JP2012175280A - Radio reception device and soft determination value generation method therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a radio reception device and a soft determination value generation method therefor, which are capable of accomplishing highly accurate soft determination error correction decoding by suppressing the influence of a large metric value while maintaining the influence of the magnitude relationship of a transmission bit string on a LLR calculation.SOLUTION: A radio reception device includes: a soft determination output part 22 for generating a soft determination value from a reception signal received via MIMO communication; and a soft determination error correction decoding part for error-correcting/decoding by a soft determination value from the soft determination output part 22. The soft determination output part 22 has a LLR calculation part 226 for subtracting a minimum metric value against a bit-1 from a minimum metric value against a bit-0, dividing the result of the aforementioned subtraction by a metric value that is a larger of the metric value against bit-0 or bit-1, thereby generating a soft determination value and outputting the value to the soft determination error correction decoding part 23, when calculating a logarithmic likelihood ratio based on a candidate list χ of candidates selected from among all candidates of transmission signal vectors on the basis of the reception signal.

Description

  The present invention relates to a radio reception apparatus that performs MIMO (Multiple-Input Multiple-Output) decoding and a soft decision value generation method thereof.

  As a technique for increasing the transmission capacity without expanding the use frequency bandwidth, a MIMO communication system in which a wireless transmission / reception apparatus includes a plurality of antennas has been attracting attention as an elemental technique of a high-speed wireless communication system, and has been adopted as an international standard IEEE802.11n. Is also adopted.

  In the MIMO communication method, radio waves from a plurality of transmission antennas of a wireless transmission device are combined in a communication path space and received by the wireless reception device, so that interference between transmission radio waves is removed and the original independent transmission data Needs to be decrypted. At that time, it is required to suppress the influence of noise as much as possible and reduce errors in the decoded data.

The linear decoding method based on ZF (Zero Forcing) and MMSE (Minimum Mean Square Error) is the most basic method as a MIMO decoding method and can be realized with a small amount of computation, but the decoding error rate characteristic is different from other methods. Inferior. There is MLD (Maximum Likelihood Detection) for searching for the maximum likelihood candidate from all the candidate patterns of the transmission signal sequence as a method for obtaining the optimum decoding result, but the amount of calculation required for searching for the maximum likelihood candidate becomes enormous. Therefore, it is difficult to implement in an actual communication system.
Therefore, LRA detection (LRA (Lattice Reduction Aided) -Detection) method, which can obtain a decoding result close to MLD with a relatively low amount of computation by combining a linear decoding method with a lattice reduction (Lattice Reduction) method, has attracted attention. Yes.

  However, in a communication system, an error correction code is used to reduce the data error rate, and a bit string that has been subjected to error correction coding is decoded after MIMO decoding. In this LRA detection method, the decoding result is a hard decision value. Therefore, a high error correction effect cannot be obtained. Therefore, error correction decoding using a soft decision value is more preferable than error correction decoding using a hard decision value because a higher error correction effect can be obtained.

  Therefore, in order to further improve the data error rate, recently, an LRA detection method for soft decision value output that enables the application of soft decision error correction decoding to the LRA detection method has been studied and proposed (for example, patents). Reference 1).

  In the wireless communication device of Patent Literature 1, transmission is completed by applying lattice basis reduction to a channel estimation value, equalizing a received signal according to a basis reduction reference channel, and selecting a set of candidate vectors based on the basis reduction reference By determining the probability that a bit has a special value, determining a transmitted symbol vector corresponding to each candidate vector, and determining the probability that each transmitted bit value will be transmitted based on the received signal, It is described that a soft estimated value (soft decision value) of a transmitted bit value is determined from a received signal.

Here, an example of a soft decision value generation method by a conventional radio reception apparatus will be described with reference to FIG.
For example, the candidate list shown in FIG. 5A shows correspondence between all transmission bit string candidates and metric values when the number of transmission antennas is two and the number of transmission bits per transmission signal is two bits. The transmission bit string candidates are arranged in ascending order based on the metric value. However, the noise power is equally constant for each bit.

What is used for the calculation of LLR (Log-Likelihood Ratio) is the minimum metric value when the bit is 0 and 1 in the transmission signal candidate vector of each bit in the candidate list. Is the minimum metric value. At this time, if all candidates in the candidate list are targeted for LLR calculation, an optimum result can be obtained.
For example, for bit b _1,1 , the minimum metric value when the bit is 1 is 0.36, and the minimum metric value when the bit is 0 is 7.33. The difference becomes LLR, but when normalized on the basis of LLR (b _1,2 ) so as to facilitate explanation, the following equation (1) is obtained.

