CN102594467A

CN102594467A - Receiver detection method for wireless multiple input multiple output system

Info

Publication number: CN102594467A
Application number: CN2012100411796A
Authority: CN
Inventors: 卢炳山; 伊海珂; 俞晖; 刘伟; 罗汉文
Original assignee: Shanghai Jiao Tong University
Current assignee: Leadcore Technology Co Ltd; Shanghai Jiao Tong University
Priority date: 2011-08-15
Filing date: 2012-02-22
Publication date: 2012-07-18
Anticipated expiration: 2032-02-22
Also published as: CN102594467B

Abstract

The invention provides a receiver detection method of a wireless multiple-input multiple-output system, comprising the following steps: performing QR decomposition on a channel matrix H to obtain a Q matrix and an upper triangular matrix R; converting the conjugate transposition of the Q matrix to the received signal vector Multiply y to get the equalized signal of the received signal

Establish the lookup table LUT of node expansion sequence, set the radius of spherical decoding, and the available storage space is M; calculate the search center, obtain the subscript of the lookup table according to the search center and modulation mode, and obtain the initially selected node; move from the stack Remove the selected node, and expand sibling nodes and child nodes according to the selected node; judge whether the selected node is a leaf node, and maintain the soft value table; select the node selected for the next iteration; calculate the LLR value according to the soft value table. The invention effectively reduces the complexity of the receiver under the premise of ensuring the system performance.

Description

Receiver Detection Method for Wireless MIMO System

技术领域 technical field

本发明涉及的是一种无线通信技术领域的方法，具体地涉及一种无线多输入多输出系统的接收机检测方法。The invention relates to a method in the technical field of wireless communication, in particular to a receiver detection method of a wireless multiple-input multiple-output system.

背景技术 Background technique

传统的多输入多输出(Multiple Input Multiple Output，MIMO)技术，是通过利用基站和用户端的多天线结构，实现分集和空分复用，从而增加系统吞吐量的一种技术，因为成为3GPP-LTE，IEEE802.16e WIMAX的研究热点。近年来，随着turbo译码和LDPC译码技术的发展，MIMO接收机的译码与信道译码技术的结合能够大幅降低系统的误码率。最大似然检测(ML)为最优的接收机检测算法，但ML检测的复杂度随着发射天线数和调制阶数呈指数型增长。次优的线性检测算法如迫零(ZF)和最小均方误差(MMSE)准则，虽然复杂度低，但都不能达到接收满分集度，性能远低于ML检测算法。多输入多输出系统中的软输出球形译码检测方法能够大幅降低系统的复杂度。The traditional multiple input multiple output (Multiple Input Multiple Output, MIMO) technology is a technology that increases the throughput of the system by using the multi-antenna structure of the base station and the user end to achieve diversity and space division multiplexing, because it has become a 3GPP-LTE , a research hotspot of IEEE802.16e WIMAX. In recent years, with the development of turbo decoding and LDPC decoding technology, the combination of MIMO receiver decoding and channel decoding technology can greatly reduce the bit error rate of the system. Maximum likelihood detection (ML) is the optimal receiver detection algorithm, but the complexity of ML detection increases exponentially with the number of transmit antennas and modulation order. Suboptimal linear detection algorithms such as zero-forcing (ZF) and minimum mean square error (MMSE) criteria, although low in complexity, cannot achieve full receiver diversity, and their performance is much lower than that of ML detection algorithms. The soft output sphere decoding detection method in the MIMO system can greatly reduce the complexity of the system.

经对现有文献检索发现，Studer.C等在《IEEE Transaction on Information Theory》(美国电气与电子工程师协会信息理论期刊，2010年10月第56卷第4827至第4842页)上，发表了“Soft-InputSoft-Output Single Tree-Search Sphere Decoding”(“单树搜索的软输出软输入球形译码”)，该文提出了，在多天线多输入多输出系统中的，在搜索过程中动态更新每一个比特的软值和ML硬判值，该文献证明了STS球形译码算法性能是最优的，且保证每一个节点只需被访问一次，但这种软输出球形译码仍然复杂度较高，不易实现。又经检索发现，Markus M等在《Signal Processing》(信号处理杂志，2010年10月第90卷第2863页至第2876页)上，发表了“Implementation aspects of list spheredecoder algorithms for MIMO-OFDM systems”(“在MIMO-OFDM系统中列表球形译码的应用”)，该文通过K-Best球形译码或Dijkstra(迪杰斯特拉)球形译码算法的列表球形译码计算软值，但列表球形译码不能保证列表中ML硬判值的比特极性一定有相反项，造成LLR值的计算误差，影响系统性能。After searching the existing literature, it was found that Studer.C et al. published "IEEE Transaction on Information Theory" (Information Theory Journal of the Institute of Electrical and Electronics Engineers, October 2010, Volume 56, Page 4827 to Page 4842) " Soft-InputSoft-Output Single Tree-Search Sphere Decoding" ("Single Tree Search Soft-Output Soft-Input Sphere Decoding"), this paper proposes that in a multi-antenna MIMO system, the dynamic update during the search process The soft value of each bit and the ML hard judgment value, this document proves that the performance of the STS sphere decoding algorithm is optimal, and ensures that each node only needs to be visited once, but the complexity of this soft output sphere decoding is still relatively high. high and difficult to achieve. After searching, it was found that Markus M et al. published "Implementation aspects of list spheredecoder algorithms for MIMO-OFDM systems" on "Signal Processing" (Signal Processing Journal, October 2010, Volume 90, Page 2863 to Page 2876). ("Application of Listed Sphere Decoding in MIMO-OFDM Systems"), this paper calculates soft values by K-Best Sphere Decoding or Listed Sphere Decoding of Dijkstra (Dijkstra) Sphere Decoding Algorithm, but the list Sphere decoding cannot guarantee that the bit polarity of the ML hard-judgment value in the list must have an opposite item, which will cause calculation errors in the LLR value and affect system performance.

