KR102169683B1 - Efficient precoding apparatus for downlink massive MIMO System and method thereof - Google Patents

Abstract

The present invention relates to an efficient precoding device for a downlink massive MIMO system to increase an error correction function more than an existing diagonal Neumann series (DNS) precoding method and a method thereof. According to the present invention, the precoding device comprises: a matrix generation unit calculating a Z matric (Z = G·G^H) based on a channel matrix (G) between a base station and a plurality of users; a comparison unit comparing the number (V) of diagonal dominant rows with the number (N_u) of users; a control unit selecting only one row component from the Z matrix based on a comparison result or selecting the set number (k, 1 < k < N_u) of row components; an inverse matrix generation unit calculating Z^(-1); and a precoding matrix generation unit multiplying the calculated Z^(-1) by G^H to generate a precoding matrix.

Description

Efficient precoding apparatus for downlink massive MIMO system and method thereof

The present invention relates to an efficient precoding apparatus and method for a downlink large-capacity MIMO system, and more particularly, to an efficient precoding apparatus and method for improving error performance in a downlink large-capacity MIMO system.

The multi-user-based downlink large-capacity MIMO system performs signal preprocessing, that is, precoding, to remove multi-user interference (MUI) before signal transmission. In a large-capacity MIMO system, a base station can use a myriad of antennas, remove MUI with only linear zero-forcing (ZF), and generate high beamforming gain to obtain optimal error performance.

However, due to the nature of large-capacity MIMO, since it accommodates a large number of users, the complexity of the precoding technique, ZF, is very high, and an approximated ZF based on NS (Neumann Series) was developed to solve this problem. Among many NS-based ZFs, DNS (Diagonal NS) utilizes the diagonally dominant characteristics of the gram matrix, and thus has a much lower complexity than ZF and can obtain high error performance.

However, as the number of users increases, that is, as the ratio of the number of users to the number of transmit antennas increases, the law of logarithm, which is one of the characteristics of large-capacity MIMO, does not hold, and the correlation between channels increases. Due to this, the diagonal dominant characteristic of the gram matrix disappears, and the error performance of DNS is rapidly degraded. In particular, as the number of modulation orders used increases, the performance deterioration is more severe, and the most serious problem is that it cannot be solved even by increasing the base station transmission power and the length of NS.

The technology behind the present invention is disclosed in Korean Patent Publication No. 2008-0083808 (published on September 19, 2008).

The present invention is for a downlink large-capacity MIMO system capable of adaptively performing precoding according to the diagonal dominance of users for efficient precoding and improving the error performance of DNS even in a large-capacity MIMO system with high correlation between channels. An object of the present invention is to provide an efficient precoding apparatus and method.

The present invention, in a precoding apparatus applied to a base station in a large-capacity MIMO system, a matrix generator that calculates a Z matrix (Z=G·G ^H ) based on a channel matrix (G) between the base station and a plurality of users. And, a comparison unit that calculates the number of diagonally dominant rows (V; 0≤V≤N _u ) in the Z matrix and compares the number of diagonally dominant rows (V) with the number of users (N _u ), A control unit that selects only one column component in the Z matrix or a set number (k; 1<k<N _u ) of column components based on the comparison result, and a diagonal matrix that takes only diagonal components from the Z matrix. An inverse matrix generator that combines the selected column components to (D) to obtain a Φ matrix, which is an initial matrix of NS (Neumann Series), and calculates Z ^-1 by substituting the Φ matrix into an approximate formula of NS-based Z ^-1 And a precoding matrix generator for generating a precoding matrix by multiplying the calculated Z ⁻¹ by G ^{H. A} precoding apparatus for a downlink high-capacity MIMO system is provided.

In addition, the approximate expression and the convergence condition of the NS-based Z ^-1 for computing the Z ^-1 can be defined by the equation below.

Here, L is the length of LS, I _i is an i×i identity matrix, and O _i is an i×i zero matrix.

Further, the inverse matrix generation unit, wherein the diagonal matrix (D) after the operation of the inverse matrix Φ ^-1 is obtained by combining the selected heat elements to the matrix Φ in the following, the value Φ ^-1 multiplied by the Z (Φ ^- a ¹ · Z) may be substituted for the term of the approximate equation of the Z ^-1 calculate the Z ^-1.

In addition, when selecting the column component, the control unit may select each column in the Z matrix in order of high power.

In addition, when the number of diagonally dominant rows is the same as the number of users (V=N _u ), the control unit is 1 having the maximum power among each column in a state in which all the diagonal components in the Z matrix are treated as 0. When the DCNS (Diagonal plus Column NS)-1 technique for selecting two column components is applied and the number of diagonally dominant rows is 0 (V=0), all of the diagonal components in the Z matrix are treated as 0 The DCNS-k technique can be applied to select the k number of heat components in the order of the highest power size among each column.

In addition, the inverse matrix generator, when V = N _u , combines the selected one column component to the diagonal matrix (D) to obtain an initial matrix Φ matrix, and then converts the Φ matrix back to the diagonal matrix (D) and atomic Calculate Φ ^-1 , which is the inverse matrix of Φ, using the result of separating it into a matrix, and then multiply the Φ ^-1 by the Z (Φ ^-1 ·Z) and substitute it in the corresponding term of the approximation formula of Z ^-1 . Thus, Z ^-1 can be calculated.

In addition, when V=O, the inverse matrix generator may convert the j-th (j=(1,...,k)) selected column component among the k column components and the diagonal matrix D to Sherman Morrison (Sherman Morrison). ) The operation applied to the formula is repeated k times from j=1 to k, calculates the inverse matrix (Φ ^k ) ^-1 for the initial matrix Φ ^k , and then converts the (Φ ^k ) ^-1 to Z and Z ^-1 can be calculated by substituting the multiplied value ((Φ ^k ) ^-1 ·Z) into the corresponding term of the approximation of Z ^-1 .

In addition, when the number of diagonally dominant rows is smaller than the number of users and greater than 0 (0<V<N _u ), the control unit is DCNS-1 in response to the diagonally dominant V rows in the Z matrix. It is possible to select a hybrid technique that applies the technique and applies the DCNS-U technique (k=U) to the remaining U rows that are not diagonally dominant in the Z matrix.

In addition, the inverse matrix generator, when 0<V<N _u , calculates a diagonal dominance size for each row in the Z matrix, rearranges the rows of the G matrix in the order of the highest diagonal dominance size, and rearranges Z _S matrix using a G _S by calculating the _{(Z S = G S · G} S H) , and divided into four areas: the Z _S matrix based on the V value, in a clockwise direction from upper left to lower left V separating the D ₁ matrix, E ₁ matrix of V × U, D ₂ matrix and D ₂ matrix of U × V of U × U of × V with any of the following, DCNS-1 scheme to the D ₁ matrix (D ₁ ) to obtain a matrix, ^-1, and then to apply the technique to DCNS-U D ₂ matrix obtained for the (D ₂₎ ^-1 matrix, the ^{_{(D 1) -1, (D}} 2) -1, E 1 and E ₂ substituting the matrix equation below, it is possible to calculate the inverse matrix of the Z _S Z _S ^-1.

Here, L is the length of LS, and I _i is an i×i identity matrix.

In addition, for the calculation of Z _S ^-1 , the D ₁ matrix, the zero matrix, the D ₂ matrix, and the zero matrix may be allocated respectively to four regions of the Φ matrix included in the equation.

In addition, in the present invention, in a precoding method performed by a base station in a large-capacity MIMO system, calculating a Z matrix (Z=G·G ^H ) based on a channel matrix (G) between the base station and a plurality of users And, calculating the number of diagonally dominant rows (V; 0≤V≤N _u ) in the Z matrix and comparing the number of diagonally dominant rows (V) with the number of users (N _u ), and comparison Based on the result, selecting only one column component within the Z matrix or selecting a set number (k; 1<k<N _u ) of column components, and a diagonal matrix taking only diagonal components from the Z matrix ( D) combining the selected column components to obtain a Φ matrix, which is an initial matrix of NS (Neumann Series), and calculating Z ^-1 by substituting the Φ matrix into an approximate formula of NS-based Z ^-1 , and the It provides a precoding method for a downlink large-capacity MIMO system including generating a precoding matrix by multiplying the calculated Z ^-1 by G ^H.