JP 2007-74729 A

A.K.Lenstra, H.W.Lenstra, Jr., L.Lovasz, "Factoring Polynomials with Rational Coefficents,", Math.Ann., Vol.261, no.4,1982, p.515-534 Y.H.Gan, W.H.Mow, "Complex Lattice Reduction Algorithm for Low-Complexity MIMO Detection,", Proc.IEEE GLOBECOM 2005, vol.5, St. Louis, USA, Dec.2005, p.2953-2957 V. Ponnampalam, D. McNamara, A. Lillie, M. Sandell, "On Generating Soft Outputs for Lattice-Reduction-Aided Detectors,", Proc. IEEE ICC2007, Glasgow, Scotland, Jun. 2007, p, 4144-4149

However, a problem may arise when a candidate list is created by selecting a small number of candidates by a predetermined method in order to reduce the amount of LLR computation.
For example, when only the candidates shaded from the candidate list shown in FIG. 5A are obtained by a predetermined method and the LLR is calculated as nine candidates shown in FIG. As shown. Similar to equation (1), equation (2) is also normalized based on LLR (b _1,2 ).

Comparing Equations (1) and (2), in Equation (2), which is calculated using the candidate list of decimal candidates shown in FIG. 5B, in the LLR calculation of transmission bits b _1,2 , Among all the transmission bit string candidates shown in FIG. 5A, although there is still a small metric value, it is located fifth from the bottom of the candidate list of all transmission bit strings in FIG. 5A. Since a large metric value is used, the LLR of the other bits for LLR (b _1,2 ) is ½ or less compared to Equation (1). As a result, the bit likelihood information is lost, which adversely affects soft decision error correction decoding.

  In view of this, the present invention provides a radio receiving apparatus and a soft receiver capable of high-accuracy soft decision error correction decoding by suppressing the influence of a large metric value while maintaining the influence of the magnitude relationship of transmission bit strings in the LLR calculation. It is an object to provide a determination value generation method.

The radio reception apparatus of the present invention includes a soft decision output unit that generates a soft decision value from a received signal by MIMO communication, and a soft decision error correction decoding unit that performs MIMO decoding using the soft decision value from the soft decision output unit, The soft decision output unit calculates a log-likelihood ratio based on a candidate list based on candidates selected based on a received signal from among all transmission signal vector candidates, and calculates a minimum metric value for bit 0. LLR operation for subtracting the minimum metric value for bit 1 and dividing the result based on the larger metric value of either bit 0 or bit 1 to generate a soft decision value and output it to the soft decision error correction decoding unit It has the part.
The method for generating a soft decision value of a radio reception apparatus according to the present invention includes a step of generating a candidate list based on candidates selected based on a received signal from among all transmission signal vector candidates in MIMO communication, and a candidate list Subtracting the minimum metric value for bit 1 from the minimum metric value for bit 0 and dividing the result by the larger metric value for either bit 0 or bit 1 Generating a determination value.

  The present invention provides a soft decision value for error correction decoding when calculating a log-likelihood ratio based on a candidate list based on a candidate selected based on a received signal from among all transmission signal vector candidates. Is calculated by subtracting the minimum metric value for bit 1 from the minimum metric value for bit 0 and dividing the result by the larger metric value of either bit 0 or bit 1. By doing so, it is possible to suppress the influence of a large metric value while preserving the magnitude relationship of the log likelihood ratio of each transmission bit.

It is desirable to provide a fixed perturbation unit that generates a candidate list to be used when the LLR calculation unit calculates the log likelihood ratio from candidates of transmission signal vectors by a fixed perturbation method.
When generating a candidate list based on candidates selected from all transmission signal vector candidates, the candidate list can be generated by a fixed perturbation method.

  The channel response matrix is transformed by the transformation matrix, and a reduced basis matrix, which is a quasi-orthogonal matrix to be regarded as a new channel response matrix, is input to perform MMSE detection (MMSE (Minimum Mean Square Error) Detection), and the received signal vector An MMSE detection unit that calculates a signal vector multiplied by a weight matrix, and a determination unit that determines a candidate point having the minimum Euclidean distance based on the signal vector from the MMSE detection unit and obtains a hard decision signal, The fixed perturbation unit preferably generates a transmission signal vector candidate when generating the candidate list based on the hard decision signal from the decision unit. When the MMSE detection unit and the determination unit generate a hard decision signal, noise enhancement is suppressed and decoding accuracy is improved by using an LRA-MMSE detection method in which MMSE detection is performed using a reduced basis matrix as a new channel response matrix. Can do.

  It is preferable to include a lattice base reduction unit that calculates a new channel response matrix to the MMSE detection unit as a reduced basis matrix and calculates a transformation matrix using the channel response matrix as a quasi-orthogonal matrix by an LLL algorithm. Since the soft decision value generated by the LLR calculation unit preserves the magnitude relationship of the log likelihood ratio of each transmission bit, the influence of a large metric value can be suppressed, so that the lattice basis reduction unit uses the LLL algorithm to generate a reduced basis matrix. A high accuracy can be ensured with a small number of repetitions in the calculation.

