U.S. patent application number 12/615077 was filed with the patent office on 2010-03-04 for systems and methods that provide frequency domain supplemental training of the time domain equalizer for dmt.
This patent application is currently assigned to AWARE, INC.. Invention is credited to Bindu Chandna, Arnon Friedmann, Jelena Jovin, Stuart D. Sandberg.
Application Number | 20100054321 12/615077 |
Document ID | / |
Family ID | 22911663 |
Filed Date | 2010-03-04 |
United States Patent
Application |
20100054321 |
Kind Code |
A1 |
Sandberg; Stuart D. ; et
al. |
March 4, 2010 |
Systems and Methods that Provide Frequency Domain Supplemental
Training of the Time Domain Equalizer for DMT
Abstract
Using a known or later developed time domain equalizer
coefficient training algorithm, a least square solution for the
time domain equalizer coefficients is taken at a starting point and
iteratively improved on. In particular, the improvement is directed
towards maximizing number of bits per frame loaded over the time
domain equalizer coefficient choice. This can be accomplished by
maximizing capacity directly rather than setting a goal to shorten
the channel and hoping that the capacity will be maximized as a
result.
Inventors: |
Sandberg; Stuart D.;
(Arlington, MA) ; Friedmann; Arnon; (Marlboro,
MA) ; Jovin; Jelena; (Somerville, MA) ;
Chandna; Bindu; (Burlington, MA) |
Correspondence
Address: |
Jason H. Vick;Sheridan Ross, PC
Suite # 1200, 1560 Broadway
Denver
CO
80202
US
|
Assignee: |
AWARE, INC.
Bedford
MA
|
Family ID: |
22911663 |
Appl. No.: |
12/615077 |
Filed: |
November 9, 2009 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
12013874 |
Jan 14, 2008 |
7636389 |
|
|
12615077 |
|
|
|
|
11616630 |
Dec 27, 2006 |
|
|
|
12013874 |
|
|
|
|
09982065 |
Oct 19, 2001 |
7180938 |
|
|
11616630 |
|
|
|
|
60241664 |
Oct 19, 2000 |
|
|
|
Current U.S.
Class: |
375/231 |
Current CPC
Class: |
H04L 25/03019 20130101;
H04L 25/03038 20130101; H04L 2025/03477 20130101; H04L 2025/03611
20130101; H04L 2025/03617 20130101; H04L 2025/0377 20130101; H04L
2025/03414 20130101; H04L 2025/03764 20130101 |
Class at
Publication: |
375/231 |
International
Class: |
H04L 27/01 20060101
H04L027/01 |
Claims
1. A system that performs supplemental frequency domain training of
a time domain equalizer comprising: a training module that uses an
initial solution for time domain equalizer coefficients and
determines updated time domain equalizer coefficients by maximizing
the number of bits per frame; and a time domain equalizer that
receives the updated time domain equalizer coefficients.
2. The system of claim 1, wherein the supplemental training is
performed during medley.
3. The system of claim 1 wherein, the updated time domain equalizer
coefficients are used during showtime.
4. The system of claim 1, wherein the training module receives a
data matrix from an echo canceller.
5. The system of claim 1 wherein additional training is performed
based on the updated time domain equalizer coefficients.
6. The system of claim 5, wherein the additional training comprises
frequency domain equalizer training and signal to noise ratio
measurements for bit loading.
7. The system of claim 1, wherein the system is located in one or
more of a DSL, VDSL, SDSL, HDSL, HDSL2, discrete multi-tone,
discrete wavelet multi-tone DSL or wireless OFDM system.
8. The system of claim 1, wherein the training module further
determines a mean squared signal value for each bin, for the given
time domain equalizer coefficients.
9. The system of claim 1, wherein the training module further
determines an average error squared value over a predetermined
number of frames for each bin.
10. The system of claim 1, further comprising using the updated
time domain equalizer coefficients as the initial solution for the
time domain equalizer coefficients.
11. A method that performs supplemental frequency domain training
of the time domain equalizer comprising: receiving a solution of
initial time domain equalizer coefficients; determining updated
time domain equalizer coefficients by maximizing the number of bits
per frame; and forwarding the updated time domain equalizer
coefficients to a time domain equalizer.
