KR20090020767A - Apparatus and method for low-complexity detection based on unified iterative tree searching in multiple input multiple output systems - Google Patents

Abstract

A low-complexity signal detection apparatus based on the UITS in an MIMO system and a method thereof are provided to make easy combination with different tree search algorithms possible. A detector forms a tree structure corresponding to the degree of modulation by applying the QR decomposition to a channel matrix(203), determines the number of candidate symbols according to an SNR(Signal-to-Noise Ratio)(205), separates the whole tree depth into the first tree depth and the rest tree depth(207), and then determines candidate symbols as many as the number determined in the first tree depth(209).

Description

Apparatus and method for detecting low complexity signals based on iterative tree search in multi-input multi-output system

TECHNICAL FIELD The present invention relates to a multiple input multiple output system, and more particularly, to an apparatus and method for detecting a low complexity Near-ML signal based on Unified Iterative Tree Searching (UITS).

As one of the technologies for satisfying the demand for high speed data transmission, Multiple Input Multiple Output (MIMO) technology using multiple antennas has gained great attention as a core technology of the next generation wireless transmission. Is being studied.

The MIMO technology can be broadly classified into an open loop MIMO and a closed loop MIMO. First, the open loop MIMO transmits data in a state in which a transmitter does not know information about a channel, for example, space-time coding or V-BLAST (Vertical-Bell Labs LAyered Space-Time). ) And the like. Next, the closed loop MIMO is a method in which a transmitting end acquires information about a channel and then transmits data using the same. For example, singular value decomposition (SVD), spatial division multiple access (Spatial Division) Multiple Access: SDMA).

The V-BLAST transmission technique is a technique for improving data transmission efficiency per unit frequency by simultaneously transmitting different signals through a plurality of transmit antennas without using additional bandwidth in the MIMO and multi-user systems. . When the V-BLAST transmission technique is applied to the system, signals passing through different channels are received at the receiving side, and thus there is interference between antennas. Various techniques have been proposed to detect a desired signal while effectively removing the V-BLAST transmission technique.

Among the various techniques, the maximum approximation (ML) technique that detects a signal in consideration of all cases is the best, but the complexity increases exponentially according to the modulation order and the number of transmitting antennas. Because of this, the actual implementation is impossible. As another technique, the simplest conventional techniques include zero forcing based on inversion of a channel according to the number of transmit / receive antennas, or a minimum mean squares error. MMSE '). The ZF and MMSE techniques are poor in performance, and a SIC (Successive Interference Cancellation) technique for continuously detecting a signal while removing interference has been proposed to improve performance. In the SIC scheme, error propagation is generated according to the accuracy of the first detected signal. In order to prevent the error propagation, a technique of changing the order of the detected signals has been applied. However, the process of calculating Inversion continuously causes a problem of increasing complexity. In order to improve the above problem, ZF-QRD, MMSE-QRD, etc. have been proposed based on QR Decomposition (hereinafter referred to as 'QRD') of the channel matrix. Similarly, in the case of the QRD, the signal detection order may be changed to minimize error propagation (Sorted QRD: SQRD). However, the techniques described above show significant performance differences from the ML technique.

As described above, there is a tradeoff in terms of performance and complexity in accordance with the proposed technique. That is, while ZF-SQRD and MMSE-SQRD techniques have low complexity, there are many performance differences with the ML technique, and the ML technique shows the best performance but the complexity is difficult to actually implement.

Accordingly, there is a need for a signal detection technique having a low complexity and high performance in the MIMO system.

Accordingly, an object of the present invention is to provide an apparatus and method for detecting a low complexity Near-ML signal based on Unified Iterative Tree Searching (UITS) in a MIMO system.

In order to achieve the above object, according to an embodiment of the present invention, a signal detection method based on iterative tree search in a multi-antenna / user system, by applying a QR decomposition to the channel matrix to form a tree structure according to the modulation order Determining the number of candidate symbols according to a process, a signal-to-noise ratio (hereinafter referred to as 'SNR'), and the total tree depth as the first tree depth and the remaining tree depth. And determining the as many candidate symbols as the determined number at the first tree depth.

In order to achieve the above object, according to an embodiment of the present invention, a signal detection apparatus based on repetitive tree search in a multi-antenna / user system applies a QR decomposition to a channel matrix to form a tree structure according to a modulation order. A tree structure forming unit, a candidate symbol number determining unit for determining the number of candidate symbols according to a signal-to-noise ratio (hereinafter referred to as 'SNR'), and a total tree depth as the first And a candidate symbol determiner that separates the first tree depth from the remaining tree depth and determines the determined number of candidate symbols from the first tree depth.

