This application claims priority under 35 U.S.C. § 119 to an application entitled “Method for Detecting and Decoding a Signal in a MIMO Communication System” filed in the Korean Intellectual Property Office on Mar. 22, 2005 and assigned Serial No. 2005-23795, the contents of which are incorporated herein by reference.
1. Field of the Invention
The present invention generally relates to a wireless communication system, and more particularly to a method for detecting and decoding a signal in a Multiple-Input Multiple-Output (MIMO) communication system.
2. Description of the Related Art
A Multiple-Input Multiple-Output (MIMO) communication system transmits and receives data using multiple transmit antennas and multiple receive antennas. A MIMO channel formed by Nt transmit antennas and Nr receive antennas is divided into a plurality of independent spatial subchannels. Because the MIMO system employs multiple transmit/receive antennas, it outperforms a Single-Input Single-Output (SISO) antenna system in terms of channel capacity. Conventionally, the MIMO system undergoes frequency selective fading that causes Inter-Symbol Interference (ISI). The ISI causes each symbol within a received signal to distort other successive symbols. This distortion degrades the detection accuracy of a received symbol, and it is an important noise factor affecting a system designed to operate in a high Signal-to-Noise Ratio (SNR) environment. To remove the ISI, a stage at the receiving end has to perform an equalization process for a received signal. This equalization requires high processing complexity.
On the other hand, Vertical Bell Labs Layered Space-Time (V-BLAST) architecture, which is one of space division multiplexing schemes, offers an excellent tradeoff between performance and complexity. The V-BLAST scheme uses both linear and non-linear detection techniques. In other words, the V-BLAST scheme suppresses interference from a received signal before detection and removes interference using a detected signal.
When an Orthogonal Frequency Division Multiplexing (OFDM) scheme is used, an equalization process for the received signal is possible at low complexity. An OFDM system divides a system frequency band into a plurality of subchannels, modulates data of the subchannels, and transmits the modulated data. The subchannels undergo different frequency-selective fading according to transmission paths between transmit and receive antennas. The ISI incurred due to this fading phenomenon can be effectively removed by prefixing each OFDM symbol with a cyclic prefix. Therefore, when the OFDM scheme is applied to the MIMO system, the ISI is not considered for all practical purposes.
For this reason, it is expected that the MIMO-OFDM system based on a detection algorithm of the V-BLAST scheme will be selected as a next-generation mobile communication system. However, the conventional V-BLAST scheme has a severe drawback. There is performance degradation due to error propagation, which is inherent in a decision feedback process. Various methods are being studied and proposed to overcome this performance degradation. However, these methods create new problems, such as increased processing complexity of a receiving stage. This complexity increases according to a modulation level and the number of antennas. The currently proposed methods are based on an iterative process between detection and decoding without significantly increasing the overall processing complexity.
Accordingly, the present invention has been designed to solve the above and other problems occurring in the prior art. It is an object of the present invention to provide a method for detecting and decoding a signal that can improve the reliability of a received signal by detecting the signal while considering a decision error in an equalization process for the received signal.
It is another object of the present invention to provide a method for detecting and decoding a signal that can improve system performance by optimizing a signal detection order for channel-by-channel layers.
It is yet another object of the present invention to provide a method for detecting and decoding a signal that can reduce complexity by setting a signal detection order for one channel and applying the set signal detection order to all channels.
In accordance with an aspect of the present invention, there is provided a method for detecting and decoding a signal in a communication system based on MIMO-OFDM, including the steps of receiving a signal through multiple receive antennas; considering a decision error occurring at a symbol decision time and detecting a symbol from the received signal; and recovering original data transmitted from the detected symbol.
Preferably, the symbol is detected using a Minimum Mean Square Error (MMSE)-based equalization matrix. The equalization matrix is expressed by
where Hi is a channel matrix for an i-th signal, * is a complex conjugate, e is an estimation error, Qe is a decision error covariance matrix of
and I is an identity matrix.
The equalization matrix is designed such that a mean square value of the error e=xi−Gyi is minimized.
The decision error covariance matrix Qe is computed by Equation (2):
where E[emen*|{circumflex over (x)}m,{circumflex over (x)}n] corresponding to a conditional expectation value indicates that errors em and en occur due to inaccurate decisions associated with {circumflex over (x)}m≠xm and {circumflex over (x)}n≠xn.
Diagonal elements E[∥em∥2|{circumflex over (x)}m] of the decision error covariance matrix Qe indicate a mean square error value of the detected symbol.
A position of a component with a smallest value among diagonal elements of the decision error covariance matrix Qe determines a signal detection order.
The step of detecting the symbol includes the steps of estimating a previously transmitted symbol using decoded original data in a previous decoding process; and removing a component of the estimated symbol from the received signal.