发明内容 Contents of the invention

本发明的目的在于克服现有技术的上述不足，提供一种低复杂度的软输出球形译码算法，降低接收机的复杂度。本发明结合Dijkstra球形译码算法推广到单树搜索的软输出球形译码，通过构造查找表LUT(Look Up Table)，改进了Dijkstra球形译码的节点扩展搜索算法，快速地得到ML硬判值和LLR值，减少搜索空间，降低了接收机复杂度，且性能与ML检测一致。本发明具有在准ML性能且复杂度低的特点，并且适用各种多输入多输出系统的特点。The purpose of the present invention is to overcome the above-mentioned shortcomings of the prior art, provide a low-complexity soft-output sphere decoding algorithm, and reduce the complexity of the receiver. The present invention combines Dijkstra's spherical decoding algorithm and extends it to the soft output spherical decoding of single tree search, improves the node expansion search algorithm of Dijkstra's spherical decoding by constructing a lookup table LUT (Look Up Table), and quickly obtains the ML hard judgment value and LLR values, the search space is reduced, the receiver complexity is reduced, and the performance is consistent with ML detection. The invention has the characteristics of quasi-ML performance and low complexity, and is applicable to various multi-input and multi-output systems.

根据本发明的一个方面，所述无线多输入多输出系统的接收机检测方法包括以下步骤：According to one aspect of the present invention, the receiver detection method of the wireless MIMO system includes the following steps:

步骤1：对信道矩阵H进行QR分解，得到Q矩阵和上三角矩阵R；将Q矩阵的共轭转置与接收信号向量y相乘，得到接收信号的均衡信号

Step 1: Perform QR decomposition on the channel matrix H to obtain the Q matrix and upper triangular matrix R; multiply the conjugate transpose of the Q matrix by the received signal vector y to obtain the balanced signal of the received signal

步骤2：建立节点扩展顺序的查找表LUT(Look Up Table)；Step 2: Establish a lookup table LUT (Look Up Table) for the node expansion sequence;

步骤3：计算搜索中心，根据搜索中心及调制方式得到查找表的下标，得到初始选中节点，压入栈中；Step 3: Calculate the search center, obtain the subscript of the lookup table according to the search center and the modulation method, obtain the initially selected node, and push it into the stack;

步骤4：从栈中移除选中节点，并根据选中节点的扩展兄弟节点，更新半径，判断当前节点的半径是否小于当前搜索半径，是则砍掉当前节点及其所有分支，否则将兄弟节点压入栈中；Step 4: Remove the selected node from the stack, and update the radius according to the extended sibling nodes of the selected node, and judge whether the radius of the current node is smaller than the current search radius. into the stack;

步骤5：判断选中节点是否为叶子节点，如果是叶子节点，则进行软值表的维护，否则向下一层扩展搜索，计算搜索中心及调制方式得到查找表的下标，得到子节点，判断当前节点的半径是否小于当前搜索半径，是则砍掉当前节点及其所有分支，否则将子节点压入栈中；Step 5: Determine whether the selected node is a leaf node, if it is a leaf node, maintain the soft value table, otherwise expand the search to the next layer, calculate the search center and modulation method to obtain the subscript of the lookup table, obtain the child nodes, and judge Whether the radius of the current node is smaller than the current search radius, if so, cut off the current node and all its branches, otherwise push the child nodes into the stack;

步骤6：断当前栈是否为空，如果为空，则到步骤7；否则根据栈中的的存储空间和权值选择下一次迭代选中的节点，返回步骤5；Step 6: Determine whether the current stack is empty, if it is empty, go to step 7; otherwise, select the node selected for the next iteration according to the storage space and weight in the stack, and return to step 5;

步骤7：据软值表进行LLR值计算。Step 7: Calculate the LLR value according to the soft value table.

优选地，在所述步骤2中，查找表建立步骤为：Preferably, in said step 2, the look-up table establishment step is:

假设树搜索到第i层，则部分向量

已知，定义c_i为第i层的搜索中心Assuming that the tree searches to the i-th level, the partial vector

Known, define c _i as the search center of the i-th layer

$c_{i} = \frac{1}{r_{ii}} ({\overset{\cdot}{y}}_{i} - Σ_{n = i + 1}^{N_{R}} r_{in} s_{n})$ 式一 $c_{i} = \frac{1}{r_{i}} ({\overset{&Center Dot;}{the y}}_{i} - Σ_{no = i + 1}^{N_{R}} r_{in} {the s}_{no})$ formula one

则所述式一改写为：Then the formula one is rewritten as:

$d_{i} = d_{i + 1} + r_{ii}^{2} {| c_{i} - s_{i} |}^{2}$ 式二 $d_{i} = d_{i + 1} + r_{i}^{} {| c_{i} - {the s}_{i} |}^{2}$ formula two

节点首先访问距离搜索中心c_i最近的星座点s_i，然后按照与c_i由近到远的次序进行排序搜索访问；当MIMO系统的调制方式确定，调制符号集即确定，因而可以根据c_i所属位置区域对星座点的扩展顺序进行排列，建立查找表LUT；其中，将每个星座点的区域分成4块，按照空间的距离由近到远对星座点的访问次序进行排列，4-QAM需要的查找表大小为16，16-QAM需要的查找表大小为64，64-QAM需要的查找表大小为256。The node first visits the constellation point _si closest to the search center _ci , and then searches and visits according to the order from closest to far from _ci ; when the modulation mode of the MIMO system is determined, the modulation symbol set is determined, so it can be determined according to _ci Arrange the expansion order of the constellation points in the location area to which they belong, and establish a lookup table LUT; among them, the area of each constellation point is divided into 4 blocks, and the access order of the constellation points is arranged according to the spatial distance from near to far. 4-QAM The required lookup table size is 16, the required lookup table size for 16-QAM is 64, and the required lookup table size for 64-QAM is 256.