The present invention proposes a signal detection algorithm based on an integrated iterative tree retrieval scheme (UITS) in a MIMO system. have. In addition, the algorithm according to the present invention can be easily combined with other tree search algorithms (pano algorithm or QRD-M algorithm), and it is expected that the same effect can be obtained by applying to a terminal in a multi-user environment as well as a multi-antenna system. .

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description of the present invention, when it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted.

Hereinafter, an apparatus and method for detecting a low complexity signal based on repetitive tree search in a MIMO system according to the present invention will be described.

The present invention proposes an integrated iterative tree retrieval scheme that can achieve near ML performance while minimizing the receiver's effort in a MIMO system under a fading channel. In particular, the following two main ideas are presented together to minimize receiver effort. First, the number of iterative tree searches can be assigned variably, and the number of iterations (Ni) is determined based on two rules presented later. Second, two early termination techniques are used in the iterative tree search process.

The integrated iterative tree search proposed in the present invention is composed of two steps of signal processing, that is, preparation and iterative tree searching for iterative tree search, which will be described in detail later with reference to FIG. 2. Let's do it.

The present invention will be described by taking a MIMO system as an example, but of course, it is applicable to both a multi-user system as well as a multi-antenna system.

1 is a view schematically showing the structure of a MIMO system according to the present invention.

As shown, the MIMO system consists of a transmitter and a receiver. Here, the transmitter includes a modulator 101, an encoder 103, a plurality of transmit antennas, for example, a first transmit antenna Tx1 to an mth transmit antenna Txm, The receiver includes a plurality of receiving antennas, for example, the first receiving antenna Rx1 to the nth receiving antenna Rxn, a detector 111, and a demodulator 113. In the following description, it is assumed that the number of transmit antennas included in the transmitter and the number of receive antennas included in the receiver are different, but the number of transmit antennas of the transmitter and the number of receive antennas of the receiver may be the same.

Referring to FIG. 1, first, a modulator 101 of the transmitter modulates input information data bits using a predetermined modulation scheme to generate modulation symbols, and then generates the encoder. 103). Here, the modulation scheme may include a Binary Phase Shift Keying (BPSK) scheme, a Quadrature Phase Shift Keying (QPSK) scheme, a Quadrature Amplitude Modulation (QAM) scheme, a Pulse Amplitude Modulation (PAM) scheme, a Phase Shift Keying (PSK) scheme, and the like. have.

는 하기 <수학식 1>과 같이 나타낼 수 있다. The encoder 103 encodes the serial modulation symbols input from the modulator 101 by using a predetermined encoding scheme and transmits the same through the first transmitting antenna to the mth transmitting antenna. Here, the encoding method is a method of parallel conversion of the serial modulation symbols input from the modulator 101 corresponding to the number of the transmitting antennas. Here, a transmission signal vector consisting of signals transmitted through the m transmission antennas

May be represented as in Equation 1 below.

는 하기 <수학식 2>와 같이 나타낼 수 있다. Next, the detector 111 of the receiver receives a signal transmitted through the m transmit antennas from the transmitter through each of the first to nth receiving antennas, detects the received signals, and detects the demodulator ( Output to 113). In this case, the detector 111 uses a signal detection algorithm proposed by the present invention for detecting the signal. Here, the received signal vector consisting of the signals received through each of the first to nth receiving antenna

May be represented as in Equation 2 below.

는 하기 <수학식 3>과 같이 나타낼 수 있다.In addition, the received signal vector

May be represented as in Equation 3 below.

는 상기 제1수신 안테나 내지 제n수신 안테나 각각의 채널 응답(channel response)들로 구성된 채널 응답 벡터를 나타내고, 상기

는 상기 제1수신 안테나 내지 제n수신 안테나 각각을 통해 수신된 잡음(noise) 신호들로 구성된 잡음 벡터를 나타낸다. 여기서, 상기 채널 응답 벡터

는 n×m 크기의 행렬로 나타낼 수 있다.Where

Denotes a channel response vector composed of channel responses of each of the first to nth receiving antennas.

Denotes a noise vector composed of noise signals received through each of the first to nth receiving antennas. Where the channel response vector

Can be expressed as a matrix of size n × m.

The demodulator 113 demodulates the signal input from the detector 111 in a demodulation scheme corresponding to the modulation scheme applied by the modulator 101 of the transmitter and restores the original information data bits.

2 is a flowchart illustrating a signal detection method based on an iterative tree search in a MIMO system according to an embodiment of the present invention. Here, the repetitive tree search consists of a preparation process for repetitive tree searching (steps 201 to 211) and an iterative tree searching (steps 213 to 229). Here, the tree search process may be performed by various techniques, for example, a stack algorithm, a pano algorithm, a QR decomposition (hereinafter referred to as 'QRD')-M algorithm, or the like.