  When the LLR calculation target bit of the candidate list does not include either the bit 0 or the bit 1, the soft decision value output unit calculates the minimum metric value for the bit in the log likelihood ratio calculation. By calculating using the maximum metric value in the candidate list and generating a soft decision value, even if one bit is not included in the candidate list, the magnitude relationship of the log likelihood ratio of each transmission bit is saved. However, the influence of a large metric value can be suppressed.

  The present invention preserves the magnitude relationship of the log likelihood ratio of each transmission bit by dividing the conventional LLR calculation value by dividing a large metric value of either bit 0 or bit 1 to generate a soft decision value. However, since the influence of the large metric value can be suppressed, the influence of the large metric value can be suppressed while maintaining the influence of the magnitude relationship of the transmission bit strings in the LLR calculation. Therefore, the present invention can achieve soft decision error correction decoding with high accuracy.

It is a figure which shows the radio | wireless communications system (wireless transmitter and radio | wireless receiver) which concerns on embodiment of this invention. It is a figure which shows the structure of the soft decision value output part of the radio | wireless receiver shown in FIG. It is a figure which shows constellation in 16QAM, distance (alpha) between signal points, and offset value (beta). (A) And (B) is a figure which shows an example of the basis vector which extends the same grating | lattice. (A) and (B) are an example table (list) showing a transmission bit string and a metric value corresponding to the transmission bit string, (A) is a list showing all candidates, and (B) is based on a fixed perturbation method. It is a list which shows the candidate produced | generated. It is a graph which shows SNR and BER in Example 1 of this invention. It is a graph which shows the transmission distance characteristic in Example 2 of this invention. It is a graph which shows the SNR and BER when performing the iterative process by the LLL algorithm in Example 2 of this invention, and shows the example using the hard decision value output and the hard decision error correction decoding. It is a graph which shows the SNR and BER when performing the iterative process by the LLL algorithm in Example 2 of this invention, and shows the example using soft decision value output and soft decision error correction decoding.

  A wireless communication system according to an embodiment of the present invention will be described with reference to the drawings. Note that FIG. 1 does not show a modulation / demodulation unit, a transmission radio unit that converts a transmission signal (electric signal) into a radio signal (radio wave), a reception radio unit that converts a radio signal into a reception signal, or the like. .

  As shown in FIG. 1, the wireless communication system 1 includes a wireless transmission device 10 and a wireless reception device 20. The wireless transmission device 10 and the wireless reception device 20 are wireless signals by two transmission antennas 11 a and 11 b provided in the wireless transmission device 10 and two reception antennas 21 a and 21 b provided in the wireless reception device 20. Are sending and receiving.

The wireless transmission device 10 is provided with an error correction encoding unit 12 that adds an error correction code to transmission data.
The wireless reception device 20 is provided with a soft decision value output unit 22 and a soft decision error correction decoding unit 23. The soft decision value output unit 22 outputs a soft decision value based on the received signal. The soft decision error correction decoding unit 23 performs error correction based on the soft decision value from the soft decision value output unit 22 and outputs received data.

Next, the configuration of the soft decision value output unit 22 will be described with reference to FIG.
The soft decision value output unit 22 includes a channel estimation unit 221, a lattice base reduction unit 222, an MMSE detection unit 223, a determination unit 224, a fixed perturbation unit 225, and an LLR calculation unit 226. Yes.
The channel estimation unit 221 calculates a channel response matrix H from the radio signal from the radio transmission device 10 and outputs it to the lattice basis reduction unit 222.
The lattice basis reduction unit 222 calculates a reduced basis matrix H ′ from the received signal vector and the channel response matrix H, outputs the reduced basis matrix H ′ to the MMSE detection unit 223, and converts the channel response matrix H into a reduced basis matrix H ′. T is output to determination unit 224 and fixed perturbation unit 225.
The MMSE detection unit 223 outputs a signal vector z obtained by multiplying the received signal vector by the weight matrix W _{H ′} .
The determination unit 224 determines the signal vector z from the MMSE detection unit 223 as a candidate point (transmission signal candidate point) with the minimum Euclidean distance, thereby _obtaining the hard determination signal Z _rep and outputs it to the fixed perturbation unit 225. .
The fixed perturbation unit 225 creates and outputs a transmission bit string candidate list χ for the transmission signal vector based on the hard decision signal Z _rep (hard decision value) from the MMSE detection unit 223.
The LLR calculation unit 225 calculates the LLR from the transmission bit string candidate list χ and outputs the LLR to the soft decision error correction decoding unit 23 as a soft decision value.