优选地，在所述步骤3和步骤5中，搜索中心计算过程为：Preferably, in said step 3 and step 5, the search center calculation process is:

假设树搜索到第i层，则部分向量

Known, define c _i as the search center of the i-th layer

${c c}_{i i} = = \frac{11}{{r r}_{ii i}} (({\overset{\cdot &Center Dot;}{y the y}}_{i i} - - {Σ Σ}_{n no = = i i + + 11}^{{N N}_{R R}} {r r}_{in in} {s the s}_{n no})) . .$

优选地，在所述步骤3和步骤4中，节点扩展过程为：Preferably, in the step 3 and step 4, the node expansion process is:

假设当前栈要向下一层扩展的节点为N_c＝(s＝s⁽ⁱ⁾，d(s)，p，q，i)，栈中剩余的空间为M，其中s为当前的解向量，i为当前节点位于的层数，p和q分别为在当前所选择查找表的下标和当前节点在第i层的星座点的下标，其中，节点扩展方法包括如下子步骤：Assume that the node to be extended to the next layer of the current stack is N _c = (s=s ⁽ⁱ⁾ , d(s), p, q, i), and the remaining space in the stack is M, where s is the current solution vector , i is the number of layers where the current node is located, p and q are respectively the subscript of the currently selected lookup table and the subscript of the constellation point of the current node at layer i, wherein the node expansion method includes the following substeps:

1)从栈中移除节点N_c＝(s＝s⁽ⁱ⁾，d(s)，p，q，level＝i)，M＝M+1，扩展第level＝i层与节点N最近的节点，即令s_f＝(LUT(p，q+1)，s⁽ⁱ⁺¹⁾)，计算d(s_f)，如果d(s_f)＜R₀将节点N_f＝(s＝s_f，d(s_f)，p，q＝q+1，level＝i)存入栈中，M＝M-1；1) Remove node N _c =(s=s ⁽ⁱ⁾ , d(s), p, q, level=i) from the stack, M=M+1, extend the level=i layer closest to node N node, that is, let s _f =(LUT(p,q+1), s ⁽ⁱ⁺¹⁾ ), calculate d(s _f ), if d(s _f )<R ₀ will node N _f =(s=s _f , d(s _f ), p, q=q+1, level=i) are stored in the stack, M=M-1;

2)如果节点N_c是叶子节点即i＝1时，进行LLR值更新和ML解更新，否则将向下一层扩展，通过下式2) If the node N _c is a leaf node, that is, when i=1, update the LLR value and update the ML solution, otherwise it will expand to the next layer, through the following formula

${d d}_{i i} = = {d d}_{i i + + 11} + + {r r}_{ii i}^{22} {| | {c c}_{i i} - - {s the s}_{i i} | |}^{22}$

计算搜索中心c_i-1，得到查找表下标p，令s_exp＝(LUT(p，0)，s⁽ⁱ⁾)，计算d(s_exp)，如果d(s_exp)＜R₀将节点N_exp＝(s＝s_exp，d(s_exp)，p，q＝0，i-1)存入栈中，M＝M-1。Calculate the search center c _i-1 to obtain the subscript p of the lookup table, let s _exp =(LUT(p, 0), s ⁽ⁱ⁾ ), calculate d(s _exp ), if d(s _exp )＜R ₀ will The node N _exp =(s=s _exp , d(s _exp ), p, q=0, i-1) is stored in the stack, and M=M-1.

优选地，在所述步骤3和步骤4中，半径更新过程为：Preferably, in the step 3 and step 4, the radius update process is:

半径更新分为两部分：The radius update is divided into two parts:

1)对于已知部分向量，即k≥i，m＝1，2，…，Q，半径R₀与

的

相关，即1) For known partial vectors, namely k≥i, m=1, 2, ..., Q, radius R ₀ and

of

related, namely

${R R}_{00} = = max max {{d d (({s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}})) | | &ForAll; &ForAll; k k &GreaterEqual; &Greater Equal; i i,, m m = = 1,2 1,2 . . . . . .,, Q Q,, {s the s}_{k k,, m m} = = {s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}}}};;$

2)对于还没搜索的部分向量，即k＜i，m＝1，2，…，Q，半径R₀与所有的

相关，即2) For the partial vectors that have not been searched, that is, k<i, m=1, 2, ..., Q, the radius R ₀ and all

related, namely

${R R}_{00} = = max max {{d d (({s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}})) | | &ForAll; &ForAll; k k < < i i,, m m = = 1,2 1,2,, . . . . . .,, Q Q}} . .$

优选地，在所述步骤4中，软值表更新过程为：Preferably, in said step 4, the soft value table update process is:

初始化为 $d (s^{ML}) = d (s_{k, m}^{\overset{&OverBar;}{ML}}) = \infty (&ForAll; k, m),$ 分两种情况：initialized to $d ({the s}^{ML}) = d ({the s}_{k, m}^{\overset{&OverBar;}{ML}}) = \infty (&ForAll; k, m),$ There are two situations:

1)当得到一个新的ML解时，即d(s)＜d(s^ML)，则将所有符合

比特位的更新为

并且更新s^ML＝s和d(s^ML)＝d(s)；1) When a new ML solution is obtained, that is, d(s)<d(s ^ML ), then all

Bits update to

And update s ^ML =s and d(s ^ML )=d(s);

2)若d(s)＞d(s^ML)，此时只需更新

的值，当

且

时，更新2) If d(s)＞d(s ^ML ), only update

value, when

and

when, update

$d d (({s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}})) = = d d ((s the s)) . .$

优选地，在所述步骤6中，节点选择方法为：Preferably, in said step 6, the node selection method is:

令n_max＝min(M+2，N_T)，将栈中的节点根据d(s)进行从小到大排序，选择具有最小的d(s)且满足i＜n_max的节点，设选中的节点为N_c。Let n _max = min(M+2, N _T ), sort the nodes in the stack according to d(s) from small to large, select the node with the smallest d(s) and satisfy i<n _max , and set the selected The node is N _c .