12. The method of claim 11, wherein the supplemental training is
commenced during medley.
13. The method of claim 11, further comprising determining a mean
squared signal value for each bin, for the given time domain
equalizer coefficients.
14. The method of claim 11, further comprising determining the
average error squared value over a predetermined number of frames
for each bin.
15. The method of claim 11 further comprising repeating the
determining step using the updated time domain equalizer
coefficients as the initial time domain equalizer coefficients.
16. The method of claim 11, further comprising estimating a channel
frequency response.
17. The method of claim 11, wherein the updated time domain
equalizer coefficients are used during showtime.
18. The method of claim 11, wherein the method is performed in one
or more of a DSL, VDSL, SDSL, HDSL, HDSL2, discrete multi-tone,
discrete wavelet multi-tone DSL or wireless OFDM system.
19. The method of claim 11, further comprising receiving a data
matrix.
20. The method of claim 11, further comprising performing
additional training is performed based on the updated time domain
equalizer coefficients.
21.-30. (canceled)
Description
RELATED APPLICATION DATA
[0001] This application claims benefit of and priority to U.S.
Application Ser. No. 60/241,664 filed Oct. 19, 2000, entitled
"Frequency Domain Supplemental Training of the Time Domain
Equalizer for DMT," which is incorporated herein by reference in
its entirety.
BACKGROUND OF THE INVENTION
[0002] 1. Field of the Invention
[0003] In general, the systems and methods of this invention relate
to time domain equalizer training. In particular, this invention
relates to systems and methods for supplemental frequency domain
training of the time domain equalizer in a discrete multi-tone
system.
[0004] 2. Description of Related Art
[0005] In Discrete Multi-tone Modulation (DMT) systems, the time
domain equalizer (TDQ) is a finite impulse response (FIR.) filter
located at the receiver side of a DSL modem. The TDQ is used to
reduce the intersymbol interference (ISI). If the channel is
shortened in time to have a length no greater than the length of
the cyclic prefix, the intersymbol interference can be eliminated.
Thus, a common method for training the TDQ in a DMT system is to
jointly optimize the numerator and denominator of the
autoregressive (AR) model for the channel where the order of the
numerator is equal to the cyclic prefix length and the denominator
is used as the TDQ setting. The training is based on transmission
and reception of a known reference signal, such as the reverb
signal in ADSL systems, using a least squares fit of the AR channel
model.
SUMMARY OF THE INVENTION
[0006] The supplemental training according to the exemplary systems
and methods of this invention starts with the least squares
solution of the time domain equalizer coefficients outlined above
as its starting point, and iteratively improves on it.
[0007] Specifically, the medley-based supplemental training which
is the subject of this application takes as input the output of a
reverb-based TDQ training algorithm. Examples of reverb-based
training algorithms are described in Stuart Sandberg and Michael
Tzannes, "Overlapped Discrete Multitone Modulation for High Speed
Copper Wire Communications," IEEE JSAC, vol 13, no. 9, December
1995, pg 1571-1585, incorporated herein by reference in its
entirety, and include channel shortening schemes based on an AR fit
to the transmission channel.
[0008] The improvement is geared towards maximizing the number of
bits per frame loaded over the TDQ choice. In particular, capacity
is maximized directly rather than setting a goal to shorten the
channel and hoping that the capacity would be maximized as a
result. The supplemental training operates in medley transmission
mode, and requires a number of pseudo-random data frames.
[0009] Medley operation is selected in that the reverb data
transmission, which is the repetitive transmission of the same
reference frame, would not produce ISI in the received signal, and
the SINR (Signal-to-Interference and Noise) determined in this way
would not take into account the very component of error that the
TDQ is intended to reduce. From the medley data, the SINR can be
estimated over the bins used in the actual data transmission mode,
and therefore, the number of bits per frame loaded.
[0010] The systems and methods of this invention use a directed
search method on the capacity function to obtain an improved TDQ.
The function in question is highly non-linear, and after
linearization, e.g., using the first two terms in a Taylor series
expansion around the starting TDQ point, a local extremum is
sought. Since this does not guarantee the best solution, the same
supplemental training can be repeated one or more times, each time
starting with the TDQ solution resulting from the previous run.
