ocean and struc-
of both digital and analog technologies, spanning a tural acoustics and geophysics. These problems
broad range of applications, bandwidths, and realiza- require the combination of digital signal processing
tions. The Digital Signal Processing group carries out tools with a knowledge of wave propagation to
research on algorithms for signal processing and develop systems for short-time spectral analysis,
their applications. Current application areas of inter- wavenumber spectrum estimation, source localiza-
est include signal enhancement and active noise tion, and matched field processing. We emphasize
cancellation; speech, audio, and underwater acoustic the use of real-world data from laboratory and field
signal processing; advanced beamforming for radar experiments such as the Heard Island Experiment for
and sonar systems; and signal processing and cod- Acoustic Monitoring of Global Warming and several
ing for wireless and broadband multiuser communi- Arctic acoustic experiments conducted on the polar
cation networks. ice cap.
In some of our recent work, we have developed new A major application focus of the group involves signal
methods for signal enhancement and noise cancella- processing and coding for wireless multiuser sys-
tion with single or multisensor measurements. We tems and broadband communication networks. Spe-
have also been developing new methods for repre- cific interests include commercial and military mobile
senting and analyzing fractal signals. This class of radio networks, wireless local area networks and per-
signals arises in a wide variety of physical environ- sonal communication systems, digital audio and tele-
ments and also has potential use in problems involv- vision broadcast systems, and multimedia networks.
ing signal design. We are also exploring potential Along with a number of other directions, we are cur-
uses of nonlinear dynamics and chaos theory of sig- rently exploring new code-division multiple-access
(CDMA) strategies, new techniques for exploiting
1 Associate Professor, Boston University, College of Engineering, Boston, Massachusetts.
2 Department of Electrical Engineering, Systems Division, Faculty of Engineering, Tel-Aviv University, Israel; Adjunct Scientist, Depart-
ment of Applied Ocean Physics and Engineering, Woods Hole Oceanographic Institution, Woods Hole, Massachusetts.
3 Professor, Department of Electrical Engineering and Computer Science, University of Toronto, Toronto, Canada.
281
Part III, Section 2, Chapter 1. Digital Signal Processing Research
antenna arrays in wireless systems, and new meth- We hope to improve upon these existing multi-unit
ods for modeling and management of traffic in high- separation schemes and will then pursue related
speed packet-switched networks. applications (e.g., action potential coding) of our new
knowledge.
Much of our work involves close collaboration with
the Woods Hole Oceanographic Institution, MIT Lin- 1.3 Dual-Channel Signal Processing
coln Laboratory, and a number of high technology
companies in the Boston Area. Sponsors
1.2 Neural Signal Processing Sanders, a Lockheed-Martin Corporation
Contract BZ4962
Sponsor U.S. Army Research Laboratory
National Science Foundation Graduate Cooperative Agreement DAAL01-96-2-0001
Research Fellowship Project Staff
Project Staff Richard J. Barron, Professor Alan V. Oppenheim
Anthony J. Accardi, Professor Gregory W. Wornell Many models for signal estimation systems assume
In order to understand the detailed interworkings of only statistical information about the source signal to
many neurological processes, it is necessary to mea- be recovered and the channel through which the
sure the firing patterns realized by individual neu- source is sent. In some scenarios, however, there
rons. Current measuring techniques involve inserting also exists deterministic side information about the
desired signal which can be used jointly with channel
one or more electrodes into the region of interest, observations to assist recovery. For example, an
which make extracellular voltage recordings derived existing full-band, noisy analog communications
from the action potentials of nearby neurons. The dif- infrastructure may be augmented by a low-bandwidth
ficulty is that firing patterns from many different neu- digital side channel. Our research is a study of a
rons are superimposed at the electrodes, but we are hybrid channel that is the composition of two chan-
interested in individual neuron behavior. Deriving this nels: a noisy analog channel through which a signal
information from such measurements is referred to source is sent unprocessed and a secondary rate-
as separating multiple single-unit spike trains from a constrained digital channel. The source is processed
multi-unit recording. prior to transmission through the digital channel.
