The present invention relates to a MIMO receiving method, and particularly to an MIMO receiving method particularly using QR decomposition-maximum likelihood detection (MLD) in a receiver employing multi-input multi-output (MIMO) in a radio communication. According to the invention, the performance closer to that of the MLD in which throughput is large can be realized by the QR decomposition-MLD which is an easy processing.
In the radio communication, multi-input multi-output (MIMO) using a plurality of antennas is used. In the MIMO, respective different signals are transmitted from a plurality of transmitter antennas at the same time, and a signal combined in space is received by a plurality of receiver antennas. The received signal is decomposed in a manner of solving an equation to reproduce an original stream.
In IEEE802.16 that is one standards body, a radio system based on an OFDM has been proposed, and a system using the MIMO is defined.
In 3GPP that is another standards body, a radio system based on orthogonal frequency division multiplexing (OFDM) has been proposed as long term evolution (LTE), and a system using the MIMO is defined.
Similarly, in a CDMA system, there is a tendency to define a system related to the MIMO.
Even in standardization such as 802.16m or LTE-Advance assuming the fourth generation, the MIMO of 4×4 or more has been proposed according to a requirement, and a reduction in signal throughput and pursuit of performance are continuously required.
In a method of solving the MIMO, there has been known a minimum mean squared error (MMSE) obtaining log-likelihood ratio (LLR) after space separation has been conducted in advance. Assuming Gaussian noise, likelihood is represented by a distance between a receiving point and a replica in a code space. In general, it is conceivable that noise is, for example, thermal noise applied by a receiver during amplification or interference from another communication. A digital communication is intended to transmit information of 0 or 1 by code, in which likelihood is representative of a probability (speciousness) that 0 or 1 determined at the receiver side is assumed to be transmitted. A ratio (likelihood ratio) of a probability P0 that 0 is assumed to be transmitted to a probability P1 that 1 is assumed to be transmitted in which P1 is a denominator can be replaced with probability information that if the likelihood ratio is larger than 1, 0 would be probably transmitted as a transmission code, or if the likelihood ratio is smaller than 1, 1 would be probably transmitted as the transmission code. In a Gaussian distribution, a probability distribution is represented by an exponential to the above distance. Accordingly, there has been known that the likelihood can be evaluated by only treatable product-sum operation with execution of logarithmic arithmetic on the likelihood. The operation result is called “log likelihood ratio”. A positive value of the log likelihood ratio represents that the probability that 0 is assumed to be transmitted is higher. Conversely, a negative value of the log likelihood ratio represents that the probability that 1 is assumed to be transmitted is higher. Certainty that 0 or 1 has been received is higher as an absolute value of positive or negative values is higher, and used as an input when conducting decoding of soft decision. In decoding the receive signal of the MIMO, a method in which the likelihood of soft decision is evaluated on all of codes without conducting space separation in advance, and a transmission line is estimated by a decoder is called “maximum likelihood decision (MLD)”.
However, the MLD is required to calculate the likelihood with respect to all of replicas. This means a process in which taking the combinations of all patterns where respective information is 0 or 1 into consideration, all of the replicas corresponding to the combinations are generated, a distance between the receiving point and each replica is calculated, and the likelihood of each information is computed with execution of the probability operation. Accordingly, there has been known that the amount of computation is factorially increased when the number of candidate replicas is large such as an increase in the number of antennas, or 64 quadrature amplitude modulations (QAM).
In order to solve the problem on the amount of computation, a method called “QR decomposition-MLD” has been introduced in, for example, Non Patent Literature 1. The QR decomposition-MLD means a method of conducting pre-operation in which a channel matrix is subjected to QR decomposition to provide an upper triangular matrix. For example, in a configuration of 2×2 transmitter/receiver antennas, four terms (h11, h12, h21, h22) appear in the channel matrix, and the receiving point is affected by respective two codes transmitted at the same time For example, when the respective antennas transmit transmission codes in a quadrature phase shift keying (QPSK) having four kinds of code points, the 4×4=16 kinds of replicas occur. Since the number of receiver antennas is two, there is required a process of calculating the 16×2=32 kinds of replicas, and calculating a distance to the receiving point. When a transmission code is 64 QAM, there are 64 code point candidates for each transmitter antenna. Therefore, 64×64×2=8192 kinds of distance calculations occur, and the calculations become enormous. In the QR decomposition-MLD, the channel matrix is subjected to QR decomposition to reduce the number of transmitter antennas involved in the signals received by the respective receiver antennas, resulting in a reduction in the amount of computation. Also, in the QR decomposition-MLD, metrics calculated at the time when the number of involved antennas is small are ranked, candidate points are narrowed, and the amount of subsequent computation is largely reduced. Attention is paid to the QR decomposition-MLD, particularly, as a method in which the performance deterioration can be suppressed while remarkably reducing the amount of calculation when the number of antennas is increased.