[0011] In accordance with an exemplary embodiment of the invention,
an aspect of the invention relates to performing frequency domain
supplemental training of a time domain equalizer.
[0012] Additionally, aspects of the invention also relate to
performing frequency domain supplemental training of a time domain
equalizer in a discrete multi-tone environment.
[0013] Additional aspects of the invention also relate to
performing the supplemental training numerous times, with each
instance of the supplemental training using the last determined
time domain equalizer coefficients to improve the quality of the
results.
[0014] These and other features and advantages of this invention
are described in, or are apparent from, the following detailed
description of the embodiments.
BRIEF DESCRIPTION OF THE DRAWINGS
[0015] The embodiments of the invention will be described in
detail, with reference to the following figures wherein:
[0016] FIG. 1 is a functional block diagram illustrating a DMT DSL
modem according to this invention;
[0017] FIG. 2 illustrates how supplemental training fits into a
sequence of ADSL receiver initialization/training tasks; and
[0018] FIG. 3 is a flowchart detailing the perform supplemental
training step of FIG. 2.
DETAILED DESCRIPTION OF THE INVENTION
[0019] The exemplary embodiments of this invention will be
described in relation to the application and the invention to an
ADSL transceiver environment. However, it should be appreciated
that in general, the systems and methods of this invention will
work equally well for any multi-carrier communication system
including, but not limited to, DSL, VDSL, SDSL, HDSL, HDSL2, or any
other discrete multi-tone, discrete wavelet multi-tone DSL or
wireless OFDM system.
[0020] As discussed above, the supplemental training according to
an exemplary embodiment of this invention commences with a Least
Squares solution for the TDQ and improves on the accuracy.
Specifically, the number of bits per frame loaded is maximized over
the TDQ choice. The function to maximize is the sum of the number
of bits that can be loaded in the bins that are used for
transmission, and the maximization is over the TDQ setting:
Max.sub.(a)(SUM.sub.(k)(log.sub.10(SINR.sub.k)),
where: [0021] a is the TDQ vector of size Lx1, [0022] k is the bin
index (out of N used bins, while N<M where M is the size of the
receiver Fourier Transform), and [0023] SINR.sub.k is the signal to
noise and interference ratio in bin k, expressed as a function of
TDQ coefficients, a. The above is the equivalent of minimizing the
following:
[0023] min.sub.(a)(SUM.sub.(k)(ln E[|e.sub.k|.sup.2]-ln
E[|s.sub.k|.sup.2])),
where:
[0024] E[|e.sub.k|.sup.2] is the mean square error in bin k,
[0025] E[|s.sub.k|.sup.2] is the mean square signal in bin k,
and
[0026] s.sub.k=u.sub.k H.sub.k A.sub.k a,
[0027] where: [0028] u.sub.k is the medley 4-QAM reference symbol
in bin k, [0029] H.sub.k is an estimated (during reverb training)
complex channel frequency response at bin k, [0030] A.sub.k is a
Fourier basis row vector of length L, having frequency 2.pi.k/M,
and e.sub.k=s.sub.k-F.sub.k B a,
[0031] where: [0032] F.sub.k is a Fourier basis row vector of
length M, having frequency 2.pi.k/M, [0033] B is the received data
matrix of size M.times.L, and [0034] each column of which is the
received data frame (before the TDQ block), shifted by one sample
as to represent the time passing operation. In the following, the
dependence of B, u.sub.k and s.sub.k on frame will sometimes be
shown explicitly as B(n), u.sub.k, (n), and s.sub.k(n), where n is
the frame index.
[0035] This function is highly nonlinear, and only the portion
around the TDQ starting point is modeled by taking the first two
terms of the Taylor series expansion. As a result, the function to
minimize (over TDQ setting a) is:
SUM.sub.(k)(w.sub.k.sup.eE[|e.sub.k|.sup.2]-w.sub.k.sup.sE[|s.sub.k|.sup-
.2]),
where the weights are:
[0036] w.sub.k.sup.e=1/E[|e.sub.k,0|.sup.2],
[0037] w.sub.k.sup.s=1/E[|s.sub.k,0|.sup.2], and
[0038] e.sub.k,0 and s.sub.k,0 are e.sub.k and s.sub.k, evaluated
for the initial TDQ setting.