In addition, the step of calculating Z ^-1 may include calculating the Φ matrix by combining the selected column components with the diagonal matrix (D), calculating Φ ^-1, which is an inverse matrix, and then calculating the Φ ^-1 with the Z a multiplied value (Φ ^-1 · Z) may be substituted for the term of the approximate equation of the Z ^-1 calculate the Z ^-1.

In the selecting of the column component, when the column component is selected, each column in the Z matrix may be selected in an order of high power.

In addition, the step of selecting the column component may include, if the number of diagonally dominant rows is the same as the number of users (V=N _u ), in a state in which all the diagonal components in the Z matrix are treated as 0, Applying the DCNS (Diagonal plus Column NS)-1 technique for selecting one column component with the maximum power, and when the number of diagonally dominant rows is 0 (V=0), all of the diagonal components in the Z matrix In the state of being treated as 0, the DCNS-k technique can be applied to select k number of column components in the order of the highest power level among each column.

Here, in the calculating of Z ^-1 , when V = N _u , the selected one column component is combined with the diagonal matrix (D) to obtain an initial matrix Φ matrix, and the Φ matrix is again a diagonal matrix Calculate Φ ^-1 , which is the inverse matrix of Φ, using the result of dividing into (D) and atomic matrix, and then multiplying the Φ ^-1 with the Z (Φ ^-1 ·Z) is an approximate formula of Z ^-1 Z ^-1 can be calculated by substituting in the corresponding term of.

Here, in the calculating of Z ^-1 , when V=O, the j-th (j=(1,...,k)) selected column component among the k column components and the diagonal matrix (D) are The operation applied to the Sherman Morrison formula is repeated k times from j=1 to k, calculating the inverse matrix (Φ ^k ) ^-1 for the initial matrix Φ ^k matrix, and then (Φ ^k ) ^- Z ^-1 can be calculated by substituting a value ((Φ ^k ) ^-1 ·Z) multiplied by ¹ by Z in the corresponding term of the approximation formula of Z ^-1 .

In addition, the step of selecting the column component may include, when the number of diagonally dominant rows is smaller than the number of users and greater than 0 (0<V<N _u ), in the Z matrix, the diagonally dominant V rows are Correspondingly, the DCNS-1 technique may be applied and the DCNS-U technique (k=U) applied to the remaining U rows that are not diagonally dominant in the Z matrix may be selected.

Here, in the calculating of Z ^-1 , when 0<V<N _u , a diagonal dominance size is calculated for each row in the Z matrix, and the rows of the G matrix are in the order of the highest diagonal dominance size. After reordering and calculating the Z _S matrix (Z _S =G _S ·G _S ^H ) using the rearranged G _S , the Z _S matrix is divided into 4 regions based on the V value, from the top left to the bottom left. one to remove the D ₂ matrix of D ₁ matrix of V × V in the clockwise direction, V × U of E ₁ matrix, U × U D ₂ matrix and U × V in the following, the DCNS-1 scheme to the D ₁ matrix Applying to obtain the (D ₁ ) ^-1 matrix, applying the DCNS-U technique to the D ₂ matrix to obtain the (D ₂ ) ^-1 matrix, the (D ₁ ) ^-1 , (D ₂ ) ^-1 , E ₁ and E ₂ is substituted for the matrix equation below, it is possible to calculate the inverse matrix of the Z _S Z _S ^-1.

Here, L is the length of LS, and I _i is an i×i identity matrix.

According to the present invention, for efficient precoding in a large-capacity MIMO system with high correlation between channels, precoding is performed adaptively according to the diagonal dominance of users, thereby having a very low complexity and similar error performance than the conventional ZF precoding technique. It provides and has the advantage of improving error performance over existing DNS precoding techniques.

1 is a block diagram of a large-capacity MIMO system according to an embodiment of the present invention.
FIG. 2 is a diagram illustrating a configuration of a precoding apparatus applied to the base station of FIG. 1.
3 is a diagram illustrating a precoding method using FIG. 2.
4 is a diagram illustrating an aligned matrix Z _S and an initial matrix Φ in a hybrid technique.
5 is a schematic flowchart of a precoding method according to an embodiment of the present invention.
6 is a diagram illustrating a multiplication amount required to calculate Z ^-1 in the conventional technique and the technique proposed by the present invention.
7 is a diagram showing channel estimation MSE (Mean Square Error) performance in a 200×10, 200×20, and 200×30 large-capacity MIMO system.
8 is a diagram showing the frequency of use of DCNS-1, DCNS-k, hybrid DCNS-1 and DCNS-U according to the number of users.
9 is a diagram showing BER (Bit Error Rate) performance of a conventional technique and a technique proposed in the present invention in a 200×10 large-capacity MIMO system.
10 is a diagram showing BER performance of a conventional technique and a technique proposed in the present invention in a 200×20 large-capacity MIMO system.
11 is a diagram showing BER performance of a conventional technique and a technique proposed in the present invention in a 200×30 large-capacity MIMO system.
FIG. 12 is a diagram showing BER performance according to nulling user selection of the scheme proposed in the present invention when 16-QAM (Quadrature Amplitude Modulation) modulation is used in 200×20 and 200×30 large-capacity MIMO systems.
13 is a diagram showing an existing technique and a multiplication amount of the present invention according to the number of users.

Then, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement the present invention.

1 is a block diagram of a large-capacity MIMO system according to an embodiment of the present invention. A large-capacity MIMO system includes a base station 10 (BS) and a plurality of users 10 (User).

As shown in Fig. 1, an embodiment of the present invention is a base station 10 (BS) having N _t transmit antennas, and N _u users 20 (User 1, ..., User) having one receive antenna. Consider a large-capacity MIMO system including N _u ) (N _t ≫N _u ). Here, the user may mean a user terminal.

The received signal y _i of the i-th user is expressed as in Equation 1 below.

Here, P is the transmission power of the base station, W is the precoding matrix, w _i is the i-th column of w, x _i is a modulated transmission signal with an average of 0 and variance of 1 for the i-th user, and n _i is the i-th AWGN (Additive White Gaussian Noise) with an average of 0 and a variance of 1, g _i is a Rayleigh channel vector from all transmit antennas to the i-th user, T is a transpose matrix, and F is a Frobenius norm. Represents.

The channel matrix G between the base station 10 and the plurality of users 20 is expressed as in Equation 2.

Here, g _ij has an average of 0 and a variance of 1, and represents a channel coefficient from the j-th transmit antenna to the i-th user (1 ≤ _i ≤ N _u , 1 ≤ _j ≤ N _t ).

The existing ZF (Zero-Forcing) precoding technique completely removes Multi-User Interference (MUI) by multiplying the transmitted signal by a pseudo inverse matrix (precoding matrix), assuming that the base station knows the channel information accurately. The vector condition for ZF precoding is shown in Equation 3.

And the precoding matrix (W) satisfying Equation 3 is the same as Equation 4.

Here, (·) ^H represents the Hermitian transform. The precoding matrix (W) of Equation 4 corresponds to a pseudo inverse matrix to the channel matrix (G).

에 대한 역행렬인

의 연산 복잡도 차수는

로 매우 높다.However, the existing ZF precoding technique shows optimal error performance in large-capacity MIMO systems, but the gram matrix

Which is the inverse matrix for

The computational complexity order of is

Is very high.

Although the ZF technique is a linear precoding technique, it is difficult to implement because it has very high complexity in a large-capacity MIMO system accommodating a large number of users. An embodiment of the present invention proposes a precoding technique that can reduce computational complexity in a large-capacity MIMO system.

)을 다음과 같이 Z로 정의한다.In the following, for simplicity of expression, the gram matrix (

) Is defined as Z as follows.

Using Equation 5, Equation 4 is simply expressed as W = G ^H ·Z ^-1 .

It can be seen that in order for the base station 10 to precode a transmission signal, the precoding matrix W of Equation 4 must be obtained, and in order to obtain W, G ^H and Z ^-1 must be calculated and then multiplied.