The operation of the wireless communication system according to the embodiment of the present invention configured as described above will be described with reference to the drawings. In the present embodiment, a QAM (Quadrature Amplitude Modulation) method generally used in the MIMO communication method will be described as an example.
In the QAM system, information is transmitted by changing the amplitude and phase values of a signal. Therefore, a transmission symbol subjected to QAM modulation is represented on a complex plane as shown in FIG. The real part is the in-phase component and the imaginary part is the quadrature component. A transmission constellation of 2 ^Q -QAM modulation having information of Q bits per symbol is generally expressed by Expression (3). However, set A is a set of transmission signal candidate points, vector s is a transmission signal vector, complex number R (•) is a real part, complex number J is an imaginary part, and α and β are distances between adjacent signal points of the transmission constellation. And offset values (see FIG. 3).

In MIMO communication, transmission is performed in parallel from a plurality of transmission antennas (transmission antennas 11a and 11b in FIG. 1). The transmission signal vector at this time can be represented by a vector s = [s ₁ , s ₂ ,..., S _M ] ^T. However, M is the number of transmitting antennas, and element s _k ∈ set A (1 ≦ K ≦ M). The transmission bit string corresponding to the transmitted signal _{vector, [b 1,1, b 1,2,} ···, b 1, Q, b 2,1, b 2,2, ···, b M, Q ].

In MIMO communication, since radio waves from a plurality of transmission antennas are combined in a communication path space, the relationship between transmission and reception signals in a flat fading environment is as follows: transmission signal vector s (M × 1), reception signal vector y (N × 1). ), Channel response matrix H (N × M), and noise signal vector n (N × 1), it is expressed by Equation (4). N is the number of receiving antennas.

  This equation (4) indicates that each element of the MIMO reception signal vector y is a composite signal of each element of the transmission signal vector s. Further, the off-diagonal element of the channel response matrix H represents a component that becomes an interference signal. The radio reception apparatus 20 (MIMO receiver) performs processing for removing the interference signal component and decoding the original transmission data.

First, the channel estimation unit 221 first estimates a channel from a pilot signal included in a radio signal from the radio transmission apparatus 10 and outputs a channel response matrix H.
Next, LRA-MMSE detection (Lattice Reduction Aided MMSE Detection) is performed by the lattice basis reduction unit 222 and the MMSE detection unit 223.
The lattice basis reduction unit 222 converts the channel response matrix H from the following equation (5) into a reduced basis matrix (quasi-orthogonal matrix) H ′ by lattice basis reduction, and outputs the reduced matrix to the MMSE detection unit 223 and the channel response matrix H. Is converted to a reduced basis matrix H ′ and output to the fixed perturbation unit 225 and the determination unit 224.

Here, the reduced basis matrix H ′ and the transformation matrix T calculated by the lattice basis reduction unit 222 will be described in detail.
The lattice basis reduction is, for example, a reduced basis vector having a same norm from a set of vectors {h ₁ , h ₂ ,... H _M } and having a smaller norm and high orthogonality. It shows that {h ′ ₁ , h ′ ₂ ,... H ′ _M } is calculated to obtain a reduced basis.
Here, the lattice is defined by Expression (6) for linearly independent vectors h ₁ , h ₂ ... H _M ∈ set R ^N (N ≦ M). Here, the set ^RN is a set made up of all N-dimensional vectors whose elements are real numbers, the set Z is a set made up of whole integers, and L (.) Shows a lattice based on the elements in (.).

It is known that there are an infinite number of basis vectors for one lattice. FIG. 4 shows an example of basis vectors extending the same lattice. The arrows shown in FIGS. 4A and 4B are basis vectors, and the intersections of the straight lines are lattice points. In FIG. 4A and FIG. 4B, the arrangement of lattice points is the same, but the basis vectors are different. The orthogonality of the basis vectors is evaluated by the orthogonality d d (matrix H) according to the equation (7). Here, the matrix H is {h ₁ , h ₂ ,... H _M }, det (·) is the determinant of the matrix (·), and H ^H is the Hermitian transpose of the matrix H.

Non-patent document 1 describes the orthogonality d d (matrix H) represented by equation (7) as an LLL (Lenstra-Lenstra-Lovasz) algorithm that calculates a reduced basis satisfying equation (8) in polynomial time. Has been. However, M is the number of transmitting antennas.

  The constant δ∈ (1/4, 1) gives a trade-off between the degree of reduction of the obtained reduced basis and the amount of computation of the LLL algorithm, and the basis vector with a higher degree of reduction as it is set to a value closer to 1 However, the amount of computation increases. Non-Patent Document 2 describes that the LLL algorithm is extended to a base vector of a complex vector.

For the channel response matrix H, a reduced basis matrix H ′ of the channel response matrix H and a transformation matrix T from H to H ′ satisfying H ′ = HT can be obtained by the LLL algorithm.
Since it is known that bases extending the same lattice are connected by unimodular transformation, the relationship shown in Expression (9) is established.
Therefore, the transformation matrix T calculated by the lattice base reduction unit 222 by the LLL algorithm is a unimodular matrix whose elements are all integers and whose determinant value is ± 1.

