KR100716584B1 - Adaptive k-best detection method for mimo systems - Google Patents

Abstract

의 평균값을 구한다. 이렇게 구해진 평균값으로 각 탐색 레벨별로 해당하는

을 나누어 그 값에 따라 다른 K값을 적용한다. 그 후, 각 탐색 레벨의 전송 가능한 심볼들에 대한 부분 유클리디언 거리 계산이 수행되며 초기화 과정에서 계산된 T 와 그에 대응되는 부분 심볼 벡터 후보들을 최대 K개까지 선택한다. 마지막으로 탐색 레벨을 거듭하면서 다른 K값을 적용하는 과정을 반복하여 최종 심볼이 검출된다. 본 발명은 채널 이득을 고려하기 위해,

의 평균값으로

를 나누는 부분과 나눠진 값의 크기에 따라 각 탐색 레벨별로 다른 K값을 적용해서 심볼을 검출하는 부분으로 구성된다.The present invention relates to an adaptive K-best detection method in a multiple transmit / receive antenna system, and detects maximum-likelihood (ML) that can be used by a receiver when data is transmitted by a spatial multiplexing technique in order to maximize data rate in the multiple transmit / receive antenna system. An adaptive K-best detection method is disclosed, which is based on a method, wherein a calculation performed in advance to perform symbol detection is performed at initialization and corresponds to a channel gain.

Find the average value of. This is the average value obtained for each search level.

Divide by and apply a different K value depending on the value. Then, partial Euclidean distance calculation is performed on the transmittable symbols of each search level, and up to K T and corresponding partial symbol vector candidates calculated in the initialization process are selected. Finally, the final symbol is detected by repeating the process of applying different K values while repeating the search level. In order to consider the channel gain,

As the mean of

It is composed of a part that detects a symbol by applying a different K value for each search level according to the division part and the magnitude of the divided value.

 Multiple transmit / receive antenna system, maximum-likelihood, K-best, error propagation, partial Euclidean distance

Description

Adaptive K-best detection method for MIMO systems

 1 is a flowchart illustrating an adaptive K-best detection method in a multiple transmit / receive antenna system according to the present invention.

Figure 2a is a graph comparing the bit error probability performance by the adaptive k-best detection algorithm according to the present invention and the conventional k-best detection technique.

Figure 2b is a graph comparing the average amount of calculation by the adaptive k-best detection algorithm according to the present invention and the conventional k-best detection technique.

The present invention relates to an adaptive K-best detection method in a multiple transmit / receive antenna system, and more particularly, an ML that can be used at a receiving end when transmitting data using a spatial multiplexing technique in order to maximize data rate in the multiple transmit / receive antenna system. The present invention relates to an adaptive K-best detection method based on a maximum-likelihood detection method.

The detection method based on the ML detection method that can be used at the receiving end of the multiple transmit / receive antenna system in which data transmission is performed using a spatial multiplexing technique includes an ML detection method, an old detection method, and a K-best detection method. Among them, the K-best detection method has a small amount of calculation required for symbol detection compared to the conventional ML detection method, but the maximum calculation amount is fixed unlike the conventional detection method. Therefore, the actual implementation of the detection method based on the ML detection method is performed. It is the most suitable in.

The ML detection method measures the Euclidean distance between the symbol vector received through the receiving antenna and the transmittable symbol vectors and estimates the symbol vector having the smallest value as the transmitted symbol vector. In order to reduce the symbol vector set to be considered in the ML detection technique, the detection is performed by limiting the search range of the symbol vector to which Euclidean distance calculation is performed to a sphere having a specific radius. First, in order to calculate the Euclidean distance in real units, a process of converting the complex matrix y = Hs + n into the real matrix r = Mu + w is performed as follows.

Where s, y, H, and n are symbol vectors transmitted from M _t transmit antennas, symbol vectors received through M _r receive antennas, channel matrices of M _r × M _t dimensions, and a complex Gaussian with an average of zero. The noise vector of the component. From now on, for convenience, 2M _t = 2M _r = n is assumed.

