Members:
Optimizing filterbanks with respect to coding gain
Vector quantization and LSF computation/quantization
VQ image coding and IA over memory channels
·Lossless
signal compression
Lossless
signal compression (audio, ECG, images)
Compression techniques (Variable-to-fixed coding)
·Estimation,
Optimization and Adaptive filtering
Inference and prediction
from gene expressions
Convex optimization in signal processing
Underwater signal processing and Active noise control
·Robust index assignment for channels with memory, and applications for VQ ofspeech and images
·Quantization of LSF parameters using faster and less complex VQ
·Fast
and reliable methods for optimum FIR and IIR compaction filters
·Very effective lossless compression of audio, speech and medical signals
·New variable-to-fixed codes for Markov sources
·Applications of MDL based inference for gene expression prediction and ECG segmentation
·Parameterization of positive real transfer functions suitable for semidefinite programming estimation and filter design
VQ image coding and IA over memory channels
Optimizing filterbanks with respect to coding gain
At rates in the range 4-32 Kbits/s many speech coders model the vocal short-time spectral envelope (for each frame of sampled signal) using autoregressive (AR) predictors. In order to transmit the spectral information, the model parameters ought to be quantized, but since the coefficients of the AR polynomial are sensitive to quantization noise, they are first transformed to a set of equivalent parameters, named line spectral frequencies (LSF) which are subsequently quantized and transmitted.
The major problems we investigate are the fast and reliable computation of the LSF parameters, the design of algebraic quantization methods, and finally the index assignment techniques for improving significantly the error resilience when transmitting the LSF parameters over noisy channels (possibly channels with memory). For this wide spectrum of problems we proposed and experimented new solutions, which may be very useful for speech coding, but may be also applied to other areas, e.g. for image quantization and coding.
Vector quantization has been successfully used for source coding for decades. However, the implementation of an optimal vector quantizer is confronted with two main problems: the encoding/decoding complexity and the storage requirements. In order to overcome these problems, several sub-optimal, constrained, VQ schemes have been proposed: multiple-stage VQ, split VQ, tree-structured VQ, gain-shape VQ. Lattice VQ has long been used in speech coding and image coding.
To benefit of the lattice quantizer advantages, while having a codebook whose spatial distribution is suitably adapted to the probability distribution of the source, we introduce new VQ structerd, themultiple-scale lattice vector quantizers and the multiple-scale leader-lattice vector quantizer, which are very flexible structures of regular points.
When we specialize to finite-memory channels we introduce a new performance index, very robust to the changes in the parameters of the finite memory contagion model.By maximizing the new performance index one obtains very good index assignments for typical situations, e.g. low correlation of errors, high correlation of errors, memory of channel larger than the block length.
Lossless compression of high quality audio signals has become a significant topic of research, being considered adesirable feature which was implemented in the new high quality audio applications, e.g. digital versatile disks. Lossless audio compression may appear even more necessary in audio and speech archiving or whenever the audio signal undergoes multiple encoding-decoding operations. We examine severalsolutionsfor the application of Context algorithm to the compression of audio signals, but also for other signals, e.g. ECG signals.
The purpose of the so-called variable-to-fixed codes for data compression is to obtain fast coding operations, which is made possible by assigning a fixed length codeword to the ordinal of a source string identified with a leaf in a complete tree.In our research, alphabet extensions for Markov sources are studied such that each extension tree is grown by splitting the node with the maximum value for a weight generalizing the leaf probability inTunstall's algorithm. We show that the optimal asymptotic rate of convergence of the per symbol code length to the entropy does not depend on the proportional allocation of the sizes of the extension trees at the states, without imposing restrictive conditions on the weight by which the trees are extended.
Estimation, Optimization and Adaptive
Filtering
Inference
and prediction from gene expressions
Convex
optimization in signal processing
Underwater signal processing
and Active noise control
A well-established fact is the ability of the"codelength" to be a universal yardstick to measure the complexity of models. The minimum description length MDL principleis a basic axiomto the theory of modeling: given the data and the model class, select the model in the model class which achievesthe shortest codelength for the data and the model.One important problem in genomics is to find groups of genes (and/or factors) which are very likely to determine the activity of a target gene. By applying compression techniques we seek for the (universal) code giving the shortest description of the ensemble of gene expression data. By our modelling techniques (data analysis) we seek for a description of the degree of dependency of one gene given the othergenes and the conditions in the available data.