Here, the repetition of the lattice base reduction of the LLL algorithm based on Non-Patent Document 2 will be described.
The QR decomposition of the N × M channel response matrix H to which the LLL algorithm is applied is represented by Equation (10).

Non-Patent Document 2 is an algorithm that performs unimodular transformation of a matrix that satisfies both Equation (11) and Equation (12). However, the set R is a set of all real numbers.

The algorithm is composed of the following three steps (a) to (c).
(A) QR decomposition of the matrix H is performed.
(A) The column vector of the matrix R is converted so as to satisfy the expression (11).
(C) If the equation (12) is not satisfied between the continuous column vectors of the matrix R, the columns are exchanged, and the operation returns to step (c).
Thus, (a) and (c) are repeated until the condition of the expression (12) is satisfied. The LLL algorithm is an iterative algorithm, and is repeatedly performed until the stop condition is satisfied. Therefore, the number of iterations varies depending on the value of the constant δ in Expression (12) and the value of the channel response matrix H that is an input matrix.

For example, the lattice base reduction unit 222 may employ a Fixed-LLL algorithm. The Fixed-LLL algorithm is described in Non-Patent Document 3.
This Fixed-LLL algorithm is a modification of the LLL algorithm in a form suitable for hardware implementation. When the number of iterations of the algorithm is limited to a small number, a base with a higher degree of reduction than the LLL algorithm can be obtained. .

  The lattice basis reduction unit 222 obtains a reduced basis matrix H ′ whose degree of reduction has been increased by this iterative process and a conversion matrix T from the channel response matrix H to the reduced basis matrix H ′ that satisfies Equation (5).

Next, MMSE detection by the MMSE detection unit 223 will be described.
MMSE detection is a MIMO decoding method that can be performed with a relatively low amount of computation by linear computation. In this method, interference cancellation is performed by multiplying the received signal vector y by a weighting matrix that maximizes the signal-to-interference noise power ratio, in other words, a weighting matrix represented by Expression (14) that minimizes Expression (13). . Where σ is the average power of noise, and W _H is the MMSE weight matrix for the channel response matrix H.

When the signal is σ ² ≈0, the weight matrix W _H can be regarded as a pseudo inverse matrix H ⁺ = (H ^H H) ⁻¹ H ^H of the channel response matrix H. Hereinafter, it is treated as W _H = H ⁺ . At this time, the signal s after interference cancellation is expressed by Expression (15).

Since the ideal decoded signal is s, it is desirable to suppress the influence of the noise n as much as possible. Therefore, since the magnitude of the noise n depends on the orthogonality of the channel response matrix H, the noise is emphasized and the decoding accuracy is lowered as the orthogonality of the channel response matrix H is lowered simply by performing MMSE detection. Cause.
Therefore, the MMSE detection unit 223 converts the channel response matrix into a quasi-orthogonal matrix, regards this as a new channel response matrix, and detects MMSE, thereby suppressing noise enhancement and improving decoding accuracy.

Here, the MMSE detection performed by the MMSE detection unit 223 will be described in detail.
Equation (4) is rewritten into Equation (16) by lattice basis reduction performed by the lattice basis reduction unit 222 when z = T ⁻¹ s.

When MMSE detection is performed with H ′ regarded as a new channel response matrix, Equation (17) is obtained, and a signal vector z obtained by multiplying the received signal vector by the weight matrix W _{H ′} can be obtained. The signal vector z is output to the determination unit 224.

The determination unit 224 determines this z as a candidate point of z having the minimum Euclidean distance, so that z _rep (estimated value of z = T ⁻¹ s obtained by hard determination of the signal vector z, hereinafter, a hard determination signal) Is calculated). Since H ′ is a quasi-orthogonal matrix, a highly reliable z _rep is obtained, and the decoded signal s _rep is calculated from equation (18) by multiplying this by the transformation matrix T from the lattice base reduction unit 222. .

Here, the transformation matrix T by lattice reduction is a unimodular matrix, and its inverse matrix T ⁻¹ is also a unimodular matrix. When the influence of the offset value β (see FIG. 3) is removed from the transmission constellation shown in Expression (4), the transmission signal vector s and the candidate points of z = T ⁻¹ s are arranged in the same grid.
Since the constellation of the transmission signal vector signal s is known, the determination unit 224 may obtain the z hard decision signal z _rep on the constellation of the transmission signal vector s. The calculation formula is shown in Formula (19). However, “·” indicates rounding to a Gaussian integer with the minimum Euclidean distance for each element of the vector (both real and imaginary parts are integer values), and 1 _M is an M × 1 column vector in which all elements are 1 + j ( j = √−1).

