JP4675255B2 - Multi-user detection device - Google Patents

Description

The present invention relates to a digital radio communication, to a multi-user detector in particular 3GPP (3 ^rd Generation Partnership Project) conform to the TDD communications system receiver.

It called TDD (Time Division Duplex) using TD-CDMA (Time Division-Code Division Multiple Access) or TD-SCDMA (Time Division-Synchronous Code Division Multiple Access) as a recovery signal method in the Universal Mobile Telecommunication system (UMTS) There is a method. In these UMTS systems that use TDD (hereinafter referred to as UMTS-TDD), for the purpose of reducing inter-cell interference and the like, specifications that operate after securing frame timing synchronization between base stations are defined as options ( For example, refer nonpatent literature 1).

  The 3GPP standard assumes that the same timing at the same frequency can be shared by up to 16 users on the premise of the frame timing synchronization, but joint detection (or joint estimation) is a method for detecting each user individually. There is a technology called.

  Joint detection is a known multi-user detection technique, and it is known that good performance can be obtained. However, application to a WCDMA system is difficult due to the problem of spreading code length and asynchronous operation. However, in UMTS-TDD, since the transmission / reception timing of each user is synchronized and the number of users is limited, it is realistic to implement a multi-user detection technique by joint detection.

  The joint detection is realized by an equalizer called BLE (Block Liner Equalizer). As a specific implementation method, a system corresponding to convolution of transmission paths of all users and user identification codes using Cholesky decomposition. A method for calculating an inverse matrix (Cholesky decomposition method), a method for diagonalizing a system matrix using a DFT (Decrement Fourier Transform) matrix and performing equalization processing in the frequency domain (block FFT method), etc. Is known (see Non-Patent Document 2).

  According to Non-Patent Document 2, when the two methods are compared, the block FFT method has an advantage that the processing amount is smaller than the Cholesky decomposition method.

Next, the block FFT method will be described.
The system matrix corresponding to the convolution of the transmission paths and user identification codes of all users is C _{(P, K)} , the number of users is K, and the impulse response length is P (value is uniquely determined by the number of users. If the response length becomes smaller (Ref. 3: 3GPP TS 25.221)), the system matrix is DP x DK ("DP row DK column" means the same, and so on) It becomes a block cyclic matrix. The subscript _{(P, K)} of the system matrix C _{(P, K)} indicates that the matrix C is DP × DK.

Since the system matrix C _{(P, K)} is a cyclic matrix, it can be rewritten into the following quadratic form using the DFT matrix D _(n) . Note that the DFT matrix D _(n) is a matrix of (D × n) rows and (D × n) columns, and n is the number of symbols, and can take K or P, as already established.

ただし、Ｄ_(n)はＤｎ×ＤｎのブロックＤＦＴ行列であり、次式で定義される。

Here, D _(n) is a Dn × Dn block DFT matrix and is defined by the following equation.

ただし、Ｉ_(n)はｎ×ｎの単位行列（ｉｄｅｎｔｉｔｙ ｍａｔｒｉｘ）、ＤはＤ×ＤのＤＦＴ行列であるクロネッカ積である。

Here, I _(n) is an n × n identity matrix, and D is a Kronecker product, which is a D × D DFT matrix.

The frequency characteristic matrix Λ _{(P, K)} is calculated by block Fourier transform of the first block sequence of C _{(P, K)} .

なお、Ｃ_(P,K)（：，１：Ｋ）は行列Ｃの第一ブロック列（先頭のＫ列）を意味しており、周波数特性行列ΛはＰ×Ｋのブロックサイズをもつブロック行列Λ_d（１≦ｄ≦Ｄ）で構成されるＤＰ×ＤＫの行列となる。

C _{(P, K)} (:, 1: K) means the first block column (first K column) of the matrix C, and the frequency characteristic matrix Λ is a block matrix having a block size of P × K. This is a DP × DK matrix composed of Λ _d (1 ≦ d ≦ D).

  Next, when the received signal sequence is x and the estimated transmission sequence is d, the following equation is established from Equation 1.

従って推定送信系列は次式で求めることができる。

Therefore, the estimated transmission sequence can be obtained by the following equation.

つまり、周波数特性行列Λ_(P,K)が計算できれば、Ｐ個のＤポイントＦＦＴと周波数特性行列Λ_(P,K)の逆行列演算、およびＫ個のＤポイントＩＦＦＴにより、ｄを解くことができる。

In other words, if the frequency characteristic matrix Λ _{(P, K)} can be calculated, d can be solved by performing inverse matrix operation of _P D-point FFT and frequency characteristic matrix Λ _{(P, K)} and K D-point IFFT. it can.

  FFT, which is frequently used in the block FFT method, is a very general and highly useful signal processing. Therefore, mounting in FPGA, LSI, etc. is common, and there is a high possibility that existing design assets can be used. Also, when the hardware is configured using a DSP, the FFT operation is a process mainly including a product-sum operation that the DSP is good at, so that high-speed processing is possible.

