This freely distributable toolbox provides a graphical user interface and
functions in Matlab to design four different multicarrier equalizer structures:
conventional, dual-path, per tone, and filter bank.
A total of 18 design methods to compute the equalizer coefficients have
implemented: 13 for conventional, two for dual-path, two for per tone,
and one for filter bank equalizers.
Default parameters are from the G.DMT ADSL standard for downstream transmission.
This toolbox was initially released in Fall 2000.
To run the graphical user interface, type teqdemov3.
The user will then be asked to choose one of four equalizer
architectures to design.
New features of version 3.1 vs. version 3.0 include
Conventional equalizer methods
New symmetric maximum shortening SNR method for time-domain
equalizer design (from Cornell-UT Austin papers)
The target impulse response is displayed only for those methods
that define one (minimum mean-squared error training methods)
when the shortened impulse response is displayed.
Per-tone equalizer equalizer design methods
Trained with 511 symbols
(from KU Leuven recommendation)
New minimum mean squared error method per tone
(from KU Leuven paper)
New time-domain filter bank equalizer
(from UT Austin paper)
Toolbox-wide improvements
Improved spectral estimation (sliding FFT with a Hamming window
and 50% overlap of samples in consecutive blocks)
Information window gives references for all training methods
New button to save equalizer designs as Matlab text files
Known bugs and limitations include
The MMSE per-tone and time-domain filter bank equalizer training methods
do not by default search for the optimal transmission delay,
due to the amount of execution time required, and use a heuristic
to determine the transmisson delay parameter instead.
As a result, their bit rates are 10-20% below optimal.
There is an option that can be enabled to search for the
optimal transmission delay.
In Matlab 6.0 under Redhat Linux, the initial popup window
from running teqdemov3 does not display the logo for
the UT Austin Embedded Signal Processing Laboratory nor provide
scroll bars for the text in the right-hand window.
This version adds design methods for two new DMT equalizer architectures,
dual-path TEQ [8] and frequency-domain per-tone equalizer [9], to the
existing support for a conventional single-TEQ architecture [1-6].
To run the graphical user interface, type teqdemov3.
The user will then be asked to choose one of the three
equalizer architectures.
The bit rate calculations for per tone equalizer are based on
the estimation of mean square error between input-output data for
each subband, whereas the bit rate calculations of the other two
architectures are based on a matched filter bound for a single TEQ.
This version chooses default simulation parameters to be consistent
with the G.DMT ADSL standard, fixes several bugs including one in
the bit rate calculation, and works in Matlab 6.
To run the graphical user interface, type teqdemo.
This is the user's manual for versions 1.0 and 2.0.
It covers most of the features of version 3.0.
Introduction
The MATLAB DMTTEQ Toolbox is a collection of MATLAB functions to
design and test time domain equalizer (TEQ) design methods.
For the conventional equalizer architecture of a single FIR TEQ,
the toolbox implements the following TEQ design methods:
Minimum mean squared error -- unit energy constraint [1]
Minimum mean squared error -- unit tap constraint [1]
Maximum shortening signal to noise ratio (SSNR) method [2]
Maximum geometric SNR method [3]
Divide and conquer -- cancellation [4]
Divide and conquer -- minimization [4]
Maximum bit rate method [5]
Minimum intersymbol interference method [5]
Matrix pencil design method [4]
Modified matrix pencil design method [4]
Eigen approach [6]
Autoregressive Moving Average
Symmetric maximum shortening SNR [10]
The toolbox has a graphical user interface (GUI) which enables the design
of a TEQ by one of the methods above and the testing of its performance.
A snapshot of the GUI for designing single FIR TEQs is shown below.
Description
In the upper right of the control window is a pulldown menu
from which a design method can be chosen.
Below this pulldown menu are the following editable text windows
which are used to set the design and simulation parameters:
Shortened impulse response (SIR) length. This is the desired length
of the channel after equalization. For example, it should be set to 33
(one plus the cyclic prefix length) for the ANSI and G.DMT ADSL standards.
Time domain equalizer (TEQ) length. Defines the number of taps of the
TEQ.
Fast Fourier transform (FFT) size. Sets the FFT size used in DMT
modulation. It is twice the number of subchannels.
Coding gain (dB). Defines a coding gain in dB which is used during
capacity calculations [7]
Margin (dB). Sets the desired system margin in dB. This is also
used in capacity calculations [7]
Dmin and Dmax. The interval of Delta [Dmin Dmax]
in which to search for the optimal delay value.
Input power (dBm). Defines the input signal power in dBm.