Lossy
signal compression
1.Bogdan
Dumitrescu, Ioan Tabus, Predictive LSF computation, submitted to Signal
Processing, January 2000, revised November 2000.
2. Razvan Iordache, Ioan Tabus, Jaakko Astola, Robust index assignment for transmitting vector quantized LSF parameters over finite-memory contagion channels using Hadamard transform, submitted to IEEE Trans. Communication, May 2000.
3.Adriana
Vasilache, Bogdan Dumitrescu, Ioan Tabus, Multiple-Scale Leader-Lattice
VQ with Application to LSF Quantization, submitted to Signal Processing,
July 2000.
4.Ioan
Tabus, Corneliu Popeea, Jaakko Astola, A Design Procedure for Optimal Energy
Compaction IIR Filters, submitted to IEEE Trans. Circuits and Systems II,
September 1999, revised July 2000.
5.Razvan
Iordache, Ioan Tabus, Jaakko Astola, Robust index assignment for
quantization of LSF parameters transmitted over finite-memory contagion
channels
First IEEE Balkan Conf. on Signal Processing, Communications, Circuits
and Systems, Istanbul, Turkey, pp. 8-11, June 2-3, 2000
6..Jianqin
Chen, Ioan Tabus, and Jaakko Astola, Quantization of LSF by Lattice Shape-Gain
Vector Quantizer, First IEEE Balkan Conf. on Signal Processing, Communications,
Circuits and Systems, Istanbul, Turkey, pp. 13-16, June 2-3, 2000
7.B.Dumitrescu,
C.Popeea - Optimum FIR Compaction Filters with Regularity Constraints,
submitted to IEEE Trans. Signal Processing, Feb. 2000.
8.B.Dumitrescu
- Parametrization of Positive Real Transfer Functions, with Fixed Poles,
submitted to IEEE Trans. Circ. Syst. I, Nov. 2000.
9.C.Popeea,
B.Dumitrescu - Optimal Compaction Gain By Eigenvalue Minimization,
to appear in Signal Processing.
10.Dan
Stefanoiu, Ioan Tabus, Degenerate Eigenvalues - A Method to Design Adaptive
Discrete Time Wavelets, EUSIPCO-2000, Tampere, Vol. 3, 1767-1770, September
4-8, 2000.
11.Bogdan
Dumitrescu, Ioan Tabus, Predictive LSF Computation, EUSIPCO-2000, Tampere,
Vol. 2, 829-832, September 4-8, 2000.
12.R.
Iordache, A. Beghdadi, I. Tabus, Vector Quantization with Edge Reconstruction
2000 International Conference on Image Processing, Vancouver, BC, Canada,
pp. 167-170, September 10 - 13, 2000.
13.Razvan
Iordache, Ioan Tabus, Robust Index Assignment for Finite-Memory Contagion
Channels using Hadamard Transform, 2000 International Symposium on Information
Theory and Its Applications, Sheraton Waikiki Hotel, Honolulu, Hawaii,
pp. 382-385, November 5-8, 2000
14.Adriana
Vasilache, New Distortion Bounds of Lattice VQ for Gaussian Sources
2000 International Symposium on Information Theory and Its Applications,
Sheraton Waikiki Hotel, Honolulu, Hawaii, pp. 81-84, November 5-8,
2000
15.Ioan
Tabus, Riitta Niemistö, Jaakko Astola, A line spectrum approach for
the design of optimum compaction FIR filters, EUSIPCO-2000, Tampere, September
4-8, Vol. 3,1779-1782, 2000.
16.B.
Dumitrescu, C.Popeea, A Low Complexity SDP Method for Designing Optimum
Compaction Filters, ICASSP 2000, Istanbul, Turkey, vol. 1, pp. 516-519,
2000.
17.Beghdadi,
A., Iordache, R., Block Truncation Coding Using Edginess Information
ICIP 2000 Proceedings, International Conference on Image Processing,
Vancouver, Canada September 10-13, pp. 175-178, 2000,
18.Iordache,
R. , Beghdadi, A., Robust Image Vector Quantization Transmission System
with an Embedded Degradation Control Mechanism, In: Gabbouj, M. & Kuosmanen,
P. (eds). Eusipco 2000, X European, Signal Processing Conference, Tampere,
Finland, September 4-8, pp. 2113-2116, 2000.
