U.S. patent application number 10/527783 was filed with the patent office on 2010-11-11 for method and apparatus for dc offset compensation in a digital communication system.
Invention is credited to Shousheng He.
Application Number | 20100284496 10/527783 |
Document ID | / |
Family ID | 31896960 |
Filed Date | 2010-11-11 |
United States Patent
Application |
20100284496 |
Kind Code |
A1 |
He; Shousheng |
November 11, 2010 |
Method and apparatus for dc offset compensation in a digital
communication system
Abstract
A method of compensating for dc offset of a received signal
transmitted over a channel having a plurality of paths, the
received signal comprising a modulated data signal and a modulated
known training sequence signal, the method comprising the steps of:
constructing (104) from the known training sequence signal a first
regression matrix; path-combining (106) the incrementally rotated
elements of the first regression matrix to produce the elements of
a trend matrix; deriving (108) a neutralized second regression
matrix from the first regression matrix and the trend matrix;
utilising the neutralized second regression matrix to compensate
(110, 112) for dc offset of the received modulated data signal.
Inventors: |
He; Shousheng; (Sodra
Sandby, SE) |
Correspondence
Address: |
ERICSSON INC.
6300 LEGACY DRIVE, M/S EVR 1-C-11
PLANO
TX
75024
US
|
Family ID: |
31896960 |
Appl. No.: |
10/527783 |
Filed: |
September 15, 2003 |
PCT Filed: |
September 15, 2003 |
PCT NO: |
PCT/EP03/10245 |
371 Date: |
September 14, 2006 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
60413798 |
Sep 25, 2002 |
|
|
|
Current U.S.
Class: |
375/319 |
Current CPC
Class: |
H04L 25/025 20130101;
H04L 25/061 20130101; H04L 25/0244 20130101 |
Class at
Publication: |
375/319 |
International
Class: |
H04L 25/06 20060101
H04L025/06 |
Foreign Application Data
Date |
Code |
Application Number |
Sep 16, 2002 |
EP |
02256376.1 |
Claims
1. A method of compensating for dc offset of a received signal
transmitted over a channel having a plurality of paths, the
received signal comprising a modulated data signal and modulated
known training sequence signal bits, the method comprising the
steps of: constructing from the known training sequence signal, a
first regression matrix; constructing from the first regression
matrix, a trend matrix wherein each column of the trend matrix is a
path-trend vector; deriving a neutralized second regression matrix
from the first regression matrix and the trend matrix; and
utilising the neutralized second regression matrix to compensate
for dc offset of the received modulated data signal.
2. A method according to claim 1, wherein the path-trend vectors
are derived by .PSI. k = .OMEGA. ( n - m + 1 ) .PHI. k = .omega.
.omega. * ( n - m + 1 ) .PHI. k ##EQU00007## wherein .PSI..sub.k is
a path-trend vector .OMEGA. is a Toeplitz matrix generated by a
rotation vector .omega. (.omega.* is the de-rotation vector)
.PHI..sub.k is the corresponding element of the first regression
matrix, n is the number of symbols in the training sequence and m
is the number of paths of the channel.
3. A method according to claim 1, wherein the neutralized second
regression matrix comprises the difference between the first
regression matrix and the trend matrix.
4. A method according to claim 1, wherein the neutralized second
regression matrix comprises the difference between the first
regression matrix and the real part of the elements of the trend
matrix.
5. A method according to claim 4, wherein the real part of the
elements of the trend matrix are scaled by a suppression
factor.
6. A method according to claim 1, wherein the dc offset is
estimated from a trend vector of the received signal, the trend
matrix and channel estimation.
7. A method according to claim 6, wherein the channel estimation is
derived using Least-Squares technique.