As described in the above conventional art, in the MIMO, there have been known the MMSE in which the performance is deteriorated, but the signal throughput is small, and the MLD in which the performance is high, but the signal throughput is large. Also, there has been known the QR decomposition-MLD that suppresses the performance deterioration while reducing the signal throughput. However, even in the QR decomposition-MLD, a code error ratio may be increased when a specific condition is met.
In view of the above, the present invention aims at preventing the performance deterioration of the QR decomposition-MLD while suppressing an increase in required throughput. For that reason, in the present invention, for example, it is detected whether the condition in which the QR decomposition-MLD is deteriorated is met, or not, and full MLD is implemented only when it is detected that the condition is met.
The above problem can be solved by an MIMO receiving system employing the QR decomposition-MLD. The MIMO receiving system includes: a step 1 of subjecting a receive channel matrix of N×N, which is obtained from N or more antennas, to QR decomposition to provide an upper triangular matrix for each symbol of the receive signal; a step 2 of extracting an M-th submatrix of the obtained receive channel matrix after the QR decomposition, and calculating candidate metrics of selectable replicas for the submatrix; a step 3 of ranking the metrics calculated in the step 2 in an increasing order when selecting a subsequent submatrix; and a step 4 of removing K-th and subsequent replicas having lower evaluation in the ranking from the candidates of the subsequent submatrixes. In the MIMO receiving system, if the largest metric obtained in the ranking of the step 3 is smaller than a specific threshold value, the above step 4 is bypassed, and the candidate of the replica is not selected.
Also, the above problem can be solved by the above MIMO receiving system in which an average value of the largest metrics obtained in the step 3 is obtained, and a value obtained by multiplying the average value by a predetermined coefficient is set as the threshold value.
Also, the above problem can be solved by the above MIMO receiving system in which the average value or the threshold value calculated previously is accumulated in an accumulator, and the accumulated value is used.
Also, the above problem can be solved by an MIMO receiving system employing the QR decomposition-MLD. The MIMO receiving system includes: a step 1 of subjecting a receive channel matrix of N×N, which is obtained from N or more antennas, to QR decomposition to provide an upper triangular matrix for each symbol of the receive signal; a step 2 of extracting an M-th submatrix of the obtained receive channel matrix after the QR decomposition, and calculating candidate metrics of selectable replicas for the submatrix; a step 3 of ranking the metrics calculated in the step 2 in an increasing order when selecting a subsequent submatrix; and a step 4 of removing K-th and subsequent replicas having lower evaluation in the ranking from the candidates of the subsequent submatrixes. In the MIMO receiving system, if the largest metric obtained in the ranking of the step 3 is smaller than a specific threshold value, a log likelihood ratio of an appropriate symbol is set to 0.
Also, the above problem can be solved by the above MIMO receiving system in which an average value of the largest metrics obtained in the step 3 is obtained, and a value obtained by multiplying the average value by a predetermined coefficient is set as the threshold value.
Also, the above problem can be solved by the above MIMO receiving system in which the average value or the threshold value calculated previously is accumulated in an accumulator, and the accumulated value is used.
Also, the above problem can be solved by an MIMO receiving system employing the QR decomposition-MLD. The MIMO receiving system includes: a step 1 of subjecting a receive channel matrix of N×N, which is obtained from N or more antennas, to QR decomposition to provide an upper triangular matrix for each symbol of the receive signal; a step 2 of extracting an M-th submatrix of the obtained receive channel matrix after the QR decomposition, and calculating candidate metrics of selectable replicas for the submatrix; a step 3 of ranking the metrics calculated in the step 2 in an increasing order when selecting a subsequent submatrix; and a step 4 of removing K-th and subsequent replicas having lower evaluation in the ranking from the candidates of the subsequent submatrixes. In the MIMO receiving system, degeneracy is detected from the channel matrix of an appropriate symbol, and if the degeneracy is detected, the above step 4 is bypassed, and the candidate of the replica is not selected.
