JP2012095322A - Signal decoder and signal decoding method - Google Patents

Abstract

An object of the present invention is to reduce the amount of calculation during signal decoding and reduce processing delay.
A received symbol and an output from a QR decomposing unit are supplied to a multiplying unit, and the received symbol is multiplied by an output from the QR decomposing unit. A plurality of multipliers 21 and QR decomposition units 22 are provided corresponding to the number of subgroups. The multiplier 22 multiplies the conjugate transpose matrix and the received symbol, thereby orthogonalizing the received symbol. The permutation channel matrix from the AGS unit 23 is supplied to the QR decomposition unit 22. The AGS unit 23 is supplied with the channel response matrix from the channel estimation unit 14. Received symbols divided into a plurality of subgroups from the multiplier 21 and orthogonalized are supplied to the local detector 24. The local detector 24 performs signal detection by a hierarchical algorithm (for example, M algorithm) using the fact that each subgroup is orthogonalized.
[Selection] Figure 7

Description

  The present invention relates to a signal decoding apparatus and a signal decoding method that enable high speed and high capacity in, for example, mobile communication.

  In recent years, attention has been focused on MIMO (Multiple Input Multiple Output) in order to improve frequency utilization efficiency in wireless communication. In the next generation mobile radio access system, a maximum transmission rate of several hundred Mbps to several Gbps is being realized. As a technique for realizing such a high transmission rate together with high frequency utilization efficiency, space division multiplexing (SDM) has attracted attention.

  SDM is a transmission method in which a transmitter uses a plurality of antennas to transmit different signals (substreams obtained by dividing a data stream) on the same frequency and spatially multiplex them. By such multiplexing, it is basically possible to obtain a transmission rate proportional to the number of transmission antennas without expanding the use frequency band.

  One of the factors affecting the transmission capability in SDM is a signal detection algorithm in the receiver. This algorithm can be broadly classified into two types: spatial filtering based on ZF (Zero-Forcing) and minimum mean square error (mmse) norms, and maximum likelihood detection (MLD).

  Spatial filtering is spatial filtering with an active array that uses a number (L) of receive antennas greater than the number of transmit antennas.

The MLD will be briefly described. Since the possible values of each substream at a certain time include only the modulation multilevel number, when the modulation multilevel number is S and the number of transmission antennas (the number of substreams) is L, all the substreams are possible. There are S ^L combinations. The probability density calculated when one sub-stream therein is sent to the received reception signal, the largest transmission signal probability density chosen from S ^L Street is MLD.

  SDM and MLD are described in Non-Patent Document 1 below.

IEICE Transactions B Vol.J87-B No.9 pp.1162-1173 September 2004

  A method has been proposed for reducing the amount of calculation for MLD while maintaining transmission capability as much as possible. The present inventor has proposed a signal detection algorithm for separating SDM signals by a plurality of QR decompositions (Multi-QR-Decomposition: Multi-QRD).

  Spatial filtering can be realized with a small amount of calculation, but has a drawback that transmission capability is inferior to that of MLD. On the other hand, MLD has an excellent transmission capability, but has a problem of requiring a huge amount of calculation. As described above, the amount of calculation increases with an index of the number N of transmission antennas.

  Accordingly, an object of the present invention is to provide a signal decoding apparatus and a signal decoding method based on a signal detection algorithm in a spatial multiplexing system that can reduce the amount of calculation and have a small processing delay.

In order to solve the above-described problem, the present invention emits L independent transmission signals obtained from a predetermined information signal from L transmission antennas and mixes the L transmission signals. A signal input unit to which a mixed signal of
A channel estimation unit that performs channel estimation and estimates a channel response matrix of M rows and L columns;
A symbol decoding unit that receives a received symbol and a channel response matrix and detects a transmission signal;
The symbol decoding unit
Including a processing unit that divides the channel response matrix from the channel estimation unit into a plurality of different permutation channel matrices to form subgroups, and performs parallel processing for each subgroup,
The processing unit is supplied with the QR decomposition processing unit for the permutation channel matrix, the multiplication unit that multiplies the received symbol by the conjugate transpose matrix and orthogonalizes the received symbol, and the output of the multiplication unit, and detects the signal for each subgroup. And a local detector that performs
In the multiplication unit, the signal decoding apparatus multiplies received symbols of a plurality of subgroups together with a conjugate transpose matrix.