[0039] E[.] is evaluated as an average, over medley frames.
After some manipulation, the function to be optimized can be
rewritten as:
min.sub.(a)(E[a'G.sub.ea-a'G.sub.sa])=min.sub.(a){a'E[G.sub.e]a-a'E[G.su-
b.s]a}
where:
[0040]
G.sub.e=G.sub.e.sub.--.sub.mat.sup.+W.sub.eG.sub.e.sub.--.sub.mat,
is a matrix of size L.times.L, and where + is the conjugate
transpose,
[0041]
G.sub.s=G.sub.s.sub.--.sub.mat.sup.+W.sub.sG.sub.s.sub.--.sub.mat,
is a matrix of size L.times.L,
[0042] G.sub.s.sub.--.sub.mat=D.sub.uD.sub.HA, is a matrix of size
N.times.L,
[0043] D.sub.u=diagonal (u) and D.sub.H=diagonal(H), are both
matrices of size N.times.N,
[0044] A is a Fourier basis matrix of size N.times.L, consisting of
previously described vectors A.sub.k,
[0045] G.sub.e.sub.--.sub.mat=G.sub.s.sub.--.sub.mat-FB, is a
matrix of size N.times.L,
[0046] F is a Fourier basis matrix of size N.times.M, consisting of
previously described vectors F.sub.k,
[0047] W.sub.e=diagonal(w.sub.k.sup.e),
W.sub.s=diagonal(w.sub.k.sup.s), are both matrices of size
N.times.N, and
[0048] B is a received data matrix of size M.times.L, as discussed
above.
The directed search for the minimum starts with the initial TDQ
vector, a.sub.o, and for each iteration the TDQ is updated:
a.sub.i=min eigenvector{E[G.sub.e]-E[G.sub.s]}
where E[G.sub.e]-E[G.sub.s] has been linearized/localized about
a.sub.i-1 as described above.
[0049] In practice, as discussed hereinafter, the iterations of the
supplemental training are continued until arriving at a TDQ with
satisfactory performance, or for some other predetermined number of
iterations. Note that to obtain the TDQ for a new iteration, the
TDQ from the previous iteration is used to estimate the signal, the
error, and to obtain the updated matrix E[G.sub.e]-E[G.sub.s].
[0050] FIG. 1 illustrates an exemplary DSL modem 5 according to
this invention. In particular, the DSL modem 5 comprises a bit
loading module 10, an encoder 20, an Inverse Fast Fourier Transform
module 30, a cyclic prefix module 40, an echo canceller 50, a
digital-to-analog converter 60, an analog-to-digital converter 70,
a time domain equalizer 80, a training module 90, a cyclic prefix
module 100, a Fast Fourier Transform module 110, a frequency domain
equalizer 120, a decoder 130 and a bit loading module 140. As will
be appreciated by one of ordinary skill in the art, various other
components may be present in a DSL modem, however have been omitted
for the sake of clarity.
[0051] While the exemplary embodiment illustrated in the FIG. 1
shows the modem 5 and various components collocated, it is to be
appreciated that the various components of the modem can be
combined or located at distant portions of a distributed network,
such as a local area network, a wide area network, an intranet
and/or the Internet, or within a modem. Thus, it should be
appreciated, that the components of the modem 10 can be combined
into one device or collocated on a particular node of a distributed
network or combined into one or more of a CO or CPE modem. Thus, it
will be appreciated from the following description, and for reasons
of computational efficiency, that the components of the modem 10
can be arranged any location, such as in a general purpose computer
or within a distributed network or dedicated modern without
affecting the operation of the system. Furthermore, the term module
as used herein is to be understood to include, but is not limited
to, one or more of hardware components and/or associated software
for performing a given function.
[0052] In operation, the encoder, in cooperation with the bit
loading module 10, receives the input data bit stream and encodes
it into M QAM constellation points. This encoding is accomplished
in accordance with a bit loading table that is stored in the bit
loading module 10. The bit loading table defines the number of bits
carried by each tone.
[0053] The IFFT module 30 receives the encoded data and determines
a sum of N carriers each modulated by a predetermined phase and
amplitude. Specifically, the input to the IFFT module 30 is a
vector of QAM constellation points -N complex numbers, defining the
amplitude and phase of each carrier.