优选地，在所述步骤7中，LLR值计算方法为：Preferably, in said step 7, the method for calculating the LLR value is:

对于 $&ForAll; k &Element; {1, . . ., N_{T}},$ $&ForAll; m &Element; {1, . . ., Q},$ 利用下式计算LLR值for $&ForAll; k &Element; {1, . . ., N_{T}},$ $&ForAll; m &Element; {1, . . ., Q},$ Calculate the LLR value using the following formula

$L L (({b b}_{k k,, m m})) = = \{\begin{matrix} d d (({s the s}^{ML ML})) - - d d (({s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}})),, {s the s}_{k k,, m m}^{ML ML} = = - - 11 \\ d d (({s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}})) - - d d (({s the s}^{ML ML})),, {s the s}_{k k,, m m}^{ML ML} = = + + 11 \end{matrix} . .$

附图说明 Description of drawings

图1所示为本发明方法查找表建立的原理图；Fig. 1 shows the schematic diagram that the inventive method look-up table is set up;

图2所示为在4×4MIMO系统下，在QPSK调制下，不同接收机检测方法的误码率对比示意图；Figure 2 is a schematic diagram of the bit error rate comparison of different receiver detection methods under QPSK modulation in a 4×4 MIMO system;

图3所示为在4×4MIMO系统下，在16QAM调制下，不同接收机检测方法的误码率对比示意图；Figure 3 is a schematic diagram of the bit error rate comparison of different receiver detection methods under 16QAM modulation in a 4×4 MIMO system;

图4所示为在4×4MIMO系统下，在QPSK调制下，不同接收机检测方法的搜索空间对比及复杂度对比示意图；Figure 4 is a schematic diagram of the comparison of search space and complexity of different receiver detection methods under QPSK modulation in a 4×4 MIMO system;

图5所示为在4×4MIMO系统下，在16QAM调制下，不同接收机检测方法的搜索空间对比及复杂度对比示意图；Figure 5 is a schematic diagram of the search space comparison and complexity comparison of different receiver detection methods under 16QAM modulation in a 4×4 MIMO system;

图6所示为无线多输入多输出系统框图；Figure 6 is a block diagram of a wireless MIMO system;

图7所示为本发明的实现步骤框图。Fig. 7 is a block diagram showing the implementation steps of the present invention.

具体实施方式 Detailed ways

下面参照附图对本发明的实施例进行详细说明，在描述过程中省略了对于本发明来说是不必要的细节和功能，以防止对本发明的理解造成混淆。下面给出本发明的具体实施例，适用于长期演进系统及进阶长期演进系统。需要说明的是，本发明不限于实施例中所描述的应用，也可适用于其它使用多输入多输出技术和接收检测技术的无线通信系统。Embodiments of the present invention will be described in detail below with reference to the accompanying drawings, and unnecessary details and functions for the present invention will be omitted during the description to prevent confusion in the understanding of the present invention. Specific embodiments of the present invention are given below, which are applicable to a long term evolution system and an advanced long term evolution system. It should be noted that the present invention is not limited to the applications described in the embodiments, and is also applicable to other wireless communication systems using multiple input multiple output technology and reception detection technology.

实施例Example

为使本发明的目的，技术方案和优点更加清楚，下面将结合附图和具体实施实例对本发明进行详细描述。In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be described in detail below in conjunction with the accompanying drawings and specific implementation examples.

本实施例采用的系统模型为具有N_T个发送天线N_R个接收天线的MIMO系统，不失一般性，令N_R＝N_T，发射信息比特通过信道编码模块，交织器模块和串并转换调制模块后得到N_T×1发射信号向量

其中Ω为调制符号集，|Ω|＝2^Q(Q为调制阶数)。对应的N_R×1接收信号向量

所以MIMO系统的模型为：The system model adopted in this embodiment is a MIMO system with _NT transmitting antennas and _NR receiving antennas. Without loss of generality, let _NR = _NT , and transmit information bits through channel coding module, interleaver module and serial-to-parallel conversion Obtain N _T × 1 transmit signal vector after modulating the module

Among them, Ω is the modulation symbol set, |Ω|=2 ^Q (Q is the modulation order). The corresponding N _R ×1 received signal vector

So the model of the MIMO system is:

y＝Hs+n (1)y＝Hs+n

式(1)中的H为信道矩阵，n表示均值为0，方差为σ²的高斯加性白噪声。H in formula (1) is the channel matrix, and n represents Gaussian additive white noise with mean value 0 and variance ^σ2 .

对信道矩阵H进行QR分解得到H＝QR，其中R是上三角矩阵，Q为正交矩阵，式(1)式可以写成Carry out QR decomposition to the channel matrix H to obtain H=QR, wherein R is an upper triangular matrix, Q is an orthogonal matrix, and the formula (1) can be written as

$\overset{\cdot &Center Dot;}{y the y} = = Rs Rs. + + \overset{\cdot \cdot}{n no} - - - - - - ((22))$

其中

定义

为向量

的第i个元素，r_ij为上三角矩阵R的第(i，j)个元素。定义部分信号向量

向量s⁽ⁱ⁾看成一棵树的节点，设i＝N_T+1层为根节点，i＝1层为叶子节点，每个节点共有2^Q个子节点，每个叶子节点s⁽¹⁾都为一个解向量。欧氏距离

可以通过部分欧氏距离(PED)迭代计算得到：in

definition

as a vector

The ith element of , r _ij is the (i, j)th element of the upper triangular matrix R. define partial signal vector

The vector s ⁽ⁱ⁾ is regarded as a node of a tree, and the i=N _T +1 layer is set as the root node, and the i=1 layer is the leaf node, and each node has 2 ^Q sub-nodes in total, and each leaf node s ⁽¹⁾ has is a solution vector. Euclidean distance

It can be calculated iteratively by Partial Euclidean Distance (PED):

d_i＝d_i+1+|e_i|²，i＝N_T，N_T-1，…，1 (3)d _i ＝d _i+1 +|e _i | ² , i=N _T , N _T -1,...,1 (3)

${| | {e e}_{i i} | |}^{22} = = {| | {\overset{\cdot \cdot}{y the y}}_{n no} - - {Σ Σ}_{n no = = i i + + 11}^{{N N}_{R R}} {r r}_{in in} {s the s}_{n no} - - {r r}_{ii i} {s the s}_{i i} | |}^{22} - - - - - - ((44))$

其中d(s)＝d₁，初始化为0。where d(s)=d ₁ , Initialized to 0.

