CN110730022A - MIMO precoding codebook construction method based on MacQueen clustering - Google Patents

Abstract

The invention discloses a method for constructing a MIMO precoding codebook based on MacQueen clustering. Step 1: randomly select Q channel samples H; Step 2: select the first N (Q>>N) channels in the Q channel samples The matrix H is decomposed by SVD to obtain the corresponding N precoding matrices f, and these N precoding matrices are used as the initial codebook F0={f1,f2,...,fN}, each codeword fi (i=1, 2, . After the channel matrix H is classified, the centroids are updated, so that the classification is more accurate, and better system bit error rate performance can be achieved, thereby improving the reliability of the communication system.

Description

A MIMO precoding codebook construction method based on MacQueen clustering

technical field

The present invention relates to the technical field of mobile communication systems, in particular to a method for constructing a MIMO precoding codebook based on MacQueen clustering.

Background technique

The Multiple Input Multiple Output (MIMO) system architecture has been widely used as a key technology in 4G-LTE, and the latest 5G standard still continues to use MIMO technology as its physical layer architecture. When using MIMO technology, because multiple antennas are deployed at the base station and there are a large number of users in the cell, the MIMO system has serious inter-stream interference and inter-user interference. Precoding is an effective means to solve this problem. It has become the core functional module of the MIMO system architecture and has been widely used, and has also been deeply studied by academia and major communication companies.

There are currently two main types of precoding: linear precoding and nonlinear precoding. Nonlinear precoding has high complexity and is difficult to implement, while linear precoding has low complexity and simple principle, and is widely used in practical communication systems. In the existing linear precoding schemes, it can be divided into non-codebook and codebook-based precoding. However, due to the limitation of feedback link bandwidth in practical communication systems, ideal channel state information (CSI) feedback is often not achieved. For the feedback of partial CSI, 3GPP stipulates a codebook-based precoding scheme. The codebook-based precoding scheme is mainly divided into two parts: codebook construction and codeword selection. There are many methods of codebook construction, such as DFT codebook, Householder codebook, layer structure codebook, and Lloyd vector quantization codebook Wait. Among them, the Lloyd vector quantization codebook uses the Lloyd clustering algorithm to construct the codebook, but the performance of the bit error rate of the codebook generated by this method is limited by the selection of the initial codebook and the setting of the error threshold, which leads to the error of the codebook. The bit rate performance is unstable.

SUMMARY OF THE INVENTION

The object of the present invention is to provide a MIMO precoding codebook construction method based on MacQueen clustering. The present invention replaces the Lloyd clustering algorithm in the Lloyd vector quantization codebook by using the MacQueen clustering algorithm, so that the calculation process does not It involves the error threshold setting similar to the Lloyd iteration, which avoids the influence of the selection of the error threshold on the bit error rate performance of the codebook. At the same time, after each channel matrix H is classified, the centroid is updated to classify the next channel matrix. A more accurate classification basis is provided, so the method further improves the bit error rate performance of the MIMO system and improves the reliability of the communication system.

For achieving the above object, the invention provides following technical scheme: a kind of MIMO precoding codebook construction method based on MacQueen clustering, comprises the steps:

Step 1: randomly select Q channel samples H;

Step 2: Perform SVD decomposition on the first N (Q>>N) channel matrices H in the Q channel samples to obtain the corresponding N precoding matrices f, and use the N precoding matrices as the initial codebook F ₀ ={f ₁ ,f ₂ ,...,f _N }, each codeword f _i (i=1,2,...,N) is the initial centroid of N classes;

Step 3: Select the next channel matrix H, calculate the square of the Frobenius norm multiplied by each codeword matrix fi, and assign the channel matrix H to the codeword matrix fi that maximizes the square of the Frobenius norm. in the class;

Step 4: recalculate the centroid f _i of the class updated in step 3;

Step 5: Repeat steps 3 and 4 until all the Q channel samples are allocated;

Step 6: Take the centroids corresponding to the final N classes as the precoding codebook F.

Preferably, the first N (Q>>N) channel matrices H in the Q channel samples are subjected to SVD decomposition to obtain corresponding N precoding matrices f, wherein the precoding matrices are obtained according to the following formula:

H=U∑V

f=V(j)

The U, Σ, and V matrices are the left singular matrix obtained by SVD decomposition of the channel matrix, the diagonal matrix composed of singular values, and the right singular matrix, and V(j) is the matrix composed of the first j columns of the matrix V.

Preferably, the channel matrices are classified one by one in sequence, and after each channel matrix H is allocated, the centroid f of the class to which the new channel matrix is added must be recalculated, where the centroid f is calculated according to the following formula:

表示中每个信道矩阵H进行SVD分解得到的预编码矩阵f组成的类。in

express The class composed of the precoding matrix f obtained by the SVD decomposition of each channel matrix H in .