The problem is therefore one of signal separation, Using a signal processing framework for low latency
and many approaches have been attempted based and low complexity, we derive optimal encoder and
on pattern matching and feature clustering. In many receiver structures for hybrid channels.
of these approaches, the inaccurate assumption that 1.4 Batch-Iterative Channel
different neurons exhibit action potentials with unique
waveforms is made. A new instrument consisting of Equalization
four very closely spaced electrodes, called the tet- Sponsors
rode was developed in 1994. This instrument allows
us to drop the assumption and therefore perform a U.S. Army Research Laboratory
more reliable separation. The best existing separa- Cooperative Agreement DAAL01-96-2-0002
tion schemes for the tetrode are computer assisted; U.S. Navy - Office of Naval Research
they present waveform parameters in a graphical
manner so that a well-trained user can visually clus- Grant N00014-96-1-0930
ter the feature arising from separate neurons. These Project Staff
techniques necessarily prevent a full exploitation of
the information available in the tetrode measure- Albert Chan, Professor Gregory W. Wornell
ments, since decisions must be made in a low
enough dimension for human visualization. The goal of channel equalization is to minimize the
probability of error by compensating for channel dis-
tortion. One basic approach to channel equalization
is to use the decision-feedback equalizer (DFE). The
portion of the DFE that cancels postcursor intersym-
282 RLE Progress Report Number 141
Part III, Section 2, Chapter 1. Digital Signal Processing Research
bol interference (ISI) is nonlinear; as a consequence nal, called the “embedded signal” or “digital
of this, the postcursor equalizer portion does not watermark,” within another signal, called the “host
enhance noise. By contrast, the portion of the DFE signal.” The host signal is typically a speech, audio,
that cancels precursor ISI is linear, limiting its capa- image, or video signal, and the embedding must be
bilities and leaving behind a significant amount of done in such a way that the host signal is not
residual precursor ISI. degraded unacceptably. At the same time, the digital
watermark must be difficult to remove without caus-
We are currently working on a simple yet effective ing significant damage to the host signal and must
equalizer that cancels both precursor and postcursor reliably survive common signal processing manipula-
ISI in a nonlinear fashion, based on the iterative tions such as lossy compression, additive noise, and
4
equalizer described in Beheshti and Wornell (1997). resampling. Applications include copyright protec-
The equalizer suppresses both intersymbol and inter- tion, authentication, transmission of auxiliary infor-
user interference in spread-signature code-division mation, and covert communication.
5
multiple-access systems, but we have adapted that
equalizer to the single-user, high ISI, fixed-channel In our work, we are developing a general framework
scenario. We have shown theoretically and in simula- for designing digital watermarking systems, evaluat-
tions that our equalizer has a lower probability of ing their performance, and understanding their fun-
error than the DFE. In fact, at high signal-to-noise damental performance limits. In the process we have
ratio (SNR), our iterative equalizer requires 2.507 dB developed a class of digital watermarking techniques
less power to achieve the same probability of error called quantization index modulation, along with a
as the DFE. convenient realization called dither modulation, that
have considerable performance advantages over
We are now investigating the applicability of our previously proposed methods. More information can
equalizer to low-ISI channels and to the estimation of 6
be found in Chen and Wornell (1998 and 1999).
the channel in addition to the data symbols.
1.6 Multiple Descriptions for Soft
1.5 Information Embedding and Digital Memory Systems
Watermarking
Sponsors
Sponsors
Intel Corporation
National Defense Science and Engineering Fellowship
Fellowship U.S. Air Force - Office of Scientific Research
U.S. Air Force - Office of Scientific Research Grant F49620-96-1-0072
Grant F49620-96-1-0072
U.S. Navy - Office of Naval Research Project Staff
Grant N00014-96-1-0930 Stark C. Draper, Professor Gregory W. Wornell
Project Staff The multiple descriptions problem is a classic ques-
Brian Chen, Professor Gregory W. Wornell tion of information theory in which a data source is
encoded into two data streams. Each stream of data
Digital watermarking and information embedding, independently describes the original source to a cer-
which are also referred to as data hiding and stegan- tain fidelity, but together the streams describe the
ography, refer to the process of embedding one sig-
4 S. Beheshti and G. Wornell, “Iterative Interference Cancellation and Decoding for Spread-signature CDMA Systems,” Procedure Vehic-
ular Technology Conference, Phoenix, Arizona, May 1997.