MIMO接收机输出发端比特数据b_k，m，k∈{1，…，N_T}，m∈{1，…，Q}的软值估计，max-log的软值LLR(Log Likelihood Ratio)的定义式为：The MIMO receiver outputs the soft value estimation of the transmitting bit data b _{k, m} , k∈{1,..., _NT }, m∈{1,...,Q}, and the soft value of the max-log LLR (Log Likelihood Ratio) The definition formula is:

$L L (({b b}_{k k,, m m})) \approx \approx \frac{11}{{σ σ}_{n no}^{22}} [[\underset{s the s &Element; &Element; {S S}_{k k,, m m}^{- - 11}}{min min} {| | | | y the y - - Hs Hs | | | |}^{22} - - \underset{s the s &Element; &Element; {S S}_{k k,, m m}^{+ + 11}}{min min} {| | | | y the y - - Hs Hs | | | |}^{22}]] - - - - - - ((55))$

$\approx \approx \frac{11}{{σ σ}_{n no}^{22}} [[\underset{s the s &Element; &Element; {S S}_{k k,, m m}^{- - 11}}{min min} {| | | | \overset{\cdot \cdot}{y the y} - - Rs Rs. | | | |}^{22} - - \underset{s the s &Element; &Element; {S S}_{k k,, m m}^{+ + 11}}{min min} {| | | | \overset{\cdot \cdot}{y the y} - - Rs Rs. | | | |}^{22}]]$

式(5)中，L(b_k，m)表示发射向量符号s_k第m个比特的LLR值，

表示向量s中的符号s_k第m个比特的值为±1的集合。由式(5)观察可以得到，其中一个最小值为ML解s^ML的欧氏距离

d (s^{ML}) = {| | \overset{\cdot}{y} - {Rs}^{ML} | |}^{2},

其中In formula (5), L(b _{k, m} ) represents the LLR value of the mth bit of the transmitted vector symbol s _k ,

Indicates the set of values of the mth bit of the symbol s _k in the vector s ±1. From the observation of formula (5), it can be obtained that one of the minimum values is the Euclidean distance of ML solution s ^ML

d ({the s}^{ML}) = {| | \overset{&Center Dot;}{the y} - {Rs.}^{ML} | |}^{2},

in

${s the s}^{ML ML} = = \underset{s the s &Element; &Element; {Ω Ω}^{{N N}_{T T}}}{arg arg min min} {| | | | \overset{\cdot &Center Dot;}{y the y} - - Rs Rs. | | | |}^{22} - - - - - - ((66))$

对应的另一个最小值为：Another corresponding minimum value is:

$(({s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}})) = = \underset{{s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}}}{min min} {| | | | \overset{\cdot \cdot}{y the y} - - Rs Rs. | | | |}^{22} - - - - - - ((77))$

为发射向量符号s_k第m个比特的极性是s^ML的相反项。所以式(2)可以改写为：

The polarity of the mth bit of the transmit vector symbol s _k is the opposite term of s ^ML . So formula (2) can be rewritten as:

$L L (({b b}_{k k,, m m})) = = \{\begin{matrix} d d (({s the s}^{ML ML})) - - d d (({s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}})),, {s the s}_{k k,, m m}^{ML ML} = = - - 11 \\ d d (({s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}})) - - d d (({s the s}^{ML ML})),, {s the s}_{k k,, m m}^{ML ML} = = + + 11 \end{matrix} - - - - - - ((88))$

设集合

为符合发射向量符号s_k第m个比特的极性是s^ML的相反项条件的集合，则MIMO最大似然检测和max-log LLR值计算过程可以等价位一棵代价树的搜索过程，即在集合

和集合

中搜索具有最小欧氏距离的叶子节点，得到d(s^ML)和N_T×Q个

值。Set up

In order to meet the set of conditions that the polarity of the mth bit of the transmitted vector symbol s _k is the opposite term of s ^ML , the MIMO maximum likelihood detection and max-log LLR value calculation process can be equivalent to the search process of a cost tree, namely in collection

and collection

Search for the leaf node with the minimum Euclidean distance, get d(s ^ML ) and N _T ×Q

value.

本发明实施例采用了基于度量优先的Dijkstra软输出球形译码检测方法，如图7所示，包括如下步骤：The embodiment of the present invention adopts the Dijkstra soft output spherical decoding detection method based on metric priority, as shown in Figure 7, comprising the following steps:

步骤201：对信道矩阵H进行QR分解，得到Q矩阵和上三角矩阵R。Step 201: Perform QR decomposition on the channel matrix H to obtain a Q matrix and an upper triangular matrix R.

步骤202：将Q矩阵的共轭转置与接收信号向量y相乘，得到接收信号的均衡信号即

通过对等效矩阵R和均衡信号构建的搜索树来进行软输出球形译码。Step 202: Multiply the conjugate transpose of the Q matrix by the received signal vector y to obtain the equalized signal of the received signal Right now

By pairing the equivalent matrix R and equalizing the signal Constructed search tree for soft output sphere decoding.

步骤203：建立查找表LUT，设置球形译码的半径R₀＝∞，可用的存储空间大小为M。Step 203: Create a lookup table LUT, set the radius R ₀ =∞ of spherical decoding, and the available storage space is M.

具体的查找表LUT建立过程为，假设树搜索到第i层，则部分向量

已知，定义c_i为第i层的搜索中心The specific process of establishing the lookup table LUT is as follows. Assuming that the tree searches to the i-th layer, the partial vector

Known, define c _i as the search center of the i-th layer

${c c}_{i i} = = \frac{11}{{r r}_{ii i}} (({\overset{\cdot \cdot}{y the y}}_{i i} - - {Σ Σ}_{n no = = i i + + 11}^{{N N}_{R R}} {r r}_{in in} {s the s}_{n no})) - - - - - - ((99))$

则式(9)改写为：Then formula (9) can be rewritten as:

${d d}_{i i} = = {d d}_{i i + + 11} + + {r r}_{}^{ii i} {| | {c c}_{i i} - - {s the s}_{i i} | |}^{22} - - - - - - ((1010))$