Also, the above problem can be solved by an MIMO receiving system employing the QR decomposition-MLD. The MIMO receiving system includes: a step 1 of subjecting a receive channel matrix of N×N, which is obtained from N or more antennas, to QR decomposition to provide an upper triangular matrix for each symbol of the receive signal; a step 2 of extracting an M-th submatrix of the obtained receive channel matrix after the QR decomposition, and calculating candidate metrics of selectable replicas for the submatrix; a step 3 of ranking the metrics calculated in the step 2 in an increasing order when selecting a subsequent submatrix; and a step 4 of removing K-th and subsequent replicas having lower evaluation in the ranking from the candidates of the subsequent submatrixes. In the MIMO receiving system, degeneracy is detected from the channel matrix of an appropriate symbol, and if the degeneracy is detected, a log likelihood ratio of an appropriate symbol is set to 0.
According to the first means for solving of the present invention, there is provided an MIMO receiving method employing a QR decomposition-MLD, the method comprising:
a step 1 of subjecting a channel matrix of N×N, which is obtained from N (N is an integer of two or more) or more antennas, to QR decomposition to provide an upper triangular matrix for each symbol of a received signal;
a step 2 of extracting an M-th submatrix of the obtained channel matrix after the QR decomposition with an initial value of M as N, and calculating candidate metrics of selectable replicas for the submatrix;
a step 3 of ranking the metrics calculated in the step 2 in an increasing order;
a step 4 of removing predetermined K-th and subsequent replicas having lower evaluation in the ranking from the candidates of the subsequent submatrixes when the largest metric obtained in the ranking of the step 3 is larger than a predetermined specific threshold value;
a step 5 of decrementing M by 1, and repeating the step 2, the step 3, and the step 4 until M=1; and
a step 6 of bypassing the step 4 and shifting to the step 5 without selecting the candidate of the replica when the largest metric obtained in the ranking of the step 3 is smaller than the predetermined specific threshold value.
According to the second means for solving of the present invention, there is provided an MIMO receiving method employing a QR decomposition-MLD, the method comprising:
a step 1 of subjecting a channel matrix of N×N, which is obtained from N (N is an integer of two or more) or more antennas, to QR decomposition to provide an upper triangular matrix for each symbol of a received signal;
a step 2 of extracting an M-th submatrix of the obtained channel matrix after the QR decomposition with an initial value of M as N, and calculating candidate metrics of selectable replicas for the submatrix;
a step 3 of ranking the metrics calculated in the step 2 in an increasing order;
a step 4 of removing predetermined K-th and subsequent replicas having lower evaluation in the ranking from the candidates of the subsequent submatrixes when the largest metric obtained in the ranking of the step 3 is larger than a predetermined specific threshold value;
a step 5 of decrementing M by 1, and repeating the step 2, the step 3, and the step 4 until M=1; and
a step 6 of setting a log likelihood ratio of an appropriate symbol to zero when the largest metric obtained in the ranking of the step 3 is smaller than the predetermined specific threshold value.
According to the third means for solving of the present invention, there is provided an MIMO receiving method employing a QR decomposition-MLD, the method comprising:
a step 1 of subjecting a channel matrix of N×N, which is obtained from N (N is an integer of two or more) or more antennas, to QR decomposition to provide an upper triangular matrix for each symbol of a received signal;
a step 2 of extracting an M-th submatrix of the obtained channel matrix after the QR decomposition with an initial value of M as N, and calculating candidate metrics of selectable replicas for the submatrix;
a step 3 of ranking the metrics calculated in the step 2 in an increasing order;
a step 4 of detecting degeneracy from the channel matrix of an appropriate symbol, and removing predetermined K-th and subsequent replicas having lower evaluation in the ranking from the candidates of the subsequent submatrixes when the degeneracy is not detected;
a step 5 of decrementing M by 1, and repeating the step 2, the step 3, and the step 4 until M=1; and
a step 6 of detecting degeneracy from the channel matrix of the appropriate symbol, bypassing the step 4 and shifting to the step 5 without selecting the candidate of the replica when the degeneracy is detected.