Referring to FIG. 2, the detector 111 of the receiver sets the cumulative branch metric value Γ for a full-length sequence of m (the same as the number of transmitting antennas) as infinity (∞) in step 201. Initialize

In operation 203, the detector 111 forms a tree structure according to the modulation order M by applying QR decomposition (hereinafter referred to as 'QRD') to the channel matrix. Here, the QRD may be performed based on the MMSE, and the channel matrix extended according to the QRD may be represented by Equation 4 below.

는 확장된 채널 행렬이고, 상기

는 잡음 분산을 나타낸다. 상기

는 n×m

와 m×m

로 이루어진 ((n+m)×m) 유니터리 행렬로서, 확장된 직교정렬 열로 이루어진 행렬이며, 상기

는 상부 삼각 행렬(upper triangular matrix)을 나타낸다. 여기서, 상기 상부 삼각 행렬의 대각 성분은 양의 실수를 가진다. 이는 상기 QRD가 수정된 그램 슈미트(modified Gram-Schmidt)에 기반하고 있기 때문이다. Where

Is an extended channel matrix,

Represents noise variance. remind

Is n × m

And m × m

((N + m) × m) unitary matrix consisting of the extended orthogonal matrix,

Denotes an upper triangular matrix. Here, the diagonal component of the upper triangular matrix has a positive real number. This is because the QRD is based on a modified Gram-Schmidt.

In Equation 4, Equation 3 may be expressed as Equation 5 below.

는

이며, 상기

는 전치 공액(transpose conjugate)을 나타낸다. 상기

는 허미션 전치(Hermitian transpose)를 나타내고,

는 모든 원소가 0인 m×1 백터를 나타낸다. 결과적으로, 원하는 수신 신호 벡터는 최종적으로 하기 <수학식 6>과 같이 나타낼 수 있다. here,

Is

And said

Denotes a transpose conjugate. remind

Represents Hermitian transpose,

Denotes an m × 1 vector where all elements are zero. As a result, the desired received signal vector can be finally expressed as in Equation 6 below.

Here, the k th component of the received signal vector may be represented by Equation 7 below.

는 상기

의 (k,i)번째 성분을 의미한다. Where

Above

Means the (k, i) th component of.

Here, the ML signal detection techniques according to Equations 3 and 6 may be expressed as Equation 8 below.

는 각 심볼 조합의 비용(cost)을 나타낸다. 여기서, 상기 <수학식 8>을 만족하는 최적의 심볼 열

에 대한 누적된 브랜치 매트릭은 하기 <수학식 9>와 같이 나타낼 수 있다. Where ∥ represents Frobenius norm, and

Denotes the cost of each symbol combination. Here, an optimal symbol string satisfying Equation (8)

The accumulated branch metric for may be expressed as in Equation 9 below.

를 결정한다. 연속적인 신호 검출 기법에서는 에러 전파가 성능에 지대한 영향을 준다. 이러한 관점에서 볼 때, 상부 삼각 행렬에서 마지막 대각 성분의 값은 처음 검출되는 신호뿐 아니라 이어지는 신호 검출에 따른 성능에 영향을 주게 된다. 한편, SNR이 낮은 영역에서는 아무리 많은 신호 처리 과정이 부가되어도 그에 따른 성능 개선을 기대하기 힘들다. 이러한 두 가지 관점에 의거하여 아래 두 가지 법칙을 제안하였으며, 이를 적용하여 상기 후보 심볼의 수(반복횟수와 동일)

를 가변적으로 결정할 수 있다. In step 205, the detector 111 determines the number of candidate symbols according to a signal-to-noise ratio (hereinafter referred to as 'SNR') (the same as the number of repetitions).

Determine. In continuous signal detection, error propagation has a significant impact on performance. From this point of view, the value of the last diagonal component in the upper triangular matrix affects not only the first detected signal but also the performance of subsequent signal detections. On the other hand, no matter how many signal processing processes are added in the low SNR region, it is difficult to expect a performance improvement. Based on these two perspectives, the following two laws are proposed, and the number of candidate symbols is the same as the number of iterations

Can be determined variably.