By the way, G ^H is easily obtained by Hermitian transformation of Equation 2, but the inverse matrix of Z (Z ^-1 ) has a very high computational complexity.

Embodiment of the invention the NS method for reducing the computational complexity of the Z ^-1; uses the approximate expression of the (Neumann Series Neumann water) based on Z ^-1.

The approximate formula of NS-based Z ^-1 is defined by Equation 6 below.

Here, L is the length of LS, I _i is the i×i identity matrix, and Φ is the initial matrix of NS.

Equation 7 below represents a convergence condition for calculating NS-based Z ^-1 of Equation 6 for Z, which is a reversible matrix.

Here, O _i represents an i×i zero matrix.

In this manner, the operation and the inverse matrix Z ^-1 in the approximation expression of the NS Z Z ^-1 based on a lower complexity means the approximation equation to minimize operation errors. In the following embodiment, for low complexity, L=1 in Equations 6 and 7 is fixed. Of course, the present invention is not necessarily limited thereto.

The setting conditions of Φ for NS with excellent performance are summarized as follows. The first should converge Equation 7 quickly, and the second should be able to reduce the computational complexity of Φ ^-1 in Equation 6.

Among the NS-based ZF techniques, DNS (Diagonal NS) utilizes the diagonally dominant characteristic of the gram matrix (Z) to obtain high error performance while having a much lower complexity than ZF.

Specifically, the DNS precoding technique lowers the computational complexity of Equation 6 by constructing (substituting) the Φ part of Equation 6 with only the diagonal components of the gram matrix Z of Equation 5 (Φ=D). Hereinafter, for convenience, this is referred to as a diagonal matrix (D).

The diagonal matrix (D) for the Z matrix is represented by Equation 8 below. That is, the diagonal matrix D corresponds to taking only the diagonal component from the gram matrix Z and processing the remaining components as 0.

If Φ is used as Z in Equation 6, the computational complexity is very high because Φ ^-1 must be calculated using all the elements (components) in the gram matrix Z.

However, in the case of the DNS precoding technique, the computational complexity can be greatly reduced by using Φ as D, since it is only necessary to calculate the inverse matrix of Φ, that is, Φ ^-1 using only the diagonal components, not all components in the gram matrix (Z).

In this way, the DNS technique does not use all of the Z values based on the diagonal dominant characteristic of the gram matrix (Z), but takes only the diagonal component of the Z value and uses it as the Φ value, thereby reducing the computational complexity of Φ ^-1 in Equation 6 Lower it.

)의 역행렬인 Z^-1를 수학식 6의 근사식을 통해 구하는 과정에서, 수학식 6의 Φ를 어떤 값으로 사용(적용)하느냐에 따라 복잡도가 달라짐을 알 수 있다.Based on this, in order to obtain the precoding matrix W of Equation 4 in the base station 10, the gram matrix (

In the process of obtaining Z ^-1 , the inverse matrix of) through the approximation of Equation 6, it can be seen that the complexity varies depending on what value Φ of Equation 6 is used (applied).

If the number of users is smaller than the number of transmit antennas of the base station (the larger ρ is), the correlation between channels of a large-capacity MIMO system is lowered. In this case, the DNS precoding technique quickly converges Equation 7 and the initial matrix (Φ) of NS is replaced with a diagonal matrix (D) (Φ=D), so the computational complexity of Φ ^-1 is low, and the NS-based ZF Fit in a way.

However, if the number of users compared to the number of transmission antennas of the base station is larger (the smaller ρ), the correlation of the large-capacity MIMO system is high.In this case, when using the NS precoding technique, Equation 7 is not satisfied even if L is increased numerous times Can lead to serious error performance degradation.

That is, in order to reduce the DNS error performance degradation, the gram matrix (Z matrix) should dominate diagonally. However, in a highly correlated large-capacity MIMO system, the gram matrix may partially dominate diagonally or may not completely dominate diagonally. In addition, even if the gram matrix is diagonally dominant, if the modulation order is high, error performance deterioration occurs, and thus, the embodiment of the present invention uses an adaptive precoding technique to reduce this.

The following embodiments of the present invention propose a precoding scheme for securing the reliability of DNS even in a large-capacity MIMO system with high correlation. The proposed precoding technique adaptively performs precoding according to the diagonal dominance of each user.

)의 역행렬인 Z^-1를 수학식 6의 근사식을 통해 구하되, 유저의 대각 지배성에 따라 Φ를 달리 사용하여, 상황에 따라 복잡도와 오류 성능을 조절한다.That is, the base station 10 obtains the precoding matrix W of Equation 4, the gram matrix (

), the inverse matrix of Z ^-1 is obtained through the approximation of Equation 6, but the complexity and error performance are adjusted according to the situation by using Φ differently according to the user's diagonal dominance.

Of course, the embodiment of the present invention ultimately calculates Z ^-1 , which is the inverse matrix of Z, with low complexity and high reliability, and generates a precoding matrix W by multiplying the calculated Z ^-1 by G ^H.

That is, the base station 10 calculates the inverse matrix of Z (Z ^-1 ) based on the approximation of Equation 6, and then multiplies Z ^-1 by G ^H to generate the precoding matrix W. Further, the transmission signal is multiplied by the precoding matrix, and the precoded transmission signal is transmitted to the user 20 to remove multi-user interference (MUI).

Here, the base station 10 analyzes the diagonal dominance based on the Z matrix obtained between the base station 10 and the plurality of users 20 and compares the number of diagonal dominant rows with the number of users. Adaptively adjust the method of finding the inverse matrix (Z ^-1 ).

Hereinafter, a precoding apparatus and method according to an embodiment of the present invention will be described in more detail.

It is assumed that the precoding apparatus 100 according to the embodiment of the present invention is included in the base station 10. Of course, the precoding apparatus 100 may correspond to the base station 10 itself.

FIG. 2 is a diagram illustrating a configuration of a precoding apparatus applied to the base station of FIG. 1, and FIG. 3 is a diagram illustrating a precoding method using FIG. 2.

2 and 3, the precoding apparatus 100 according to the embodiment of the present invention includes a matrix generation unit 110, a comparison unit 120, a control unit 130, an inverse matrix generation unit 140, and And a coding matrix generator 150.

First, the matrix generator 110 calculates a Z matrix (Z=G·G ^H ) based on the channel matrix G between the base station 10 and the plurality of users 20 (S310).

The channel matrix G between the base station 10 and the plurality of users 20 is shown in Equation 4. The matrix generator 110 multiplies G and G ^H (Hermit transform for G) to calculate a Z matrix as shown in Equation 5.

In addition, the comparison unit 120 calculates the number of diagonally dominant rows (V; 0≤V≤N _u ) in the Z matrix, and compares the number of diagonally dominant rows (V) with the number of users (N _u ). (S320).

The comparison unit 120 determines whether or not diagonally dominant for each row of the Z matrix of Equation 5 using the condition of Equation 9 below.

Equation 9 is an equation for determining whether the size of the diagonal component z _ii among the elements belonging to the row is greater than the sum of the sizes of the remaining components for each row, and in that case, the row is determined as the diagonally dominant row.

That is, if the ith row of the Z matrix satisfies Equation 9, it is diagonally dominant, otherwise it is not diagonally dominant. In simple terms, a row in which the diagonal component has an advantage over the rest of the components except itself can be defined as the diagonal dominant row.

As a simple example, Z has a size of 3×3 (N _u =3), the first row is [z ₁₁ ,z ₁₂ ,z ₁₃ ] = [5,1,2] and the second row is [z ₂₁ ,z ₂₂ , z ₂₃ ] = [2,7,4], and the third row, [z ₃₁ , z ₃₂ , z ₃₃ ] = [1, 4, 6]. Here, in fact, all the z-values of the elements in the Z matrix have a complex number form, but for convenience of explanation, they are simply illustrated as positive integers.

In this case, the size 5 of the diagonal component (z ₁₁ ) of the first row is greater than 3 (5> 1+2), the sum of the remaining two components (z ₁₂ ,z ₁₃ ), so the first row is the diagonally dominant row. do. Similarly, the size 7 of the diagonal component (z ₂₂ ) of the second row is greater than the sum of the other two components, and the size 6 of the diagonal component (Z ₃₃ ) of the third row is greater than the sum of the other two components.