Thus, it is an advantage of using the lattice basis reduction that the transmission signal candidate point can be easily determined. However, in this determination method, when z is rounded to the grid point with the minimum Euclidean distance, there is a possibility that the point is a point outside the candidate point of the hard determination signal z _rep . Therefore, when there is an element that is not included in the transmission constellation in s _rep = Tz _rep , the determination unit 224 determines this element as a candidate point with the minimum Euclidean distance on the constellation of the transmission signal vector s.

Next, generation of a candidate list by the fixed perturbation unit 225 will be described.
Generally, an LLR for each bit is used as an input to the soft decision error correction decoding unit 22. When b _{m, q} is the qth bit of the transmission bit string corresponding to the transmission signal from the mth transmission antenna _, the LLR for b _{m, q} is defined by equation (20).
The LLR value has a larger absolute value in the positive direction as the likelihood when b _{m, q} is 1 is higher, and the absolute value is larger in the negative direction as the likelihood when 0 is higher. Take the value. Assuming that U ⁱ _{m, q} is a set of b _{m, q} = i (iε {0,1}) among all transmission signal vector candidates considered from the transmission constellation, b _{m, q} = The likelihood of i is obtained by equation (21).
In order to reduce the amount of computation, in this embodiment, Max-Log APP (Maximum-Logarithm A Posteriori Probability) represented by Expression (22) is adopted as an approximation of Expression (21).

When the noise is AWGN (Additive White Gaussian Noise) with an average power σ ² , LLR is expressed by Equation (23) using Max-Log APP.

The LLR of each bit as an input of the soft decision error correction decoding unit 23 only needs to have a relative magnitude relationship, and an absolute value does not matter. Therefore, equation (23) can be transformed to equation (24) under constant noise power.
In the case of directly calculating Equation (21) and also in the case of calculating Equation (23) and Equation (24) using Max-Log APP, since the likelihoods of all transmission signal vector candidates are used for the calculation, these are soft. It can be interpreted as an MLD of judgment value output. Therefore, when the number of transmission signal vector candidates is large, the amount of calculation is enormous, and it is difficult to implement in an actual system.

According to Non-Patent Document 3, which is a prior art document related to the fixed perturbation method, based on the hard decision value output of LRA-MMSE detection, a candidate list χ of certain K transmission signal vectors represented by Expression (25) Is created, and the LLR is calculated from the candidates only in the candidate list χ using Max-Log APP.

For this purpose, first, the 4M + 1 fixed perturbation vectors of the equation (26) are added to the z hard decision signal z _rep from the decision unit 224 with the equations (27), respectively, and the vector z (k) is obtained. Generate.

Then, using this, a candidate list χ composed of transmission signal candidate points s ^{(k) is} generated by Expression (28).
However, since the obtained s ^(k) may include a point outside the transmission signal candidate point, such s ^(k) adds the transmission signal candidate point with the minimum Euclidean distance to the candidate list χ instead. .
In this way, the candidate list χ is generated by the fixed perturbation unit 225 and output to the LLR calculation unit 226.

Next, generation of the soft decision value by the LLR calculation unit 226 will be described.
The LLR obtained using the Max-Log APP for the generated candidate list χ is expressed by Equation (29).
However, _{Em, q} (i) is the following formula | equation (30).
If the noise power is constant, Equation (29) can be changed to Equation (31).

What is used for the LLR calculation is the minimum metric value when the bit is 0 and the minimum metric value when the bit is 1 in the transmission signal candidate vector s ^(k) for each bit. The metric value indicates a value that serves as a scale represented by Expression (32) when calculating the LLR.

Therefore, the candidate list χ in the LLR calculation is preferably a set having a metric value as small as possible among all candidates of the transmission signal vector s.
However, in the fixed perturbation method, 4M + 1 transmission signal candidate points s ^(k) having the smallest metric value are not always obtained, and transmission signal candidate points s ^(k) having a large metric value are added to the candidate list χ. Cases arise.
At this time, if the metric value represented by Expression (33) is used or if there is no candidate represented by Expression (34) other than the metric value having such a large value, the metric value having this large value is used for the LLR calculation. As a result, the absolute values of some LLRs become extremely large.

Therefore, at the time of decoding by the soft decision error correction decoding unit 23, the influence of the LLR having a large value becomes dominant, and the LLRs of other bits are hardly reflected in the error correction decoding result. That is, the likelihood information of each bit is lost, which causes a reduction in error correction effect. An example of this can be seen from the LLR calculated by the equations (1) and (2) based on the candidate list of FIG.
In Expression (1), LLR calculation is performed on all candidates shown in FIG. 5A using Max-Log APP. That is, the optimum result is obtained because it conforms to the equation (29).
The fixed perturbation unit 225 of the wireless reception device according to the present embodiment generates a list χ including nine candidates by the fixed perturbation method from the candidate list shown in FIG. 5A (FIG. 5B). reference). By doing so, the amount of calculation can be reduced.
Expression (2) is an LLR of each bit calculated from a list χ composed of nine candidates.
In this way, unlike the equation (1) in which all candidates shown in FIG. 5A are targets of the LLR calculation, the equation (2) for nine candidates using the fixed perturbation method has the bit likelihood. Information will be lost.