에 의해 이루어진다. 여기서 low SNR 영역이라 함은 원하는 성능에서 적절한 에러 확률을 수용하면서 필요 이상의 수신기 노력을 요하는 영역으로 정의한다. 간단한 예로서, 일반적인 시스템에서, 송신기의 스케줄러가 수신기로부터 피드백 받은 CSI 정보에 따라 MCS(modulation and coding scheme)를 선택하 면, 해당 MCS에 대해 요구 SNR이 정해져있으므로, 수신기에서는

(= 해당 MCS의 요구 SNR-

)이상의 영역을 high SNR로 선택하고, 그 이하의 영역을 low SNR로 선택할 수 있다. 여기서, 상기

는 채널환경의 차이에 따른 요구SNR 차이나 문턱값 마진 등에 따른 보정 상수를 나타낸다.Rule 1: The total SNR to be observed is divided into a low SNR and a high SNR region. Where the delimiter

Is done by. In this case, the low SNR region is defined as an area requiring more receiver effort than necessary while accepting an appropriate error probability at a desired performance. As a simple example, in a typical system, if the scheduler of the transmitter selects a modulation and coding scheme (MCS) according to the CSI information fed back from the receiver, the required SNR is defined for the MCS.

(= SNR

The higher region can be selected as the high SNR, and the lower region can be selected as the low SNR. Where

Denotes a correction constant according to the required SNR difference or threshold margin according to the channel environment.

값이 미리 정의된 특정 값(

)보다 작은 경우 상기

를 1로 결정하고, 그 외의 경우는 상기

를 M으로 결정한다.Rule 2-1: For the Low SNR area,

A specific value with a predefined value (

Above if less than

Is determined to be 1, otherwise

Determine M as.

값에 상관없이 상기

를 M으로 결정한다. Rule 2-2: for the High SNR region,

Regardless of the value

Determine M as.

가 감소하면 일반적으로 오류가 증가하여 성능이 나빠지나, 제1문턱값인

보다 낮은 low SNR 영역에서는 상기

의 증/감소에 따른 오류 확률의 차이가 크지 않다는 것을 확인할 수 있다. 따라서, low SNR 영역에서는 제2의 문턱값

을 기준으로 그 이하인 경우 성능의 개선이 크지 않으므로 상기

를 굳이 증가시킬 필요 없이 1로 설정해도 무방하다. Here, Figure 3 provides the basis for the two rules. As shown in FIG. 3, the

Decreases, the error generally increases, resulting in poor performance, but the first threshold,

In the lower low SNR region,

It can be seen that the difference in error probability due to the increase / decrease of is not large. Therefore, the second threshold in the low SNR region

If the value is lower than the above, the performance improvement is not significant.

You can set it to 1 without increasing it.

와

를 어떻게 결정할 지가 중요한 관건이 된다.

와

를 특정 SNR

에서 ML과 제안된 알고리즘에 의한 전체 에러 확률이라 하고 이 정보가 이용 가능하다고 한다면, 상기

는 하기 <수학식 8>과 같이 결정할 수 있다. Where

Wow

How to decide is important.

Wow

Specific SNR

Is the overall error probability by ML and the proposed algorithm, and if this information is available,

May be determined as in Equation 8 below.

은 수용 가능한 에러 확률(tolerable error probability)로 정의된다. 그렇지만 임의의 채널 행렬과 낮은 SNR에 대해서 트리 검색에 기반한 신호 검출 기법에 대한 이론적 분석은 알려져 있지 않을뿐더러 정확히 구하기 힘들다. 이러한 이유로 실험적으로 상기

와

를 선정하도록 한다. 비록 두 변수에 대한 설정이 경험에 기반하고는 있지만, 다음과 같은 방법에 의거해서 상기 두 변수를 결정할 수 있다. 먼저 관찰하고자 하는 전체 SNR을 두 부분 즉, low SNR과 high SNR로 구분하는데, 이는 low SNR영역에서는 복잡도가 상당히 크면서 제일 좋은 성능을 도출하는 ML 알고리즘을 이용한다고 해도 그에 따른 성능 개선을 기대하기 힘들기 때문이다. 특히, 트리 검색 알고리즘을 이용하는 경우는 이러한 현상을 뚜렷이 볼 수 있으며, 이에 대한 복잡도 개선을 위해서 두 영역으로 구분 지어 서 서로 다른 반복 횟수를 결정하고자 한다. 여기서, low SNR 영역과 high SNR 영역을 구분하기 위해서, 도 4를 참조하여

값이 상기

값 이하의 값을 갖는 경우에

를 1로 설정하여 그 성능을 관찰하였다. 상기 도 4에서는

값이 0.4, 0.7, 1.0, 1.3의 경우에 대해서 평균 비트 에러 율(bit error rate : 이하 'BER'이라 칭함) 커브를 관찰할 수 있으며, 이를 이용하여

를 결정할 수 있다. 이때, 시스템에서 수용할 수 있는 에러율

이 주어졌다고 가정한다면, 각 변조 기법에 대해서 서로 다른

에 따라 두 영역이 구분 지어진다. 덧붙여 설명하면, 처음 검출되는 신호(트리 구조에서는 첫 번째 깊이에 해당)인

에 직접적인 영향을 주는

의 값이 에러확률에 중요한 영향을 미치게 된다.