을 만족하는 반경

의 구로 한정시킨다.

라 정의하면

은 의 양자화된 값을 의미한다.As mentioned earlier in the pre-processing of the K-best detection technique, we first determine the range of symbol vectors to consider for Euclidean distance calculation.

Radius to satisfy

To the sphere of.

If you define

Is the quantized value of.

는 R의 각 성분

에 의해 아래 식과 같이 정의된다.Next, the Cholesky factorization via M ^T M the upper triangular matrix R as represented by R ^T R with the image of the new component of the upper triangular matrix, each Q

Each component of R

Is defined as

라 하면 다음과 같은 수식 전개가 이루어진다.In this case, we define a vector e that satisfies e = p-u and calculate each component of the vector.

In this case, the following equation development is performed.

인 범위 내에서

을 추정하기 위해 각 탐색 레벨 별로 K개의 부분 심볼 벡터셋을 유지하게 되는데 이러한 과정에서 부분 유클리디언 거리 계산을 위해 아래의 두 식이 사용된다.K-best detection technique is from the above equation

Within the range of

K partial symbol vectors are maintained for each search level to estimate. In this process, the following two equations are used to calculate the partial Euclidean distance.

으로 설정된다. 또한 (8)식에서

는 i번째 탐색 레벨에서의 부분 유클리디언 거리를 나타내며 초기에

로 설정한다. 케이-베스트 검출 기법에서 전송된 심볼을 검출하는 과정은 다음과 같다.In equation (7) above, P _i represents the i th component of P

Is set. Also in (8)

Denotes the partial Euclidean distance at the i th search level

Set to. The process of detecting the transmitted symbol in the K-best detection technique is as follows.

First to i = n, so 8 by the formula to calculate the partial Euclidean distance for all U _n possible n-th search level K starting with a small value one in the second step, i.e., T _n-1 value is a positive Among the ones having a value of K, K is taken from the larger one, and corresponding candidate symbols are stored. After setting i = i-1 in the second step, use all possible U _n-1 at the n-1 th search level for each of the K candidate symbol sets U _n stored in the first step (8). After the equation is calculated, K is selected from the larger T _n-2 value having a positive value and the corresponding partial symbol vector is stored. Repeating the above method, in the last step, T _o is obtained by applying all possible u _1s in the current step to each of the K partial symbol vector set candidates in the previous step, and then the total that satisfies the largest of these T _O values. The symbol vector is finally detected.

The K-best detection technique employs a sequential detection scheme in which the candidate vector sets selected in the previous search step influence the determination of the partial symbol vector set candidates in the current search step. Therefore, selecting the wrong candidates in the previous step leads to continuous error propagation in determining the candidate vector sets until the final step because the wrong information is transmitted in determining the candidate vector sets in the next step. This error propagation problem causes a performance degradation in terms of bit error probability.

Accordingly, the present invention has been made in view of the above circumstances, and an object of the present invention is to employ a conventional K-best detection technique by applying a different K value for each search level according to channel gain in symbol detection using a K-best detection technique. Compared with the proposed method, the average operation amount is lowered by improving the bit error probability performance and reducing the unnecessary operation required for unnecessary symbol detection.

In the multiple transmit / receive antenna system for achieving the above object, the adaptive K-best detection method comprises the steps of: a) determining the decoding order of each search level and performing calculations for Q, p, C ,

이며,

Is,

은 P의 양자화된 값, C는 유클리디언 거리, 및 M은 채널을 의미하며;Here, Q-Chol is a function of obtaining each component q _ij of Q from M ^T M through Cholesky factorization,

Is the quantized value of P , C is the Euclidean distance, and M is the channel;

,

으로 초기값을 설정하고, q행렬의 대각 성분에 대해 평균값 a를 연산하는 단계,b) i = n,

,

Setting an initial value with and computing the mean value a for the diagonal components of the q matrix,

, 여기서 n은 1, 2, 3, . . .,2M _T 이며;