The step of detecting the symbol includes the step of setting a detection order for layers in which signals are received through an identical subchannel.
The detection order for the layers is set in descending order from a layer with a highest channel capacity.
The channel capacity is computed by Equation (3):
where Cnk is defined as channel capacity for an n-th layer in a k-th subchannel, Cnk being computed by Equation (4):
Cnk=log2(1+SINRnk). (4)
The detection order is set in ascending order from a layer in which a metric Mn for the n-th layer is smallest.
The metric Mn is computed by Equation (5):
where H is a channel matrix, ρ is a mean received power to noise ratio in each receive antenna, and I is an identity matrix.
The detection order among layers is determined only for one particular subchannel, and the same order is applied to all subchannels.
The above and other objects and advantages of the present invention will be more clearly understood from the following detailed description taken in conjunction with the accompanying drawings, in which:
The present invention will be described in detail herein below with reference to the accompanying drawings.
In
In
In the present invention, it is assumed that channel state information (CSI) is predetermined for the receiver. The present invention considers a baseband signal model based on a zero-mean complex value and a discrete-time frequency selective fading MIMO-OFDM channel model.
When an N-dimensional complex transmission signal vector and an N-dimensional complex reception signal vector are defined by xk and yk, a signal received through the k-th subcarrier is expressed by Equation (6):
Assuming that total power of xk for obtaining the maximum capacity is P and a transmitter does not know a channel state, transmission signal power must be equally distributed between N transmit antennas according to variance σS2. A covariance matrix of xk is defined by Equation (7):
where E[•] and (•)† denote an expectation value and a complex conjugate transpose matrix, respectively, IN is an identity matrix of a size N the additional term of nk has variance σn2, and is complex Gaussian noise of an independent and identical distribution.
A channel coefficient hji,k of {overscore (H)}k denotes a path gain from the i-th transmit antenna to the j-th receive antenna. The path gain is modeled as a sample of independent complex Gaussian random parameters having the variance of 0.5 on a dimension-by-dimension basis. If antennas of each stage on a communication link are divided according to more than a half wavelength, independent paths are maintained.
A signal model of a layered space-time OFDM system considering error propagation is newly introduced into the present invention. Transmission symbols are defined by xn representing a symbol transmitted from the n-th antenna and x=[x1x2 . . . xN]T representing a vector signal with (•)T representing the transpose of a vector. For convenience, the decision order {{circumflex over (x)}1 {circumflex over (x)}2 . . . {circumflex over (x)}i−1} is designated by an optimum detection order of the V-BLAST scheme proposed by Foschini.
{circumflex over (x)}n denotes a symbol detected for Layer n, and hn denotes the n-th row of {overscore (H)}.
xi=[xixi+1 . . . xN]T, Hi=[hihi+1 . . . hN], {circumflex over (x)}i−1=[{circumflex over (x)}1{circumflex over (x)}2 . . . {circumflex over (x)}i−1]T, and Ĥi−1=[h1h2 . . . hi−1].
In the conventional V-BLAST algorithm, a symbol vector {circumflex over (x)}i−1 pre-detected until the (i−1)-th step is removed from a vector signal received in the i-th step. As a result, a corrected received vector yi can be expressed by Equation (8):
In Equation (8), it is assumed that the previous decisions are correct (i.e., {circumflex over (x)}n=xn for n=1,2, . . . , i−1). This signal detection process regards undetected signals {xi, xi+2, . . . , xN} as interference, and is performed using a linear nulling process as in a Minimum Mean Square Error (MMSE) scheme. Equation (8) requires the accuracy of the pre-detected vector symbol {circumflex over (x)}i−1. In a situation in which a decision error is present, Equation (8) is rewritten as Equation (9):
where êi−1=[e1e2 . . . ei−1]T and en=xn−{circumflex over (x)}n.
Next, an MMSE algorithm based on a new signal model of Equation (9) will be described.
The present invention uses a nulling matrix based on an MMSE criterion considering a decision error. In the MMSE criterion, an equalization matrix G is designed such that a mean-square value of an error e=xi−Gyi is minimized, and can be obtained using the well-known orthogonality principle in mean-square estimation as expressed in Equation (10):
E[eyi†]=E[(xi−Gyi)yi†]=0 (10)
The equalization matrix G satisfies Equation (11).
E[(xi−Gyi)yi†]=Qx
where a covariance matrix is defined by QAB=E[AB†] and QA=E[AA†].
and G can be expressed from Equation (9) and Equation (11) as Equation (12):
where Qx
Therefore, a decision error variance matrix Qê
where * denotes the complex conjugate and a conditional expectation value E[emen*|{circumflex over (x)}m,{circumflex over (x)}n] is used to indicate that errors em and en occur due to inaccurate decisions associated with {circumflex over (x)}m≠xm and {circumflex over (x)}n≠xn, respectively.