값이 1보다 작은 값을 갖는다는 것은 신호의 세기를 감쇄시키고 노이즈는 강화시키게 된다. 따라서, 서로 다른 변조 기법에 대해서 low SNR 영역을 구분 지으려면,

=

이 1 또는 작은 근처의 값에 대한 성능과 원하는 성능(실험에서는 ML 성능)과의 차이가 현저하게 보이는 SNR지점을

로 결정하면 된다. 실제 AMC를 적용한 시스템에서는 SNR에 따라 이미 MCS가 정해져 있으므로 이 정보에 기반하여 위에서 제시한 방법을 적용하면 된다.Where

Is defined as the acceptable error probability. However, the theoretical analysis of signal detection based on tree search for arbitrary channel matrices and low SNR is unknown and difficult to obtain accurately. For this reason, experimentally reminded

Wow

To be selected. Although the setting for both variables is based on experience, the two variables can be determined based on the following method. First, the total SNR to be observed is divided into two parts, low SNR and high SNR. In the low SNR region, even if the ML algorithm that yields the best performance is very complicated, it is difficult to expect a performance improvement. Because. In particular, when the tree search algorithm is used, this phenomenon can be clearly seen, and in order to improve the complexity, it is divided into two areas to determine different repetition times. Here, to distinguish between the low SNR region and the high SNR region, referring to FIG. 4.

Value above

Has a value less than or equal to

Was set to 1 to observe its performance. In FIG. 4

For the values 0.4, 0.7, 1.0, and 1.3, the average bit error rate curve (hereinafter referred to as BER) can be observed.

Can be determined. The error rate that the system can accept

Given this is different for each modulation scheme

The two areas are divided accordingly. In addition, the first detected signal (corresponding to the first depth in the tree structure)

Directly affects

Has a significant effect on the error probability.

Having a value less than 1 attenuates the signal strength and enhances the noise. Thus, to distinguish the low SNR region for different modulation schemes,

=

Find an SNR point where the difference between the performance for this value of 1 or a small value and the desired performance (ML performance in the experiment) is significant.

You can decide. In the AMC-applied system, the MCS is already determined according to the SNR, so the above-described method can be applied based on this information.

개의 후보 심볼을 결정한다.In step 207, the detector 111 separates the entire tree depth into a first tree depth and a remaining tree depth, and the flow proceeds to step 209 where the first tree depth is determined.

Candidate symbols are determined.

개의 후보 심볼에 대해서 유클리디안 제곱 거리(Squared Euclidean Distance : 이하 ‘SED’라 칭함)가 작은 순으로 정렬하고, 상기 정렬된 후보 심별들에 상응하는 브랜치 매트릭을 정의하면, 하기 <수학식 11>과 같이 나타낼 수 있다. 이때, 상기 순서적으로 정렬된 후보 심볼에 대한 인덱스 k를 1로 설정한다. Then, the detector 111 is determined in step 211

If the Euclidean square distance (SED) is arranged in ascending order of the candidate symbols, and defines a branch metric corresponding to the sorted candidate discriminations, Equation 11 It can be expressed as At this time, the index k for the ordered candidate symbols is set to one.

와 같은지 여부를 검사한다. 상기 k가 상기 후보 심볼의 수(반복횟수와 동일)

와 같을 시, 상기 검파기(111)는 229단계로 진행하여 상기

의 해당 심볼열을 최적 심볼열로 결정한 후, 본 발명에 따른 알고리즘을 종료한다. 반면, 상기 k가 상기 후보 심볼의 수(반복횟수와 동일)

와 같지 않을 시, 상기 검파기(111)는 215단계에서 k번째 후보 심볼에 이어지는 다음 트리 깊이까지의 심볼열을 검색하고, 상기 검색된 심볼열에 대한

을 계산한다. In step 213, the detector 111 k is the number of candidate symbols (same as the number of repetitions).

Checks whether K is the number of candidate symbols (the number of repetitions)

If the same as, the detector 111 proceeds to step 229

After the corresponding symbol string is determined as the optimal symbol string, the algorithm according to the present invention is terminated. In contrast, k is the number of candidate symbols (same as the number of repetitions).

If not equal to, the detector 111 retrieves the symbol string up to the next tree depth following the kth candidate symbol in step 215, and searches for the found symbol string.