, Where n is 1, 2, 3,. . ., 2 M _T ;

c) determining the K value based on the value a in step b),

이면 K=c,

이면 K=d,

이면 K=e, 여기서 w, y는 실수이고, c, d, e은 정수이며;

If K = c,

If K = d,

If K = e, where w and y are real and c, d and e are integers;

d) Compute the partial Euclidean distance for all transmittable symbols u _i from the i th search level and store up to K T _i-1 values corresponding to T _i-1 > 0 and corresponding candidate symbol sets Steps,

이며;

Is;

e) determining whether the number Z of the candidate symbol set u _i selected in step d) is 0;

을 검출된 심볼로 결정하고, 아닌 경우, 다음의 i=i-1값에 대한 e_i 및 S_i-1을 연산하고, 상기 단계 c), d), e)의 과정을 반복하는 단계,f) when the value of Z is 0 as a result of the determination in step e),

Determining as a detected symbol, if not, calculating e _i and S _i-1 for the following i = i-1 values and repeating the steps c), d), and e),

; 및

; And

g) if the i value becomes 1 by step f), finally detecting the entire symbol vector u ₁ corresponding to the largest T _O value.

의 평균값을 구한다. 이렇게 구해진 평균값으로 각 탐색 레벨별로 해당하는 q_ii을 나누어 그 값에 따라 다른 K값을 적용한다. In the multiple transmit / receive antenna system according to the present invention, in the adaptive K-best detection method, a calculation performed in advance to perform symbol detection is performed at initialization, and corresponds to a channel gain.

Find the average value of. The average value thus obtained is divided by q _ii for each search level, and a different K value is applied according to the value.

Then, partial Euclidean distance calculation is performed on the transmittable symbols of each search level, and up to K T and corresponding partial symbol vector candidates calculated in the initialization process are selected.

Finally, the final symbol is detected by repeating the process of applying different K values while repeating the search level. The invention comprises, as an average value of q _ii for each search level according to the size of the part and the divided value divides the q _ii a technique for detecting the symbols by applying different K values to account for channel gain.

Existing K - best detection method is because taking only a maximum of K partial symbol vector for each navigation level in the symbol detection process can was a significantly reduces the complexity compared to the ML detection scheme, but a C channel state calculated to determine the search radius In the case of having a large value due to bad or high noise power, candidate vector sets that do not need to be considered in each search step are maintained, and partial Euclidean distance calculations for these unnecessary candidate vector sets are accumulated, which increases the amount of computation.

의 비교를 통해 K개의 후보 심볼셋을 취한다고 가정한다.Taking the first step of the K-best detection technique as an example, the partial Euclidean distance at the C and n th search levels with large values, with reference to Eq. (8)

Suppose we take K candidate symbol sets through the comparison of.

각각의 값들이 모두 C보다 작을 가능성이 크다. In this case, since C is the Euclidean distance between the entire received symbol vector and the entire symbol vector estimated by the ZF method, the partial Euclidean distance is obtained for one component of the vector.

Each value is likely to be less than C.

As a result, it moves to the next stage while maintaining the maximum number of K, including symbols that are not likely to be transmitted as stochastic.

To solve this problem, considering the channel condition, use high value K at the low level gain level and low value K at the high level gain level to reduce the number of unnecessary passes and reduce the amount of computation required. Can be reduced.

Looking at the above description through the formula is as follows.

First, the average value of the diagonal components of the q matrix corresponding to the channel gains at each search level is obtained.

The value obtained by dividing q _ii corresponding to each search level by the average value ( a ) is compared. If the value is less than 0.2, K = 8, and if the value is between 0.2 and 2.0, K = 3, and if K = 1 applies.

The average value of the diagonal components of the matrix is obtained because the K value is changed by looking at the relative channel gain for each search level even if the channel condition is very good or very bad, thereby guaranteeing a constant bit error probability performance and a reduction in computation amount.

Since the bit error probability and the amount of calculation are in a trade-off relationship, the K value that varies according to the range or range of changing the K value can be adjusted in consideration of the required bit error probability performance and the amount of calculation.