According to the fourth means for solving of the present invention, there is provided an MIMO receiving method employing a QR decomposition-MLD, the method comprising:
a step 1 of subjecting a channel matrix of N×N, which is obtained from N (N is an integer of two or more) or more antennas, to QR decomposition to provide an upper triangular matrix for each symbol of a received signal;
a step 2 of extracting an M-th submatrix of the obtained channel matrix after the QR decomposition with an initial value of M as N, and calculating candidate metrics of selectable replicas for the submatrix;
a step 3 of ranking the metrics calculated in the step 2 in an increasing order;
a step 4 of detecting degeneracy from the channel matrix of an appropriate symbol, and removing predetermined K-th and subsequent replicas having lower evaluation in the ranking from the candidates of the subsequent submatrixes when the degeneracy is not detected;
a step 5 of decrementing M by 1, and repeating the step 2, the step 3, and the step 4 until M=1; and a step 6 of detecting degeneracy from the channel matrix of the appropriate symbol, and setting a log likelihood ratio of the appropriate symbol to zero when the degeneracy is detected.
According to the present invention, the performance of the QR decomposition-MLD can be improved, and the performance closer to the full MLD can be realized without largely increasing the amount of computation. In the present invention, for example, in the QR decomposition-MLD, it is detected whether the condition in which the performance is deteriorated is met, or not, and full MLD is implemented only when it is detected that the condition is met, with the result that the above advantages can be realized.
As described above, in a radio communication, multi-input multi-output (MIMO) using a plurality of antennas is employed. In the MIMO, signals different from each other are transmitted from a plurality of transmitter antennas at the same time, and a signal combined in space is received by a plurality of receiver antennas. The received signal is decomposed in a manner of solving an equation to reproduce an original stream.
In a method of solving the MIMO, there has been known a minimum mean squared error (MMSE) obtaining log-likelihood ratio (LLR) of bit after space separation has been conducted in advance, with the use of the estimated channel matrix. Also, a method of conducting the space separation called “maximum likelihood decision (MLD)” in combination with the likelihood calculation at the same time has been known as a derivation of an optimum solution. However, the MLD is required to calculate metric calculation for all of replicas (distance calculation between a receiving point and the replica: calculation related to the likelihood of a candidate transmission code). There has been known that the amount of computation is factorially increased when the number of candidate replicas is large such as an increase in the number of antennas, or 64 QAM. In order to solve the above problem on the amount of computation, a method called “QR decomposition-MLD” has been known.
The QR decomposition-MLD indicates a method in which a channel matrix is subjected to QR decomposition to provide an upper triangular matrix, the likelihood is calculated with the use of the partial matrix, and replicas are ranked according to the likelihood calculation results to narrow the candidate points. Attention is paid to the QR decomposition-MLD, particularly, as a method in which the performance deterioration can be suppressed while remarkably reducing the amount of calculation when the number of antennas is increased. However, similarly, in the QR decomposition-MLD, a code error ratio may be increased when a specific condition is met. In the present invention and the embodiments, it is detected whether the condition in which the QR decomposition-MLD is deteriorated is met, or not, and full MLD is implemented only when it is detected that the condition is met, with the result that the performance deterioration of the QR decomposition-MLD can be prevented while suppressing an increase in the amount of computation as required.
A flow of the QR decomposition-MLD will be described with reference to
The respective steps of each flowchart are executed by the MLD processor 115 or the baseband chip 142. Hereinafter, the respective steps will be described.
In
The respective transmitted codes (signals) pass through a propagation channel (for example,
where x is the receive signal, s is the transmit signal, h is a channel representative of the propagation channel, and n is a noise power. In this example, because the receiver receives the signals by the two antennas, the receive signal is expressed by a two-dimensional vector. Because the transmitter also transmit the signals by the two antennas, the transmit signal is expressed by a two-dimensional vector. Symbol h that is the propagation channels represents four channels from the two antennas to the two antennas, and are expressed by a matrix of 2×2. Because the noise mainly includes thermal noise of the receiver, the noise is expressed by a vector added to each of the two antennas of the receiver. In order to generate the estimated receiving point, there is a need to estimate the above propagation channel h. For that reason, the transmitter transmits a signal obtained by embedding the pilot signal which is known information in an appropriate symbol. The receiver detects the pilot signal to estimate the propagation channel. In a time or a frequency where there is no pilot, the propagation channel can be estimated by interpolating the result of the propagation channel estimation conducted with the symbol having the pilot signal. As a result, the receiver can estimate the channel matrix expressed by H in Expression 1.