Accordingly, in the above example, since all rows constituting the Z matrix are diagonally dominant rows, the number of diagonally dominant rows (V) is 3, which is also equal to N _u , so V=N _u =3.

The precoding scheme according to the present embodiment is divided into three scenarios according to the number (V) of diagonally dominant rows.

Specifically, the scenario where V=N _u (when all rows in the Z matrix dominate diagonally; Case 1), the scenario where V=0 (when all rows are not diagonally dominant; Case 2), and 0<V< It can be divided into a scenario with N _u (where some rows dominate diagonally; Case 3). An embodiment of the present invention applies different precoding techniques for each scenario.

Specifically, in the case of V=N _u (Case 1), that is, when all users are diagonally dominant, the DCNS (Diagonal plus Column NS)-1 technique is applied to improve error performance while having a low complexity similar to the DNS technique. Precode.

And, when V=0 (Case 2), that is, when all users are not diagonally dominant, the channel correlation is high, so DCNS-k precoding is applied even if the complexity is slightly increased to improve the error performance of DNS. . Of course, the DCNS-k technique has a much lower complexity than the ZF technique and has similar error performance.

Next, when 0<V<N _u , that is, when some users partially dominate diagonally, DCNS-1 is applied to users who dominate diagonally, and DCNS-U is applied to users who are not diagonally dominant. -1 and DCNS-U) Apply precoding technique.

As will be described later, DCNS-1 represents a special situation where k=1 in DCNS-k. In addition, in the DCNS-k method, a Φ matrix is obtained by combining the diagonal matrix D with respect to the Z matrix and k columns selected from the Z matrix. That is, when obtaining the Φ matrix, the DCNS technique does not use only the diagonal components in the Z matrix, as in the DNS technique, but combines and uses a predetermined number of column components included in the Z matrix.

Hereinafter, for convenience of explanation, Case 1 and Case 2 will be first described.

The comparison unit 120 compares the number of diagonally dominant rows (V) with the number of users (N _u ), and transmits the V value and the comparison result to the controller 130.

Then, based on the comparison result performed by the comparison unit 120, the control unit 130 selects only one column component in the Z matrix of Equation 5, or a set number (k; 1<k<N _u ) Select the thermal component of (S330).

In fact, the controller 130 serves to adaptively control the precoding technique in the embodiment of the present invention, and may also control each of the units 110, 120, 140, and 150.

Here, when selecting the column component, the control unit 130 may select the column component in the order of the highest power among each column in the Z matrix of Equation (5).

If only one column component needs to be selected in the Z matrix of Equation 5, one column with the maximum power can be selected among all columns, and when k column components are selected, the higher power k among all columns. Just select the heat component of the dog

Specifically, when the number of diagonally dominant rows (V) is equal to the number of users (N _u ) (V=N _u ; Case 1), the control unit 130 processes all diagonal components in the Z matrix as 0 DCNS-1 technique is applied to select one column component with the maximum power among each column. At this time, when the squares of each component in the column are summed, the power level of the column is obtained.

In addition, when the number of diagonally dominant rows (V) is smaller than the number of users (N _u ) but is 0 (V=0; Case 2), all of the diagonal components in the Z matrix are processed as 0. The DCNS-k technique is applied, which selects k heat components from each column in the order of the highest power level.

Here, k may be determined within a range of 1 < k < N _u in consideration of target performance and channel environment of the system. Of course, if k=N _u, it is effectively the same as the existing ZF technique, so it is preferable that k is determined to be smaller than N _u (for example, k=0.3N _u , 0.6N _u ). In fact, the controller 130 adaptively controls the precoding technique in the embodiment of the present invention.

Then, the inverse matrix generator 140 combines the column components selected in step S330 with respect to the diagonal matrix (D) taking only the diagonal components from the Z matrix to obtain a Φ matrix, which is an initial matrix of NS (Neumann Series), and obtains a Φ matrix. Z ^-1 is calculated by substituting in the NS-based approximation of Z ^-1 (Equation 6) (S340).

Here, the inverse matrix generator 140 combines the column components selected by the control unit 130 to the diagonal matrix D to obtain a Φ matrix, and then calculates Φ ^-1 by performing an inverse matrix operation on the obtained Φ. Then, calculates a Z ^{-1 -1} Φ by substituting the re-multiplied by the Z value (Φ · Z ^-1) in the term of the equation (6) (approximate expression of Z ^-1).

Thereafter, the precoding matrix generator 150 multiplies Z ^-1 calculated in step S340 by G ^H , and finally generates a precoding matrix W (S350).

Accordingly, the base station 10 precodes the transmission signal using the precoding matrix W and transmits it to the user 20. That is, the transmission signal is multiplied by the precoding matrix W to provide the precoded transmission signal to the user.

Hereinafter, step S340 will be described in more detail.

In the DCNS-k technique, for simple operation, a Hollow matrix (E) is used to obtain a Φ matrix, which is an initial matrix of DS. Here, E is defined as in Equation 10 below.

This matrix E is obtained by processing all diagonal components as 0 in the Z matrix of Equation 5, and can be easily calculated by subtracting the D matrix from the Z matrix.

In addition, in the DCNS-k technique, the initial matrix Φ ^k is set as in Equation 11 below.

Here, E ^k is composed of the k-th column selected from E and the remaining N _u -k zero vectors.

Using Equations 10 and 11, a Φ matrix formed by combining the column component and the diagonal matrix D selected in step S330 can be obtained.

First, for better understanding of the invention, an example of generating the initial matrix Φ (=Φ ¹ ) of DCMS-1 using DCNS-1 when N _u =3 in V=N _u (Case 1) is first described. Explain.

At this time, if it is assumed that the first column is the highest power among all three columns in the E matrix in the state where the diagonal component is removed from the Z matrix, E ¹ and Φ ¹ are as in Equation 12 below.

That is, E ¹ of Equation 12 is obtained by selecting only the first column component having the maximum power intensity in the E matrix of Equation 10. And if E ¹ is combined with the diagonal matrix D using Equation 11, Φ ¹ is obtained as in Equation 12.

Comparing with the Z matrix of Equation 5, it can be seen that the initial matrix Φ ¹ is composed only of the diagonal component and the first column component of the Z matrix. That is, when the DCSN-1 technique is applied, a Φ matrix is obtained by combining the diagonal component in the Z matrix and one column component selected in the Z matrix.

In this way, when V = N _u (Case 1), the inverse matrix generator 140 combines one column component selected by the control unit 130 with the diagonal matrix D, and the initial DCNS-1 The matrix Φ(=Φ ¹ ) is first created.

Then, the inverse matrix generator 140 calculates Φ ^-1 , which is an inverse matrix of Φ, using the result of separating the Φ matrix into a diagonal matrix D and an atomic matrix again. Then, by substituting the product of the calculated value Z again and Φ ^-1 (Φ · Z ^-1) in the term of the equation (6) (approximate expression of Z ^-1) and the last calculated Z ^-1.

Here, the reason for using the atomic matrix is to simplify the process of calculating the inverse matrix for the Φ matrix.

That is, if Φ ¹ of Equation 12 is separated into a diagonal matrix (D) and an atomic matrix, it is the same as Equation 13, and the inverse matrix of Equation 13 is simply summarized as Equation 14.

Here, Equation 13 is simply expressed as Φ ¹ =D·M, and the inverse matrix thereof is expressed as (Φ ¹ ) ^-1 = M ^-1 ·D ^-1 .

However, in general, the inverse matrix operation of the diagonal matrix (D) and the inverse matrix operation of the atomic matrix (M) are performed in a very easy way. That is, as shown in Equation 14, M ^-1 may be performed by dividing the inner elements of the matrix M, and D ^-1 may be expressed as the reciprocal of the elements of the D matrix.

As such, it can be seen that when the Φ matrix is separated using an atomic matrix, the calculation of the inverse matrix becomes very simple.