19.Riitta
Niemistö, Bogdan Dumitrescu, and Ioan Tabus, SDP Design Procedure
for Energy Compaction IIR Filters, ICASSP-2001, Salt Lake City,
Utah, May 7-11, 2001.
20.Adriana
Vasilache and Ioan Tabus, Indexing and entropy coding of lattice codevectors,
ICASSP-2001, Salt Lake City, Utah, May 7-11, 2001.
Lossless
signal compression
21.Ciprian
Doru Giurcaneanu, Ioan Tabus, Jaakko Astola, Adaptive context based sequential
prediction for lossless audio compression, Signal Processing, vol 80, pp.2283-2294,
2000.
22.Ioan
Tabus, Jorma Rissanen, Asymptotics of Greedy Algorithms for Variable-to-Fixed
Coding of Markov Sources, submitted to IEEE Trans. Information Theory,
April 2000.
23.Ciprian
Doru Giurcaneanu, Ioan Tabus, Serban Mereuta, Using contexts and R-R interval
estimation in lossless ECG compression, Computer Methods and Programs in
Biomedicine, Section I: Methodology, accepted Dec. 2000.
24.Ioan
Tabus, Gergely Korodi, Jorma Rissanen, Text compression based on variable-to-fixed
codes for Markov sources, DCC'2000, Data Compression Conference, Snowbird,
Utah, pg.133-141, March 28-30, 2000.
25.Ciprian
Doru Giurcaneanu, Ioan Tabus, and Jorma Rissanen, MDL based digital signal
segmentation, EUSIPCO-2000, Tampere, Vol. 1, pp. 339-342, September 4-8,
2000.
26.Ciprian
Doru Giurcaneanu, Ioan Tabus, Escape Sequences for Lossless Audio Compression,
2000 International Symposium on Information Theory and Its Applications,
Sheraton Waikiki Hotel, Honolulu, Hawaii, vol. 1, pp. 386-389, November
5-8, 2000
27.Ciprian
Doru Giurcaneanu, Ioan Tabus, On the sign of kurtosis, The Second International
Workshop on Independent Component Analysis and Blind Signal Separation,
ICA-2000, Helsinki, Finland, pp. 499-502, June 19-22, 2000.
28.Ciprian
Doru Giurcaneanu, Ioan Tabus, Low complexity integer to integer transform
coding, SPECLOG- 2000, Tampere, June 2-3, 2000.
29.Ciprian
Doru Giurcaneanu, Ioan Tabus, and S. Mereuta, Long term prediction using
QRS detection for lossless ECG compression. NSIP01, Baltimore, Maryland,
USA, June 3-6, 2001.
30.Ciprian
Doru Giurcaneanu and Ioan Tabus, Low-complexity transform coding with integer-to-integer
transforms. ICASSP-2001, Salt Lake City, Utah, May 7-11, 2001.
Estimation,
Optimization and Adaptive filtering
31.Bogdan
Dumitrescu, Ioan Tabus, Petre Stoica, On the parameterization of positive
real sequences and MA parameter estimation, submitted to IEEE Transactions
on Signal Processing, December 2000.
32.Ioan
Tabus , Ciprian Doru Giurcaneanu, and Jaakko Astola, Optimal predictive
design of Boolean and order statistics based filters, EUSIPCO-2000, Tampere,
Vol. 1, pp. 393-396, September 4-8, 2000.
33.Jari
Kataja, Ioan Tabus, Application of the LS-LMS algorithm in Active Noise
Control in ducts, IASTED Signal and Image Processing, Las Vegas, pp. 343-348,
Nov. 19-23, 2000.
34.Timo
Suojoki, Ioan Tabus, Nonlinear Filter Based Normalization Techniques For
Sonar Detection, NSIP01, Baltimore, Maryland, USA, June 3-6, 2001.
35.Ioan
Tabus and Jaakko Astola, Using MDL for Gene Expression Prediction from
Microarray Measurements, NSIP01, Baltimore, Maryland, USA, June 3-6,
2001.
36.Sari
Peltonen, Ioan Tabus, Jaakko Astola, Edward Dougherty, Nina Hirata
Robust optimization of stack filters, NSIP01, Baltimore, Maryland,
USA, June 3-6, 2001.
37.Ioan
Tabus and Jaakko Astola, MDL Optimal Design for Gene Expression
Prediction from Microarray Measurements Tampere University of Technology,
Technical Report,ISBN.952-15-0529-X,
November 2000.