节点首先访问距离搜索中心c_i最近的星座点s_i，然后按照与c_i由近到远的次序进行排序搜索访问。当MIMO系统的调制方式确定，调制符号集即确定，因而可以根据c_i所属位置区域对星座点的扩展顺序进行排列，建立查找表LUT。为降低查找表的存储空间，本发明将每个星座点的区域分成4块，按照空间的距离由近到远对星座点的访问次序进行排列，如图1所示，以16QAM调制为例，当搜索中心c_i落入斜线阴影区域时，16个星座点按照先后顺序和阴影区域的距离排序，其中，该顺序可以根据实际需要进行变化，这并不影响本发明的实质内容，例如图1(a)中，星座点“9”和“10”的顺序就有可能调换，然后仿真结果表明，这种近似处理对性能几乎没有影响。为了覆盖QAM调制中所有可能的星座点，4-QAM需要的查找表大小为16，16-QAM需要的查找表大小为64，64-QAM需要的查找表大小为256。The node first visits the constellation point _si closest to the search center _ci , and then searches and visits according to the order from closest to far away from _ci . When the modulation mode of the MIMO system is determined, the modulation symbol set is determined, so the expansion sequence of the constellation points can be arranged according to the location area to which c _i belongs, and a look-up table LUT can be established. In order to reduce the storage space of the lookup table, the present invention divides the area of each constellation point into 4 blocks, and arranges the access order of the constellation points according to the spatial distance from near to far, as shown in Figure 1, taking 16QAM modulation as an example, When the search center _ci falls into the shaded area of the oblique line, the 16 constellation points are sorted according to the sequence and the distance of the shaded area, wherein, the order can be changed according to actual needs, which does not affect the essence of the present invention, for example Fig. In 1(a), the order of constellation points "9" and "10" may be exchanged, and the simulation results show that this approximation has almost no impact on performance. In order to cover all possible constellation points in QAM modulation, 4-QAM requires a lookup table size of 16, 16-QAM requires a lookup table size of 64, and 64-QAM requires a lookup table size of 256.

步骤204：计算

根据

及调制方式得到查找表的下标p，得到初始点

设存储节点将

N_{c} = (s = s^{(N_{T})}, d (s), p, q = 0, i = N_{R}),

M＝M-1。Step 204: Calculate

according to

and the modulation method to get the subscript p of the lookup table, and get the initial point

Let the storage node be

N_{c} = (the s = {the s}^{(N_{T})}, d (the s), p, q = 0, i = N_{R}),

M=M-1.

步骤205：从栈中移除节点N_c＝(s＝s⁽ⁱ⁾，d(s)，p，q，i)，M＝M+1，扩展第i层与节点N_c最近的节点，令s_f＝(LUT(p，q+1)，s⁽ⁱ⁺¹⁾)，计算d(s_f)根据当前向量s_f更新球形译码的半径R₀。如果d(s_f)＜R₀，将N_f＝(s＝s_f，d(s_f)，p，q＝q+1，i)存入栈中，M＝M-1；否则转到步骤206。Step 205: remove node N _c =(s=s ⁽ⁱ⁾ , d(s), p, q, i) from the stack, M=M+1, expand the i-th layer closest node to node N _c , Let s _f =(LUT(p,q+1), s ⁽ⁱ⁺¹⁾ ), calculate d(s _f ) to update the radius R ₀ of spherical decoding according to the current vector s _f . If d(s _f )<R ₀ , store N _f =(s=s _f , d(s _f ), p, q=q+1, i) into the stack, M=M-1; otherwise go to Step 206.

本步骤的球形译码的半径更新方法为：The radius update method of the spherical decoding in this step is:

当搜索树得到部分解向量

时，所以球形译码的半径更新分为两部分When the search tree gets a partial solution vector

, so the radius update of spherical decoding is divided into two parts

(1)对于已知部分向量，即k≥i，m＝1，2，…，Q，半径R₀与

的

相关，即(1) For known partial vectors, namely k≥i, m=1, 2, ..., Q, radius R ₀ and

of

related, namely

${R R}_{00} = = max max {{d d (({s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}})) | | &ForAll; &ForAll; k k &GreaterEqual; &Greater Equal; i i,, m m = = 1,2 1,2,, . . . . . .,, Q Q,, {s the s}_{k k,, m m} = = {s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}}}}$

(2)对于还没搜索的部分向量，即k＜i，m＝1，2，…，Q，半径R₀与所有的

相关，即(2) For the partial vectors that have not been searched, that is, k<i, m=1, 2,..., Q, the radius R ₀ and all

related, namely

${R R}_{00} = = max max {{d d (({s the s}_{k k,, m m}^{\overset{&OverBar; &OverBar;}{ML ML}})) | | &ForAll; &ForAll; k k < < i i,, m m = = 1,2 1,2,, . . . . . .,, Q Q}}$

步骤206：如果节点N_c为叶子节点(i＝1)，更新软值表

和s^ML；否则向下一层扩展，通过式(10)计算搜索中心c_i-1，得到查找表下标p，令s_exp＝(LUT(p，0)，s⁽ⁱ⁾)，计算d(s_exp)，根据s_exp更新半径R₀(半径更新方法与步骤205一致)。如果d(s_exp)＜R₀，将N_exp＝(s＝s_exp，d(s_exp)，p，q＝0，i-1)存入栈中，M＝M-1；否则转到步骤207。Step 206: if the node N _c is a leaf node (i=1), update the soft value table

and s ^ML ; otherwise, expand to the next layer, calculate the search center c _i-1 through formula (10), and get the subscript p of the lookup table, let s _exp =(LUT(p, 0), s ⁽ⁱ⁾ ), calculate d(s _exp ), update the radius R ₀ according to s _exp (the radius update method is consistent with step 205). If d(s _exp )<R ₀ , store N _exp =(s=s _exp , d(s _exp ), p, q=0, i-1) in the stack, M=M-1; otherwise go to Step 207.

本步骤中的更新软值表的过程为：The process of updating the soft value table in this step is:

初始化

当树搜索到达叶子节点时即得到新的解向量s⁽¹⁾才进行软值表更新，分两种情况：initialization

When the tree search reaches the leaf node, the new solution vector s ⁽¹⁾ is obtained before the soft value table is updated. There are two cases:

(1)当得到一个新的ML解时，即d(s)＜d(s^ML)，则将所有符合

比特位的

更新为

并且更新s^ML＝s和d(s^ML)＝d(s)。这样保证了在当前在所有与s^ML相反的比特位的

都为当前最小值。(1) When a new ML solution is obtained, that is, d(s)<d(s ^ML ), then all

Bits

update to

And update s ^ML =s and d(s ^ML )=d(s). This ensures that at the current time in all bits opposite to s ^ML

are the current minimum values.