8. A method of calculating an unbiased channel estimation for a
multi-path propagation channel, the method comprising the steps of:
constructing a first regression matrix from a known training
sequence signal of an input signal; constructing from the first
regression matrix, a trend matrix wherein each column of the trend
matrix is a path-trend vector; deriving a neutralized second
regression matrix from the first regression matrix and the trend
matrix; and calculating the unbiased channel estimation using the
neutralized second regression matrix.
9. A method according to claim 8, wherein the path-trend vectors
ark derived by .PSI. k = .OMEGA. ( n - m + 1 ) .PHI. k = .omega.
.omega. * ( n - m + 1 ) .PHI. k ##EQU00008## wherein .PSI..sub.k is
a path-trend vector .OMEGA. is a Toeplitz matrix generated by a
rotation vector .omega. (.omega.* is the de-rotation vector)
.PHI..sub.k is the corresponding element of the first regression
matrix, n is the number of symbols in the training sequence and m
is the number of paths of the channel.
10. A method according to claim 8, wherein the neutralized second
regression matrix comprises the difference between the first
regression matrix and the trend matrix.
Description
TECHNICAL FIELD
[0001] The present invention relates to DC offset (or biased noise)
compensation in digital communication systems. In particular, but
not exclusively, it relates to biased noise/DC offset compensation
in digital communication systems where a'known training sequence of
limited length is transmitted together with data burst for the
estimation of a multi-path (or multi-tap) signal propagation
channel, such as in a normal burst of TDMA/GSM/EDGE systems.
BACKGROUND OF THE INVENTION
[0002] The problem with biased noise/DC offset exists in,' among
others, homodyne receivers that convert radio frequency signal
directly into base band signal. Various factors such as components
mismatch, local oscillator leakage and interferences may contribute
to this distortion. When the modulation of the transmitted signal
consists of a rotation operation, for example in GSM/EDGE systems,
the DC offset will causes a single frequency trend in the received
signal after de-rotation (demodulation). If this frequency trend is
left uncompensated, the DC offset can cause significant receiver
performance degradation.
[0003] Several methods are known and are currently used for
compensation of the DC offset.
[0004] One such known method is blind DC estimation. This is the
simplest and most straightforward method. The received signal is
averaged before de-rotation. When this is applied to TDMA systems
which have limited symbols in a burst, this method is not accurate
due to uncertainty of the data symbols in the transmission.
[0005] Another such known method is joint channel and DC
estimation. This is used when the DC offset has been treated as an
extra tap in the multi-tap channel estimation, utilising the
constellation rotation as a reference, for example the .pi./2
rotation in GSM and the 3.pi./8 rotation in the EDGE modulation. It
has been observed, however, that the performance of this method
depends on the training sequence used. It does not perform well
when a training sequence has high amplitude at the trend frequency.
Further, the accuracy of channel estimation is compromised due to
the fact that an extra parameter needs to be estimated with the
same training sequence. In addition, the performance is also
affected by the form of burst synchronisation.
[0006] A further known method is referred to as classical trend
elimination as disclosed for example in "System Modelling and
Identification"; R. Johansson, in particular pages 83 to 85, pages
126 and 127 and pages 464 and 465. This is a method in system
identification. When a stimulating sequence {t.sub.k} of length n
is applied to a linear system with m parameters, a system response
{x.sub.k} of length n is collected for the system identification.
Trend elimination modifies the system model where neutralized
sequences are used for both stimulus and observation. The
neutralized sequence of the stimulus s.sub.k and the neutralized
sequence of the observation (input signal) y.sub.k are derived as
follows:
s k = t k - .tau. .tau. = 1 n i = 0 n - 1 t i ##EQU00001## y k = x
k - .rho. .rho. = 1 n i = 0 n - 1 x i ##EQU00001.2##
[0007] This method, however, can be employed only if either the
training sequence, the stimulus, is significantly longer than the
model order n>>m, or just a single parameter is sufficient
for system identification. It cannot be applied in a digital
communication system where a multi-path channel is required to be
estimated with a limited training sequence.
