Patent Grant
8218694
The invention relates to a method for transmitting a digital signal, a method for receiving a digital signal, a transmitter and a receiver
In mobile communications, high user capacities and high data rates are desirable. To achieve this, mobile radio systems have to be highly spectral efficient. Using multicarrier modulation according to OFDM (orthogonal frequency division multiplexing) robust performance and high spectral efficiency can be achieved.
Before the OFDM modulation, a pre-transform can be carried out, resulting in a so-called PT-OFDM (pre-transform OFDM) system.
In [1] (and also in [2]), an iterative detection algorithm for a PT-OFDM system is described. This will be described in the following.
An iteration (corresponding to an iteration index i) of the iterative detection algorithm corresponds to three stages, a reconstruction step, a linear filtering step and a decision step.
In the ith reconstruction step, i.e. in the reconstruction step of the iteration corresponding to the iteration index i, the mith component of the received signal r (received signal vector) is estimated. This is done by using the previously detected symbol {circumflex over (x)}i−1 (i.e. the signal vector detected in the previous iteration). mi corresponds to the frequency domain channel with the ith smallest amplitude. In the filtering step, the cross interference of the data is removed by a linear filter denoted by G. In the detection step, a tentative (hard or soft) decision (denoted by dec(.)) is made to generate the symbol detected in the ith iteration, {circumflex over (x)}i.
The algorithm is initialized with r0=r, {tilde over (x)}=Gr0 and {circumflex over (x)}0=dec({tilde over (x)}0).
The ith iteration is given by:
ri=1m
{tilde over (x)}i=Gri
{circumflex over (x)}i=dec({tilde over (x)}i)
where 0m is defined as a diagonal matrix with value 1 on its mth diagonal term and 0 otherwise, and 1m as a diagonal matrix with value 0 on its mth diagonal term and 1 otherwise.
In [1], the pre-transform is based on the Walsh-Hadamard transform (WHT) or on other standard transforms.
An object of the invention is to increase the performance of existing transmitting methods.
The object is achieved by the method for transmitting a digital signal, the method for receiving a digital signal, the transmitter and the receiver with the features according to the independent claims.
A method for transmitting a signal comprising a plurality of signal values is provided, the signal values being grouped to at least one signal value block wherein each signal value block is processed by a pre-transformation to produce a block of modulation symbols. The pre-transformation comprises a phase rotation of the signal block values, which corresponds to the multiplication of the signal value block with a phase rotation matrix. At least one carrier signal is modulated based on the modulation symbols and the modulated carrier signal is sent.
Further, a method for receiving a signal is provided wherein a modulated carrier signal is received and the modulated carrier signal is demodulated to produce a block of modulation symbols. The block of modulation symbols is processed by an inverse pre-transformation, wherein the inverse pre-transformation comprises a phase rotation of the modulation symbols, which corresponds to the multiplication of the block of modulation symbols with a phase rotation matrix.
Further, a transmitter according to the method for transmitting a digital signal described above and a receiver according to the method for receiving a digital signal described above are provided.
Illustratively, a pre-transform is used that comprises a phase rotation. By the phase rotation, errors are evenly spread and error propagation is reduced. Especially when the transform size is small, i.e. when the dimension of the signal value block is small and signal-to-noise ratio is high, performance in terms of BER (bit error rate) is improved with respect to transmitting methods according to prior art.
Embodiments of the inventions arise from the dependent claims. Embodiments of the invention which are described in the context of the method for transmitting a digital signal are also valid for the method for receiving a digital signal the transmitter and the receiver.
The pre-transformation can further comprise a domain transformation of the signal value block. In this case, the pre-transformation can correspond to a multiplication of the signal value block with a product of a phase rotation matrix and a domain transformation matrix, for example a FFT (fast Fourier transform) matrix.
In one embodiment, the domain transformation is performed after the phase rotation.
The domain transformation can be a discrete sine transformation, a discrete cosine transformation or a discrete Fourier transformation. The pre-transformation can also comprise a Walsh-Hadamard-transformation (WHT).
In one embodiment, the phase rotation rotates at least one of the components of the signal value block by an angle that is not zero. In another embodiment, the phase rotation rotates all or all but one of the components of the signal value block by an angle that is not zero. In one embodiment, the absolute values of the components of the signal value block are not changed by the phase rotation.