In Step 302 of
When it is assumed that a first term starts from the left of Expression 2, a vector Y of the first term represents a converted receive signal. A conversion equation is a second term which is obtained by multiplying the vector X of the receive signal by a conversion matrix G. In the term, G is a transformation operator that realizes the upper triangular matrix which is not limited to a unique operator but various operators may be conceivable. For example, a Givens rotation matrix has been also known as one of the transformation operators that realize the upper triangular matrix. A fourth term represents that GH is converted into R through the operator G. That is, when the respective terms of Expression 1 are multiplied by G, since GX=GHS+GN is satisfied, R=GH is met as compared with Expression 2. In this expression, the feature of R resides in that an element r21 the left side of the second expression is 0. Because of this format, the channel matrix is called “upper triangular”.
where * is complex conjugate.
With the use of Expression 3, R=GH can be rewritten as follows.
In this step, with the use of Expression 3 as one example, operation for obtaining Y in Expression 2 and R in Expression 4 is implemented.
The processing is shifted to Step 303 in
y
2
=r
22
s
2
+ñ
2 (Ex. 5)
In the second expression, terms related to s1 are erased by the upper triangular. The constellation is concentrated in four points as indicated by • in
This is computed with respect to all candidates of R2. The metric represents the probability likelihood, that is, probabilistic certainty (in this example, the likelihood is higher as the metric is smaller). The metric can be also obtained by using an appropriate index corresponding to a distance between the estimated receive point and the real receive point, which is obtained by the receive signal, for example.
The processing is shifted to Step 304 in
y
1
=r
11
s
1
+r
12
s
2
+ñ
1 [Ex. 7]
The processing is shifted to Step 305 in
In
The processing is returned to Step 303 (second) in
where attention needs to be paid to a fact that the candidates of R2 are narrowed in Step S305. When it is assumed that R2 is narrowed to only the point (11), R2 has only one candidate. Therefore, the number of combinations of (R1, R2) is only four, and the amount of calculation is reduced to ¼.
The processing is shifted to Step 304 in
The processing is shifted to Step 306 in
When it is assumed that transmit information (s1, s2) is divided into bit information, and expressed as four bits such as ((b0, b1) (b2, b3)) each bit means that attention is paid to one bit among those four bits. For example, when attention is paid to only the bit of b1, all of eight combinations of the other bits (b0, b2, b3) are taken into consideration, and the probabilities for P0 and P1 are calculated. Because it is heavy to calculate the probabilities for eight kinds of combinations, for example, MAX log MAP approximation has been well known for the purpose of reducing the amount of calculation. This is a method in which although the 8 kinds of combinations should be originally taken into consideration, only the combinations of bits where the metric becomes smallest are selected, and P0 or P1 is approximated with the probability of the bit combinations. As other algorithms, for example, sphere decoding and sequential Gaussian approximation (SGA) have been also known, and appropriate algorithms can be used.
If it is assumed that noise has the Gaussian distribution, the probability is expressed as exp(−x2). A part of x2 in this expression corresponds to the metric calculated up to this time. Accordingly, the likelihood ratio can not only obtain an advantage that P0/P1 is simply replaced with a difference such as log(P0)−log(P1), but also can eliminate the operation of exp required for calculation of P0 or P1, through a logarithmic arithmetic. Consequently, the log likelihood ratio is obtained by selecting, when it is assumed that a bit to which attention is paid is 0 or 1, a combination in which the metric is smallest from all the combinations of the other bits, and calculating a difference between log (P0) and log (P1) with the use of a fact that the minimum metric becomes log (P0) or log (P1). This operation is conducted on all of the four bits.