Calculate

가 상기 계산된

보다 작은지 여부를 검사한다. 상기

가 상기 계산된

보다 작을 시, 상기 검파기(111)는 상기 229단계로 진행한다. 반면, 상기

가 상기 계산된

보다 작지 않을 시, 상기 검파기(111)는 219단계에서 상기 검색된 심볼열이 전체 트리 깊이의 심볼열인지 여부를 검사한다. 여기서, 상기 검색된 심볼열이 전체 트리 깊이의 심볼열이 아닐 시, 상기 검파기(111)는 221단계로 진행하여 상기 검색된 심볼열에 이어지는 다음 트리 깊이까지의 심볼열을 검색하고, 상기 검색된 심볼열에 대한

을 계산한 후, 상기 217단계로 돌아간다. 반면, 상기 검색된 심볼열이 전체 트리 깊이의 심볼열일 시, 상기 검파기(111)는 223단계로 진행하여 상기

를 상기

로 갱신한다. Then, the detector 111 is the step (217)

Calculated above

Check if it is less than remind

Calculated above

If smaller, the detector 111 proceeds to step 229. In contrast,

Calculated above

If not smaller, the detector 111 checks in step 219 whether the found symbol string is a symbol string of the entire tree depth. In this case, when the found symbol string is not a symbol string of the entire tree depth, the detector 111 searches for a symbol string up to the next tree depth following the found symbol string in step 221, and searches for the found symbol string.

After the calculation, the process returns to step 217. On the other hand, when the searched symbol string is a symbol string of the entire tree depth, the detector 111 proceeds to step 223.

Remind

Update to.

가 (k+1)번째 후보 심볼의 브랜치 매트릭보다 작은지 여부를 검사하고, 상기

가 (k+1)번째 후보 심볼의 브랜치 매트릭보다 작을 시, 상기 229단계로 진행한다. 반면, 상기

가 (k+1)번째 후보 심볼의 브랜치 매트릭보다 작지 않을 시, 상기 검파기(111)는 227단계에서 상기 k를 상기 k에 1을 더한 수로 갱신하고, 상기 213단계로 돌아가 이하 단계를 반복 수행한다. Then, the detector 111 is the step (225)

Determines whether is less than the branch metric of the (k + 1) th candidate symbol, and

If is smaller than the branch metric of the (k + 1) th candidate symbol, the flow proceeds to step 229. In contrast,

If is not smaller than the branch metric of the (k + 1) th candidate symbol, the detector 111 updates the k to the number of k plus 1 in step 227 and returns to step 213 to repeat the following steps. .

에 이어지는 전체 트리 깊이의 심볼열(full-length sequence)

을 검색하고, 이렇게 찾아진 최초의 full-length sequence의 누적된 브랜치 매트릭을

이라 할 때, 이 값은 처음에 정한

값보다 작기 때문에 상기

를 상기

로 갱신할 수 있다. 상기 갱신된

와 두 번째 후보 심볼의 브랜치 매트릭 값

와의 비교를 통해, 두 번째 반복을 수행할지 조기에 중단할지 여부를 결정하게 된다. 즉,

이면, 신호 검출 과정을 중단하고, 그렇지 않은 경우는 두 번째 후보 심볼에 이어지는 full-length sequence를 검색하게 된다. 이것은 두 번째 반복에서 검색된 full-length sequence가 첫 번째 반복에서 찾은 full-length sequence보다 더 작은 누적 매트릭 값을 가질 가능성을 배제할 수 없기 때문이다. 한편,

인 경우, 두 번째 반복이 진행이 되는데 full-length가 아니더라도 임의의 검색된 심볼열의 누적된 매트릭 값이

보다 큰 경우가 발생할 수 있다. 이런 경우는 더 이 상의 검색 과정은 의미가 없으므로 신호 검출 과정을 중단한다. 지금까지 설명한 과정에 의거하여 전송 신호에 가장 유사한 full-length sequence를 검색한다. For example, the smallest SED among the candidate symbols ordered through the alignment.

Full-length sequence of full tree depths following

Search for the cumulative branch matrix of the first full-length sequence

This value is set initially

Because it is less than the value

Remind

Can be updated with Updated above

And branch metrics of the second candidate symbol

By comparing with, you decide whether to perform the second iteration or stop it early. In other words,

If so, the signal detection process is stopped, otherwise, the full-length sequence following the second candidate symbol is searched. This is because the possibility that the full-length sequence retrieved in the second iteration has a smaller cumulative metric than the full-length sequence found in the first iteration cannot be excluded. Meanwhile,

If is a second iteration, even if it is not full-length, the accumulated metric value of any found symbol string is

Larger cases may occur. In this case, the further detection process is meaningless, so the signal detection process is stopped. Based on the procedure described so far, the full-length sequence most similar to the transmitted signal is searched.