AWGN power (dBm/Hz). Sets the amount of additive white Gaussian noise
in dBm/Hz. AWGN is added to the near-end crosstalk noise.
CSA loop #(1-8). Selects the desired ADSL channel on which to run
the simulation. Currently the eight standard CSA loops are supported.
Below the editable text windows is another pull-down menu which
is used to select the desired graph to be displayed.
The following graphics can be selected:
Target & shortened channel. Displays the shortened channel impulse response and the target channel impulse response for the minimum mean-squared
error (MMSE) and geometric SNR methods. For all other methods, the location
of the target window is displayed instead of a target impulse response.
TEQ impulse response.
Shows the impulse response of the TEQ.
TEQ frequency response.
Shows the frequency response of the TEQ.
SNR & MFB.
The SNR and matched filter bound (MFB) to the SNR is displayed as a
function of frequency (subchannels).
Original & shortened channel.
Displays the channel impulse response before and after equalization.
Noise power spectrum.
Shows the power spectrum of the noise which consists of NEXT noise plus AWGN.
Delay plot.
Displays the performance measure (i.e., MSE, SSNR, and channel capacity) of
the method with respect to the delay.
Equalized channel frequency response.
Displays the frequency response of the channel after equalization.
The two remaining buttons in the control frame are
Info. Displays information on how to use the GUI.
Calculate. Starts the calculation and performance evaluation of the
TEQ.
The following performance measures are calculated and listed in the
table:
Rate. Gives the achievable bit rate with the given channel and TEQ
settings.
SNR. Shows the SNR at the output of the equalizer in dB.
SSNR. Shows the shortening SNR in dB. This is defined as the
ration of the energy of the shortened channel impulse response in the
target window the the energy outside the target window.
MSE. Gives the MSE for the MMSE and geometric SNR methods.
Delay. Shows the optimal delay for the system.
Max Rate. Shows the absolute maximum achievable bit rate given the
channel and equalizer settings. It is calculated from the MFB.
Once all of the design and simulation parameters are set to the desired values
and the design method is chosen, the user hits the ``Calculate'' button
to start the calculations.
The simulator first loads the channel information and generates the
channel noise according to the parameter values.
Then, it will generate a transmit sequence and pass it through the
channel to obtain a received signal.
In the next step, the simulator estimates the power spectra of the
transmitted signal and channel noise.
It also estimates the magnitude square of the channel frequency response.
Based on these estimates, the SNR in each subchannel is estimated.
The simulator then calls the desired TEQ design function to calculate
the equalizer taps, target impulse response (if it exists for that method),
and optimal delay.
All of the results are then passed to a performance evaluation function
which returns the six performance measures.
The selected graph is plotted and the results are written in the table.
For different graphs the simulations does not need to be run again, all
results are saved.
References
[1] N. Al-Dhahir and J. M. Cioffi, "Efficiently computed reduced-parameter
input-aided MMSE equalizers for ML detection: A unified approach,"
IEEE Trans. on Info. Theory, vol. 42, pp. 903-915, May 1996.
[2] P. J. W. Melsa, R. C. Younce, and C. E. Rhors, "Impulse response
shortening for discrete multitone transceivers," IEEE Trans. on
Communications, vol. 44, pp. 1662-1672, Dec. 1996.
[3] N. Al-Dhahir and J. M. Cioffi, "Optimum finite-length equalization for
multicarrier transceivers," IEEE Trans. on Communications, vol. 44,
pp. 56-63, Jan. 1996.
[5] G. Arslan ,
B. L. Evans,
and S. Kiaei,
"Equalization for Discrete Multitone Receivers To Maximize Bit Rate,"
IEEE Trans. on Signal Processing,
vol. 49, no. 12, pp. 3123-3135, Dec. 2001.
[6] B. Farhang-Boroujeny and
Ming Ding,
"Design Methods for Time Domain Equalizer in DMT Transceivers",
IEEE Trans. on Communications, vol. 49, pp. 554-562, March 2001.
[7] J. M. Cioffi, A Multicarrier Primer. Amati Communication
Corporation and Stanford University, T1E1.4/97-157, Nov. 1991.
[9] K. V. Acker, G. Leus, M. Moonen, O. van de Wiel, and T. Pollet,
"Per tone equalization for DMT-based systems,"
IEEE Trans. on Communications,
vol. 49, no. 1, pp. 109-119, Jan 2001.
[10] R. K. Martin, C. R. Johnson, Jr., M. Ding, and B. L. Evans,
"Exploiting Symmetry in Channel Shortening Equalizers",
Proc. IEEE Int. Conf. on Acoustics, Speech, and Signal Proc.,
April 6-10, 2003, Hong Kong, China.