The four log likelihood ratios corresponding to the obtained four bits are real numbers of positive or negative values. This means an index indicating information that the probability that 0 is conceivably transmitted is higher if the real number is a positive value, and means that the transmission information of 0 is more probable as the positive value is larger. Conversely, this is an index meaning information that a probability that 1 is conceivably transmitted is higher if the real number is a negative value, and means that the transmission information of 1 is more probable as the negative value is smaller. In the above example, the obtained log likelihood ratio is accumulated in a memory or the like as positive or negative real numbers in turn for each four bits.
Hereinafter, the above description will be supplemented with one specific example.
It is assumed that
In this situation, let us consider a first bit of s1. When it is assumed that (s1, s2)=(“0x”,“xx”) where x is arbitrary is transmitted, as P0, eight kinds of combinations in total including two kinds of combinations in s1 and four kinds of combinations in s2 illustrated in
Likewise, P1 is calculated. Eight kinds of replicas as y1 and four kinds of replicas as y2 can be created as indicated in black circles of
The log likelihood ratio is represented as follows.
In this example, through Expression 11, the log likelihood ratio is positive, and b0 bit indicates information that the probability that 0 has been transmitted is higher. Likewise, the log likelihood ratio is calculated for each of the bits b1, b2, and b3.
The processing is shifted to Step 307 in
In the QR decomposition-MLD, when the channel matrix is close to degeneracy, or when noise of an appropriate symbol is increased, the metric operation result related to Expression 6 becomes small wholly, and in this case, the performance may be deteriorated. When degeneracy is conducted, for example, in the constellation of y2 in
The degeneracy is, for example, a condition for satisfying the following expression in Expression 1, and a condition in which an equation consisting of two expressions is not solved.
h
11
h
22
−h
12
h
21=0
For that reason, it is determined to meet the above condition, and in that case, the selection of the candidates conducted in Step 305 is stopped to obtain a performance closer to that of the MLD.
Hence, a flow of the QR decomposition-MLD described above with the use of
A difference between an embodiment of
Step 304 in
The processing is shifted to Step 402 in
If the processing is shifted to Step 403 in
In Step 404 of
With the above correction, a case in which the performance is deteriorated by the QR decomposition-MLD is predicted, and the processing can be conducted as the full MLD. In most cases, because the processing is conducted as the QR decomposition-MLD, an increase in the amount of computation can be also suppressed by about several times at a maximum. Hence, the problem is solved.
Incidentally, the threshold value may be determined in advance or can be created on the basis of a calculation result. Because there has been known that degeneracy or a status in which a noise level is high occurs in only a specific symbol, the threshold value that can detect it needs to be calculated. Specifically, the threshold value can be created on the basis of the last ranked metric in conducting the past operation such as a foregoing subframe or OFDM symbol.
That is,
Also,
In the first embodiment, the description is given of a novel algorithm that allows the symbol to operate as the full MLD when the last ranked metric becomes the threshold value or lower. However, as the symbol operates as the full MLD, the amount of computation is increased by several times. Under the circumstances, there is a method in which the log likelihood ratio of the symbol is set to 0 (symbol not positive and not negative and having no information) originally assuming that the error correction operates. The occurrence of the code error is originally caused by provision of a step in which, for example, although a specific symbol is degenerated from the propagation status, and low-ranked, and the reliability is remarkably reduced, the symbol is subjected to QR decomposition to forcedly decide a transmit symbol of a specific antenna. Despite the symbol of no reliability, the correction of the incorrect result becomes difficult, which is problematic. Therefore, the degenerated and low-ranked symbol is set to 0 without calculation of the log likelihood ratio (information indicating that the probability that the transmission information is 0 is high if the ratio is plus, and indicating that the probability that the transmission information is 1 is high if the ratio is minus) , and does not affect the computation of the other bits. This enhances the performance.
With the above operation, occurrence of an error caused by the error propagation can be prevented by the QR decomposition-MLD without operating the full MLD. Hence, the problem can be solved.
As in the first embodiment, in the second embodiment, as a method of creating the threshold value, the last ranked metrics are averaged as illustrated in Step 801 of
Also,
Also,
According to the present invention, particularly in a cellular communication based on an OFDMA, the performance of the QR decomposition-MLD can be improved. An increase in throughput required at this time can be suppressed to be small.
Filing Document | Filing Date | Country | Kind | 371c Date |
---|---|---|---|---|
PCT/JP2009/054796 | 3/12/2009 | WO | 00 | 12/27/2011 |