5 G.W. Wornell, “Spread-Signature CDMA: Efficient Multiuser Communication in the Presence of Fading,” IEEE Trans. Info. Theory, Sep-
tember 1995.
6 B. Chen and G.W. Wornell, “Dither Modulation: A New Approach to Digital Watermarking and Information Embedding,” Proceeding of
SPIE: Security and Watermarking of Multimedia Contents (part of Electronic Imaging ’99), San Jose, California, January 1999, forthcom-
ing; B. Chen and G.W. Wornell, “Digital Watermarking and Information Embedding Using Dither Modulation,” Proceeding of 1998 IEEE
Second Workshop on Multimedia Signal Processing (MMSP-98), Redondo Beach, California, December 7-9, 1998, pp. 273-78.
283
Part III, Section 2, Chapter 1. Digital Signal Processing Research
source to a higher fidelity. An example is an imper- In our research so far, we have been able to general-
fectly packetized network where two packets are ize this work to the case of computations occurring in
transmitted. 8
semigroups and semirings, and to outline a proce-
dure that reflects such algebraically-based ABFT
Either or both packets may be received. We want to design into hardware. Currently, we are exploring
encode the data into the packets so that if only one extensions of our approach to sequences of compu-
packet is received the most fundamental data is tations associated with the evolution of dynamic sys-
extractable but, if both arrive, further refinements are tems in particular algebraic settings, such as linear
also available. systems over groups, or rings, or semirings, or finite
As described above, this problem was originally con- automata and discrete-event systems. Along these
ceived of as a transmission problem with packet lines, we have obtained an illuminating characteriza-
drops. We are extending the use of the multiple tion of all possible redundant linear time-invariant
descriptions coding paradigm to memory systems. (LTI) state-space embeddings of a given LTI state-
These systems can easily trade off storage space for space model. We have also illustrated a method of
quality of the stored signal. They facilitate the con- constructing fault-tolerant finite automata using a
struction of lower-fidelity representations of the combination of group and semigroup homomorphic
source and, in general, ease the computational mappings. In our future work, we intend to fold prob-
requirements of memory management. abilistic models for failures and errors into the design
and analysis of ABFT systems.
1.7 Algebraic and Probabilistic 1.8 Low-Complexity Diversity
Structure in Fault-Tolerant Transmission for Fading Channels
Computation
Sponsors
Sponsor
National Science Foundation
Sanders, a Lockheed-Martin Corporation Graduate Research Fellowship
Contract BZ4962 Grant MIP-9502885
Project Staff U.S. Navy - Office of Naval Research
Christoforos N. Hadjicostis, Professor George C. Grant N00014-96-1-0930
Verghese Project Staff
The traditional approach towards fault-tolerant com- Nicholas J. Laneman, Professor Gregory W. Wornell
putation has been modular redundancy. Although
universal and simple, modular redundancy is inher- In the mobile wireless communications setting, trans-
ently expensive and inefficient in its use of resources. mission quality is severely degraded by fading—fluc-
Recently developed algorithm based fault tolerance tuations in received signal energy induced by
(ABFT) techniques offer more efficient fault cover- multipath propagation and relative motion of the
age, but their design is specific to each application. A transmitter and receiver. Diversity transmission via
particular class of ABFT techniques involves the multiple transmit antennas, bandwidth expansion, or
design of arithmetic codes that protect elementary coding mitigates the effects of fading by essentially
computations. In the case of computations that can repeating information on independent realizations of
be represented as operations in a group, the doctoral the channel and averaging the received signal. This
7 most basic form of diversity transmission in space,
dissertation by Beckmann has shown how to obtain
a variety of useful results and systematic construc- frequency, or time corresponds to a repetition code,
tive procedures. and more powerful codes for fading channels have
been developed over the years. Unfortunately, the
7 P.E. Beckmann, Fault-Tolerant Computation Using Algebraic Homomorphisms, RLE TR-580 (Cambridge, MIT Research Laboratory for
Electronics, 1993).
8 C.N. Hadjicostis, Fault-Tolerant Computation in Semigroups and Semirings, RLE TR-594 (Cambridge, MIT Research Laboratory for
Electronics, 1995).
284 RLE Progress Report Number 141
no reviews yet
Please Login to review.
