JP2009182964A - Repeated signal detection method for mimo (multiple-input, multiple-output) system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a repeated signal detection method for an MIMO system. <P>SOLUTION: The method includes an arithmetic step of searching an antenna with the highest receiving signal reliability from a plurality of antenna candidates in the k-th detection layer by the use of an arithmetic parameter computed based on maximum posterior probability algorithm, and storing, with respect to the searched antenna, S paths of paths having the maximum posterior probability as survival paths, in which when k is 1, the arithmetic parameter is an initialized arithmetic parameter, and when k is larger than 1, the arithmetic parameter is an arithmetic parameter updated according to the detection result of the k-1-th detection layer; an antenna deletion step of updating the arithmetic parameter according to the arithmetic result, and deleting the antenna with the highest receiving signal reliability from the antenna candidates; and a repeating step of repeating both the steps until all the antenna candidates are deleted. The first survival path having the maximum posterior probability in the final detection layer is taken as a final detection result. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a wireless communication technology field, and more particularly to a communication technology of a MIMO (Multiple Input Multiple Output) system, and more particularly, to a repetitive signal detection method of a MIMO (Multiple Input Multiple Output) system.

  The contradiction between an increase in data rate and limited radio resources becomes more severe every day, and may become a bottleneck for performing radio communication in a radio fading channel in the future.

  Since MIMO (multiple input multiple output) channels have a larger capacity than SISO (single input single output) channels, wireless communication using MIMO channels has received much attention over the last decade (GJ Foschini, “Layered”). “space.time-for-architecture-for-wireless-communication-in-fading-environments-where-using-multiple-antennas”, BellLabs.Tech.J., vol.2, pp.41-59, G. Signal Processing: Principles, Algorithms, and Applicat ons 2: ed ", Macmillan Publishing Company, 1992). A MIMO system is a wireless communication system that transmits data via multiple transmission antennas and multiple reception antennas, and can provide a high data rate, a large throughput, and a wider communication distance. MIMO channels have enormous potential because they can provide large data rates and reliability with no cost in the frequency spectrum. MIMO has already become a hot technology for 4th generation (4G) communication and is a key part of most new wireless standards (eg HSDPA, 802.11n, 802.16e, 802.10, etc.). In MIMO systems, the blast (BLAST) structure has received much attention because it can obtain the highest multiple gain.

  Conventional blast (BLAST) detection methods include zero forcing (ZF) detection, minimum mean square error (MMSE) detection, serial interference cancellation (SIC) detection, maximum likelihood (ML) detection, and maximum posterior probability (MAP) detection. Etc. are included. Here, maximum posterior probability (MAP) detection and maximum likelihood (ML) detection are optimal detection algorithms. Under the condition that the transmission source transmits prior information such as prior probability, both of these two detection methods can obtain optimum performance.

  However, the complexity of these two types of detection methods is both the number of transmitting antennas and exponential power, and cannot meet the requirements for wireless communication with high real-time characteristics. For example, for the maximum posterior probability MAP detection method, a greedy search algorithm is used in the code vector tree diagram, leaving all possible paths in the tree structure, the complexity of which is the exponential power of the number of transmit antennas.

For example, if the number of transmitting antennas is 4 and QPSK modulation (the number of transmitted codes is 4) is adopted, the fourth layer transmitting antenna has four nodes, and the third layer transmitting antenna has 4 × nodes. 4 = 16, the second layer transmitting antenna has 16 × 4 = 64 nodes, and the first layer transmitting antenna has 64 × 4 = 256 nodes. It is necessary to search all 4 ⁴ = 256 paths and calculate the vector posterior probability: P (s ^j | y) of all possible 256 paths. j is a natural number in the interval [1,256]. Then, the maximum value is selected from the 256 vector posterior probabilities, and the corresponding vector is an estimated value (hard decision) of the transmission code vector.

On the other hand, if the number of transmission antennas is 4 and 16QAM modulation (the number of codes transmitted is 16) is adopted, the transmission antenna of the fourth layer has 16 nodes, and the transmission antenna of the third layer has 16 × 16 nodes. = 256, the second layer transmission antenna has 256 × 16 = 4096 nodes, and the first layer transmission antenna has 4096 × 4 = 65536 nodes. The 16QAM modulation needs to search 65536 paths and calculate the vector posterior probability: P (s ^j | y) of all possible 65536 paths. j is a natural number in the interval [1, 65536]. Then, the maximum value is selected from the 65536 vector posterior probabilities, and the corresponding vector is an estimated value (hard decision) of the transmission code vector.

  Since such an operation has a high degree of complexity and is not tolerated by the communication system, the following optimal algorithm is usually used in an actual communication system.

  The next optimal detection methods (eg, ZF detection, MMSE detection, SIC detection, etc.) are less complex than the ML algorithm, but their performance is clearly inferior to ML detection.

  In the next optimal detection method, serial interference cancellation (SIC) is a decision feedback technique and Foschini adopts the technique in V-BLAST system (GJ Foschini, “Layered space-time architecture for wireless communication). in fading envi- ronments when using multiple antennas ", Bell Labs. Tech. J., vol. 2, pp. 41-59, 1996). The data stream is separated into N unrelated substreams, each substream is transmitted by one of N transmit antennas, and in each iteration, the receiver detects one substream (antenna). Thus, the detection complexity is simplified by subtracting interference from the received signal.

を検出する。受信信号ｒ⁽¹⁾＝ｒから

ここで、

は、チャネル行列Ｈ⁽¹⁾のａ₁番目列である。システムモデル（１）は（Ｎ−１）×ＭのＭＩＭＯと簡素化される：ｒ⁽²⁾＝Ｈ⁽²⁾ｘ⁽²⁾＋ｎ
ここで、新しいチャネル行列Ｈ⁽²⁾はＨ⁽¹⁾からａ₁番目列を引いて得られるものである。ｘ⁽²⁾はＮ−１項を備える列ベクトルである。 Assume that each element _ak represents an antenna number according to the following antenna order O = (a ₁ ,..., A _N ) during detection. In the first detection layer, the first substream (antenna) a ₁ and the code corresponding thereto through a substream detection method such as a detection method such as ZF or MMSE

Is detected. From received signal r ⁽¹⁾ = r

here,

Is the a ₁ th column of the channel matrix H ⁽¹⁾ . The system model (1) is simplified to (N-1) × M MIMO: r ⁽²⁾ = H ⁽²⁾ x ⁽²⁾ + n
Here, the new channel matrix H ⁽²⁾ is obtained by subtracting the a _1st column from H ⁽¹⁾ . x ⁽²⁾ is a column vector with N-1 terms.

を一つだけ残し、ツリー検索から見れば、ノード（符号）序列

を生き残り経路と言う。ここで、１≦ｋ≦Ｎ。 Similarly, the detector of the receiver can delete an antenna for each book with linear complexity. In fact, the SIC algorithm is a special case tree search algorithm that grows exponentially according to the number of transmit antennas. The SIC detection algorithm is a node (code) for each layer.