The process of applying the adaptive K-best algorithm of the present invention to the conventional K-best detection method is as follows.

Step 1. Initialization Step

The decoding order of each search level is determined and calculations for Q, p, and C are performed.

Step 2. First Step

A.

이면 K=c,

이면 K=d,

이면 K=e 을 적용한다.B.

If K = c,

If K = d,

Then K = e is applied.

C. Perform the following calculations for all possible things.

D. Select up to K from the largest _one among positive T _n-1 and store the corresponding candidate symbol sets.

E. Perform the following calculation for each symbol selected.

Step. 3. Repeat process

A. i = i-1

B. For each of the K partial symbol vectors selected in the previous step, perform the following calculation by substituting all possible u _i in the current step.

C. T _i-1 with a positive value The maximum K is selected from the larger ones and the corresponding partial symbol vector set candidates are stored.

D. Perform the following calculation for each symbol selected.

E. Repeat step 3 until i = 1 to detect the entire symbol vector corresponding to the largest value of the final T _O.

In the K-best detection technique, there is a trade-off relationship between the bit error probability and the amount of calculation according to the K value.

In other words, increasing the value of K increases the number of passes taken to a higher search level, resulting in performance improvements in terms of bit error probability, but at the same time increasing the amount of computation.

Conversely, a smaller value of K means a smaller number of passes to a higher search level, which leads to a performance degradation in the bit error probability of decreasing computation amount.

As described above, the method of applying the adaptive K-best algorithm according to the present invention uses the K value for each search level differently from the conventional K-best detection method in consideration of the effect of the channel gain, so that the average computation amount required for symbol detection is determined. It has a reducing effect.

In addition, the higher K-value is applied at the search level with lower channel gain, which reduces the propagation of error, thereby improving performance in terms of bit error probability.

Hereinafter, with reference to the accompanying drawings will be described the present invention in more detail.

1 is a flowchart illustrating an adaptive K-best detection method in a multiple transmit / receive antenna system according to the present invention.

Referring to FIG. 1, in step S1, an initialization operation is performed. In FIG. 1, "Q-Chol" is a function for outputting each component q _ij of Q from the input M ^T M through Cholesky factorization, and "Quant" is a function for outputting quantized values through the ZF technique.

U _i denotes a partial symbol vector selected from the i th level to the n th level, and Z denotes the number of partial symbol vectors actually selected.

First, assuming that the receiving end perfectly knows the information on the channel M through channel estimation, the received symbol vector r and the initial K value are given as inputs, and in step S1, preprocessing for performing the algorithm is performed (S1). .

Subsequently, an initial radius determination step is performed. In the initial radius determination step, the calculation for the initial radius is performed (S2).

,

으로 초기값이 설정된 후 q행렬의 대각성분에 대해 평균값을 구한다(S3).Then, the channel average gain calculation step is performed. In the channel average gain calculation step,

,

After the initial value is set, the average value of the diagonal components of the q matrix is obtained (S3).

In step S4, the K value changed according to the result of dividing q _ii by the average value a obtained by step S3 is applied to partial symbol detection (S4).

Then, in step S5, the partial Euclidean distance is calculated. In the partial Euclidean distance calculation, the partial Euclidean distance calculation is performed on all transmittable symbols U _n from the nth search level.

를 구하고 초기 반경 C 대신

을 만족하는 최대 K개의 T_n-1값과 그에 대응되는 후보 심볼셋이 저장되고, 단계 S7에서는 선택된

의 개수를 판단해서 0인 경우에는

을 검출된 심볼로 결정하게 되며, 그렇지 않은 경우 단계 S8로 진행되어, i값 판단 단계로 넘어간다.In step S6, the partial Euclidean distance at the th search level in the T _i-1 generation and storing the candidate symbol set.

Find and replace the initial radius C

Up to K T _{n -1} values and candidate symbol sets corresponding thereto are stored, and in step S7,

If the number of times is 0,

Is determined as the detected symbol, otherwise, the process proceeds to step S8, where the process proceeds to the i value determination step.

