IPapers from ICASSP 2005

1. L. Borup, M. Nielson, and R. Gribonval, "Nonlinear approximation
with redundant dictionaries", ICASSP 2005. pp IV-261-IV-265.  
The paper talks about implementing a nonlinear method (redundant
wavelets bi-frames) for retrieving and storing the data in a compact
form.

2. Y. Cao and M. Ibnkahla, "Exact symbol error rate and total
degradation performance of nonlinear M-QAM fading channels", ICASSP
2005, pp III-985 - III-988.
The paper studies the nonlinear model for the fading channels most
suitable for Satellite mobile communication. Based on the nonlinear
distortion and the efficiency of power amplifiers, the system
degradation parameters are optimized. But, the five systems studied do
not behave the same over the same settings for this optimality.

3. M. Chen, D. Mandic, T. Gautama, and M. Hulle, "On nonlinear modular
neural filters," ICASSP 2005, V-317 - V-320.  
Pipelined recurrent neural networks architecture is implemented based
on the nonlinear predictor calculation on modular nested recurrent
NNs. The nonlinear implementation helps in improving on the
computational efficiency over the single network. The approach
concludes that for the nonlinear system, PRNN approximates the
nonlinearities much better than the other variants.

4. J. Jeraj and V. Mathews, "Identification of nonlinear, memoryless
systems using Chebyshev nodes," ICASSP 2005, I-93 - I-96 
The paper gives a approach in finding the nonlinearities from the
input-output measurements using the minimax approximation. The
algorithm finds the first estimate of output at the Chebyshev nodes
using a localized linear representation and then steps ahead to solve
the first moment. The system overcomes the influence of measurement
errors on the nonlinearities.

5. F. Kuech, A. Mitnacht, W. Kellermann, "Nonlinear acoustic echo
cancellation using adaptive orthogonalized power filters," ICASSP
2005, III-105 - III-108.
The paper explains the implementation of nonlinear echo cancellation
technique to over the nonlinear distortions caused by the combination
of acoustic echo cancellation and the loudspeakers and their
amplifiers. The speech signal is inheretantly non-stationary, but
implementing a nonlinear acoustic echo cancellation technique on
mobile communication receivers improves the quality. The paper
presents the simulation results.

6. S. Goh and D. Mandic, "A class of gradient-adaptive step size
algorithms for complex-valued nonlinear neural adaptive filters,"
ICASSP 2005, V-253 - V-256.
Using an approach of gradient-adaptive step size algorithms for the
nonlinear system, the signal can be tracked yielding lesser errors and
faster response. The same system can be used for predicting a signal.

7. W. Leong, and J. Homer, "EKENS: A learning on nonlinear blindly
mixed signals," ICASSP 2005, IV-81 - IV-84.  
The paper proposes a technique - EKENS - to perform source separation
in nonlinear mixtures. EKENS does a good job related to mapping in
nonlinear domain by adapting to the actual statistical distribution of
the sources. The kernel density distribution at the output signal in a
noisy environment is estimated to achieve nonlinear mapping.

8. W. Ling, Y. Ho, J. Reiss, and X. Yu, "Nonlinear behaviors of band
pass sigma-delta modulators with stable system matrices,", ICASSP
2005, IV-73 - IV-76.  
Using state-space variables for representing the signals with rich
frequency spectra is feasible. The approach can be used to above
unwanted tones generated by the quantizers.  Paper brings out the
relationship between fractal patterns for periodic output sequence.

9. Y. Mahgoub and R. Dansereau, "Voicing-state classification of
co-channel speech using nonlinear state-space reconstruction," ICASSP
2005, pp I-409 - I-412.
Under the notion that the nonlinear state-space reconstruction would
result in better classification of co-channel speech signal into
unvoiced-voiced, unvoiced-unvoiced, and voiced-voiced. Even if the
speech training data is not used, the approach excels the other
methods - Bayesian and SAPVR methods. The applications sorted out are
the areas in which the training data is not available. Usually, it is
seen that the training data can make the system speaker- and
environment- dependent.

10. A. Malipatil, Y. Huang, S. Andra, and K. Bennett, "Kernelized
set-membership approach to nonlinear adaptive filtering," ICASSP 2005,
pp IV-449 - IV-152.  
The approach is tested on equalization of nonlinear Inter-Symbol
interference (ISI) channels and predistortion of nonlinear high power
amplifiers to check suitability of the SM-nonlinear filtering. As the
filter coefficients can be updated online. BER or SER (symbol error
rate) performance for SM-NLMS is far superior as compared to SVMs and
SM-BC algorithms.

11. G. Sicuranzo and A. Carini, "Nonlinear Multichannel active noise
control using partial updates," ICASSP 2005, pp III-109 - III-112.
The approach refers to reducing the computational complexity for
modeling nonlinear multichannel active noise controllers. The affine
projection adaption algorithm is modified to take advantage of
truncated Volterra filters in an idea to achieve suitable partial
update strategies. The end result is reduced computation.