  However, on the other hand, the Cholesky decomposition process necessary for the inverse matrix operation includes square root operation and division, which makes hardware implementation difficult.

  Next, FIG. 3 shows a hardware configuration example for realizing the block FFT method realized by Equations 1 to 5, and the processing procedure will be described. The functional block shown in the upper part of FIG. 3 has a function of reproducing the estimated transmission sequence d of each user. That is, the received signal is input to the frequency domain equalization circuit 11 and the propagation path estimation unit 12. The frequency domain equalization circuit includes a D-point FFT operation unit 11-1, a matrix multiplication unit 11-2, and a D-point IFFT operation unit 11-3. It has a function of reproducing the estimated transmission sequence d.

On the other hand, the functional block shown in the lower part of FIG. 3 has a function of calculating and outputting a correction value of a frequency characteristic used in the frequency domain equalization circuit using the above-described received signal.
First, to estimate the channel characteristics in the propagation path estimation unit 12, the calculation of the frequency characteristic matrix lambda _{(P, K)} in the frequency characteristic estimation unit 13 based on the result, the inverse of the frequency characteristic matrix lambda _{(P, K)} Perform matrix operation calculation. The propagation path estimation can be realized using a known technique such as using a known sequence such as a mid ensemble.

The frequency characteristic estimator 13 that performs the inverse matrix operation of the frequency characteristic matrix Λ _{(P, K)} includes the system matrix head K column extractor 13-1, D point as three functional blocks for realizing the operation of Equation 3. It is comprised from the FFT calculating part 13-2 and the inverse matrix calculating part 13-3, and a series of calculation results are output from an inverse matrix calculating part.

In the inverse matrix calculation unit, the frequency characteristic matrix Λ _{(P, K)} is a block diagonal matrix, and there are many elements having “0” as a component. Therefore, the inverse matrix calculation is performed for the entire DP × DP matrix. do not have to. If the number of users is 4 or more, it is practical to apply Cholesky decomposition or the like to the inverse matrix operation. If the number of users is 1 or 2, it is preferable to directly calculate the inverse matrix.

  Although it is necessary to determine the number of FFT / IFFT points in the frequency domain equalization circuit based on the balance between the calculation accuracy and the number of data to be decoded, there is a research report that an FFT operation of 16 points or more is desirable from the viewpoint of calculation accuracy.

  Next, consider a case where the number of FFT points is examined in consideration of the number of data to be handled.

  In TD-SCDMA, there are 44 data symbols in a burst, and 22 data symbols are transmitted in a state of being divided into the latter half and the first half with the midamble in between. Therefore, if 32-point FFT is applied, the decoding process is performed in two steps before and after, so that each process in the first half is completed, and there is an advantage that there is no increase in overhead due to the FFT division process.

  As described above, in TD-SCDMA, it is optimal to apply batch processing by applying 32-point FFT.

  According to the conventional method, when it is assumed that 32-point FFT is adopted, the number of FFT points used in Equation 3 and Equation 5 needs to be unified.

  Next, FIG. 4 shows the amount of calculation required for the complex FFT calculation, and the relationship between the FFT calculation amount and the number of FFT points will be described. This calculation amount is a value using a five butterfly routine in which the radix of the FFT calculation is 2, the complex number multiplication is calculated by four real number multiplications and two real number additions, and all unnecessary multiplications are removed.

  As is clear from FIG. 4, as the number of FFT points increases, the amount of computation increases exponentially. In other words, reducing the number of FFT points has a great effect on the amount of processing.

  The number of FFT points affects not only the processing amount of the FFT calculation itself in the multi-user detection processing by the block FFT method but also the processing amount of the inverse matrix calculation unit of the frequency characteristic estimation unit. The proportion of the above may exceed 50% depending on the mounting method.

  In the above description, the case of TD-SCDMA has been described as an example. Next, the case of TD-CDMA will be briefly described. TD-CDMA has a plurality of frame configuration specifications. For example, in the case of frame TYPE1 and spreading factor 16, the number of data of 1/2 burst is 65. However, the optimum number of FFT points at this time is not 128 points, and it is more advantageous in terms of processing amount to execute an FFT of about 32 points by dividing it multiple times. Therefore, the number of FFT points suitable for processing for correcting frequency characteristics together with TD-CDMA and TD-SCDMA is about 32.

3GPP TS 25.402 M.M. Volmer, et. al. “Comparative Study of Joint-Detection Technologies for TD-CDMA Based Mobile Radio Systems” IEEE Journal on Selected area in Communications. 19, NO. 8, aug. 2001

  The amount of processing required for multi-user detection by the block FFT method realized by Equations 1 to 5 is very large. Furthermore, the amount of processing tends to increase exponentially with the increase in the number of FFT points and the number of users, and there is a problem in terms of hardware scale and power consumption in mounting.