Since Equation 6 requires the multiplication between the calculated Φ ^-1 and Z, if (Φ ¹ ) ^-1 ·Z is calculated using Equation 14, the first column of the result is the unit of [1 0 0] ^T It becomes a vector. This means that the first thermal component has been supressed, and through this, it can be seen that all MUI (interference) for the first user with the greatest power has been removed.

In this way, if the MUI for a specific user is completely nulled, the signal to noise plus interference ratio (SINR) after precoding may be increased, thereby improving error performance. In this way, in order to maximize error performance, the MUI component with the largest power is nulled.

Next, in case 2 of V=0, that is, when the DCSN-k technique is applied, a Φ ^k matrix is obtained using the diagonal components in the Z matrix and k column components selected in the Z matrix.

In this case, in the case of using the DCSN-1 technique with k=1, (Φ ¹ ) ^-1 was calculated with low complexity using the structure of the atomic matrix, but as k increases, the complexity of (Φ ^k ) ^-1 becomes nonlinear. Increases.

To solve this, in the embodiment of the present invention, when V = 0, (Φ ^k ) ^-1 is calculated using the Sherman Morrison formula.

That is, when V=0, the inverse matrix generator 140 shermanes the j-th (j=(1,...,k)) selected column component and the diagonal matrix D among k column components selected in the order of power magnitude. The operation applied to the Morrison formula is repeated k times from j=1 to k, and the inverse matrix (Φ ^k ) ^-1 for the initial matrix Φ ^k is calculated. That is, iteratively uses the Sherman Morrison formula to calculate (Φ ^k ) ^-1 .

Thereafter, the inverse matrix generator 140 substitutes a value ((Φ ^k ) ^-1 ·Z) multiplied by (Φ ^k ) ^-1 by Z into the corresponding term of Equation 6 (approximate expression of Z ^-1 ), and thus Z ^-1 is finally calculated.

In a system with N _u users and k = K, when DCNS-k is applied, an example of calculating (Φ ^k ) ^-1 by repeatedly using the Sherman Morrison formula is as follows.

For convenience of calculation, it is assumed that the MUI size of the k-th column among the total K columns is the k-th largest. First, for the reversible matrix A of m×m and the column vectors u and v of m×1, the Sherman Morrison formula is as follows.

In general, the inverse matrix is not easy to distribute, but as shown in Equation 15, separation is easy using the Morrison formula.

An embodiment of the present invention repeatedly uses Equation 15 to calculate (Φ ^k ) ^-1 .

The structure of Equation 15 is a structure applicable to the inverse matrix operation of the present invention. That is, the inverse matrix of Equation 11 can be expressed as Equation 16.

When Equation 15 and Equation 16 are correlated, it can be seen that the diagonal matrix (D) of Equation 16 corresponds to A and E ^k corresponds to uv ^T.

Using this, Equation 16 can be expressed in the form of Equation 17.

와

는 각각

와

의 j번째 선택된 열을 나타낸다.here,

Wow

Are each

Wow

Represents the j-th selected column of.

The embodiment of the present invention repeatedly uses Equation 15 to calculate (Φ ^k ) ^-1 , and iteratively calculates from (Φ ¹ ) ^-1 for the reversible matrix D.

If (Φ ¹ ) ^-1 is used to calculate (Φ ² ) ^-1 in the same manner, it is as shown in Equation 18 below.

That is, if the Sherman Morrison method is repeated k times from k=1 to k, the inverse matrix (Φ ^k ) ^-1 for the Φ ^k matrix can be obtained with lower complexity. This is because A in Equation 16 can be set to D in the first Sherman Morrison equation as in Equation 17.

Finally, an embodiment of a case where the number of diagonally dominant rows is greater than 0 and less than the number of users (0<V<N _u ; Case 3) will be described.

When 0<V<N _u , the controller 130 applies the DCNS-1 technique in response to the diagonally dominant V rows in the Z matrix, and DCNS- for the remaining U rows that are not diagonally dominant in the Z matrix. The U technique (k=U) is applied. This will be described in detail as follows.

In order to apply the hybrid DCNS-1 and DCNS-U, the inverse matrix generator 140 first calculates a diagonal dominance size for each row in the Z matrix and rearranges the rows of the G matrix in the order of the highest diagonal dominance size. . Then, the Z _S matrix (Z _S =G _S ·G _S ^H ) is calculated using the rearranged G _S.

A method of obtaining the size of the i-th user's diagonal dominance d _i is as shown in Equation 19.

,

인 경우, Z의 첫번째 행에서 d₁=5-(6+7)=2, 두 번째 행에서 d₂=8-(6+3)=-1, 그리고 세 번째 행에서 d₃=16-(7+3)=6이 각각 구해진다. 따라서 Z의 대각 지배성 크기 벡터 d=[d₁ d₂ d₃]=[2 -1 6]이 된다.E.g,

,

If, in the first row of Z, d ₁ =5-(6+7)=2, in the second row d ₂ =8-(6+3)=-1, and in the third row d ₃ =16-( 7+3)=6 are obtained respectively. Therefore, the diagonal dominant magnitude vector of Z becomes d=[d ₁ d ₂ d ₃ ]=[2 -1 6].

Here, the first and third rows whose diagonal dominant size is greater than 0 correspond to the diagonal dominant rows, while the second row whose diagonal dominance size is greater than 0 corresponds to the non-diagonal dominant row. Hence, the number of dominant rows V=2. This corresponds to the 0<V<N _u condition (Case 3), and a hybrid precoding technique is required.

In addition, from these results, it can be seen that the third user has the highest diagonal dominance and the second user has the lowest diagonal dominance.

이 된다. 또한 이러한 G_S를 이용하여 Z 행렬을 변형하면 수학식 20이 된다.If we sort the G matrix in diagonal dominance order,

Becomes. In addition, if the Z matrix is transformed using G _S , Equation 20 is obtained.

4 is a diagram illustrating an aligned matrix Z _S and an initial matrix Φ in a hybrid technique.

As shown in FIG. 4, the inverse matrix generator 140 divides the Z _S matrix calculated through Equation 20 into four regions using V values (the number of diagonally dominant rows).

That is, in FIG. 4, the Z _S matrix is a D ₁ matrix of V×V, an E ₁ matrix of V×U, a D ₂ matrix of U×U, and U×V in a clockwise direction from top left to bottom left based on values. It can be seen that the D ₂ matrix is separated. Here, U=N _u -V, which corresponds to the number of non-dominant rows diagonally.

The situation of N _u =3 and V = 2 presented above corresponds to a very simple example proposed for convenience of explanation. For example, for the case of N _u =20 and V = 5, the Z _S matrix is a 5×5 D _It will be split into ₁ matrix, 5×15 E ₁ matrix, 15×15 D ₂ matrix, and 15×5 D ₂ matrix.

Then, the inverse matrix generator 140 to apply the DCNS-1 Method D ₁ to obtain a matrix (D ₁₎ ^-1 matrix, obtained by (D ₂₎ ^-1 matrix by applying the technique to the U-DCNS D ₂ matrix and ^{_{then, (D 1) -1, (}} D 2) -1, by substituting the E ₁ and E ₂ equation (21) below the matrix to calculate the Z _S ^-1 is the inverse matrix of Z _S.

Here, of course, L is the length of LS, and I _i is an i×i identity matrix.

That is, in order to obtain Z _S ^-1 in Equation 21, (D ₁ ) ^-1 and (D ₂ ) ^-1 must be calculated.

To this end, referring to the Φ matrix of FIG. 4, the inverse matrix generator 140 corresponds to the four regions of the Φ matrix included in the equation for calculation of Z _S ^-1 in Equation 21, and each D It can be seen that ₁ matrix, zero matrix, D ₂ matrix, and zero matrix are allocated (set).