Further, according to the present invention, L transmission signals independent from each other obtained from a predetermined information signal are emitted from L transmission antennas, and M mixed signals obtained by mixing the L transmission signals are input. A signal input step;
A channel estimation step of performing channel estimation and estimating a channel response matrix of M rows and L columns;
A symbol decoding step for receiving a received symbol and a channel response matrix and detecting a transmission signal;
The symbol decoding step consists of
Dividing the channel response matrix into a plurality of different permutation channel matrices to form subgroups, and performing parallel processing for each subgroup,
The processing steps include a QR decomposition processing step for the permutation channel matrix, a multiplication step for multiplying the received symbol by a conjugate transpose matrix and orthogonalizing the received symbol, and an output of the multiplication step, and signal detection for each subgroup. And a local detection step that performs
In the multiplication step, the received symbols of a plurality of subgroups are collectively multiplied by a conjugate transpose matrix.

  In the present invention, a received signal is divided into subgroups, and signal detection is performed for each subgroup by a hierarchical algorithm or the like. The parallel processing for each subgroup can reduce the calculation amount and reduce the processing delay. Furthermore, since the multiplying unit collectively multiplies the received symbols of a plurality of subgroups with the conjugate transpose matrix, there is an advantage that the amount of calculation can be reduced.

It is a block diagram which shows the structure of the transmitter in one embodiment of this invention. It is a basic diagram which shows the structure of the transmission signal in one embodiment of this invention. It is a block diagram which shows the structure of the receiver in one embodiment of this invention. It is a block diagram which shows the structure of an example of the symbol decoding part by this invention. It is a basic diagram which shows reduction of the metric calculation amount by M algorithm of a local detector. It is a flowchart which shows the flow of the process of one Embodiment of this invention. It is a flowchart which shows the partial modification of the flow of the process of one Embodiment of this invention. It is a flowchart which shows the process of the M algorithm which is an example of the local detection in one embodiment of this invention. It is a flowchart which shows the process of sphere decoding which is another example of the local detection in one embodiment of this invention. It is a graph of the result of the computer simulation which shows the effect of adaptive group control. It is a graph of the result of the computer simulation which shows the effect of the number of surviving paths. It is a graph of the result of the computer simulation which shows the comparison of the BER characteristic regarding QPSK. It is a graph of the result of the computer simulation which shows the comparison of BER characteristic regarding 16QAM.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings. 1A and 1B show an exemplary configuration of a transmitter in a communication system according to an embodiment. The transmission bit sequence is input to an FEC (Forward Error Correction) encoder 1 and subjected to error correction coding. The output of the FEC encoder 1 is supplied to the block interleaver 2, and interleave processing is performed in units of OFDM symbols. The output of the interleaver 2 is supplied to the S / P converter 3 and converted from serial data to parallel data. The S / P converter 3 generates transmission data sequences for transmission data such as L, for example, four transmission antennas AT1, AT2, AT3, and AT4.

  Note that the S / P converter 3 may be provided in the input stage, and the FEC encoder 1 and the interleaver 2 may be provided for each of four parallel data series.

  The data series is divided into a plurality of data series by the S / P converters 4a, 4b, 4c, and 4d connected to the S / P converter 3. Data series from the S / P converters 4a to 4d are supplied to the mappers 5a, 5b, 5c and 5d, respectively. The mappers 5a to 5d perform modulation of digital modulation (QAM (Quadrature Amplitude Modulation), PSK (Phase Shift Keying), etc.), respectively.

  The outputs of the mappers 5a to 5d are supplied to Inverse Fast Fourier Transform (IFFT) units 6a, 6b, 6c, and 6d to generate time domain waveforms. The IFFT units 6a to 6d are supplied with pilot symbols from the pilot symbol generator 7 and added with pilot symbols.