(2)若d(s)＞d(s^ML)，此时只需更新

的值，当

且

时，跟新

d (s_{k, m}^{\overset{&OverBar;}{ML}}) = d (s) .

(2) If d(s)＞d(s ^ML ), only update

value, when

and

when, update

d ({the s}_{k, m}^{\overset{&OverBar;}{ML}}) = d (the s) .

步骤207：令n_max＝min(M+2，N_T)，将栈中的节点根据d(s)进行从小到大排序，选择具有最小的d(s)的节点且满足层数i＜n_max，设选中的节点为N_c。如果当前栈不为空即M≠0，，转到步骤3；否则转到步骤205。Step 207: Let n _max = min(M+2, N _T ), sort the nodes in the stack according to d(s) from small to large, select the node with the smallest d(s) and satisfy the level i<n _max , let the selected node be N _c . If the current stack is not empty, ie M≠0, go to step 3; otherwise go to step 205.

步骤208：根据式(5)计算每一个比特的LLR值L(b_k，m)，输出结果。Step 208: Calculate the LLR value L(b _k,m ) of each bit according to formula (5), and output the result.

图2、图3、图4、图5为本发明实施例的球形译码算法与传统的几种球形译码算法性能和复杂度的比较。Fig. 2, Fig. 3, Fig. 4 and Fig. 5 are comparisons of the performance and complexity of the sphere decoding algorithm of the embodiment of the present invention and several traditional sphere decoding algorithms.

仿真参数如下：信道采用郊区宏小区SCM信道，载频为2GHz，带宽为3MHz，信道编码采用turbo编码，调制方式为QPSK和16QAM，码率为378/1024，发射天线数：4，接收天线数：4。The simulation parameters are as follows: the channel adopts the SCM channel of the suburban macro cell, the carrier frequency is 2GHz, the bandwidth is 3MHz, the channel coding adopts turbo coding, the modulation method is QPSK and 16QAM, the code rate is 378/1024, the number of transmitting antennas: 4, the number of receiving antennas :4.

图2、图3给出了不同方法下软判决的误码率曲线比较，本发明提出的译码方法与STS球形译码方法法的性能几乎一致，相比于K-BEST列表球形译码方法，性能有较大的提升，并且在高阶调制16QAM下，性能提升越明显。Fig. 2, Fig. 3 have provided the bit error rate curve comparison of soft decision under different methods, and the performance of decoding method proposed by the present invention and STS sphere decoding method method are almost identical, compared with K-BEST list sphere decoding method , the performance is greatly improved, and the performance improvement is more obvious under the high-order modulation 16QAM.

图4、图5给出了不同接收机检测方法的复杂度的比较，复杂度的比较是通过搜索过程中搜索节点的数目比较，由图4、图5可见，本发明提出的译码方法搜索的节点数目与STS-SD译码方法的下降趋势类似，但较STS-SD译码方法，搜索空间即访问过的节点数减小，并且访问过的节点数的变化幅度与STS-SD译码方法相比较小，利于估计系统的硬件实现开销，译码时延和吞吐率。与利于硬件实现的K-BSET列表球形译码相比，本发明提出的球形译码方法在低阶调制(QPSK)下，性能较K-BSET(64)球形译码方法更好，且搜索节点数更少。在高阶调制(16QAM)下，性能远好于K-BSET(16)球形译码方法，且搜索节点数增长幅度很小，在硬件实现和性能上有较好的折中。Fig. 4, Fig. 5 have provided the comparison of the complexity of different receiver detection methods, the comparison of complexity is through the number comparison of search nodes in the search process, as seen from Fig. 4, Fig. 5, the decoding method that the present invention proposes searches The decrease trend of the number of nodes is similar to that of the STS-SD decoding method, but compared with the STS-SD decoding method, the search space, that is, the number of visited nodes is reduced, and the change range of the number of visited nodes is similar to that of the STS-SD decoding method The method is relatively small, which is beneficial to estimate the hardware implementation cost, decoding delay and throughput of the system. Compared with the K-BSET list sphere decoding that is beneficial to hardware implementation, the sphere decoding method proposed by the present invention has better performance than the K-BSET (64) sphere decoding method under low-order modulation (QPSK), and the search node Fewer. Under high-order modulation (16QAM), the performance is much better than the K-BSET (16) spherical decoding method, and the increase in the number of search nodes is small, and there is a good compromise between hardware implementation and performance.

从图2、图3、图4、图5对比说明，本发明实例在保持高性能的前提下，有效的降低了系统的复杂度。From the comparison of Fig. 2, Fig. 3, Fig. 4 and Fig. 5, it is shown that the example of the present invention effectively reduces the complexity of the system under the premise of maintaining high performance.

Claims

1. the receiver detection method of a wireless multiple-input-multiple-output systems is characterized in that, comprising:

Step 1: channel matrix H is carried out QR decompose; Obtain Q matrix and upper triangular matrix R; The conjugate transpose and the received signal vector y of Q matrix are multiplied each other, obtain receiving the equalizing signal of signal

Step 2: set up the look-up table LUT of node expansion order, the radius of globular decoding is set, available storage size is M;

Step 3: calculate search center, obtain the subscript of look-up table LUT, initially chosen node, be pressed in the stack according to search center and modulation system;

Step 4: from stack, remove and choose node; And, upgrade the radius of globular decoding according to the expansion brotgher of node of choosing node, whether the radius of judging present node is less than the current search radius; Be then to cut down present node and all branches thereof, otherwise the brotgher of node is pressed in the stack;

Step 5: judge and choose whether node is leaf node, if leaf node then carries out the maintenance of soft value table; Otherwise downward one deck expanded search; Calculate the subscript that search center and modulation system obtain look-up table LUT, obtain child node, whether the radius of judging present node is less than the current search radius; Be then to cut down present node and all branches thereof, otherwise child node is pressed in the stack;

Step 6: judge that whether current stack is empty, if be empty, then arrives step 7; Otherwise according in the stack memory space and the weights node of selecting next iteration to choose, return step 5;

Step 7: carry out the LLR value based on soft value table and calculate.