[0054] The cyclic prefix module 40 receives the output of the IFFT
module 30 and separates the received symbols in time in order to
decrease the intersymbol interference (ISI). As is well known, the
signal passing through the line is linearly convolved with the
impulse response of the line. If the impulse response is shorter
than the duration of the cyclic prefix as discussed above, each
symbol can be processed separately, thereby eliminating the
intersymbol interference.
[0055] The echo canceller 50 generates a replica of the transmitted
signal that leaks back into the receiver. Upon subtraction of the
near-end echo-replica, the received far-end signal can be processed
as if its only impairment has been the channel induced noise
sources. In general, the echo cancellation in DSL systems considers
the asymmetric upstream/downstream nature that results in different
sampling rates for upstream and downstream communications. However,
many variations and methods for reducing echo are well known to one
of ordinary skill in the communications arts and will not be
discussed herein.
[0056] The time domain equalizer module 80 is a filter designed to
minimize the intersymbol interference and interchannel interference
(ICI). This is done by reducing the total impulse response of the
line to the length of the cyclic prefix, as discussed above, such
that one symbol does not interfere with the next symbol and
accordingly intersymbol interference can be reduced or
eliminated.
[0057] The cyclic prefix module 100 complements the cyclic prefix
module 40 and forwards its output to the FFT module 110. The FFT
module 110 complements the operation of the IFFT module 30 by
transforming the received N carriers back into amplitude and phase
information, which is then decoded back into bits in cooperation
with the decoder 130 and the bit loading module 140.
[0058] The training module 90 manages a number of training features
that are present in the ADSL modem system 5. However, for the sake
of clarity, only the training related to the application of this
invention will be described. Clearly, one of ordinary skill in the
art will appreciate that additional training will be present during
the training and/or operating condition of a typical DSL modem.
[0059] In particular, during a portion of initial training, the DSL
modem 5 enters into reverb. During this reverb training, and in
conjunction with the training module 90, an estimate of the channel
frequency response (H.sub.k) is determined. For example, as
discussed in the Sandberg article referenced above, the channel
frequency response can be estimated.
[0060] Next, a reverb based TDQ training algorithm, such as the one
discussed the Sandberg article referenced above, is used to
determine the initial TDQ coefficients. Upon determination of these
coefficients, which are stored in a memory (not shown) in the
training module 90, medley is commenced. During medley, the
training module 90, operating on data received from the echo
canceller 50, performs one or more supplemental TDQ training
sessions according to the systems and methods of this invention.
Then, the updated time domain equalizer coefficients are provided
from the training module 90 to the time domain equalizer 80.
[0061] Based on the determined TDQ, additional medley training is
performed such as, but not limited to, FDQ training and SNR
measurements for bit loading. At this point, the DSL modem 5 is
ready to enter showtime.
[0062] FIG. 2 outlines an exemplary method of performing
supplemental training to determine updated time domain equalizer
coefficients according to an exemplary embodiment of this
invention. In particular, control begins in step S100 and continues
to step S110. In step S110, reverb is commenced. Next, in step
S120, the channel frequency response (H.sub.k) is estimated. Then,
in step S130, a reverb based TDQ training algorithm is used to
determine the initial TDQ coefficients. Control then continues to
step S140.
[0063] In step S140, medley is commenced. Next, in step S150, the
supplemental TDQ training in accordance with this invention
determines the improved time domain equalizer coefficients by
maximizing the number of bits per frame. Then, in step S160, the
updated time domain equalizer coefficients are provided to the time
domain equalizer for use during showtime. Control then continues to
step S170.
[0064] In step S170, based on the determined TDQ, additional medley
training such as frequency domain equalizer training and
signal-to-noise ratio measurements for the determined time domain
equalizer coefficients that are used for the bit loading is
completed. Then, in step S190, the modem enters the showtime.
Control then continues to step S200, where the control sequence
ends.