SUMMARY OF THE INVENTION
[0008] A generalised trend elimination processing is proposed,
where a single frequency trend, possibly a pure DC offset when the
trend frequency is zero, is taken into consideration in an unbiased
multi-path channel estimation to compensate for the biased noise/DC
offset.
[0009] The DC offset estimation/compensation of the present
invention may be utilised in a radio receiver working in a
multi-path channel environment with or without rotation modulation.
The estimation/compensation method of the present invention can
provide better channel estimation in the presence of biased noise
or DC offset since the method of the present invention reduces the
sensitivity to the distortion.
[0010] This is achieved by adjusting the trend suppression level
such that the channel estimation is less variant when different
training sequences, or different training sequence segments are
used.
[0011] The method of the present invention is computationally
simpler since the initial channel estimation is separated from DC
offset estimation. In addition, path-trend vectors can be
pre-calculated, which further reduces the computational load.
[0012] The invention extends trend elimination (bias/DC offset
compensation) to include single frequency trends and enables
multi-path channel estimation with relatively short training
sequences. The term DC offset is generalised to include an additive
single frequency component which causes static (zero frequency) or
incremental (non-zero frequency) phase offset. Depending on the
rotation scheme in the modulation, this phase offset increment can
be considered as the normalised trend frequency. The objective of
the invention is to acquire channel estimation with
eliminated/suppressed trend and to obtain trend estimation for DC
offSet compensation.
BRIEF DESCRIPTION OF DRAWINGS
[0013] FIG. 1 is a flow diagram illustrating the method according
to an embodiment of the present invention.
DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS
[0014] An embodiment of the present invention will now be described
with reference to FIG. 1. The digital communication system
comprises a modulator and transmitter (not shown here). The
modulator modulates a signal for transmission using known
techniques. The signal to be transmitted comprises a plurality of
data bits and a plurality of training sequence bits which
constitute a normal burst. The training sequence being a sequence
of bits known by the receiver of the digital communication system
allows the receiver to determine precisely the position of the data
bits within a burst, and to enable the receiver to derive the
distortion caused by transmission etc. In the digital communication
system in which the method of the present invention is utilised,
transmission of the burst is made over a multi-path propagation
channel comprising m-taps. The distortion is derived, in part, from
an estimation of the propagation channel. Application of the
training sequence to the channel estimation provides the distortion
when the result is compared to the actual received training
sequence.
[0015] In general, linear regression with Least-Squares (LS) error
criterion is used to obtain channel estimation in digital
communication systems. In LS estimation, the estimation model can
be expressed as
x=.PHI.h+v
which includes a channel vector (m is the span of the channel),
h=[h.sub.0h.sub.1 . . . h.sub.m-1].sup.T
a received signal vector
x=[x.sub.0s.sub.1 . . . x.sub.n-m].sup.T
a noise vector
v=[v.sub.0v.sub.1 . . . v.sub.n-m].sup.T
and a regression matrix
.PHI. = [ .PHI. 0 .PHI. 1 .PHI. m - 1 ] = [ t m - 1 t m - 2 t 1 t 0
t m t m - 1 t n - 1 t n - m ] ##EQU00002##
where each column vector in the regression matrix is defined as
the-path-regression vector
.PHI..sub.k=[t.sub.m-1-kt.sub.m-k . . .
t.sub.n-2-kt.sub.n-1-k].sup.T
[0016] Therefore, the regression matrix .PHI. is constructed 104
from samples of a known training sequence of a received signal by
minimising the least-square error, without DC compensation, channel
estimation with LS results in
h=(.PHI..sup.T.PHI.).sup.-1.PHI..sup.Tx
[0017] In the presence of DC offset, this equation gives erroneous
results.