For example, diagonal elements E[emeN*|{circumflex over (x)}m,{circumflex over (x)}n] indicate the variance of the decision error em due to the inaccurate decision associated with {circumflex over (x)}m. Because non-diagonal elements E[emen*|{circumflex over (x)}m,{circumflex over (x)}n] do not have a correlation between errors where m≠n, E[emen*|{circumflex over (x)}m,{circumflex over (x)}n] is the same as E[em|{circumflex over (x)}m] E[en*|{circumflex over (x)}n].
When it is assumed that previously detected signals are perfect and error propagation does not occur, the equalization matrix G proposed in the present invention is equal to the conventional MMSE matrix. In other words, Qê
Next, a method for deciding an optimum detection order on the basis of a new equalization matrix G in accordance with the present invention will be described.
A covariance matrix Qe of an estimation error e=xi−Gyi can be computed after the equalization matrix G is set. Using Equation (12), the covariance matrix Qe is expressed by Equation (14):
Qe=Qx
Diagonal elements indicate mean-square error (MSE) values of detected symbols. Therefore, the successive detection order depends on a position of the smallest diagonal element of Qe. This is equal to a position of the largest diagonal element GHi of Equation (14).
Next, the operation of a demapper applied for the signal detection and decoding method of the present invention will be described.
It is well known that the use of a soft output demapper and a soft input channel decoder significantly improves system performance. First, an optimum soft bit metric considering a detection error is computed after several assumptions are made in a detected vector signal {circumflex over (x)}i−1.
The index t denotes the position on the main diagonal of the matrix Qe where the MSE is minimized. In other words, {circumflex over (x)}l is selected as a decision at the i-th step where i≦t≦N. gl is defined as the row of the equalization matrix G associated with an equalizer for {circumflex over (x)}i. Applying this equalizer vector into Equation (4) yields Equation (15):
For analytical conveniences, it is assumed that the terms of w follow a complex Gaussian distribution. An error probability of an MMSE detector can be easily assessed under an assumption that output interference and noise are Gaussian noise.
Since each term in w is independent of other terms, the variance of w can be computed by Equation (16):
In Equation (16), the second term corresponds to the decision error up to the (i−1)-th step, and it affects system performance significantly. After a biased term is properly scaled, the input to the unbiased demapper can be written as Equation (17):
{tilde over (x)}i={tilde over (z)}i/β=xi+v, (17)
where v is complex noise with the variance σv2=σw2/∥β∥2.
Next, the computation of a Log Likelihood Ratio (LLR) for soft bit information will be briefly described.
Let S and s be a set of constellation symbols and an element of the set S, respectively. Then the conditional probability density function (pdf) of {tilde over (x)}i in Equation (17) is given by Equation (18):
When the i-th bit of xi is defined as bli and two mutually exclusive subsets are defined as S0o={s:bli=0} and S1i={s:bli=1} where i=1,2 . . . log2 Mc and Mc is defined as the constellation magnitude |S|, a posteriori LLR of bli can be defined as Equation (19):
Equation (19) can be rewritten through slight manipulation as shown in Equation (20):
In order to compute σv2, E[∥ej∥2|{circumflex over (x)}j] must be computed for j=1,2, . . . , i−1 in Equation (16) and these quantities are related to the probability of decision error at the j-th step.
Next, a method for computing the error probability will be described.
The error probability associated with a Maximum Likelihood (ML) demapper is invariant to any rotation of a signal constellation. This means that the error probability depends on only a relative distance between signal points within the signal constellation. Let us define Pe as the error probability between two neighboring QAM signal points. Also, the minimum distance of the Mc−QAM constellation is given by Equation (21):
The error probability Pe between two signals separated by minimum distance dmin is computed by Equation (22):
where
and σ2 corresponds to the noise variation in an in-phase or 4-quadrature phase direction. Plugging dmin into Equation (22) yields Equation (23):
where the fact that σ2 is a half the noise variance σv2 for the QAM symbols is utilized. An accurate approximate value of the Q function has been found over the range of 0<x<∞ as Equation (24):
where a=0.344 and b=5.334.
This error function is used to estimate conditional expectation values E[el|{circumflex over (x)}l] and E[∥el∥2|{circumflex over (x)}l].
In
A process for computing E[el|{circumflex over (x)}l] and E[∥el∥2|{circumflex over (x)}l] values using Equation (23) is described with reference to a conditional probability mass function P(s|{circumflex over (x)}l). The conditional probability mass function P(s|{circumflex over (x)}l) depends on a hard decision value {circumflex over (x)}l. It is only required to consider the following three cases in order to cover all the possible outcomes of {circumflex over (x)}l:
When {circumflex over (x)}l belongs to the set of corner points, the conditional probability P(s|{circumflex over (x)}l) of erroneous detection into each neighbor signal point is shown in Table 1.