In order to suppress such loss of likelihood information, the LLR calculation unit 226 divides by the larger value of the metric values used in the LLR calculation according to the equation (23). This is shown in equation (35).
Similarly, in the equation (35), the equation (36) can be obtained under a condition where the noise power is constant.

When the LLR is calculated from the candidate list χ shown in FIG. 5B based on Expression (36) and normalized, Expression (37) is obtained.

As can be seen from this result, since the influence of a large metric value can be suppressed while preserving the LLR size relationship of each transmission bit, LLRs of transmission bits other than b _1,2 are also prevented from losing likelihood information. Is done.
The soft decision error correction decoding unit 23 performs soft decision error correction decoding based on the LLR calculated by the LLR calculation unit 226 in this way. Since the soft decision error correction decoding unit 23 performs decoding based on the LLR in which the loss of likelihood information is suppressed, a high error correction effect can be obtained.

In the candidate list χ shown in FIG. 5B, the minimum value of bit 1 is 0.36 and the minimum value of bit 0 is 34.13 in transmission bits b _1,2 . In this case, since bit 0 is included, the minimum value of bit 0 is 34.13, and the denominator in equation (36) is also 34.13. However, depending on the candidate list, a candidate of bit 0 may not be included. At that time, the LLR value is calculated by subtracting the minimum metric value of bit 1 from the maximum metric value in the candidate list χ and dividing by the maximum metric value of bit 1 in accordance with the following formula (30). Can do. By doing so, the influence of a large metric value can be suppressed while preserving the magnitude relationship of the log likelihood ratio of each transmission bit even if the candidate list does not include the bit 0 bit.

Example 1
A simulation of the wireless reception apparatus according to the embodiment of the present invention was performed, and BER (Bit Error Rate) characteristics were measured. The simulation compared the conventional method using Equation (29) and the method using Equation (35) of the present invention when calculating the LLR using the fixed perturbation method.
The simulation conditions are shown in Table (1), and the simulation results are shown in FIG.

In FIG. 6, the present invention which is a soft decision output is indicated as (A1) LRA-MMSE SDD Proposed.
Further, as a comparative example, the hard decision value output LRA-MMSE is (B1) LRA-MMSE HDD, the conventional soft decision value output LRA-MMSE is (C1) LRA-MMSE SDD conventional, the soft decision value output MLD (Max-IOg). APP use) is shown as (D1) MLD SDD.
As shown in FIG. 6, in the conventional method shown in (C1), as the SNR (signal to noise ratio) increases, the BER characteristic difference from the case of the hard decision value output shown in (B1) is reduced and further reversed. You can see that In the present invention shown in (A1), it can be seen that a better characteristic than the hard decision value output shown in (B1) is obtained with the SNR of the entire area. In particular, when the BER is 10 ⁻⁴ , a gain of 5 dB is obtained. Further, even when compared with the soft decision value output MLD shown in (D1), there is a difference of about 2 dB when the BER is 10 ⁻⁴ , and the present invention shown in (A1) has a similar BER characteristic. It can be confirmed.

(Example 2)
Next, using a propagation loss model, the transmission distance characteristics in an environment outside the line-of-sight and the performance when the number of iterations of the LLL algorithm for performing lattice base reduction was limited were evaluated. Table 2 shows the simulation parameters for the evaluation.
First, the transmission distance characteristics in a non-line-of-sight environment were simulated. Moreover, the result transmission distance characteristic which is a simulation result is shown in FIG. Also in FIG. 7, the present invention, which is a soft decision output, is (A1) LRA-MMSE SDD Proposed, and as a comparative example, the hard decision value output LRA-MMSE is L (B1) RA-MMSE HDD, The soft decision value output LRA-MMSE is indicated as (C1) LRA-MMSE SDD Conventional. Further, the soft decision value output MMSE is indicated as (E1) MMSE SDD.

7, when the throughput is 500 Mbps, the hard decision value output LRA-MMSE shown in (B1), the soft decision value output LRA-MMSE of the conventional method shown in (C1), and the soft decision value of the present invention shown in (A1). The transmission distances of the output LRA-MMSE were 10.5 m, 12.0 m, and 14.0 m, respectively.
The soft decision value output LRA-MMSE (A1) of the present invention is about 33% with respect to the hard decision value output LRA-MMSE (B1), and about 17 with respect to the soft decision value output LRA-MMSE (C1) of the conventional method. % Transmission distance improvement.