At this time, when calculating the inverse matrix (D ₁ ) ^-1 for the D ₁ matrix in the Φ matrix of FIG. 4, the DCNS-1 technique is applied. For example, similar to the method of obtaining the inverse matrix of Z above, first, the diagonal matrix D for the D ₁ matrix and one column component having the maximum power are combined to form a matrix Φ ^1, and then the method of Equations 13 and 14 the inverse matrix of (Φ ¹⁾ computation of ^-1, and this product of D ₁ and D ₁ · again (Φ ¹⁾ Φ ^{1 -1} of the Calculate. And, D ^₁ · (Φ ₁₎ by substituting a ^-1 in the term of Equation 6 (Z instead replaced with D _1), to compute the inverse of the D ₁ ^-1 D ₁ in such a way obtaining the inverse matrix of Z I can.

In addition, when calculating the inverse matrix (D ₂ ) ^-1 for the D ₂ matrix in the Φ matrix of FIG. 4, the DCNS-U technique (k=U) may be applied. That is, the computation of the inverse matrix of D ₂ D ₂ ^-1 using the above-described prior-k DCNS applying k = U in technique, and a Sherman, Morris formula referred to in equation (15).

In this way, in the rearranged Z _S , D1 corresponds to the diagonally dominant rows, so DCNS-1 is used to obtain D1 ^-1 to obtain high error performance with low complexity, and D2 corresponds to rows that do not correspond to D ^-2 . To find out, DCNS-U is used to improve DNS error performance even in systems that are not dominant.

As described above, by adaptively varying the precoding technique according to V, the inverse matrix Z ^-1 of the gram matrix can be obtained as shown in Equation 6.

In addition, the algorithm of the technique proposed in the present invention is terminated by passing through a Matched Filter (MF) that multiplies G ^H to the left of Z ^-1 as shown in Equation 4.

5 is a schematic flowchart of a precoding method according to an embodiment of the present invention.

Referring to FIG. 5, the precoding apparatus 100 first obtains Z (gram matrix) by using the channel matrix G (S501). Then, the number (V) of diagonally dominant rows in the Z matrix is calculated (S502), and compared with the number of users (N _u ) (S503).

As a result of comparison, if V=N _u , (Φ ¹ ) ^-1 is obtained by applying the DCNS-1 technique (S504), and then Z ^-1 is calculated by substituting (Φ ¹ ) ^-1 into the Φ term of Equation 6 (S505), the calculated Z ^-1 is multiplied by G ^H to finally generate a precoding matrix W as shown in Equation 4 (S506).

There is not a comparison result of the step S503, V = N _u, it is determined whether V = 0 (S507). If V=0, after initial setting of k=0 and Φ ^k =D (S508), (Φ ^k ) ^-1 is obtained by applying the DCNS-k technique (S509). In the case of S509 process, (Φ ^k ) ^-1 is finally calculated by repeatedly using the Sherman Morris formula while increasing k from 1 by one. Thereafter, the calculated (Φ ^k ) ^-1 is substituted into the Φ term in Equation 6 to calculate Z ^-1 (S510), and then Z ^-1 is multiplied by G ^H to obtain the precoding matrix (W) as shown in Equation 4. It is finally generated (S511).

Then, the comparison result of the S507 step, V = 0 If not a there through Equation (21) using the hybrid technique of DCNS-1 and DCNS-U after obtaining the Z _S by reordering the Z along a diagonal dominant sex size Z ^{- 1} is immediately calculated (S512). Then, Z ^-1 is multiplied by G ^H as shown in Equation 4 to finally generate a precoding matrix W (S513).

The following describes the results of a simulation of the performance of the technique proposed in the present invention.

6 is a diagram illustrating a multiplication amount required to calculate Z ^-1 in the conventional technique and the technique proposed by the present invention.

에 비례하므로 표현의 편의성을 위해 최종 복잡도 식에서 상수값은 제외하였다. In FIG. 6, it is assumed that multiplication of two complex numbers requires one multiplication operation when calculating the multiplication operation amount. And the complexity of the large-capacity MIMO system is

Since it is proportional to, the constant value was excluded from the final complexity equation for convenience of expression.

In addition, when both the conventional method and the method proposed in the present invention obtain a precoding matrix, only the method of obtaining Z ^-1 is different and all other methods are the same, so in FIG. 6 only the amount of multiplication required to calculate Z ^-1 is considered. I did.

In the results of FIG. 6, the existing ZF scheme has the highest complexity and the DNS scheme has the lowest complexity. It can be seen that the complexity of the DCNS-k technique used in Case 2 is high after the ZF technique, followed by the hybrid technique used in Case 3, and the DCNS-1 used in Case 1.

7 to 13 are diagrams showing simulation results of the technique proposed in the present invention.

In all simulations, N _t =200 was set, and the channel was modeled as a Rayleigh distribution with an average of 0 and a variance of 1, and all users were set to generate an AWGN with an average of 0 and a variance of 1. In addition, assuming a time division duplex (TDD) system in which the coherence interval of the radio channel is sufficiently large, the uplink and downlink channels are transposed with each other through perfect calibration.

That is, for downlink channel estimation, all users transmit an uplink pilot to the base station, and the base station estimates it and uses the corresponding value as a precoding channel coefficient.

가 유저수와 같아야하므로

로 설정하였다. 상향링크 파일럿 수신 신호는 다음과 같다.In addition, in order to minimize the error performance degradation due to pilot contamination in adjacent cells, the length of the pilot sequence

Should be equal to the number of users

Was set to. The uplink pilot received signal is as follows.

는 유저 송신 전력으로 본 시뮬레이션에서는 기지국 송신 전력

보다 3dB 낮은 전력으로 설정하였다. 그리고

는

를 만족하는

의 직교 파일럿 시퀀스 행렬,

는 모든 원소가 평균이 0이고 분산이 1인

의 AWGN이다. 기지국에서 추정된 채널 행렬은 다음과 같다.here,

Is the user transmit power, and in this simulation, the base station transmit power

It was set to 3dB lower power. And

Is

Satisfying

Orthogonal pilot sequence matrix of,

Is that all elements have a mean of 0 and a variance of 1

Is AWGN. The channel matrix estimated by the base station is as follows.

는 변형된 AWGN이다.here,

Is the modified AWGN.

의 MSE를 계산해야 한다. 그렇지만,

의 각 성분들은 모두 독립적이므로

의 i행 및 j열인

의 MSE를 계산하여도 무방하고, 이는 다음과 같다.To find the theoretical channel estimation MSE

The MSE of should be calculated. nevertheless,

Since each component of is all independent

Row i and column j of

It is okay to calculate the MSE of, which is as follows.

7 is a diagram showing channel estimation MSE (Mean Square Error) performance in a 200×10, 200×20, and 200×30 large-capacity MIMO system. Here, the higher the MSE value, the worse the performance.

In FIG. 7, the theoretical result (Theory) is the same as the simulation result (Monte Carlo) regardless of the number of users. In addition, as the number of users increases, the MSE performance is excellent, because the length of the pilot increases as the number of users increases, and the transmission power is proportional to the length of the pilot as shown in Equation 22.

8 is a diagram showing the frequency of use of DCNS-1, DCNS-k, hybrid DCNS-1 and DCNS-U according to the number of users. The horizontal axis represents the number of users (N _u ), and the vertical axis represents the utilization rate (%).

In the result of FIG. 8, when N _u ≤ 13, the large-capacity MIMO system is almost dominant, so the scheme proposed in the present invention selects DCNS-1 at an average ratio of 90% or more. This means that the fewer users there is less interference, so the DCNS-1 technique is sufficient.

And when _u 14≤N ≤20 days, large MIMO system is, in part, because it is a diagonal dominant select hybrid DCNS-1 and U-DCNS a technique is to average ratio of 90% or more in the present invention.

Finally, when N _u ≥21, since the large-capacity MIMO system is hardly dominant, the scheme proposed in the present invention selects DCNS-k at an average ratio of 85% or more. As shown in FIG. 8, the frequency of precoding selection according to the number of users helps understanding the results of FIGS. 9 to 13.

In FIGS. 11 to 13, Quadrature Phase Shift Keying (QPSK) and 16-QAM modulation were used to measure BER. In addition, the number of users was set to 10, 20, and 30, respectively, because, as shown in the result of FIG. 7, significant user intervals for measuring the performance of the present invention are 9 to 27.