When {circumflex over (x)}l belongs to the set of edge points, the conditional probability P(s|{circumflex over (x)}l) of erroneous detection into each neighbor signal point is shown in Table 2.
When {circumflex over (x)}l belongs to the set of inner points, the conditional probability P(s|{circumflex over (x)}l) of erroneous detection into each neighbor signal point is shown in Table 3.
Here,
Note that Q2 term is negligible. In that case, only the closest neighbors are included.
Assuming that transmitted signals are equally likely, the conditional probability P(s|{circumflex over (x)}l) that s is transmitted when the detected signal is {circumflex over (x)}l falls into one of 3 categories described above.
When only an error between two adjacent constellation signal points is considered, the conditional expectation values E[el|{circumflex over (x)}l] and E[∥el∥2|{circumflex over (x)}l] are computed by Equation (25) and Equation (26), respectively:
where the set N{circumflex over (x)}
In the signal detection and decoding method as described above, the complexity increases due to a process for computing the equalization matrix G. In the present invention the complexity O(NM3) is lower than O(N3)+O((N−1)3)+ . . . +O(23) in the conventional method.
In
When an interference cancellation method is applied, the performance of an overall system is affected by the order in which each layer is detected. It is very efficient that interference is removed using decision feedback information estimated from a decoder's output signal of the previous step in flat fading channels. In other words, all decision values for the detected layer are transferred to the decoder when one layer is detected, and an output of the decoder is again encoded and is used for interference cancellation in the next layer.
Accordingly, all decision values detected in one layer must be transferred to the decoder in every detection step.
In accordance with the second embodiment of the present invention, the receiver decides the detection order for a total layer according to one computation during a total detection process before the interference cancellation is performed and applies the same detection order to all subchannels.
In accordance with the second embodiment of the present invention, a decision element for deciding the detection order uses a channel capacity value.
Cnk denotes Shannon capacity associated with i-th subchannel in n-th layer and it is computed by Equation (27):
Cnk=log2(1+SINRnk), (27)
where for an unbiased MMSE filtering, SINRnk can be expressed as Equation (28):
where σMMSE−LE.nk2 is an MMSE for the n-th layer in the k-th subchannel. When Equation (12) is replaced by Equation (14), σMMSE−LE.nk2 is expressed by Equation (29):
Here, [A]ij is the (i, j) element of a matrix A. In this case, the terms associated with decision errors are set to 0 (i.e., Qê
Using the ABC lemma for matrix conversion, i.e., (A+BC)−1=A−1−A−1B(CA−1B+I)−1CA−1, Equation (29) can be rewritten as Equation (30):
σMMSE−LE.nk2=[σn2({overscore (H)}m*{overscore (H)}m+α1N)−1]m (30)
When Equation (28) and Equation (30) are inserted into Equation (27), the capacity Cnk is computed by Equation (31):
The aggregate capacity Cn of the n-th layer across all subchannels is given by Equation (32):
The detection order based on Cn can be selected.
An operation for selecting a layer in which Cn is maximized is identical to the one for retrieving a layer in which a metric value Mn in Equation (33) is minimized.
After the metrics Mn for all layers are computed, the detection order among layers is determined in an ascending order of Mn. The detection order in the detection method in accordance with the present invention may be different in each step. Because a process for updating the order in every step is not useful for the overall performance improvement, the update is not performed to reduce complexity when the representative detection order is set in the first step. As illustrated in
The number of transmit antennas and the number of receive antennas are 4, a Convolutional Code (CC) at a code rate ½ is used, an OFDM scheme defined in the Institute of Electrical and Electronics Engineers (IEEE) 802.11a standard based on a 64-length FFT is used, and an OFDM symbol interval is 4 μs including a guard interval of 0.8 μs. In the simulations, a 5-tap multipath channel with an exponentially decaying profile is used. It is assumed that the frame length is one OFDM symbol interval.
When 16QAM is applied as illustrated in
This improvement is obtained through soft bit metric generation and decision error consideration in an equalization process of the signal detection and decoding method in accordance with the present invention.
As described above, the signal detection and decoding method of the present invention can significantly improve system performance in a coded bit system using a new equalization matrix G considering a decision error.
It is expected that the signal detection and decoding method of the present invention can obtain various diversity gains associated with frequency, space, and time diversities with a successive interference-canceling algorithm by introducing an optimum soft bit demapper.
Because the signal detection and decoding method of the present invention can improve system performance by correcting an equalization matrix, it is expected that the maximum system performance can be improved in a minimum increase in the complexity of a receiver.
While the present invention has been described with reference to the preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the scope of the present invention as defined by the following claims.
Number | Date | Country | Kind |
---|---|---|---|
23795-2005 | Mar 2005 | KR | national |