On the other hand, taking the case of applying the tree search process according to the present invention to the stack algorithm, for example, the average bit error rate (hereinafter referred to as "BER") performance of the prior art and the technique according to the present invention through the experimental results And comparing the computational complexity, it is as shown in Figs. For convenience, the algorithm will be referred to as a stack-based iterative detection (SBID) algorithm.

Here, the experimental environment for the comparison is as shown in Table 1 below.

Modulation technique 16-QAM, 64-QAM Transmit and Receive Antennas (4, 4) channel Rayleigh fading channel Channel state information Perfect algorithm 1.SD (Viterbo-Boutros (VB) algorithm with Babai point) 2.SBD (stack-based detector) 3.SBID (stack-based iterative detector) Stack size M / 4

,

,

) (M-ary QAM,

,

,

) (16-QAM, 10dB, 1.0, 0.015) (64-QAM, 18dB, 1.0, 0.002)

First, FIG. 5 is a diagram illustrating a comparison between the average BER performance of the signal detection algorithm according to the prior art and the present invention. Through this, all average BER performance curves are almost similar. That is, it can be seen that the BER performance similar to the prior art can be maintained.

Next, FIG. 6 shows a comparison between the computational complexity of the signal detection algorithm according to the prior art and the present invention. The algorithm based on the algorithm of the present invention has a computational complexity compared to the SD and SBD techniques according to the prior art. It can be seen that it can be reduced to a considerable level.

Meanwhile, in the detailed description of the present invention, specific embodiments have been described, but various modifications are possible without departing from the scope of the present invention. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined not only by the scope of the following claims, but also by the equivalents of the claims.

1 is a view schematically showing the structure of a MIMO system according to the present invention;

2 is a flowchart illustrating a signal detection method based on iterative tree search in a MIMO system according to an embodiment of the present invention;

3 illustrates a target error rate according to SNR;

값에 따른 평균 BER 성능을 도시한 도면, 4 is

Diagram showing average BER performance by value,

5 is a view showing a comparison between the average BER performance of the signal detection algorithm according to the prior art and the present invention, and

6 shows a comparison of the computational complexity of the signal detection algorithm according to the prior art and the present invention.

Claims (17)

A signal detection method based on iterative tree search in a multi-antenna / user system, Applying a QR decomposition to the channel matrix to form a tree structure according to the modulation order, Determining a number of candidate symbols according to a signal-to-noise ratio (hereinafter, referred to as 'SNR'), Separating the total tree depth into the first tree depth and the rest of the tree depth, Determining as many candidate symbols as the determined number at the first tree depth. The method of claim 1, wherein the determining of the number of candidate symbols according to the SNR comprises: Dividing the entire SNR into a high SNR higher than the first threshold and a low SNR lower than the first threshold according to a first threshold; When the current SNR corresponds to a high SNR, determining the number of candidate symbols to be the same value as the modulation order;  When the current SNR corresponds to the low SNR, the candidate when the lowest component value is smaller than a second threshold value by using a lowest component value of an upper triangular matrix generated by applying a QR decomposition to the channel matrix. The number of symbols is determined to be 1, and if the lowest component value is larger than the second threshold, the candidate symbol number is determined to be the same value as the modulation order. The method of claim 2, Wherein the first threshold value is a value obtained by adding a correction SRP value according to a channel environment to a required SNR value of the MCS applied to the received signal, or the required SNR value. The method of claim 2, The second threshold value is a value calculated using Equation 12 below.

Where

Denotes the overall error probability of the ML signal detection scheme, and

Denotes the overall error probability of the scheme proposed in the present invention. remind

Represents the first threshold, and

Denotes an acceptable error probability. In other words, the second threshold has an acceptable error probability and a specific SNR.