If you look at the tree search, only one node is left,

Is called the survival route. Here, 1 ≦ k ≦ N.

  It is necessary to explain that the performance of SIC depends on 1) the order of antenna deletion; and 2) the performance of the detection method for each layer (antenna). Since SIC only considers signal interference to noise ratio (SINR), it leads to error transmission by SIC, especially when channel state information (CSI) is not ideal. Due to error transmission, non-optimal ranking or substream detection methods degrade performance.

  Since the deletion order of SIC antennas has a profound influence on detection performance, for example, H-norm order (channel gain order by antenna) and DiaGR order (pairs of QR factorization of channel matrix H) can be improved. Several different ranking methods have been submitted, such as ranking by angular factor), Hinv ranking (ranking by antenna SINR), etc. (Yongmei Dai, Sumei Sun, and Zhongding Lei. “A Comprehensive Study of QRD-M Detection and Sphere Decoding for MIMO-OFDM Systems ", Personal, Indoor and Mobile Radio Communications, 2005. PIMRC 2005. EEE 16th International Symposium on, 2005). All of these ranks consider only the influence of the channel matrix H, and do not use as much information as possible of received signal information and constellation shape. However, the shape of the constellation also affects the accuracy of the ranking.

  For example, BPSK has better detection performance than 16QAM or 8PSK. In short, a received signal having a small minimum Euclidean distance is more constellation reliable than a received signal having a large minimum Euclidean distance with respect to the constellation point. All of these factors that determine reliability can be uniformly reflected in the maximum a posteriori probability (A) of the received signal.

ここで、

はアンテナａ_ｋのＭＭＳＥポスト均等化（ｐｏｓｔ‐ｅｑｕａｌｉｚａｔｉｏｎ）信号対雑音比（ＳＮＲ）で、Ｉ_ｋは瞬時的な信頼度因数（ＩＦＲ）で、無偏ユークリッド検出の関数であり、該関数は受信信号ｒの信頼性の情報を利用した（Ｊ．Ｍ．Ｃｉｏｆｆｉ， ｅｔ ａｌ. “ＭＭＳＥ ｄｅｃｉｓｉｏｎ‐ｆｅｅｄｂａｃｋ ｅｑｕａｌｉｚｅｒｓ ａｎｄ ｃｏｄｉｎｇ. I. Ｅｑｕａｌｉｚａｔｉｏｎ ｒｅｓｕｌｔｓ”, Ｃｏｍｍｕｎｉｃａｔｉｏｎｓ， ＩＥＥＥ Ｔｒａｎｓａｃｔｉｏｎｓ ｏｎ. ４３．１０ （１９９５）: ２５８２‐２５９４）。 Seethaler provides a ranking method called dynamic-nulling-and-canceling (DNC) (D. Seethaler, Artes, H. and Hlawatsch, F. “Dynamical nulling-and-cancelling with Mirner-Morfir-Mimner-Frequency-Mean-Frequency-Mrf-Nim-FrM systems ", Acoustics, Speech, and Signal Processing, 2004. Proceedings. Volume 4 (2004): Page (s): iv-777-iv-780; D. Setheler, H. arts, W. Arts. -And-Cancelin for Efficient Near-ML Decoding of MIMO Systems ", Signal Processing, IEEE Transactions on 54.12 (2006):. 4741-4752). In this method, not only SINR but also the influence of the received signal is taken into account, and the posterior probability is maximized.

here,

Is the MMSE post-equalization signal-to-noise ratio (SNR) of antenna a _k , I _k is the instantaneous reliability factor (IFR), and is a function of unbiased Euclidean detection, which function is received Information on reliability of signal r (JM Cioffi, et al. “MMSE decision-feedback equalizers and coding. I. Equalization results”, Communications, IEEE Transactions. 2594).

  The DNC technique used for MIMO detection reduces error transmission in the SIC process by maximizing the posterior probability for each detection layer. Since error transmission was greatly reduced by DNC, excellent detection performance was obtained for large multi-input multi-output MIMO such as 8 × 8 MIMO of 4QAM.

  However, as the modulation level increases, the minimum Euclidean distance of the constellation also decreases, but as the minimum Euclidean distance decreases, the influence of the IFR immediately disappears, so the performance of the DNC method is improved. The shape is limited. This method has excellent performance only with a simple modulation constellation such as 4QAM. In some cases, the performance is inferior to other methods.

  Currently, as for MIMO detection, how to balance detection performance and calculation complexity remains a big challenge.

  The main object of the present invention is to provide a method for detecting a repetitive signal in a MIMO (multiple input multiple output) system that maintains the performance of MIMO detection and reduces the computational complexity.

In order to achieve the above object, in a repetitive signal detection method of a MIMO (multiple input multiple output) system according to an embodiment of the present invention, a kth operation is performed using an operation parameter calculated based on a maximum posterior probability (MAP) algorithm. The detection layer searches for the antenna having the highest reliability of the received signal from among a plurality of antenna candidates, and stores S paths as surviving paths from the path having the maximum posterior probability for the searched antenna. Here, when S ≧ 1 and k = 1, the calculation parameter is an initialized calculation parameter, and when k> 1, the calculation parameter is updated according to the detection result of the (k−1) th detection layer. Calculation steps that are calculated parameters,
An antenna deletion step of updating the calculation parameter according to the calculation result and deleting the antenna having the highest reliability of the received signal from the antenna candidates;
repeating as k = k + 1, repeating the calculation step and the antenna deletion step by a recursive algorithm until all antenna candidates are deleted,
One surviving path having the maximum posterior probability in the detection layer of the final layer is set as the final detection result.

  The detection method of the present invention has a performance close to low calculation complexity and maximum likelihood detection.

4 is a method flowchart for implementing MIMO detection based on MAP according to an embodiment of the present invention; It is a schematic diagram which shows the SER performance of the complex number and real number MAP-DOM algorithm of this invention. It is a schematic diagram which shows the relationship between the magnitude | size of a MIMO system, and a SNR threshold value. It is a MAP-DOM algorithm in S = 2 by the present invention, and a non-dynamic ranking performance comparison schematic diagram. It is a MAP-DOM algorithm in S = 4 by this invention, and a performance comparison schematic diagram of a non-dynamic order.

  In order to further clarify the objects, technical solutions, and merits of the present invention, embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Here, the schematic embodiment of the present invention and the description thereof are to interpret the present invention, and are not limited to the present invention.

  In order to solve the contradiction between the performance and the computational complexity in the conventional MIMO detection method, the present invention provides a new SIC detection method based on the dynamic rank of the maximum posterior probability and introduces the dynamic rank algorithm. Prevent performance degradation and reduce computational complexity through recursive algorithms. The present invention further enhances the detection function by the dynamic order of multiple surviving paths. For this reason, the detection method of the present invention has low computational complexity and performance that approximates maximum likelihood detection.

  First, for the convenience of description, main symbols appearing later will be described.