[0065] FIG. 3 outlines in greater detail the perform supplemental
training block S150 in FIG. 2. In particular, control begins in
step S500 and continues to S510. In step S510, the time domain
equalizer coefficients are initialized. Next, in step S520, m is
set to zero. Then, in step S5530, the mean squared signal value is
determined for each bin, for the given time domain equalizer
coefficients. Note that the s.sub.k is equal to the medley 4-QAM
reference symbol in bin k multiplied by the estimated complex
channel frequency response at bin k, obtained during reverb
training, multiplied by a Fourier basis row vector of length L
having frequency 2.pi.k/m, multiplied by the time domain equalizer
coefficients (a). Control then continues to step S540.
[0066] In step S540, n is set equal to zero and the mean square
error is also set to zero. Next, in steps S550 through S570, the
average error squared value is evaluated, over N.sub.1 frames, for
each bin. Thus, the signal s.sub.k (n) and the received data B(n)
are frame dependent. Then, in step S580, W.sub.s and W.sub.e will
be established as diagonal matrices with elements W.sub.k.sup.s and
W.sub.k.sup.e. This allows localization of linearized metrics about
the current time domain equalizer coefficients. Control then
continues to step S590.
[0067] In step S590, n is set equal to zero, and the matrices
G.sub.e and G.sub.s are initialized to all-zeros matrices. Next, in
steps S590 through S650, G.sub.s and G.sub.e, which are functions
of the reference signal u.sub.k(n) and the received data B(n) for
each frame, are averaged over N.sub.2 frames. Note that
a.sup.+{E[G.sub.e]-E[G.sub.S]} a is a linearized/localized
approximation for
.SIGMA. log.sub.10(1/SNR.sub.k)
[0068] which is the metric to be minimized. Then, for the next TDQ
vector, the minimum Eigen vector solution is determined and the
process is repeated using the updated determined localization.
Control then continues to step S690 where the control sequence
ends.
[0069] As illustrated in FIG. 1, the time domain equalizer
coefficient determination system can be implemented either on a
single program general purpose computer, or a separate program
general purpose computer. However, the time domain equalizer
coefficient determination system can also be implemented on a
special purpose computer, a programmed microprocessor or
microcontroller and peripheral integrated circuit element, an ASIC
or other integrated circuit, a digital signal processor, a
hard-wired electronic or logic circuit such as a discrete element
circuit, a programmable logic device such as a PLD, PLA, FPGA, PAL,
a modem, or the like. In general, any device capable of
implementing a finite state machine that is in turn capable of
implementing the flowcharts can be used to implement the time
domain equalizer coefficient determination system according to this
invention.
[0070] Furthermore, the disclosed method may be readily implemented
in software using object or object-oriented software development
environments that provide source code that can be used on a variety
of computer or workstation hardware platforms. Alternatively, the
disclosed line time domain equalizer coefficient determination
system may be implemented partially or fully in hardware using
standard logic circuits or VLSI design. Whether software or
hardware is used to implement the systems in accordance with this
invention is dependent on the speed and/or efficiency requirements
of the system, the particular function, and the particular software
and/or hardware systems or microprocessor or microcomputer systems
being utilized. The time domain equalizer coefficient determination
system and methods illustrated herein, however, can be readily
implemented in hardware and/or software using any known or
later-developed systems or structures, devices and/or software by
those of ordinary skill in the applicable art from the functional
description provided herein and a general basic knowledge of the
computer and communications arts.
[0071] Moreover, the disclosed methods may be readily implemented
as software executed on a programmed general purpose computer, a
special purpose computer, a microprocessor, or the like. In these
instances, the methods and systems of this invention can be
implemented as a program embedded on a personal computer such as a
Java.RTM. or CGI script, as a resource residing on a server or
graphics workstation, as a routine embedded in a dedicated line
characterization system, a modem, a dedicated time domain equalizer
coefficient determination system, or the like. The time domain
equalizer coefficient determination system can also be implemented
by physically incorporating the system and method into a software
and/or hardware system, such as the hardware and software systems
of a time domain equalizer coefficient determination system or
modem, such as a DSL modem.
[0072] It is, therefore, apparent that there has been provided, in
accordance with the present invention, systems and methods for
determining time domain equalizer coefficients. While this
invention has been described in conjunction with a number of
exemplary embodiments, it is evident that many alternatives,
modifications and variations would be or are apparent to those of
ordinary skill in the applicable arts. Accordingly, the invention
is intended to embrace all such alternatives, modifications,
equivalents and variations that are within the spirit and scope of
this invention.
* * * * *