Next, a performance comparison when the number of iterations of the LLL algorithm for performing lattice base reduction is limited will be shown.
For the lattice reduction, the FiX-LLL algorithm is adopted in the present invention and the comparative example. The MMSE detection is a case where the comparative example performs hard decision error correction decoding for hard decision value output LRA-MMSE detection, and the present invention is a case where soft decision error correction decoding is performed for soft decision value output LRA-MMSE detection. is there. The results are shown in FIGS. 8 and 9, respectively. In FIGS. 8 and 9, repetition from 3 (3 loop) to 9 (9 loop) and no limit (No Limit) indicate the number of repetitions of the LLL algorithm when performing LRA-MMSE detection.

First, in the case of using the hard decision value output LRA-MMSE detection shown as (B1) in FIG. 8, when the limit of the number of iterations is 9 (9 loop), it is equivalent to the case of no limit (No Limit). A BER characteristic is obtained, and a gain of about 7 dB can be confirmed at BER 10 ^{−4 as} compared with soft decision value output MMSE detection (MMSE SDD) indicated by (E1). However, when the limit of the number of repetition processing is less than this, the BER characteristic deteriorates, and when the number of limit is 3, the soft decision value output MMSE detection is inferior.

On the other hand, in the case of performing soft decision error correction using the present invention indicated by (A1) in FIG. 9, when the limit of the number of iterations is six (6 loop), the soft decision value output indicated by (E1) is output. Compared to MMSE detection, a gain of about 10 dB can be confirmed at BER10 ⁻⁴ , and a gain of about 13 dB can be confirmed when the limit is 9 times (9 loops).
When considering the hardware implementation of the LLL algorithm, the required number of iterations greatly affects the hardware cost. In the present invention, even when the number of iterations is limited to 6, it is possible to obtain a good BER characteristic of 9 or more iterations of the hard decision value output LRA-MMSE detection, compared to the conventional method. This also contributes to a reduction in hardware cost of the lattice base reduction unit 222 that implements the LLL algorithm.

  The present invention is suitable for a radio reception apparatus that performs MIMO detection and a soft decision value generation method thereof.

DESCRIPTION OF SYMBOLS 1 Wireless communication system 10 Radio transmitter 11a, 11b Transmit antenna 20 Radio receiver 21a, 21b Transmit antenna 22 Soft decision value output part 221 Channel estimation part 222 Lattice base reduction part 223 MMSE detection part 224 Determination part 225 Fixed perturbation part 226 LLR Arithmetic unit 23 Soft decision error correction decoding unit

Claims (6)

A soft decision value output unit for generating a soft decision value from a received signal by MIMO (Multiple-Input Multiple-Output) communication;
A soft decision error correction decoding unit that performs error correction decoding based on a soft decision value from the soft decision output unit;
The soft decision value output unit
When calculating the log likelihood ratio based on a candidate list of candidates selected based on the received signal from among all the transmission signal vector candidates, the minimum metric value for bit 1 is calculated from the minimum metric value for bit 0. And an LLR arithmetic unit that subtracts the result and divides the result by a large metric value of either bit 0 or bit 1 to generate a soft decision value and outputs the soft decision value to the soft decision error correction decoding unit. Wireless receiver.   The radio | wireless of Claim 1 provided with the fixed perturbation part which produces | generates the candidate list used when the said LLR calculating part calculates log likelihood ratio from the candidate of a transmission signal vector by the fixed perturbation method (Fixed Perturbation). Receiver device. The channel response matrix is transformed by the transformation matrix, and a reduced basis matrix, which is a quasi-orthogonal matrix to be regarded as a new channel response matrix, is input to perform MMSE detection (MMSE (Minimum Mean Square Error) Detection), and the received signal vector An MMSE detector that calculates a signal vector multiplied by a weight matrix;
A determination unit that determines a candidate point having a minimum Euclidean distance based on a signal vector from the MMSE detection unit and obtains a hard decision signal;
The radio reception apparatus according to claim 2, wherein the fixed perturbation unit generates a transmission signal vector candidate when generating the candidate list based on a hard decision signal from the determination unit.   The radio reception apparatus according to claim 3, further comprising a lattice basis reduction unit that calculates a reduced basis matrix for the MMSE detection unit by an LLL algorithm.   When the LLR calculation target bit of the candidate list does not include either the bit 0 or the bit 1, the soft decision value output unit calculates the minimum metric value for the bit in the log likelihood ratio calculation. The radio reception apparatus according to claim 1, wherein a soft decision value is generated by performing an operation using a maximum metric value in the candidate list. Generating a candidate list based on candidates selected based on a received signal from among all transmission signal vector candidates in MIMO communication;
In computing the log likelihood ratio based on the candidate list, the minimum metric value for bit 1 is subtracted from the minimum metric value for bit 0, and the result is based on the larger metric value of either bit 0 or bit 1. And a soft decision value generating method for a wireless receiving apparatus, comprising: generating a soft decision value by dividing.