は、Ｏ＝（ａ₁,・・・,ａ_k）の順位に従ってｋ本のアンテナ（ａ₁乃至ａ_k）の受信信号が削除されたＳＩＣ検出結果である；
（ｆ₁，・・・,ｆ_n）＝ＳＣＧ^[n]Ｆは、集合Ｆからｎ個の最大値を並べて選択することを示す。ここで、ｆ_i∈Ｆ、ｆ₁≧ｆ₂≧・・・≧Ｆにおける任意の他方の値；
（ｆ₁，・・・,ｆ_n）＝ＳＣＬ^[n]Ｆは、集合Ｆからｎ個の最小値を並べて選択することを示す。ここで、ｆ_i∈Ｆ、ｆ₁≦ｆ₂≦・・・≦Ｆにおける任意の他方の値；
（・）_kはベクトル（・）のｋ番目の項；
（・）_i,jは行列（・）のｉ番目行ｊ番目列の要素；
（・）^Tは行列又はベクトルの転置；

は行列又はベクトルのエルミート共役転置（ｈｅｒｍｉｔｉａｎ ｔｒａｎｓｐｏｓｉｔｉｏｎ）；

は、構造した一つの新しいベクトルの集合である。Ａが空であると、

（１）本発明の検出方法の基本原理
本発明をより明確に理解するために、以下に本発明の最大事後確率による新しい検出方法の基本原理を説明する。 N is the number of transmit antennas;
M is the number of receiving antennas;
S is the survival route;
H is an M × N channel matrix, where h _i is the i th column of the matrix H and corresponds to the i th transmit antenna;
MIMO received signal vector where r is M × 1;
x is an N × 1 MIMO transmit signal vector;
n is an M × 1 noise vector;
C _k represents a set of constellation points of antenna k, 1 ≦ k ≦ N;

Is the SIC detection result in which the received signals of the _k antennas (a _{1 to} a _k ) are deleted according to the order of O = (a ₁ ,..., A _k );
(F ₁ ,..., F _n ) = SCG ^[n] F indicates that n maximum values are selected from the set F side by side. Where f _i εF, any other value in f ₁ ≧ f ₂ ≧... ≧ F;
(F ₁ ,..., F _n ) = SCL ^[n] F indicates that n minimum values are selected from the set F side by side. Where f _i εF, any other value in f ₁ ≦ f ₂ ≦... ≦ F;
(•) _k is the k-th term of the vector (•);
(·) _{I, j} is the element in the i-th row and j-th column of the matrix (·);
(·) ^T is a matrix or vector transposition;

Is a Hermitian transposition of a matrix or vector;

Is a new set of structured vectors. If A is empty,

(1) Basic Principle of Detection Method of the Present Invention In order to understand the present invention more clearly, the basic principle of a new detection method based on the maximum posterior probability of the present invention will be described below.

ここで、ｘはゼロ平均を有し、共分散行列

で、Ｉは単位行列である。ｎはＭ×１の複素数ガウスホワイト雑音（ＡＷＧＮ）ベクトルでゼロ平均、円形対称複素数ガウス分布を有し、ｎ＝（ｎ_{1，・・・，}ｎ_M）^T、共分散

（σ²は雑音電力）、且つｎとｘは相関していない。ＨはＭ×Ｎのチャネル行列で、ｈ_iは行列Ｈのｉ番目列で、ｉ番目の送信アンテナに対応する。 The present invention is for a MIMO system model such as a linear MIMO system with N transmit antennas and M receive antennas. The transmission signal vector x is x = (x _1,..., X _N ) ^T , the reception signal vector r is r = (r _1,..., R _M ) ^T , and (•) ^T is a vector Where M ≧ N and r and x have the following relationship:

Where x has zero mean and covariance matrix

Where I is a unit matrix. n is an M × 1 complex Gaussian white noise (AWGN) vector having zero mean, circularly symmetric complex Gaussian distribution, n = (n _1,..., n _M ) ^T , covariance

(Σ ² is noise power), and n and x are not correlated. H is an M × N channel matrix, h _i is the i-th column of the matrix H, and corresponds to the i-th transmitting antenna.

ここで、Ｃ_kは第ｋ本のアンテナのコンスタレーションポイントの集合で、１≦ｋ≦Ｎ；
集合

は、全てのあり得る送信符号の集合を含む。 For a linear MIMO system (1), the goal to detect is to search for a vector with the maximum posterior probability (APP):

Here, C _k is a set of constellation points of the k th antennas, and 1 ≦ k ≦ N;
set

Contains the set of all possible transmission codes.

を羅列的に探索するものである。実践において、リアルタイムなシステムに用いられてはいけない。特に、ＭＩＭＯが非常に大きい場合、例えば、４×４のＭＩＭＯ設定と１６ＱＡＭの場合、送信する符号の数は

である。 The direct path is the exponential computational complexity,

Are enumerated. In practice, it should not be used for real-time systems. In particular, when the MIMO is very large, for example, in the case of 4 × 4 MIMO setting and 16QAM, the number of codes to be transmitted is

It is.

の大きさをＳまで減少し、ここで、Ｓ≧１で、このＳ個のベルトルを生残り経路と言う。 As one of the basic ideas based on the MAP algorithm of the present invention,

Is reduced to S, where S ≧ 1, and these S beltles are called survival paths.

がＳ個の生残り経路の集合で、Ｐ（ｘ｜ｒ）が事後確率であるとすると、これらの生残り経路は以下の関係がある：
Ｐ（ｘ^(N,1)｜ｒ）≧Ｐ（ｘ^(N,2)｜ｒ）≧・・・≧Ｐ（ｏｔｈｅｒｓ｜ｒ）；
それに応じて、ＭＡＰの公式（２）は、

即ち、Ｓ個の生残り経路の集合から最大事後確率（ＡＰＰ）を有するベクトルをサーチし、その演算の複雑度は生残り経路と正比例する。 For example, O = (a ₁ ,..., A _N ) is the order of detection, and

Is a set of S surviving paths and P (x | r) is a posterior probability, these surviving paths have the following relationship:
P (x ^{(N, 1)} | r) ≥P (x ^{(N, 2)} | r) ≥ ... ≥P (others | r);
Accordingly, the MAP formula (2) is

That is, a vector having the maximum posterior probability (APP) is searched from a set of S surviving paths, and the complexity of the operation is directly proportional to the surviving paths.

  However, searching for S surviving paths still has exponential computational complexity.

は、ｋ−１番目の層の古い生残り経路

がさらに拡大されて得られるものである。ここで

同様に、ｋ番目の層のＳ本の新しい生残り経路の集合Ｘ^(k)＝｛ｘ^(k,1)，・・・，ｘ^(k,S)｝は、ｋ−１番目の層のＳ本の古い生残り経路の集合Ｘ^(k-1)＝｛ｘ^(k-1,1)，・・・，ｘ^(k-1,S)｝を以下のアルゴリズムで再帰にサーチして得られる。

ここで、ＳＣＧ^[S]は集合

から順にＳ個の最大の値を選択するのを表示する。 Serial interference cancellation (SIC) technology provides a recursive idea to obtain approximate MAP detection performance with approximate linear arithmetic complexity. For the traditional SIC detection method, one new survivor path for the kth layer

Is the old survival path of the k-1st layer

Is obtained by further enlarging. here

Similarly, a set X ^(k) = {x ^{(k, 1)} ,..., X ^{(k, S)} } of ^S new surviving paths in the kth layer is A set of S old surviving paths X ^(k−1) = {x ^(k−1,1) ,..., X ^{(k−1, S)} } is obtained recursively by the following algorithm. It is done.