The phase rotation matrix is for example a diagonal matrix. In one embodiment, the absolute value of all components on the diagonal of the phase rotation matrix is 1.
In one embodiment, the phase rotation matrix has the form diag(1, α, . . . , αM−1) where α=exp(−jπ/(2M)) and M is the dimension of the signal value block. Other values can be used for α or for the diagonal elements of the phase rotation matrix such that other phase rotations are realized.
The invention can for example be used in communication systems according to WLAN 11a, WLAN 11g, Super 3G, HIPERLAN 2 and WIMAX (Worldwide Interoperability for Microwave Access).
The methods according to the invention can be carried out by a computer which is supplied with the corresponding instructions.
The transmitter/receiver system 100 is formed according to a PT-OFDM (Pre-Transform Orthogonal Frequency Division Multiplexing) system. For simplicity, it is assumed that M=2k, e.g. M=32, and that M information symbols xm, m=1, 2, . . . , M are transmitted at the same time in form of one OFDM symbol. For transmitting these information symbols, the vector of information symbols, x=[x1, x2, . . . , xm]T, in the following also called the original signal vector, is fed to a pre-transform unit 101. The superscript T denotes the transpose operator.
The pre-transform unit 101 calculates a vector of modulation symbols s=[s1, s2, . . . , sM]T for the original signal vector according to
s=W·x.
W represents a PT (pre-transform) matrix of size M×M. There is no loss of code rate in terms of number of information symbols transmitted per channel use. In the case of an OFDM system, the matrix W would simply be an identity matrix.
The vector (or block) of modulation symbols s generated by the pre-transform unit 101 is then passed to an IFFT (inverse fast Fourier transform) unit 102 which carries out an inverse fast Fourier transform on the block of modulation symbols.
The inverse fast Fourier transform is used in this embodiment as an efficient realization of an inverse Fourier transform. Other domain transformations can be used instead of the inverse fast Fourier transform, for example an inverse discrete sine transform or an inverse discrete cosine transform.
The vector generated by the IFFT unit 102 is then mapped from parallel to serial, i.e. to a sequence of signal values, by a P/S (parallel to serial) unit 103. A cyclic prefix unit 104 inserts a cyclic prefix into the sequence of signal values to form a PT-OFDM symbol which is transmitted via a channel 105.
The cyclic prefix that is inserted has a duration no shorter than the maximum channel delay spread. The channel 105 is assumed to be a quasi/static frequency selective Rayleigh fading channel corrupted by additive white Gaussian noise (AWGN).
The pre-transform unit 101, the P/S unit 102 and the cyclic prefix unit 104 are part of a transmitter 106.
The PT-OFDM symbol is received by a receiver 107. A cyclic prefix removal unit 108 removes the cyclic prefix from the PT-OFDM symbol. The resulting sequence of signal values is mapped from parallel to serial by a S/P unit 109 and is domain transformed according to a fast Fourier transform by an FFT (fast Fourier transform) unit 110. Analogously to the IFFT unit 102, the FFT unit 110 can in other embodiments also be adapted to perform a discrete sine transform or a discrete cosine transform or another domain transformation.
The output vector of the FFT unit 110 is denoted by r=[r1,r2, . . . , rm]T and can be written as
r=Γ·s+n=Γ·W·x+n
where Γ=diag(h1, h2, . . . , hM) is a diagonal matrix with diagonal elements h1, . . . , hM which are the frequency domain channel coefficients and n is the AWGN vector of dimension M×1. The frequency domain channel coefficients are given by hm=Σn{tilde over (h)}nexp(−j2πn(m−1)/M), m=1, 2, . . . , M, assuming a sampled spaced Lth order FIR (finite input response) channel model {{tilde over (h)}n}n=0L.
The output vector r of the FFT unit 110 is fed to a detection unit 111. The detection unit 111 performs an iterative detection algorithm. An iteration (corresponding to an iteration index i) of the iterative detection algorithm corresponds to three stages, a reconstruction step, a linear filtering step and a decision step.
In the ith reconstruction step, i.e. in the reconstruction step of the iteration corresponding to the iteration index i, the mith component of the vector r is estimated. This is done by using the previously detected symbol {circumflex over (x)}i−1 (i.e. the signal vector detected in the previous iteration). mi corresponds to the frequency domain channel with the ith smallest amplitude. In the filtering step, the cross interference of the data is removed by a linear filter denoted by G. In the detection step, a tentative (hard or soft) decision (denoted by dec(.)) is made to generate the symbol detected in the ith iteration, {circumflex over (x)}i. When the last iteration has been performed (e.g. after a given number of iterations, e.g. 4, has been performed) the detected symbols {circumflex over (x)}i are output by decision units 112.