In step S8, if i is greater than ₁ , the process proceeds to step S8-1. After the calculation for S _i-1 is performed, the value of i is decreased by 1 in step S8-2, and the process goes back to step S4.

을 a로 나누어진 결과 값에 따라 다른 K값을 적용하고, n-1번째 탐색 레벨에서의 부분 유클리디언 거리

의 계산을 통해

를 만족하는 최대 K개의 T _n-2 값과 그에 대응되는 부분 심볼 벡터

들을 저장한다. After that, as in the previous step

Apply a different K value according to the result divided by a, and the partial Euclidean distance at the n- 1th search level.

Through the calculation of

Up to K T _n-2 Values Matching the Partial Symbol Vectors

Save them.

을 검출하게 된다.As described above, if the i value is 1 in step S8 while repeating the selection of the partial symbol vector set by applying a different K value according to the channel gain for each partial symbol detection step, the process moves to the symbol detection step of step S9 to finally obtain the largest T ₀ value. The full symbol vector corresponding to

Will be detected.

2A and 2B show the results of comparing the bit error probability performance and the average computation amount of the adaptive K-best detection technique using the technique of the present invention and the conventional K-best detection technique.

Here, the average amount of calculation is measured by the sum of the number of multiplications required for the partial Euclidean distance calculation, which is the main operation of the K-best detection process.

The experimental environment assumes a multi-antenna system using 4 transmit / receive antennas each and a 16-QAM modulation technique.

In order to compare bit error probability and average amount of computation with the existing 5-best and 6-best detection techniques, the values of w and y are 0.2 and 2.0, and the values of c, d and e are 8, 3 and 1, respectively. 8-3-1 adaptive K-best was used.

Figure 2a shows the bit error probability according to the signal-to-noise ratio. In the case of the 8-3-1 adaptive K-best detection technique, the technique of using a high K at a low channel gain reduces the error propagation and thus reduces the bit error probability. I can confirm that there is.

When the bit error probability of 10 ^-3 is used, a gain of about 1.4 dB occurs when compared to the 5-best method. Also, referring to FIG. 2b comparing the average operation amount, 2 to 23% when compared to 5-best, Compared to 6best, it can be seen that it is about 2 ~ 33% lower.

As described above, the present invention detects a symbol by applying a different K value for each search level, thereby improving performance in terms of bit error probability by reducing the effect of error propagation in the partial search step compared to the conventional K-best detection technique. In addition, in a search level with a large channel gain, a low K is applied to reduce the average amount of computation required for symbol detection.

Although the present invention has been described in detail by the above embodiments, the present invention is not limited thereto, and variations and improvements can be made without departing from the ordinary knowledge of those skilled in the art.

Claims (1)

a) determining the decoding order of each search level and performing calculations on Q, p, C ,

Is, Here, Q-Chol is a function of obtaining each component q _ij of Q from M ^T M through Cholesky factorization,

Is the quantized value of P , C is the Euclidean distance, and M is the channel; b) i = n,

,

Setting an initial value with and computing the mean value a for the diagonal components of the q matrix,

, Where n is 1, 2, 3,. . ., 2 M _T ; c) determining the K value based on the value a in step b),

If K = c,

If K = d,

If K = e, where w and y are real and c, d and e are integers; d) Compute the partial Euclidean distance for all transmittable symbols u _i from the i th search level and store up to K T _i-1 values corresponding to T _i-1 > 0 and corresponding candidate symbol sets Steps,

Is; e) determining whether the number Z of the candidate symbol set u _i selected in step d) is 0; f) when the value of Z is 0 as a result of the determination in step e),

Determining as a detected symbol, if not, calculating e _i and S _i-1 for the following i = i-1 values and repeating the steps c), d), and e),

; And g) if the i value becomes 1 by step f), finally detecting the entire symbol vector u ₁ corresponding to the largest T _O value. Best detection method.

Images