Where SCG ^[S] is a set

To select S maximum values in order.

と正比例する。このような方式により複本の経路検出は複雑度を大いに低下し、性能がわずかに低下するのに比べて指数演算を避けた。 Computation complexity and number of possible paths to search for S surviving paths for kth layer

Is directly proportional to With this method, multiple path detection greatly reduces the complexity and avoids exponential operations compared to slightly reduced performance.

  SIC error transmission causes severe detection performance degradation. In order to eliminate this effect, the most reliable antenna is first deleted. This requires dynamic ranking for the antennas.

を介してｋ−１本のアンテナが削除されるとすると、まだＮ−ｋ＋１本のアンテナが残されている。このＮ−ｋ＋１本の削除されないアンテナ（アンテナ候補）は、集合

により表示できる。最も信頼性の高いアンテナは最大のＡＰＰを必ず有する：

ここで、

は、最大のＡＰＰを有する送信ベクトルである。こうして、ｋ番目の層について、更新された順位Ｏ^(k)は

で、まだ指数の演算複雑度を有する。 Rank in k-th layer

If k−1 antennas are deleted via, N−k + 1 antennas are still left. The N−k + 1 non-deleted antennas (antenna candidates)

Can be displayed. The most reliable antenna always has the largest APP:

here,

Is the transmission vector with the largest APP. Thus, for the k th layer, the updated rank O ^(k) is

And still have exponential computational complexity.

が挙げられる。こういう場合、最も信頼性の高いアンテナは以下の標準で定義できる：

ここで、

は、アンテナａ_kの次の最適の硬判決（ｈａｒｄ‐ｄｅｃｉｓｉｏｎ）結果である。

をサーチする複雑度が生残り経路の数、コンスタレーションの大きさ｜Ｃ｜と比例するので、動的順位の総の複雑度はＳ×｜Ｃ｜×｜Ａ^(k)｜と正比例する。特にＳ＝１であると、複雑度は｜Ｃ｜×｜Ａ^(k)｜＝｜Ｃ｜×（Ｎ−ｋ＋１）と比例するまで低下する。実践において、動的順位（ＤＬＯ）について、ｓ＝１であると、当該アルゴリズムもより優れた性能に達することができる。 Similarly, as described above, the survivor path S present in the k-1 th layer in the present invention is ^{X (k-1) = {} x (k-1,1), ···, x ( ^{k-1, S)} } and the next optimal method is

Is mentioned. In such cases, the most reliable antenna can be defined by the following standards:

here,

Is the next optimal hard-decision result for antenna _ak .

Is proportional to the number of surviving paths and the size of the constellation | C |, the total complexity of the dynamic ranking is directly proportional to S × | C | × | A ^(k) |. Particularly when S = 1, the complexity decreases until it is proportional to | C | × | A ^(k) | = | C | × (N−k + 1). In practice, for dynamic ranking (DLO), if s = 1, the algorithm can also achieve better performance.

  For a high signal-to-noise ratio (SNR) and a large S, it is highly probable that the next optimal and most reliable antenna is equal to the optimal and most reliable antenna.

と再帰する。 Simply put, the k-th layer detection task combining dynamic ranking and survivor path detection is:

And recursively.

Finally, the detection result of final determination of ^{the first} surviving path x ^{(N, 1 )} in X ^(N) is x _final .

In the above description regarding detection of one or more MAP paths, the number of remaining paths is one fixed value in order to simplify the code. In fact, the number of surviving paths can be set to different values S _k so as to balance the relationship between performance and complexity for different detection layers. According to experience, S ₁ ≧ S ₂ ≧ _... ≧ S _N is set statically. As another method based on a predetermined reference or range, dynamically adjusting the value of S _k.

をサーチするとする。ここで、Ｘ^(k-1)はＳ本の生残り経路の集合である。Ｓ＝１であると、優れた性能に達することができ、低複雑度を有することもできる。 In Formula (6), a set was made to simplify the discussion.

Suppose you search for. Here, X ^(k−1) is a set of S surviving paths. When S = 1, excellent performance can be achieved, and low complexity can be achieved.

  The detection algorithm based on MAP of the present invention is an approximation based on one or more surviving paths. In accordance with high signal-to-noise ratio (SNR) environments and fairly high S-values, the performance of the multiple path method approximates MAP detection.

ここで、ｘ_kはベクトルｘの要素で、

は（ｘ₁,・・・,ｘ_k）が削除された受信信号である。ベクトルｘのＡＰＰを演算することは本発明における肝心なステップである。以下にてまずどのようにスカラー符号ｘ_kのＡＰＰを演算するかについて説明する。 (2) Calculation of APP by Gaussian Distribution Approximation It is necessary to calculate the APP of the vector x in both dynamic ranking and S survivor path detection. Use the Bayes rule recursively:

Where x _k is an element of the vector x,

Is a received signal from which (x ₁ ,..., X _k ) has been deleted. Computing the APP of the vector x is an important step in the present invention. Hereinafter, how to calculate the APP of the scalar code x _k will be described first.

を受信信号ｒのゼロフォーシング（ＺＦ）検出結果とする。Ｈがフルコラムランク行列（ｆｕｌｌ ｃｏｌｕｍｎ ｒａｎｋ ｍａｔｒｉｘ）で、Ｐ（ｘ_k｜ｒ）の事後確率はＰ（ｘ_k｜ｙ_zf）と等しい。事後確率Ｐ（ｘ_k｜ｙ_zf）に対してＢａｙｅｓ規則を使う：

ここで、スカラーｘ_kはアンテナｋの送信符号で、Ｃはコンスタレーションポイントの集合である。 About system model (1)

Is the zero forcing (ZF) detection result of the received signal r. H is a full column rank matrix (full column rank matrix), P | posterior probability of (x _k r) is P | equal to (x _k y _zf). Use the Bayes rule for the posterior probability P (x _k | y _zf ):

Here, the scalar x _k is a transmission code of the antenna k, and C is a set of constellation points.

Below we search for some results based on the distributions f (r | x _k ) and f (r | x _k ).