2. the receiver detection method of wireless multiple-input-multiple-output systems according to claim 1 is characterized in that,

In said step 2, the look-up table establishment step is:

Hypothesis tree searches the i layer, then the part vector

Known, definition c _iIt is the search center of i layer

c_{i} = \frac{1}{r_{Ii}} ({\overset{\cdot}{y}}_{i} - Σ_{n = i + 1}^{N_{R}} r_{In} s_{n})

Formula one

Then said formula one is rewritten as:

d_{i} = d_{i + 1} + r_{Ii}^{2} {| c_{i} - s_{i} |}^{2}

Formula two

Node is at first visited range search center c _iNearest constellation point s _i, then according to c _iOrder from the near to the remote carries out the sorted search visit; Confirm that when the modulation system of mimo system the modulation symbol collection is promptly definite, thereby can be according to c _iArrange the expansion of constellation point in proper order in the belonging positions zone, sets up look-up table LUT; Wherein, The zone of each constellation point is divided into 4, from the near to the remote the visit order of constellation point is arranged according to the distance in space, the look-up table size that 4-QAM needs is 16; The look-up table size that 16-QAM needs is 64, and the look-up table size that 64-QAM needs is 256.

3. the receiver detection method of wireless multiple-input-multiple-output systems according to claim 1 is characterized in that,

In said step 3 and step 5, search center computational process is:

Hypothesis tree searches the i layer, then the part vector

Known, definition c _iIt is the search center of i layer

c_{i} = \frac{1}{r_{ii}} ({\overset{\cdot}{y}}_{i} - Σ_{n = i + 1}^{N_{R}} r_{in} s_{n}) .

4. the receiver detection method of wireless many defeated people's multiple output systems according to claim 1 is characterized in that,

In said step 3 and step 4, the node expansion process is:

Suppose that it is N that current stack is wanted the node of downward one deck expansion _c=(s=s ⁽ⁱ⁾, d (s), p, q; I), remaining space is M in the stack, and wherein s is current solution vector, and i is the number of plies that present node is positioned at; P and q are respectively in the subscript of current selected look-up table and the present node subscript in the constellation point of i layer, and wherein, the node extended method comprises following substep:

1) from stack, removes node N _c=(s=s ⁽ⁱ⁾, d (s), p, q, level=i), M=M+1 expands level=i layer and the nearest node of node N, even s _f=(LUT (p, q+1), s ⁽ⁱ⁺¹⁾), calculate d (s _f), if d is (s _f)＜R ₀With node N _f=(s=s _f, d (s _f), p, q=q+1 level=i) deposits in the stack M=M-1 in;

2) if node N _cBe leaf node when being i=1, carry out the LLR value and upgrade with ML and separate renewal, otherwise downward one deck is expanded, through following formula

d_{i} = d_{i + 1} + r_{ii}^{2} {| c_{i} - s_{i} |}^{2}

Calculate search center c _I-1, obtain look-up table subscript p, make s _Exp=(LUT (p, 0), s ⁽ⁱ⁾), calculate d (s _Exp), if d is (s _Exp)＜R ₀With node N _Exp=(s=s _Exp, d (s _Exp), p, q=0 i-1) deposits in the stack M=M-1 in.

5. the receiver detection method of wireless multiple-input-multiple-output systems according to claim 1 is characterized in that, in said step 3 and step 4, the radius renewal process is:

Radius upgrades and is divided into two parts:

1) for known portions vector, i.e. k>=i, m=1,2 ..., Q, radius R ₀With

Relevant, promptly

R_{0} = Max {d (s_{k, m}^{\overset{&OverBar;}{ML}}) | &ForAll; k &GreaterEqual; i, m = 1,2, . . ., Q, s_{k, m} = s_{k, m}^{\overset{&OverBar;}{ML}}};

2) for the part vector that does not also have search, i.e. k＜i, m=1,2 ..., Q, radius R ₀With all

Relevant, promptly

R_{0} = Max {d (s_{k, m}^{\overset{&OverBar;}{ML}}) | &ForAll; k < i, m = 1,2, . . ., Q} .

6. the receiver detection method of wireless multiple-input-multiple-output systems according to claim 1 is characterized in that,

In said step 4, soft value table renewal process is:

Be initialized as

d (s^{ML}) = d (s_{k, m}^{\overset{&OverBar;}{ML}}) = \infty (&ForAll; k, m),

Divide two kinds of situation:

1) when obtaining a new ML and separate, i.e. d (s)＜d (s ^ML), then all are met

Bit Be updated to

And upgrade s ^ML=s and d (s ^ML)=d (s);

2) if d (s)＞d (s ^ML), only need this moment to upgrade

Value, when

And The time, upgrade

d (s_{k, m}^{\overset{&OverBar;}{ML}}) = d (s) .

7. the receiver detection method of wireless multiple-input-multiple-output systems according to claim 1 is characterized in that,

In said step 6, node selecting method is:

Make n _Max=min (M+2, N _T), the node in the stack is sorted according to d (s) from small to large, select to have minimum d (s) and satisfy i＜n _MaxNode, establishing the node of choosing is N _c

8. the receiver detection method of wireless multiple-input-multiple-output systems according to claim 1 is characterized in that,

In said step 7, the LLR value calculating method is:

For

&ForAll; k &Element; {1, . . ., N_{T}},

&ForAll; m &Element; {1, . . ., Q},

Utilize computes LLR value

L (b_{k, m}) = \{\begin{matrix} d (s^{ML}) - d (s_{k, m}^{\overset{&OverBar;}{ML}}), s_{k, m}^{ML} = - 1 \\ d (s_{k, m}^{\overset{&OverBar;}{ML}}) - d (s^{ML}), s_{k, m}^{ML} = + 1 \end{matrix} .