와

로 설정하였다. That is, it is easy to infer that DCNS-1 is selected when the number of users is 8 or less, and DCNS-k is selected when the number of users is 28 or more. In addition, the base station transmission power is set from -10dB to 10dB, because this is an appropriate transmission power in a large-capacity MIMO system pursuing energy efficiency. In DCNS-K to show various performances of the present invention

Wow

Was set to.

9 is a diagram showing BER (Bit Error Rate) performance of a conventional technique and a technique proposed in the present invention in a 200×10 large-capacity MIMO system. This shows the performance when there are 10 users.

In FIG. 9, the existing DNS using QPSK and the scheme proposed in the present invention have almost the same BER performance, and are almost similar to the optimal ZF BER performance. This is because when N _u =10 in the result of FIG. 8, the large-capacity MIMO system completely dominates diagonally. In this case, since only DCNS-1 is used, BER performance does not improve as K increases regardless of the modulation order.

However, DNS using 16-QAM suffered a deterioration compared to the BER performance of ZF, but the technique proposed in the present invention nulls the user with the largest MUI, so the degree of deterioration is less than that of DNS.

FIG. 10 is a diagram showing the BER performance of the conventional scheme and the scheme proposed in the present invention in a 200 × 20 large-capacity MIMO system. FIG. 11 is a BER performance of the conventional scheme and the scheme proposed in the present invention in a 200 × 30 large-capacity MIMO system. It is a diagram showing the performance. That is, FIG. 10 shows a case of 20 users (N _u =20) and FIG. 11 shows a case of 30 users (N _u =30).

In FIGS. 10 and 11, the DNS and BER performance of the scheme proposed in the present invention have a pattern similar to that of FIG. 9 according to the modulation order. However, compared to the result of FIG. 9, the deterioration of the DNS compared to the BER performance of ZF was greater. This is when N _u =20 and N _u =30 in the results of Figs. 10 and 11, the large-capacity MIMO system is partially This is because they are diagonally dominant and not completely diagonally dominant. In the results of FIGS. 10 and 11, unlike the results of FIG. 9, the BER performance is improved as K increases. As described above, it can be seen that the performance of the technique of the embodiment of the present invention is more prominent as the number of users increases.

FIG. 12 is a diagram showing BER performance according to nulling user selection of the scheme proposed in the present invention when 16-QAM (Quadrature Amplitude Modulation) modulation is used in 200×20 and 200×30 large-capacity MIMO systems. In FIG. 12, the No-selection scheme represents a case of simply selecting randomly, unlike the present invention in which the magnitude of power is considered when selecting K columns. In FIG. 12, it can be seen that the difference in BER performance remains depending on whether or not a nulling user is selected. Since the complexity of nulling user selection is very low, this improvement in BER performance is a very significant result.

13 is a diagram showing an existing technique and a multiplication amount of the present invention according to the number of users. In FIG. 13, regardless of the number of users, the complexity of the technique proposed in the present invention is always smaller than that of the conventional ZF.

On the other hand, when N _u =10, the complexity of the scheme proposed in the present invention is almost similar to that of the existing DNS, because a large-capacity MIMO system is dominant diagonally as shown in FIG. 9.

However, when N _u >10, as the number of users increases, the complexity of the DNS increases more than the complexity of the DNS.This means that when the number of users is large, the method of the present invention selects hybrid DCNS-1 and DCNS- U and DCNS- K to prevent degradation of BER performance. Because it increases

일 때의 복잡도 증가는 도 10 내지 도 11에서 보인 BER 성능 개선 대비 크지 않음을 알 수 있다.Although the complexity of the scheme proposed in the present invention is higher than that of DNS when N _u >10,

It can be seen that the increase in complexity when is not much compared to the improvement in BER performance shown in FIGS. 10 to 11.

The present invention as described above provides an approximate ZF precoding technique using NS in a downlink large-capacity MIMO system. Existing DNS precoding has a very low complexity, but in a large-capacity MIMO system with a high correlation, the error performance of DNS is very poor, which is difficult to solve even by increasing the transmission power of the base station and the length of the NS.

However, the embodiment of the present invention has high error performance even in a large-capacity MIMO system with high correlation by adaptively selecting a precoding technique according to the diagonal dominance of the large-capacity MIMO system. Here, DCNS-1, DCNS-K, hybrid DCNS-1 and DCNS-U are adaptively selected according to the number of users (V) that dominate the large-capacity MIMO system.

In other words, when V=N _u , all users are diagonally dominant, so DCNS-1 with low complexity is used, and when V=0, all users are not diagonally dominant, so DCNS-k is used to improve error performance. When <V<N _u , DCNS-1 is applied to the diagonally dominant user, and DCNS-U is applied to the diagonally non-dominant user. And, in order to improve the error performance of DCNS-k, k users with the largest MUI are selected and completely nulled.

From the simulation results, it was confirmed that the technique proposed in the present invention has a low complexity even in a high-correlation large-capacity MIMO system, has a BER performance that is much higher than that of the existing DNS, and has a BER performance that is close to the optimal ZF.

According to the present invention as described above, for efficient precoding in a large-capacity MIMO system with high correlation between channels, precoding is adaptively performed according to the diagonal dominance of users, while having a much lower complexity than the conventional ZF precoding technique. In addition to providing similar error performance, it has the advantage of improving error performance over existing DNS precoding techniques.

The present invention has been described with reference to the embodiments shown in the drawings, but these are only exemplary, and those of ordinary skill in the art will appreciate that various modifications and equivalent other embodiments are possible therefrom. Therefore, the true technical scope of the present invention should be determined by the technical spirit of the appended claims.

10: base station 20: user
100: precoding device 110: matrix generator
120: comparison unit 130: control unit
140: inverse matrix generator 150: precoding matrix generator

Claims (20)

In a precoding apparatus applied to a base station in a large-capacity MIMO system,
A matrix generator for calculating a Z matrix (Z=G·G ^H ) based on a channel matrix G between the base station and a plurality of users;
A comparison unit that calculates the number of diagonally dominant rows (V; 0≤V≤N _u ) in the Z matrix and compares the number of diagonally dominant rows (V) with the number of users (N _u );
A control unit for selecting only one column component or a set number (k; 1<k<N _u ) of column components in the Z matrix based on a comparison result;
The Φ matrix, which is the initial matrix of NS (Neumann Series), is obtained by combining the selected column components to the diagonal matrix (D) taking only the diagonal components from the Z matrix, and substituting the Φ matrix into the approximate formula of NS-based Z ^-1 An inverse matrix generator for calculating Z ^-1 ; And
A precoding apparatus for a downlink large-capacity MIMO system comprising a precoding matrix generator that multiplies the calculated Z ^-1 by G ^H to generate a precoding matrix. The method according to claim 1,
Approximate expression and the convergence condition of the NS-based Z ^-1 for computing the Z ^-1 is a pre-coding device for the downlink MIMO system capacity is defined by the equation below:

Here, L is the length of LS, I _i is an i×i identity matrix, and O _i is an i×i zero matrix. The method according to claim 1,
The inverse matrix generator,
After obtaining the Φ matrix by combining the selected column components with the diagonal matrix (D), the inverse matrix Φ ^-1 is calculated,
Pre-coding device for the downlink MIMO system capacity by substituting the Φ ^-1 Z and the multiplied value (Φ ^-1 · Z) of the corresponding term of the approximate equation of the Z ^-1 for calculating the Z ^-1. The method according to claim 1,
The control unit,
When selecting the column component, a precoding apparatus for a downlink large-capacity MIMO system that selects each column in the Z matrix in order of high power. The method according to claim 1,
The control unit,
When the number of diagonally dominant rows is the same as the number of users (V=N _u ), selecting one column component having the maximum power from each column in a state in which all the diagonal components in the Z matrix are treated as 0 DCNS (Diagonal plus Column NS)-1 technique is applied,
When the number of diagonally dominant rows is 0 (V=0), DCNS-k selects k column components in the order of power magnitude among each column in a state in which all the diagonal components in the Z matrix are treated as 0 A precoding device for a downlink large-capacity MIMO system to which the technique is applied. The method of claim 5,
The inverse matrix generator,
In the case of V=N _u , the selected one column component is combined with the diagonal matrix (D) to obtain an initial matrix Φ matrix, and the result of dividing the Φ matrix into a diagonal matrix (D) and an atomic matrix is used. And calculate Φ ^-1 , which is the inverse matrix of Φ,
Pre-coding device for the downlink MIMO system capacity by substituting the Φ ^-1 Z and the multiplied value (Φ ^-1 · Z) of the corresponding term of the approximate equation of the Z ^-1 for calculating the Z ^-1. The method of claim 5,
The inverse matrix generator,
In the case of V=O, an operation operation of applying the j-th (j=(1,...,k)) selected column component of the k column components and the diagonal matrix (D) to the Sherman Morrison formula Repeated k times from j=1 to k, calculating the inverse matrix (Φ ^k ) ^-1 for the initial matrix Φ ^k matrix,
By substituting the value ^{((Φ k) -1 · Z} ) the (Φ ^k) multiplied by ^-1 and the Z in the term of the approximate equation of the Z ^-1 a downlink MIMO system for calculating the mass Z ^-1 Precoding device for. The method of claim 5,
The control unit,
When the number of diagonally dominant rows is smaller than the number of users and greater than 0 (0<V<N _u ),
A hybrid technique in which the DCNS-1 technique is applied to the V rows that are diagonally dominant in the Z matrix and the DCNS-U technique (k=U) is applied to the remaining U rows that are not diagonally dominant in the Z matrix. A precoding device for a downlink large-capacity MIMO system to select. The method of claim 8,
The inverse matrix generator,
Wherein 0 <V <N _u is the case, Z in the above Z matrix in the diagonal controlled for each row property by calculating the amount of high diagonal dominant sex size order of rearranging the rows of the G matrix, and using the rearranged G _{S S} Compute the matrix (Z _S =G _S ·G _S ^H ),
The Z _S matrix is divided into four regions based on the V value, and a D ₁ matrix of V×V, an E ₁ matrix of V×U, a D ₂ matrix of U×U in a clockwise direction from the top left to the bottom left, and by applying the technique DCNS-1, remove the matrix of D ₂ U × V next, the D ₁ to obtain a matrix (D ₁₎ ^-1 matrix, by applying a U-DCNS techniques in the matrix D ₂ (D ₂₎ After getting ^-1 matrix,
The ^{_{(D 1) -1, (D}} 2) -1, a precoding unit for the downlink MIMO system for calculating a large Z _S ^-1 the inverse matrix of Z _S by substituting the E ₁ and E ₂ in the matrix equation below :

Here, L is the length of LS, and I _i is an i×i identity matrix. The method of claim 9,
Free for a downlink large-capacity MIMO system in which the D ₁ matrix, the zero matrix, the D ₂ matrix, and the zero matrix are assigned to each of the four regions of the Φ matrix included in the equation for the calculation of Z _S ^-1 . Coding device. In the precoding method performed by a base station in a large-capacity MIMO system,
Calculating a Z matrix (Z=G·G ^H ) based on a channel matrix G between the base station and a plurality of users;
Calculating the number of diagonally dominant rows (V; 0≤V≤N _u ) in the Z matrix and comparing the number of diagonally dominant rows (V) with the number of users (N _u );
Selecting only one column component in the Z matrix based on the comparison result, or selecting a set number (k; 1<k<N _u ) of column components;
The Φ matrix, which is the initial matrix of NS (Neumann Series), is obtained by combining the selected column components to the diagonal matrix (D) taking only the diagonal components from the Z matrix, and substituting the Φ matrix into the approximate formula of NS-based Z ^-1 Calculating Z ^-1 ; And
And generating a precoding matrix by multiplying the calculated Z ^-1 by G ^{H. A} precoding method for a downlink large-capacity MIMO system. The method of claim 11,
Approximate expression and the convergence condition of the NS-based Z ^-1 for computing the Z ^-1 is a pre-coding method for the downlink MIMO system capacity is defined by the equation below:

Here, L is the length of LS, I _i is an i×i identity matrix, and O _i is an i×i zero matrix. The method of claim 11,
The step of calculating the Z ^-1 ,
After obtaining the Φ matrix by combining the selected column components with the diagonal matrix (D), the inverse matrix Φ ^-1 is calculated,
Pre-coding method for the downlink MIMO system capacity by substituting the Φ ^-1 Z and the multiplied value (Φ ^-1 · Z) of the corresponding term of the approximate equation of the Z ^-1 for calculating the Z ^-1. The method of claim 11,
Selecting the thermal component,
When selecting the column component, a precoding method for a downlink high-capacity MIMO system in which power is selected from among each column in the Z matrix in order of high power. The method of claim 11,
Selecting the thermal component,
When the number of diagonally dominant rows is the same as the number of users (V=N _u ), selecting one column component having the maximum power from each column in a state in which all the diagonal components in the Z matrix are treated as 0 DCNS (Diagonal plus Column NS)-1 technique is applied,
When the number of diagonally dominant rows is 0 (V=0), DCNS-k selects k column components in the order of power magnitude among each column in a state in which all the diagonal components in the Z matrix are treated as 0 Precoding method for a downlink large-capacity MIMO system to which the technique is applied. The method of claim 15,
The step of calculating the Z ^-1 ,
In the case of V=N _u , the selected one column component is combined with the diagonal matrix (D) to obtain an initial matrix Φ matrix, and the result of dividing the Φ matrix into a diagonal matrix (D) and an atomic matrix is used. And calculate Φ ^-1 , which is the inverse matrix of Φ,
Pre-coding method for the downlink MIMO system capacity by substituting the Φ ^-1 Z and the multiplied value (Φ ^-1 · Z) of the corresponding term of the approximate equation of the Z ^-1 for calculating the Z ^-1. The method of claim 15,
The step of calculating the Z ^-1 ,
In the case of V=O, an operation operation of applying the j-th (j=(1,...,k)) selected column component of the k column components and the diagonal matrix (D) to the Sherman Morrison formula Repeated k times from j=1 to k, calculating the inverse matrix (Φ ^k ) ^-1 for the initial matrix Φ ^k matrix,
By substituting the value ^{((Φ k) -1 · Z} ) the (Φ ^k) multiplied by ^-1 and the Z in the term of the approximate equation of the Z ^-1 a downlink MIMO system for calculating the mass Z ^-1 For precoding method. The method of claim 15,
Selecting the thermal component,
When the number of diagonally dominant rows is smaller than the number of users and greater than 0 (0<V<N _u ),
A hybrid technique in which the DCNS-1 technique is applied to the V rows that are diagonally dominant in the Z matrix and the DCNS-U technique (k=U) is applied to the remaining U rows that are not diagonally dominant in the Z matrix. Precoding method for a downlink large-capacity MIMO system for selecting. The method of claim 18,
The step of calculating the Z ^-1 ,
Wherein 0 <V <N _u is the case, Z in the above Z matrix in the diagonal controlled for each row property by calculating the amount of high diagonal dominant sex size order of rearranging the rows of the G matrix, and using the rearranged G _{S S} Compute the matrix (Z _S =G _S ·G _S ^H ),
The Z _S matrix is divided into four regions based on the V value, and a D ₁ matrix of V×V, an E ₁ matrix of V×U, a D ₂ matrix of U×U in a clockwise direction from the top left to the bottom left, and by applying the technique DCNS-1, remove the matrix of D ₂ U × V next, the D ₁ to obtain a matrix (D ₁₎ ^-1 matrix, by applying a U-DCNS techniques in the matrix D ₂ (D ₂₎ After getting ^-1 matrix,
The ^{_{(D 1) -1, (D}} 2) -1, by substituting the E ₁ and E ₂ in the matrix equation below free for downlink MIMO system capacity for calculating calculates the Z _S ^-1 is the inverse matrix of Z _S Coding method:

Here, L is the length of LS, and I _i is an i×i identity matrix. The method of claim 19,
Free for a downlink large-capacity MIMO system in which the D ₁ matrix, the zero matrix, the D ₂ matrix, and the zero matrix are assigned to each of the four regions of the Φ matrix included in the equation for the calculation of Z _S ^-1 . Coding method.

Images