を有するガウス確率密度関数（ＰＤＦ）であることを容易に証明できる（Ｄ．Ｓｅｅｔｈａｌｅｒ， Ａｒｔｅｓ，Ｈ, ａｎｄ Ｈｌａｗａｔｓｃｈ，Ｆ． “Ｄｙｎａｍｉｃ ｎｕｌｌｉｎｇ‐ａｎｄ‐ｃａｎｃｅｌｌｉｎｇ ｗｉｔｈ ｎｅａｒ‐ＭＬ ｐｅｒｆｏｒｍａｎｃｅ ｆｏｒ ＭＩＭＯ ｃｏｍｍｕｎｉｃａｔｉｏｎ ｓｙｓｔｅｍｓ”, Ａｃｏｕｓｔｉｃｓ， Ｓｐｅｅｃｈ， ａｎｄ Ｓｉｇｎａｌ Ｐｒｏｃｅｓｓｉｎｇ, ２００４. Ｐｒｏｃｅｅｄｉｎｇｓ. Ｖｏｌｕｍｅ ４ （２００４）: Ｐａｇｅ（ｓ）: ｉｖ‐７７７‐iv‐７８０）。ここで、ｅ_kは１つの単位ベクトルで、ｋ番目の要素は１である。σ²Ｉは雑音ベクトルｎの相関係数行列である。 f (y _ZF | x _k ) is covariance with expected value μ _k = x _{k ·} e _k

Can easily be proved to be a Gaussian probability density function (PDF) having the following (D. Seethaler, Artes, H, and Hlavattsch, F. “Dynamic nulling-and-cancelling with near-ML performance for MIMO communicators, MIMO communics,” , Speech, and Signal Processing, 2004. Processedings. Volume 4 (2004): Page (s): iv-777-iv-780). Here, _ek is one unit vector, and the kth element is 1. σ ² I is a correlation coefficient matrix of the noise vector n.

を有する。Ｃ_kに対して逆行列補助定理（ｍａｔｒｉｘ ｉｎｖｅｒｓｉｏｎ ｌｅｍｍａ）により求めると、

と得られる。 W stands for Wiener equalizer.

Have For C _k , the inverse matrix theorem (matrix inversion lemma)

And obtained.

と得られる。 Substituting into f (y _ZF | x _k ) and diffusing,

And obtained.

In the formula (8), φ (x _k | r) = | y _{MMSE, k} / W _{k, k} −x _k | is the unbiased Euclidean distance of the transmitting antenna k. Such a detection method is called unbiased MMSE detection (J. G. Proakis and DG Manolakis, “Digital Signal Processing: Principles, Algorithms, and Applications 2: ed”, McCillan Public Publishing, 19). Usually the unbiased probability of unbiased MMSE detection is slightly less than traditional MMSE detection. For constant modulation, unbiased MMSE detection corresponds to normal MMSE detection.

はＭＭＳＥポスト均等化信号対雑音比（ＭＭＳＥ ｐｏｓｔ‐ｅｑｕａｌｉｚａｔｉｏｎ ＳＮＲ）という。アンテナｋの無偏ＭＭＳＥ検出の信号干渉／雑音比（ＳＩＮＲ）を明らかにした。各アンテナ単位に電力を送信するため、ＭＭＳＥ検出の平均分散誤差（ＭＳＥ）は

で、且つ

である。そのため、本発明において以下の公式を定義してｒ条件における符号ｘ_kのメトリックとする：

便利のために、以下に

にて次メトリックを表す。これらは同一の意味があることを示す。 Factor to the left of the unbiased Euclidean distance

Is called the MMSE post-equalization signal-to-noise ratio (MMSE post-equalization SNR). The signal interference / noise ratio (SINR) of the unbiased MMSE detection of antenna k was clarified. Since power is transmitted to each antenna unit, the mean dispersion error (MSE) of MMSE detection is

And

It is. Therefore, in the present invention, the following formula is defined as the metric of the code x _k in the r condition:

For convenience,

Represents the next metric. These indicate the same meaning.

と等しい。 Thus, if the receiver has no prior information about the transmission code x _k , the scalar x _k APP with log-max is approximately:

Is equal to

は、アンテナｋへの無偏ＭＭＳＥ検出の最小無偏ユークリッド距離を有するコンスタレーションポイントである。 here,

Is the constellation point having the minimum unbiased Euclidean distance for unbiased MMSE detection to antenna k.

これは目前時刻の瞬時的な符号の信頼性を記載した。瞬時的な信頼度因数の値は大きいほど、より信頼できる。 The posterior probability APP of the scalar x _k is determined by: 1) The channel statistics SINR _k , which described the long-term channel reliability. And 2) the instantaneous reliability factor (IRF),

This described the reliability of the instantaneous code at the current time. The larger the instantaneous reliability factor value, the more reliable.

(3) Dynamic ranking algorithm of the present invention As discussed above, the most reliable antenna is removed first to avoid the effects of SIC error transmission.

によるＳ本の生残り経路の集合で、且つ

であるとすると、アルゴリズム（６）によって最も信頼性の高いアンテナをサーチできる。

ここで、

最も信頼性の高いアンテナをサーチする複雑度を更に簡素化するために、ここで例えばＳ＝１の特殊場合のみを考慮すると、

受信機には送信ベクトルｘに関する事前情報（ｐｒｉｏｒｉ ｉｎｆｏｒｍａｔｉｏｎ）がないとすると、以下のように最大事後確率を有するアルゴリズムによって最も信頼性が高いアンテナをサーチする：

公式（８）を上式に代入すると、

と得られる。

はベクトルｘ^（k-1,1）が削除された受信信号で、

はアンテナａ_kの瞬時的な信頼度因数（ＩＲＦ）で、

の無偏ＭＭＳＥ検出の最適点（最小無偏ユークリッド距離を有するコンスタレーション点）で、

は次の最適点（第２の最小ユークリッド距離を有する）である。 X ^(k-1) = {x ^(k-1,1) , ..., x ^{(k-1, S)} } is a predetermined order of the ^(k-1) th layer.

A set of S surviving paths by and

If so, the most reliable antenna can be searched by the algorithm (6).

here,

To further simplify the complexity of searching for the most reliable antenna, consider only the special case where, for example, S = 1,

If the receiver does not have prior information about the transmission vector x, it searches for the most reliable antenna by an algorithm with the maximum posterior probability as follows:

Substituting formula (8) into the above equation,

And obtained.

Is the received signal with the vector x ^(k-1,1) removed,

Is the instantaneous reliability factor (IRF) of antenna a _k

The optimal point for unbiased MMSE detection (constellation point with minimum unbiased Euclidean distance),

Is the next optimal point (having the second minimum Euclidean distance).

及び削除されるアンテナ数と正比例する。 The computational complexity of searching for the most reliable antenna is linear and the size of the constellation

And directly proportional to the number of antennas to be deleted.

によりｋ−１本のアンテナが削除され、且つＳ本の生残り経路を取得したとすると、受信機に送信符号の事前情報がないと、ｋ番目の層のＳ本の生残り経路は以下の式で演算されて得られる：

のＺＦ検出結果として、ここで、

番目の要素である。ｙ_ZFは

の線形関数で、且つ期待値Ｅ｛ｙ_ZF｝＝０及び分散

を有するため、

とｙ_ZFとの間のジャコビ行列式（Ｊａｃｏｂｉ ｄｅｔｅｒｍｉｎａｎｔ）で、確率密度関数

である。 (4) Survival path (one or multiple surviving paths) data detection algorithm of the present invention Predetermined order

K-1 antennas are deleted and S surviving paths are acquired, and if there is no prior information of transmission codes in the receiver, the S surviving paths of the kth layer are Obtained by computing with an expression:

As the ZF detection result of

Is the second element. y _ZF

And an expected value E {y _ZF } = 0 and variance

To have

And the Jacobian determinant between y _ZF and the probability density function

It is.