Preferably, in the step 3, the next channel matrix H is selected, starting from the N+1 th channel matrix H, and then the N+2, N+3, . . . , Q th channel matrices H in sequence.

Preferably, in the step 3, the calculation expression for calculating the square of the Frobenius norm multiplied by H and each codeword matrix fi is:

表示质心为f_k的信道矩阵H组成的类。in

Represents a class consisting of channel matrices H whose centroids are _fk .

The invention provides a MIMO precoding codebook construction method based on MacQueen clustering, which has the following beneficial effects:

In the invention, the MacQueen clustering algorithm is used to replace the Lloyd clustering algorithm in the Lloyd vector quantization codebook, so that the error threshold setting similar to the Lloyd iteration is not involved in the calculation process, and the selection of the error threshold value is avoided. At the same time, the centroid is updated after the classification of each channel matrix H, which provides a more accurate classification basis for the classification of the next channel matrix. Therefore, this method further improves the BER performance of the MIMO system. improve the reliability of the communication system.

Description of drawings

Figure 1 is a block diagram of the MIMO-OSTBC system;

Fig. 2 is the working flow chart of precoding based on codebook;

3 is a flowchart of a method for constructing a MIMO precoding codebook based on MacQueen clustering provided by the present invention;

4 is a schematic diagram showing the comparison of the bit error rate (BER) performance curves of the MIMO precoding codebook construction method based on MacQueen clustering provided by the present invention, the Lloyd vector quantization codebook, and the DFT codebook when the size of the precoding codebook is L=64 ;

5 is a schematic diagram showing the comparison of the bit error rate (BER) performance curves of the MIMO precoding codebook construction method based on MacQueen clustering provided by the present invention and the Lloyd vector quantization codebook when the precoding codebook size L=16.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

In this example, the communication system used in the simulation is a MIMO-OSTBC system, which adopts 4-QAM modulation, Alamouti space-time coding, the number of transmit antennas Tx is 4, and the number of receive antennas Rx is 2. The coding part adopts a codebook-based precoding scheme, and its specific work flow is shown in Figure 2.

Example 1

A method for constructing a MIMO precoding codebook based on MacQueen clustering, comprising the following steps:

Step 1: randomly select Q=10000 channel matrices H;

类的初始质心，

表示质心为f_i的信道矩阵H组成的类；Step 2: SVD decompose the first N=64 channel matrices into H=UΣV, and according to f=V(2), that is, select the first two columns of the right singular matrix V as the precoding matrix, and obtain the corresponding 64 precoding matrices f, and take these 64 precoding matrices as the initial codebook F ₀ ={f ₁ ,f ₂ ,...,f ₆₄ }, the codebook size L=64, each codeword f _i (i=1,2, ...,64) is 64

the initial centroid of the class,

represents the class composed of the channel matrix H whose centroid is f _i ;

Step 3: Select the 65th channel matrix H ₆₅ , and calculate the square of the Frobenius norm multiplied by this matrix and each codeword matrix f _i , that is,

类中；Then, the channel matrix H ₆₅ is assigned to the class corresponding to the codeword matrix f that maximizes the value value. For the sake of clarity, suppose that when i=m, the value value is the largest, then H ₆₅ is assigned to

class;

类的质心f_m：Step 4: Recalculate the updated value in Step 3 according to the following formula

Class centroid f _m :

类中每个f与类中所有的信道矩阵H计算

的值，并求它们的期望，然后选出使期望值最小的f作为该类的新质心，表示

中每个信道矩阵H进行SVD分解得到的预编码矩阵f 组成的类；where the formula represents the calculation

Each f in the class is associated with All channel matrices in the class H are calculated

, and find their expectations, and then select the f that minimizes the expected value as the the new centroid of the class, express

The class composed of the precoding matrix f obtained by SVD decomposition of each channel matrix H in the

Step 5: Select the 66th, 67th, ..., 10000th channel matrix respectively and repeat steps 3 and 4;

类的质心f_i作为码字组成我们所需要的预编码码本F。Step 6: After the final assignments are made, each

The centroid f _i of the class is used as a codeword to form the precoding codebook F we need.

Example 2

A method for constructing a MIMO precoding codebook based on MacQueen clustering, comprising the following steps:

Step 1: randomly select Q=10000 channel matrices H;

类的初始质心，表示质心为f_i的信道矩阵H组成的类；Step 2: SVD decompose the first N=16 channel matrices into H=UΣV, and according to f=V(2), that is, select the first two columns of the right singular matrix V as the precoding matrix, and obtain the corresponding 64 precoding matrices f, and take these 64 precoding matrices as the initial codebook F ₀ ={f ₁ ,f ₂ ,...,f ₁₆ }, the codebook size L=16, each codeword f _i (i=1,2, ...,16) is 16

the initial centroid of the class, represents the class composed of the channel matrix H whose centroid is f _i ;