The algorithm is initialized with r0=r, {tilde over (x)}=Gr0 and x0=dec({tilde over (x)}0).
The ith iteration is given by:
ri=1m
{tilde over (x)}i=Gri
{tilde over (x)}i=dec({tilde over (x)}i)
where 0m is defined as a diagonal matrix with value 1 on its mth diagonal term and 0 otherwise, and 1m as a diagonal matrix with value 0 on its mth diagonal term and 1 otherwise.
The matrix W, which defines the pre-transformation carried out by the pre-transform unit 101 is chosen according to the following criteria
W should be unitary and
W should have elements which are of constant amplitude
The first requirement serves to preserve the capacity of the system, while the second requirement maximizes the worse post-filtered SNR (signal-to-noise ratio) at every detection step when there is no error propagation.
Given that the matrix satisfies the criteria mentioned, there exist certain transforms that lead to better performance than others when error propagation occurs. The effect is most striking when a ZF filter is used for the filtering step and when the transform size is small.
Using the Walsh Hadamard transform performs worse at high SNR condition when QPSK (quadrature phase shift keying) signal constellation is used. This is because the error propagation leads to an “error constellation” which increases the bit error performance as compared to the case when other well designed transforms are used instead.
In this embodiment, a transform is used according to a design which gives a more random-like “error constellation” since error propagation is bound to occur. This is achieved by using a pre-transformation according to the matrix W0 defined in the following formula, wherein the transform size M, as stated above, is assumed to be a power of 2.
W0=F×diag(1,α, . . . , αM−1)
where α=exp(−jπ/(2M)) and F is the FFT matrix of size M.
It can be easily verified that W0 is unitary and has elements with constant amplitude.
Coincidentally, this transform is proposed for maximum likelihood detection (MLD) in order to exploit maximum channel diversity in [3]. However, the approach that is adopted here is different since W0 is used to minimize the error propagation caused by the transform and W0 is not used for MLD which is very complex in implementation, in the order of 4M for QPSK constellation. Note also that when W0 is used for the transform, the PT-OFDM system shown in
Advantageously, this means that the peak-to-average-power ratio is reduced to the smallest possible when the signal constellation used has constant amplitude.
Simulations show that by using a pre-transform according to the matrix W0, higher performance can be achieved in terms of lower BER (bit error rate), especially when the SNR is high. Therefore, if the noise variance is not known at the receiver, the designed transform allows performance to be improved significantly.
Under the assumption that the variance of the noise corrupting the channel 105 is known, the MMSE (minimum mean square error) filter described in the following can be used by the detection unit 111 to improve the performance of the transmitter/receiver system 100.
Taking into account the MMSE criteria and assuming that previous detected symbols are correct for each reconstruction, the linear filter for the ith (i=1, 2, . . . , M) iteration can be derived:
and σ2 is the noise variance. This G is used in the filtering step of the reconstruction algorithm carried out by the detection unit 111 as described above.
For the initial iteration of the reconstruction algorithm, the MMSE filter with
βm=hm*/(|hm|2+σ2), m=1, . . . , mM.
is used.
When the matrix W is chosen as unitary and has constant amplitude elements, even when the MMSE filter is used, the choice of mi is unchanged (corresponding to the frequency domain channel with the ith smallest amplitude). That is, this choice still maximizes the worse post-filtered SNR at every detection step under the assumption that the previous detection is correct.
Simulations show that using this filter, the error floor can be reduced and superior performance can be achieved for high SNR. Note that this MMSE filter requires the knowledge that the noise variance is known at the receiver. However, simulations show that it is robust to noise variance errors.
In one embodiment, the reconstruction is extended as will be described with reference to
The receiver 200 may be used instead of the receiver 107 shown in
Analogously to the detection unit 111, a vector r, e.g. the output vector of an FFT unit performing an FFT, is fed to the detection unit 201.
A filtering unit 202 of the receiver 201 performs a filtering step of a reconstruction algorithm, e.g. the initial filtering step of the reconstruction algorithm described above. The result of the filtering step, denoted by {tilde over (x)}0 in accordance to the above description of the reconstruction algorithm is supplied to a first nonlinear detection unit 203.