そのため、

前記検討に基づき

生残り経路は最小のメトリックの和を有する。ｋ番目の層の生残り経路のメトリックの和は、スカラーｘ_kのメトリックにｋ−１番目の層の生残り経路の相応メトリックの和を加算して得られる。Ｓ本の生残り経路を探索する複雑度は

と比例になる。 This provides a recursive relationship between the probability density function and the metric.

for that reason,

Based on the above review

The surviving path has the smallest metric sum. The k-th layer survivor path metric sum is obtained by adding the sum of the k-1 th layer survivor path metric to the scalar x _k metric. The complexity of searching for S survivors is

Becomes proportional.

  As can be seen from the above, each time a route having the maximum posterior probability (APP) is selected as a survivor-paths, the equi-effect selection method selects the route having the minimum metric each time based on the argument. The surviving route.

（ここで、Ｈ^(k)はｋ−１列が削除されたチャネル行列Ｈである）を演算するために、行列乗法と逆行列を演算する必要がある。 (5) Recursive operation MMSE detection result for kth layer

(Here, H ^(k) is the channel matrix H with the k-1 column deleted), it is necessary to calculate the matrix multiplication and the inverse matrix.

を表し、直接各層のＤ^(k)を求めると、Ｄ^(k)を更新する総の複雑度はＯ（Ｎ⁴）に近似し、本発明の検出アルゴリズムにおける最大演算量の部分である。 The matrix D ^(k) is

And D ^{(k) of} each layer is directly ^obtained , the total complexity of updating D ^(k) approximates to O (N ⁴ ), which is a portion of the maximum calculation amount in the detection algorithm of the present invention.

ここで、Ｄ_11、Ｄ_12、Ｄ_21、Ｄ_22、ｄ_1、ｄ₂及びδは行列Ｄ^(k-1)における要素である。

ここで、δは行列Ｄ^(k-1)のａ_k番目行とａ_k番目列の要素に対応している。

はチャネル行列Ｈ^(k-1)のアンテナ／列のサブスクリプトである。 Therefore, a simpler algorithm for calculating D ^(k) is required. D ^(k) is obtained from D ^(k-1) by the following recursive algorithm (D. Seethaler, Artes, H, and Hlawattsch, F. “Dynamic nulling-and-cancelling with near-ML performance”. for MIMO communication systems ", Acoustics, Speech, and Signal Processing, 2004. Proceedings. Volume 4 (2004): Page (s): iv-777-iv-780):

Here, D _11, D _12, D _21, D _22, d _1, d ₂ and δ are elements in the matrix D ^(k−1) .

Here, δ corresponds to the elements of the a _k th row and the a _k th column of the matrix D ^(k−1) .

Is the antenna / column subscript of the channel matrix H ^(k-1) .

により初期化される。Ｄ^(k)を更新する複雑度はＯ（Ｎ³）まで低下する。 Recursive expression

It is initialized by. The complexity of updating D ^(k) is reduced to O (N ³ ).

は更新する前のベクトルを表し、

はアンテナ指標である。ｙ^(k)とｙ^(k＋1)との間の関係は：

ここで、

且つ

ここで、γは行列Ｃ^(k-1)のａ_k番目行とａ_k番目列の要素に対応している。

はベクトルｙ^(k)のａ_k番目の要素に対応している。 Similarly, the present invention uses a recursive algorithm to compute MMSE detection.

Represents the vector before updating,

Is an antenna index. The relationship between y ^(k) and y ^{(k + 1)} is:

here,

and

Here, γ corresponds to the elements of the a _k th row and the a _k th column of the matrix C ^(k−1) .

Corresponds to the a _k th element of the vector y ^(k) .

を一度演算するだけでよい。公式（１４）には一行と一列が削除される以外に、他の数学操作が必要ではない。ＭＭＳＥ更新の総の演算複雑度はただＯ（Ｎ⁴）からただＯ（Ｎ²）まで低下する。 In formula (13), the calculation of S surviving paths is only O (N ² ) complexity.

Need only be calculated once. The formula (14) requires no other mathematical operations other than deleting one row and one column. The total computational complexity of the MMSE update is reduced from just O (N ⁴ ) to just O (N ² ).

Based on the principle and algorithm, the flow of the MIMO detection algorithm of the present invention will be described in detail below. In the following flow, X ^(k) , Y ^(k) , L ^(k) , and O ^(k) indicate a surviving path, an MMSE result, a set of metrics and antenna ranks, respectively. FIG. 1 is a flowchart illustrating a method for implementing MIMO detection based on MAP according to an embodiment of the present invention. As shown in FIG. 1, the MIMO detection method of this embodiment includes the following steps.

ここで、ｙ^(1,1)はＭＭＳＥ検出結果を表し、Ｌ^(0,1)はｙ^(1,1)に対応するメトリックで、ｘ^(0,1)は生残り経路である。 In step S100, the parameter matrix (matrix method C and inverse matrix D), the MMSE detection result, the surviving path metric, the surviving path, the antenna rank, and the like are initialized.

Here, y ^(1,1) represents the MMSE detection result, L ^(0,1) is a metric corresponding to y ^(1,1) , and x ^(0,1) is a surviving path.

Then, the repetitive signal variable k is set to 1 to N, and the dynamic order SIC detection based on the MAP is performed. In particular,
Step S200, searching for the most reliable antenna. The steps include:

ここで、ｄｉａｇ（Ｄ^(k)）は選択方陣Ｄ^(k)の対角要素を示す。 Step S201, SINR of antenna candidates {1,..., N} \ O ^(k) is calculated for each one based on the matrix D diagonal element:

Here, diag (D ^(k) ) indicates a diagonal element of the selection square D ^(k) .

  In step S202, the instantaneous reliability factor I of the antenna is calculated for each one based on the MMSE detection result.

を更新する。 In step S203, the reliability of the antenna is calculated for each one by the formula (10), and the antenna a _k (the most reliable antenna) having the optimum reliability is selected, and the antenna is based on the calculated reliability. Rank

Update.

In step S300, a metric corresponding to the antenna _ak of the _kth layer is calculated by the formula (11), and S paths corresponding to the minimum metric are selected as surviving paths, and specifically include the following.

２）

よりＳ個の最小値をｋ番目の層のメトリックＬ^(k)として選出し、当該Ｓ個の最小値に対応するＳ本の経路を生残り経路とするとともに、相応するコンスタレーションポイント及びＭＭＳＥ検出結果を記録する。 1) A repetitive signal variable i is set to 1 to S, and a metric that each constellation point can have is calculated.

2)

The S minimum values are selected as the metric L ^(k) of the ^kth layer, and the S paths corresponding to the S minimum values are taken as surviving paths, and corresponding constellation points and MMSE detections are made. Record the result.