  An object of the present invention is to provide a UMTS-TDD multi-user detection apparatus based on the block FFT method that can reduce the processing amount while minimizing the performance degradation defined by BER or the like.

To achieve the above object, the present invention is, TD-CDMA (Time Division- Code Division Multiple Access) using TDD (Time Division Duplex) as condensate signal system in the universal mobile communication system (UMTS) or TD- In a multi-user detection device of a reception signal processing unit of a wireless communication system using time division-synchronous code division multiple access (SCDMA), a frequency domain equalization unit (frequency domain equalization circuit) for correcting a reception signal in the frequency domain, a channel estimation unit which estimates a channel condition output based on the known sequence in the received signal, based on the propagation path state obtained by the propagation path estimator, a propagation path of each user After estimating the frequency characteristics of a combination of the user identification code, an inverse characteristic of the frequency characteristic of the propagation path is calculated, and outputs the calculation result as a correction value for the frequency domain, the frequency domain equalization unit frequency A frequency estimation unit, and sets the number of FFT points applied by the frequency domain equalization unit to m times the number of FFT points applied by the frequency characteristic estimation unit (m is a power of 2), and the frequency characteristic estimation The unit interpolates the calculated frequency characteristic data by m times, thereby multiplying the frequency characteristic data number by m times, and using the frequency characteristic data obtained by multiplying the number by m times to calculate a frequency domain correction value, It outputs to a frequency domain equalization part, It is characterized by the above-mentioned.

  According to the present invention, it is possible to provide a UMTS-TDD multi-user detection apparatus based on the block FFT method that can reduce the processing amount while minimizing performance degradation.

Embodiments of the present invention will be described below. FIG. 1 is a block diagram showing an embodiment of the present invention. Since multi-user detection is a function implemented in the receiving unit of the wireless device, this description will be made on the receiving unit.
In FIG. 1, the received signal is input to a frequency domain equalization circuit 21 and a propagation path estimation unit 22.

  The frequency domain equalization circuit 21 includes a D1 point FFT calculation unit 21-1, a matrix multiplication unit 21-2, and a D1 point IFFT calculation unit 21-3. The frequency domain equalization circuit 21 executes frequency domain equalization expressed by Equation 5, It has a function of reproducing the estimated transmission sequence d of each user.

On the other hand, the functional block shown in the lower part of FIG. 1 estimates the channel characteristics in the channel estimation unit 22 based on the received signal, and the frequency characteristic matrix Λ _{(P, K)} in the frequency characteristic estimation unit 23 based on the result. And a configuration for performing the inverse matrix operation calculation of Λ _{(P, K)} .

The frequency characteristic estimation unit 23 that performs an inverse matrix calculation of the frequency characteristic matrix Λ _{(P, K)} includes a system matrix head K column extraction unit 23-1, a D2 point FFT calculation unit 23-2, and an FFT calculation result interpolation unit 23-3. The matrix calculation unit 23-4 outputs a series of calculation results to the matrix multiplication unit 21-2 from the matrix inverse calculation unit 23-4.

  In the conventional multi-user detection apparatus based on the block FFT method shown in FIG. 3, the number of FFT / IFFT calculation points of the D point FFT calculation unit 1 constituting the frequency domain equalization circuit 11 is set to D, and D of the frequency characteristic estimation unit 13 is set. The same value is applied, assuming that the number of FFT calculation points of the point FFT calculation unit 13-2 is D. In the present invention, each of the D1-point FFT calculation unit 21-1 and the D2-point FFT calculation unit 23-2 The number of FFT points is set to D1 and D2, respectively, and their relationship is set to D2 = D1 / m (m is a power of 2), thereby reducing the amount of calculation.

  For example, when D1 = 32 and m = 2, D2 = 16. At this time, the number of FFT points in the D2 point FFT calculation unit 23-2 is set to 16.

  However, in this case, there is a mismatch in the order of the matrix input to the multiplication process in the matrix multiplier 21-2 in the frequency domain equalization circuit. Therefore, in order to avoid this problem, an FFT operation is performed to match the matrix size. A result interpolation unit 23-3 is provided, and the calculation results are interpolated to match the calculation orders.

Various known methods can be applied as the interpolation method. In particular, when the zero-order interpolation is applied as the interpolation method, the processing of the inverse matrix calculator 23-4 can be simplified. This is because according to the 0th-order interpolation, the two adjacent FFT calculation results are interpolated as the same value in the FFT calculation result interpolation unit 23-3, so that if D2 = 32, m = 2, and D1 = 16, The D1 (= 32) Λ _d constituting the frequency characteristic matrix Λ _{(P, K)} have a relationship of Λ ₁ = Λ ₂ , Λ ₃ = Λ ₄ ,..., Λ ₃₁ = Λ ₃₂ . Therefore, it is sufficient to execute the inverse matrix calculation by half, and the inverse matrix calculation for the interpolation is unnecessary.