[0018] In the method of an embodiment of the present invention, a
trend matrix .PSI. of the same dimension as .PHI. is introduced
106, in which each column is defined as a path-trend vector of size
n-m+1
.PSI. k = .OMEGA. ( n - m + 1 ) .PHI. k = .omega. .omega. * ( n - m
+ 1 ) .PHI. k ##EQU00003##
[0019] where ( )* notifies conjugate transposition, and .OMEGA. is
a Toeplitz matrix generated by the rotation vector .omega., (the
de-rotation vector having the form of .omega.*), where
.omega.=[1e.sup.j.beta.e.sup.j2.beta. . . .
e.sup.j(n-m).beta.].sup.T
[0020] The phase shift (the nominal trend frequency) Pin the
rotation vector is system dependent. For example, in GSM
.beta.=.pi./2 while in EDGE .beta.=3.pi./8. When .beta.=0, i.e. the
constellation does not rotate in the modulation, .OMEGA.
degenerates to a unit matrix with every element equal to 1, and the
path-trend vector degenerates to a unit vector in the regression
matrix as
.PSI. k ( l ) .beta. = 0 = 1 n - m + 1 i = 0 n - m .phi. k ( i ) =
1 n - m + 1 i = 0 n - m t k + i , ##EQU00004## l = 0 , , n - m
##EQU00004.2##
[0021] The method according to an embodiment of the present
invention constructs 108 a new "neutralized" path-regression
vectors:
.theta..sub.k=.phi..sub.k-.psi..sub.k, k=0, 1, . . . , m-1
and a new regressor matrix
.THETA.=.PHI.-.PSI.=[.theta..sub.0, .theta..sub.1 . . .
.theta..sub.m-1]
[0022] Utilising this new regression matrix and the neutralised
receiving signal,
y = x - .rho. .rho. = .OMEGA. n - m + 1 x = .omega. .omega. * n - m
+ 1 x ##EQU00005##
where the receiving trend vector p depends also on the modulation
rotation and can degenerate to an average when .beta.=0, the new
model can then be described as
y=.THETA.h+v
and an unbiased channel estimation is obtained 110.
h=(.THETA..sup.T.THETA.).sup.-1.THETA..sup.Ty
[0023] Further, the equation above also provides an implicit DC
(i.e. the trend) estimation. The offset vector can be then
calculated 112 as
.delta..sub.DC=.alpha..sub.DC.omega.*=.rho.-.PSI.h
[0024] The amplitude of the offset can be determined as
.alpha. DC = .delta. DC .omega. n - m + 1 ##EQU00006##
[0025] In practise, complete trend elimination may be not desirable
since it can damage the quality of the cross correlation between
original and de-trended training sequence E{.phi..psi.*} which is
essential to the accuracy of the channel estimation. Computational
considerations will also prefer a real, instead.of complex,
training sequence in the operation. A compromised modification of
the new "neutralized" patin-regression vectors is thus
.theta..sub.k=.phi..sub.k-.mu.Re(.psi..sub.k)
where the real part of the trend vector is taken out and scaled
down (0<.mu.<1) before subtracted from the path regression
vector. Trend suppression, instead of trend elimination is
incorporated in the processing. For different training sequences,
which have different amplitude at the trend frequency, different
suppression level may be applied by choosing different .mu..
[0026] The method above, therefore, constructs path trend vectors
for each path from rotation vector and path regression vector. The
method provides a path trend elimination/suppression model for
channel estimation, using a modified "neutralized" regression
matrix in LS channel estimation. It separates the channel
estimation and DC estimation by eliminating/suppressing DC offset
in the channel estimation. Implicit DC offset estimation is
obtained by combination of the de-trended path--regression and
de-trended receiving signal. Application of the method of the
present invention in linear system identification can be achieved
with short stimulus and biased/single frequency offset noise.
[0027] Although a preferred embodiment of the method and apparatus
of the present invention has been illustrated in the accompanying
drawing and described in the forgoing detailed description, it will
be understood that the invention is not limited to the embodiment
disclosed, but is capable of numerous variations, modifications
without departing from the scope of the invention as set out in the
following claims.
* * * * *