In step S400, the surviving path X ^(k) having the recorded constellation point is updated, and the antenna _ak is deleted from the antenna candidates.

Step S500, the parameter matrix and the MMSE detection result are updated. In this step, y ^(k) , D ^(k) , and C ^(k) are updated by the recursive algorithm of formulas (13), (12), and (14).

It is determined whether the antenna of the Nth layer has been detected. If the Nth layer has been detected, the first surviving path X ^(N) in dynamic order is selected as the last detection result. This completes the detection of the MIMO system.

  Normally, MIMO detection is performed in the complex domain (all elements in the system model (1) are complex). Since the performance of MIMO detection is deeply influenced by the deletion order, the detection performance is higher in theory as the order is more precise.

  Due to the conversion relationship between complex numbers and real numbers, any complex number equation is represented by an equal-effect real number equation. Similarly, a complex MIMO system is represented in the same manner as the real MIMO system shown below. Therefore, the embodiment of the present invention further expands the complex system model with the real system model so as to improve the detection performance.

ここで、

は変数の実数部分（実部）と虚数部分（虚部）を表す。 In the complex number model, all x, r, n, H, etc. in the formula (1) are complex numbers. This can be expanded with a real model of equal effects:

here,

Represents the real part (real part) and imaginary part (imaginary part) of the variable.

  Because of the uncorrelation between the real part and the imaginary part, the complex code can always be divided into two uncorrelated real sub-codes for QAM modulation. For example, one 16QAM code is always separated from two 4PAM codes.

  For the detection algorithm of the present invention (the detection algorithm for multiple surviving paths in the present invention may be called a multiple path diameter detection (MAP-DOM) algorithm based on MAP dynamic order), the detection performance is improved by a real code. That is effective. Detection performance can be improved from a more precise dynamic order. Directly, this also results in computational complexity, but if the complex code constellation is large, such as 64QAM, the complexity of the real code constellation is still complex algorithm because the size of the real code constellation is much smaller than the complex code constellation Is less complex.

  The overwhelmingly large number of MIMO detection methods are complicated and include a plurality of arithmetic operations such as inverse matrix, QR decomposition, rank (Sort) and the like. Since the complexity of these operations is deeply influenced by specific algorithms, it is very difficult to reliably analyze the complexity of MIMO detection.

  For convenience, the following approximates the complexity of the MAP-DOM algorithm by overlooking some details differences. Use the following assumptions:

  1) MIMO has the same number of TX antennas and RX antennas and M = N; has the same modulation method such as 16QAM for each data stream; the same number such as S surviving paths per layer Have a surviving path.

  2) The unit of operation has complex arithmetic or division complexity, and addition or subtraction is sparse; the complexity of real operations is ¼ of the complexity of complex operations.

3) It is assumed that the complexity of the matrix calculation or inverse matrix is N ³ .

  4) When the S value is very small, the complexity of the order (or linear complexity) is neglected.

The complexity of the total MAP-DOM algorithm includes two parts: an initial operation represented by C _init and a repetitive signal operation represented by C _iter .

1) Initial operation C _init
For complex systems, C _init includes 1 × matrix arithmetic, 1 × inverse matrix, and 2 × matrix and vector arithmetic (y ^(1,1) = D ⁽¹⁾ (H ⁽¹⁾ r)). The total complexity is C _init → 2N ³ + 2N ² .

  The real MAP-DOM algorithm has the same complexity as the complex algorithm.

2) C _iter
The number of repetitive signals in the MAP-DOM algorithm is N (2N for the real number algorithm), and the complexity is 1 to N (the real number algorithm is 1 to 2N).

要するに、表１に示すように、ＭＡＰ‐ＤＯＭアルゴリズムの複雑度は以下の要素に決められる。 In general, the complexity of each iteration is as follows (n is the iteration variable):
Dynamic layer rank: Approximate to | C | · n Detection of multiple path diameters: Approximate to | C | · S Update parameter matrix and vector: Approximate to 2n ² + S · n Therefore, the total complexity is as follows estimated: complex system C _iter about 2 / 3N ³ to +1/2 (S + | C |) · in N ^2, is the real system C _iter about ^{4 / 3N 3 +1/2 (S +} √ | C |) N ²

In short, as shown in Table 1, the complexity of the MAP-DOM algorithm is determined by the following elements.

1) The number of transmit antennas is N, and the complexity approximates O (N ³ );
2) If the number of surviving paths S, the size of the constellation | C |, and the number of transmitting antennas N are constant, the complexity of the MAP-DOM algorithm is directly proportional to the sum of S and | C |;
3) The complex algorithm is more sensitive to the constellation size | C | than the real algorithm because the number of points in each constellation is reduced to √ | C | by the real number algorithm;
4) When | C | is as small as 4QAM, the complex algorithm is less complex than the real algorithm. However, as the modulation level (Level) increases, the real algorithm has a lower complexity than the complex algorithm. One of the goals of MIMO is to realize high-speed data transmission with a limited bandwidth. Therefore, the trend of technological development is to use high level modulation such as 16QAM or 64QAM in a MIMO system. At that time, the real algorithm is superior to the complex algorithm because it has higher detection performance and lower computational complexity.

  For complex and real systems, the total complexity of the MAP-DOM algorithm is close. If the number of surviving paths is small, the complexity of the real system is higher than the complexity of the complex system in the update step, but as the surviving paths increase and the constellation increases, the two types of operations are approximated together. Has complexity.

  Since each antenna of the real algorithm has constellation points of √ | C |, the complex algorithm is more sensitive to the constellation size than the real algorithm. As the constellation points increase, the complexity of the complex algorithm exceeds that of the real algorithm.

  Below is a simulated display of the performance of different MIMO settings on a computer.

1) Code error rate (SER) performance estimation SER performance estimation is performed for a complex number system and a real number system. Here, only the detection result of the 8 × 8 MIMO system having 16QAM is provided, but it has already been confirmed that an approximate result can be obtained for the other different MIMO arrangement. FIG. 2 is a schematic diagram showing the SER performance of the complex and real MAP-DOM algorithms. Here, DOM-C represents a complex MAP-DOM algorithm, and the subsequent numbers 2 and 4 are the number of surviving paths. DOM-R represents a real MAP-DOM algorithm, and the subsequent numbers 2 and 4 are the number of surviving paths. The figure corresponds to a MIMO system with N = M = 8, 16QAM, and the following conclusions can be obtained.

Different MIMO settings can achieve a performance in which both DOM-C and DOM-R approximate ML in a large range. When SER = 10 ⁻⁴ for ML detection when S = 4, the performance of DOM-R does not decrease to 0.5 dB, and the performance of DOM-C decreases only within 2 dB.

  The performance of DOM-R is superior to that of DOM-C. This merit is due to a more precise dynamic ranking. If M is small, the performance of DOM-R is 2M higher than DOM-C or the performance is the same.

  For an overwhelming number of MIMO settings, it is sufficient to set S to 4, and a larger S can only improve performance slightly.