  FIG. 2 shows a BER characteristic simulation result when the multi-user detection circuit according to the present embodiment is mounted. The simulation presupposes a simple block fading model in which a two-pass model in which the power sum is constant and the level fluctuation of each path is fixed at the beginning of the frame. The frame configuration conforms to the 3GPP TD-SCDMA specification, but error correction is not performed. In addition, 0th-order interpolation having a large calculation amount reduction effect is applied to the interpolation processing of the FFT calculation result interpolation unit.

  The graph is a plot of simulation results with the number of FFT points when calculating the frequency characteristic matrix in the calculation of Equation 3, that is, the number of FFT points D2 of the D2 point FFT operation unit 23-2 in FIG. It is. On the other hand, the FFT point number D1 of the D1 point FFT operation unit 21-1 is set to a fixed value 32. The value of the legend K indicates the number of users, the legend (FFT4) is D2 = 4, the legend (FFT8) is D2 = 8, the legend (FFT16) is D2 = 16, and the legend (FFT32) is D2 = 32. The legend (FFT 32) data corresponds to the conventional method.

  From this result, it can be confirmed that there is almost no characteristic deterioration due to D2 = 16 in the present invention. Also, depending on the system specifications, it may be acceptable to set D2 = 8. However, the characteristic deterioration is large when D2 = 4. Although the results are not shown here, since there is no frequency characteristic in a non-multipath environment, there is no characteristic difference between D2 = 4 and D2 = 32.

  As shown in FIG. 4, the processing amount required for the FFT operation can be greatly reduced by reducing the number of points. When the 32-point FFT operation is replaced with a 16-point FFT (D2 = 16), it is expressed by Equation 3. The amount of processing required to calculate the frequency characteristics of the propagation path is less than half.

  Furthermore, only when the interpolation processing of the FFT calculation result interpolation unit is zero-order interpolation, the processing amount related to the inverse matrix calculation of Equation 5 can be significantly reduced. Since the inverse matrix processing of Expression 5 occupies a dominant position in the entire processing of the block FFT method, the effect of reducing the processing amount when using the zero-order interpolation is great. Specifically, the multiplication and addition processing required for multiuser detection can be reduced to about 70%. Further, the division and square root operations required when the inverse matrix operation is realized by Cholesky decomposition can be 50%, respectively. Moreover, as described above, the characteristic deterioration when the present patent is mounted is slight.

  Also, D2 = 8 can be selected if the number of users can be limited or the permissible range of system specifications can be expanded. In this case, multiplication and addition processing in the processing amount required for multi-user detection can be reduced to 50% or less, and division and square root calculation can be compressed to 25%, respectively.

  According to the present invention, it is possible to reduce the processing amount for the propagation path estimation result by allowing a slight deterioration of the BER characteristic, and it is possible to reduce the hardware scale of the received signal processing unit and reduce power consumption. Play.

It is a block diagram which shows embodiment of this invention. It is a graph which shows the effect of this Embodiment. It is a block diagram which shows a conventional apparatus. It is the figure which showed the relationship between the number of FFT points, and the amount of calculations.

Explanation of symbols

  21 frequency domain equalization circuit, 21-1 D1 point FFT operation unit, 21-2 matrix multiplication unit, 21-3 D1 point IFFT operation unit, 22 propagation path estimation unit, 23 frequency characteristic estimation unit, 23-1 system matrix head K column extraction unit, 23-2 D2 point FFT calculation unit, 23-3 FFT calculation result interpolation unit, 23-4 inverse matrix calculation unit.

Claims (1)

In multi-user detection apparatus of the reception signal processing section of a radio communication system using a TD-CDMA or TD-SCDMA using TDD as recovery signal system in the universal mobile communication system,
A frequency domain equalizer for correcting the received signal in the frequency domain;
A propagation path estimator that estimates and outputs a propagation path condition based on a known sequence in the received signal;
Based on the propagation path state obtained by the propagation path estimation unit, after estimating a frequency characteristic of a combination of the channel and the user identification code of each user, and calculates the inverse characteristic of the frequency characteristic of the propagation path, A frequency characteristic estimation unit that outputs the calculation result as a frequency domain correction value to the frequency domain equalization unit,
While setting the number of FFT points applied in the frequency domain equalization unit to m times the number of FFT points applied in the frequency characteristic estimation unit (m is a power of 2),
The frequency characteristic estimation unit interpolates the calculated frequency characteristic data m times to multiply the frequency characteristic data number by m, and uses the frequency characteristic data obtained by multiplying the number m times to calculate a frequency domain correction value. Output to the frequency domain equalization unit.

Images