Step 3: Select the 17th channel matrix H ₁₇ , and calculate the square of the Frobenius norm multiplied by the matrix and each codeword matrix f _i , namely

类中；Then, the channel matrix H ₁₇ is assigned to the class corresponding to the codeword matrix f that maximizes the value value. For the sake of clarity, suppose that when i=m, the value value is the largest, then H ₁₇ is assigned to

class;

Step 4: Recalculate the updated value in Step 3 according to the following formula Class centroid f _m :

类中每个f与

类中所有的信道矩阵H计算

的值，并求它们的期望，然后选出使期望值最小的f作为该

类的新质心，

表示

中每个信道矩阵H进行SVD分解得到的预编码矩阵f 组成的类；where the formula represents the calculation

Each f in the class is associated with

All channel matrices in the class H are calculated

, and find their expectations, and then select the f that minimizes the expected value as the

the new centroid of the class,

express

The class composed of the precoding matrix f obtained by SVD decomposition of each channel matrix H in the

Step 5: Select the 18th, 19th, ..., 10000th channel matrix respectively and repeat steps 3 and 4;

类的质心f_i作为码字组成我们所需要的预编码码本F。Step 6: After the final assignments are made, each

The centroid f _i of the class is used as a codeword to form the precoding codebook F we need.

The bit error rate performance of Embodiment 1 of the present invention and the existing precoding codebook in the above-mentioned communication system is simulated and compared, and Figure 4 is obtained. The rate performance is simulated and compared, and Figure 5 is obtained.

in conclusion:

(1) It can be seen from FIG. 4 and FIG. 5 that, under the same conditions, compared with the existing precoding codebook scheme, the present invention can not increase the feedback overhead, that is, the same codebook size. The bit error rate of the system is further reduced, and under the condition of high signal-to-noise ratio, the present invention exhibits more excellent bit error rate performance.

(2) It can be seen from the specific examples described in this paper that the method of the present invention does not involve the setting of error thresholds similar to Lloyd iteration in the calculation process, which avoids the selection of error thresholds on the performance of codebook bit error rate. At the same time, after the classification of each channel matrix H, the centroid is updated, which makes the classification more accurate, and further improves the bit error rate performance of the MIMO system, thereby enhancing the reliability of the communication system.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, and substitutions can be made in these embodiments without departing from the principle and spirit of the invention and modifications, the scope of the present invention is defined by the appended claims and their equivalents.

Claims (5)

1. a MIMO precoding codebook construction method based on MacQueen clustering, is characterized in that, comprises the steps: Step 1: randomly select Q channel samples H; Step 2: Perform SVD decomposition on the first N (Q>>N) channel matrices H in the Q channel samples to obtain the corresponding N precoding matrices f, and use the N precoding matrices as the initial codebook F ₀ ={f ₁ ,f ₂ ,...,f _N }, each codeword f _i (i=1,2,...,N) is the initial centroid of N classes; Step 3: Select the next channel matrix H, calculate the square of the Frobenius norm multiplied by each codeword matrix fi, and assign the channel matrix H to the codeword matrix fi that maximizes the square of the Frobenius norm. in the class; Step 4: recalculate the centroid f _i of the class updated in step 3; Step 5: Repeat steps 3 and 4 until all the Q channel samples are allocated; Step 6: Take the centroids corresponding to the final N classes as the precoding codebook F. 2. a kind of MIMO precoding codebook construction method based on MacQueen clustering according to claim 1, is characterized in that: carry out SVD decomposition to first N (Q>>N) channel matrices H in Q channel samples , the corresponding N precoding matrices f are obtained, where the precoding matrix is obtained according to the following formula: H=U∑V f=V(j) The U, Σ, and V matrices are the left singular matrix obtained by SVD decomposition of the channel matrix, the diagonal matrix composed of singular values, and the right singular matrix, and V(j) is the matrix composed of the first j columns of the matrix V. 3. a kind of MIMO precoding codebook construction method based on MacQueen clustering according to claim 1, is characterized in that: the channel matrix is classified one by one in order, and after each distribution of a channel matrix H, all Recalculate the centroid f of the class to which the new channel matrix is added, where the centroid f is calculated according to the following formula:

in

express

The class composed of the precoding matrix f obtained by the SVD decomposition of each channel matrix H in . 4. a kind of MIMO precoding codebook construction method based on MacQueen clustering according to claim 1, is characterized in that: in described step 3, choose next channel matrix H, be from the N+1th channel matrix H At the beginning, it is the N+2, N+3, . . . , Q-th channel matrix H in sequence. 5. a kind of MIMO precoding codebook construction method based on MacQueen clustering according to claim 1, is characterized in that: in described step 3, calculate the square of the Frobenius norm that H and each codeword matrix fi are multiplied by The calculation expression is: in

Represents a class consisting of channel matrices H whose centroids are _fk .

Images