2) Influence of Number of Surviving Paths FIG. 3 is a schematic diagram showing the relationship between the MIMO size and the SNR threshold. The SER corresponding to the curve in the figure is 10 ⁻⁴ , and N = M and 16QAM. According to analog results, for different MIMO settings, DOM-R and DOM-C have the ability to quickly approach ML detection as the number of surviving paths increases.

  The larger the MIMO setting, the lower the SNR threshold. As the number of antennas increases, the channel capacity for each antenna increases and reaches the limit.

  FIG. 3 shows how to set the number of surviving paths so as to balance detection performance and computational complexity. In practice, for 16QAM modulation, DOM-R with S = 4 and DOM-C with S = 6 can provide performance equivalent to ML detection. Performance degradation is less than 1 dB.

3) Role of dynamic order To illustrate the role of dynamic order, FIGS. 4a and 4b are MAP-DOM (DOM-R and DOM-C) and non-moving when S = 2 and S = 4, respectively. We compared the performance of multiple path diameter algorithms (NDOM-R and NDOM-C) with the same rank. As can be seen from FIGS. 4a and 4b, the performance of (DOM-R and DOM-C) is at least about 5 dB better than that of (NDOM-R and NDOM-C). This beneficial effect is due to dynamic ordering, because a superior dynamic ordering method can significantly reduce error transmission.

  For non-dynamic ranking, NDOM-R is slightly better than NDOM-C. This is because the NDOM-R cannot obtain further gain by a more precise ranking.

  In short, the present invention provides a new MIMO detection method that approximates ML detection performance, which not only has near-maximum performance, but also has low computational complexity. The method was based on the MAP standard and improved in performance by dynamic ranking and multiple path diameter detection. Error transmission is reduced by the dynamic order, and detection reliability is improved by detecting a plurality of path diameters. The detection method of the present invention by real spreading provides a more reliable antenna ranking method and further improves the performance of the system.

  Since theoretical analysis is difficult due to multiple path detection algorithms, the present invention analogizes and estimates the MAP-DOM performance on a computer, and performs for different MIMO settings, different dynamic orders, and different numbers of surviving paths. The comparison explained that MAP-DOM has a performance that approximates ML detection (performance decrease is less than 0.5 dB) and has low computational complexity.

  The complexity of MAP-DOM MIMO detection is greatly reduced by recursive algorithms, has better surveyability for different MIMO settings, and easily adjusts the balance between complexity and performance by changing the number of surviving paths can do. Unlike SD, MAP-DOM has a fixed complexity for different SNRs and channel conditions.

  Based on the embodiment, the object, technical solution and beneficial effect of the present invention have been described in detail. The above description is only an embodiment of the present invention, and does not limit the scope of the claims of the present invention. Any amendments, equivalent replacements, improvements, etc. within the spirit and principle of the present invention are all described in the present invention. Within the scope of the following claims.

Claims (10)

In a repetitive signal detection method for a MIMO (multiple input multiple output) system,
Using the operation parameter calculated based on the maximum posterior probability (MAP) algorithm, the kth detection layer searches for the antenna with the highest reliability of the received signal from among a plurality of antenna candidates, On the other hand, S routes from the route having the maximum posterior probability are stored as surviving routes, where S ≧ 1 and k = 1, the calculation parameters are initialized calculation parameters, and k> 1 In the case of the calculation step, the calculation parameter is a calculation parameter updated according to the detection result of the k-1th detection layer, and
An antenna deletion step of updating the calculation parameter according to the calculation result and deleting the antenna having the highest reliability of the received signal from the antenna candidates;
repeating as k = k + 1, repeating the calculation step and the antenna deletion step by a recursive algorithm until all antenna candidates are deleted,
A method of detecting a repetitive signal, characterized in that one surviving path having the maximum posterior probability in the detection layer of the final layer is set as a final detection result. The repetitive signal detection method according to claim 1,
The calculation parameters include a matrix product, an inverse matrix, a minimum mean square error (MMSE) detection result, a survivor path metric, a survivor path, and an antenna rank.
The calculation step includes:
Calculate the signal-to-interference / noise ratio SINR for each antenna candidate based on the inverse matrix diagonal element,
Calculate the instantaneous reliability factor of each antenna candidate based on the minimum mean square error (MMSE) detection result,
Calculate the reliability of the antenna based on the ratio SINR of the signal to interference / noise and the instantaneous reliability factor, and select the antenna with the highest reliability of the received signal,
Storing S paths having the smallest metric corresponding to the selected antenna as surviving paths. The repetitive signal detection method according to claim 2,
The initialized matrix product, inverse matrix, minimum mean square error (MMSE) detection result, survivor path metric, survivor path, and antenna rank are expressed as follows:

Here, Y ⁽¹⁾ , L ⁽⁰⁾ , X ⁽⁰⁾ , and O ⁽⁰⁾ represent a set of minimum mean square error MMSE detection results, survivor path metrics, survivor paths, and antenna ranks, respectively. Is the channel matrix, σ ² is the noise power, I is the unit matrix, y ^(1,1) is the minimum mean square error MMSE detection result, r is the received signal vector, and L ^(0,1) is y ^{( 1,1)} is the metric, x ^(0,1) is the surviving path,
Calculate the signal ratio SINR to the interference noise of each antenna candidate by the following equation:

Calculate the reliability of the antenna using the following formula,

Where I _ak is an instantaneous reliability factor, A ^(k) is a set of antenna candidates,
Select the survival route by the following formula,

here,

Represents that the S minimum values are selected side by side from the metrics for all the constellation points in the selected antenna candidate. The repetitive signal detection method according to claim 3,
The step of selecting the surviving route is
The iteration variable i is set to 1 to S, and the metric for all constellation points is calculated by the following formula.

Here, i is a surviving path, j is a constellation point corresponding to the i th surviving path, L ^{(k−1, i)} is a metric of the i th surviving path in the k−1 th layer,

Is the metric of the survivor path x _j in the kth layer,

S minimum values are extracted as the metric L ^(k) of the k-th layer, and S paths corresponding to L ^(k) are selected as surviving paths. The repetitive signal detection method according to claim 3,
Updating the calculation parameter according to the calculation result includes updating the antenna rank according to the reliability of the calculated antenna candidate,
Update the survival route by the following formula,

The matrix multiplication, inverse matrix, and minimum mean square error MMSE detection result are updated by the following formula:

  The repeated signal detection method according to claim 1, wherein the number of surviving paths in different detection layers is the same.   The repetitive signal detection method according to claim 1, wherein the number of surviving paths in different detection layers is different, and this number is semi-fixed or dynamically adjusted and determined. In the repeating signal detection method of claim 7, and characterized in that the number of survivor paths S _k in the different detection layer is set semi-fixedly satisfies _{S 1 ≧ S 2 ≧ ··· ≧} S N To do.   2. The repetitive signal detection method according to claim 1, wherein the MIMO (multiple input multiple output) system is a system using complex signals. The repetitive signal detection method according to claim 1, wherein the MIMO (multiple input multiple output) system is a real signal system obtained by converting a complex signal as follows:

here,

Represents the real part and the imaginary part of the variable, respectively.

Images