The receiver further comprises a second nonlinear detection algorithm unit 205. The structure of the first nonlinear detection unit 203 and the second nonlinear detection unit 205 are described in the following with reference to
The nonlinear detection unit performs an ordered interference cancellation algorithm as will be described in the following.
The input vector of the non-linear detection unit 300 is a soft estimate of a transmitted signal (in case of the first non-linear detection unit, this is the output {tilde over (x)}0 of the filtering unit 202). The input vector of the non-linear detection unit 300 is fed to an ordering unit 301.
The ordering unit 301 performs an ordering step by obtaining the minimum Euclidean distance of the input to any point of the signal constellation. Then, the components of the input vector are ordered from largest to smallest (minimal) Euclidean distance and a hard decision is performed on the components of the input vector to form c1, c2, . . . , cM.
c1, c2, . . . , cM are fed to a cancellation unit 302 which performs the following algorithm:
For interference cancellation j=1, . . . , J.
(i) use {ck}k≠j to cancel from reconstructed received signal ri to obtain a soft estimate of cj
(ii) perform hard decision on the soft estimate and update the newly detected cj
(iii) increment j and continue with (i)
J is the number of cancellation steps and is for example chosen equal to M.
Illustratively, the interference cancellation algorithm uses the “best” components of the estimate, in the sense that they have minimal Euclidean distance to the signal constellation to improve the “worse” components, which have a higher Euclidean distance to the signal constellation.
The output of the first cancellation unit 302 is fed to the reconstruction unit 204. The reconstruction unit 204 performs the reconstruction step and the filtering step for the ith iteration (where i=1, 2, . . . ) according to the reconstruction algorithm described above.
The result of each iteration performed by the reconstruction unit 204 is fed to the second non-linear detection unit 205. The output of the second non-linear detection unit 205 is fed back to the reconstruction unit 204 for the next iteration to be performed except for the last iteration, when the output is supplied to the decision units 206 which generate the output of the receiver 200.
The receiver 200 can also be used with a pre-transform (according to a matrix W) according to prior art and with a filter (according to a matrix G) according to prior art. This means that the idea of ordering the signal values according to a distance measure and using the signal values which are best (in terms of smallest distance) to cancel the interference from the other signal values is independent from using a pre-transformation comprising a phase rotation matrix and from using a filter which is dependent on the variance of the noise of the channel used for data transmission.
In the above, the following documents are cited:
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/SG2005/000264 | 8/3/2005 | WO | 00 | 7/9/2008 |
| Publishing Document | Publishing Date | Country | Kind |
|---|---|---|---|
| WO2006/014141 | 2/9/2006 | WO | A |
| Number | Name | Date | Kind |
|---|---|---|---|
| 5610944 | Mau et al. | Mar 1997 | A |
| 6067292 | Huang et al. | May 2000 | A |
| 6125103 | Bäuml et al. | Sep 2000 | A |
| 6298050 | van Heeswyk et al. | Oct 2001 | B1 |
| 6377632 | Paulraj et al. | Apr 2002 | B1 |
| 6426983 | Rakib et al. | Jul 2002 | B1 |
| 6654408 | Kadous et al. | Nov 2003 | B1 |
| 6882619 | Gerakoulis | Apr 2005 | B1 |
| 7680200 | Kwun et al. | Mar 2010 | B2 |
| 20020131537 | Storm et al. | Sep 2002 | A1 |
| 20030002568 | Dabak et al. | Jan 2003 | A1 |
| 20040086054 | Corral | May 2004 | A1 |
| 20040146115 | Feng et al. | Jul 2004 | A1 |
| 20060258316 | Lei et al. | Nov 2006 | A1 |
| 20080165865 | Bar-Ness et al. | Jul 2008 | A1 |
| Number | Date | Country |
|---|---|---|
| A-H10-84329 | Mar 1998 | JP |
| WO9516259 | Jun 1995 | WO |
| WO02103974 | Dec 2002 | WO |
| WO03088540 | Apr 2003 | WO |
| WO03065635 | Aug 2003 | WO |
| WO2004025841 | Mar 2004 | WO |
| WO2004032347 | Apr 2004 | WO |
| Number | Date | Country | |
|---|---|---|---|
| 20080260070 A1 | Oct 2008 | US |