Is the smallest difference between the overall error probability and the ML signal detection method. The method of claim 2, The first threshold value is an SNR value in which the performance difference between the performance of the proposed method and the ML signal detection method for the proximity value of the second threshold value of 1 or less than 1 is remarkable. The method of claim 1, Initializing the cumulative branch metric for a symbol sequence (full-length sequence) to infinity; Ordering the determined candidate symbols in order of smallest Euclidean distance (SED), and defining a branch metric corresponding to the aligned candidate symbols; Selecting one candidate symbol sequentially for all sorted candidate symbols, and calculating a cumulative branch metric value for a symbol string of a total tree depth following the selected candidate symbol; And when the calculated cumulative branch metric value is smaller than a previous cumulative branch metric value, updating the cumulative branch metric value to the calculated cumulative branch metric value. The method of claim 6, And when the cumulative branch metric value is calculated for all of the sorted candidate symbols, determining a symbol string corresponding to the cumulative branch metric value as an optimal symbol string. The method of claim 6, wherein the calculating of the cumulative branch metric value for the symbol string of the entire tree depth following the selected candidate symbol comprises: Calculating a cumulative branch metric value for the symbol string up to the next tree depth following the selected candidate symbol until the symbol string up to the next tree depth is calculated from the symbol string up to the next tree depth following the selected candidate symbol; When the cumulative branch metric value for the symbol string up to the next tree depth is greater than the previous cumulative branch metric value, the cumulative branch metric value calculation process is terminated, and the symbol string corresponding to the previous cumulative branch metric value is optimal symbol. And determining the heat. The method of claim 6, wherein after the cumulative branch metric value update process: When a branch metric value for a candidate symbol arranged next to the selected candidate symbol is larger than the updated cumulative branch metric value, a symbol string of a total tree depth of the selected candidate symbol corresponding to the updated cumulative branch metric value is obtained. And determining the optimal symbol string. A signal detection apparatus based on repetitive tree search in a multi-antenna / user system, A tree structure forming unit for forming a tree structure according to modulation order by applying QR decomposition to the channel matrix; A candidate symbol number determiner that determines the number of candidate symbols according to a signal-to-noise ratio (hereinafter, referred to as a "SNR"); And a candidate symbol determiner for dividing an entire tree depth into a first tree depth and a remaining tree depth, and determining the determined number of candidate symbols from the first tree depth. The method of claim 10, wherein the number of candidate symbols determiner, According to a first threshold, the total SNR is divided into a high SNR higher than the first threshold and a low SNR lower than the first threshold, and when the current SNR corresponds to a high SNR, The number of candidate symbols is determined to be the same as the modulation order, and when the current SNR corresponds to the low SNR, the lowest component value of the upper triangular matrix generated according to the application of QR decomposition to the channel matrix is used. When the lowest component value is smaller than the second threshold, the candidate symbol number is determined to be 1, and when the lowest component value is larger than the second threshold, the candidate symbol number is determined to be the same value as the modulation order. Device. The method of claim 11, wherein Wherein the first threshold value is a required SNR value of an MCS applied to a received signal, or a required constant SNR value plus a correction constant according to a channel environment. The method of claim 11, wherein Wherein the second threshold value is a value calculated using Equation (13).

Where

Denotes the overall error probability of the ML signal detection scheme, and

Denotes the overall error probability of the scheme proposed in the present invention. remind

Represents the first threshold, and

Denotes an acceptable error probability. In other words, the second threshold has an acceptable error probability and a specific SNR.

Is the smallest difference between the overall error probability and the ML signal detection method. The method of claim 11, wherein The first threshold value is an SNR value in which the performance difference between the performance of the method proposed in the present invention and the ML signal detection method with respect to the proximity value of the second threshold is less than 1 or less than 1 is significant. The method of claim 10, An alignment unit for sorting the determined candidate symbols in order of decreasing Euclidean square distance (SED), and defining a branch metric corresponding to the aligned candidate symbols; A cumulative branch metric calculation unit for sequentially selecting one candidate symbol for all sorted candidate symbols and calculating a cumulative branch metric value for a symbol string of a total tree depth following the selected candidate symbol; When the cumulative branch metric value for a full-length sequence is initialized to infinity and the calculated cumulative branch metric value is smaller than a previous cumulative branch metric value, the cumulative branch metric value is converted into the calculated cumulative branch metric value. Apparatus comprising an update unit for updating to. The method of claim 15, After the cumulative branch metric value is updated, when the branch metric value for the next aligned candidate symbol of the selected candidate symbol is larger than the updated cumulative branch metric value, the selected cumulative branch metric value corresponds to the selected cumulative branch metric value. When the symbol string of the total tree depth of a candidate symbol is determined as the optimal symbol string, and the cumulative branch metric value is calculated for all the aligned candidate symbols, the symbol string corresponding to the cumulative branch metric value is an optimal symbol. And an optimum symbol string determination unit for determining a column. The method of claim 15, wherein the cumulative branch metric calculation unit, Compute a cumulative branch metric value for the symbol string up to the next tree depth following the selected candidate symbol until the symbol string up to the next tree depth following the selected candidate symbol is computed. And if the cumulative branch metric value for the symbol string up to the next tree depth is greater than the previous cumulative branch metric value, ending the cumulative branch metric calculation process.

Images