  The outputs of the IFFT units 6a to 6d are supplied to guard interval (GI) adding units 8a, 8b, 8c, and 8d, and the guard intervals are added in symbol time units. The outputs of the guard intervals 8a to 8d are respectively supplied to the transmission antennas AT1, AT2, AT3, and AT4 through a power amplifier (not shown) and transmitted as radio waves.

The S / P converters 4a to 4d divide data into channels corresponding to the number of subcarriers in the OFDM modulation scheme. A configuration related to the S / P converter 4a is extracted and shown in FIG. 1B. If the number of OFDM subcarriers is n, the S / P converter 4a forms n-channel parallel data. Data series for each channel are supplied to the mapper 5a ₁ ~5a _n, is the OFDM signal by the IFFT section 6a, the guard interval adding unit 8a, a guard interval is added to each OFDM symbol.

  FIG. 2 shows a configuration of output signals output from the L transmission antennas. The transmission signal is composed of known symbols (referred to as pilots) used for channel estimation and data symbols storing transmission data. In FIG. 2, the vertical direction indicates the number (1, 2,..., L) of the transmitting antenna, the horizontal axis indicates time, and the data configuration of one packet is shown. The first P symbol (which means OFDM symbol) of one packet is a pilot, and the subsequent D symbol is a data symbol. Note that the packet configuration of FIG. 2 is an example, and other packet configurations may be adopted.

  FIG. 3 shows the configuration of the receiver in one embodiment of the present invention. M, for example, received signals from four receiving antennas AT11, AT12, AT13, and AT14 are supplied to the guard interval removing units 11a, 11b, 11c, and 11d, respectively. The guard interval removal units 11a to 11d remove the guard interval according to the timing of the received OFDM symbol.

  One embodiment of the present invention is a technique that uses L transmitting antennas and M receiving antennas to form a plurality of communication paths and improves frequency use efficiency and communication quality. Multiple Input (Multiple Input Multiple Output), which performs SDM transmission in which signals of different data series can be transmitted from each antenna at the same time and at the same frequency.

  Output signals from the guard interval removing units 11a to 11d are supplied to FFT (Fast Fourier Transform) units 12a, 12b, 12c, and 12d, respectively. The time domain signals are converted into frequency domain signals by the FFT units 12a to 12d. Each output of the FFT units 12a to 12d is composed of a parallel data sequence of n subcarriers in OFDM.

  Data symbols being output from the FFT units 12 a to 12 d are supplied to the symbol decoding unit 13, and pilot symbols are supplied to the channel estimation unit 14. The channel estimation unit 14 performs channel estimation using the received pilot signal (known symbol) and the transmission known symbol replica. Channel estimation is performed by an appropriate method depending on the configuration of transmission known symbols.

  After the channel estimation, the symbol decoding unit 13 performs transmission signal detection using the outputs of the FFT units 12a to 12d and the channel response matrix estimated by the channel estimation unit 14. That is, in the SDM transmission technique, a channel response matrix (also called a transfer coefficient matrix) H of a MIMO channel formed by a combination of all antennas between the transmission side and the reception side is estimated on the reception side based on a known signal or the like. Processing for separating the data series transmitted from the individual transmission antennas is performed.

  In one embodiment of the present invention, as a signal detection algorithm in the symbol decoding unit 13, as will be described later, a plurality of QR decomposition (Multi-QR-Decomposition: Multi-QRD) and group detection (group detection: GD) A signal detection algorithm is used. The output of the channel estimation unit 14 is supplied to the symbol decoding unit 13, and the data series transmitted from each transmission antenna by the symbol decoding unit 13 is obtained separately.

  These transmission data sequences are respectively supplied to demappers 15a, 15b, 15c, and 15d that perform the reverse demodulation process of the mapper, and the digital modulation performed on the transmission side is demodulated. As the demapping process of the demappers 15a to 15d, a hard decision or a soft decision can be made. When an FEC code is applied, it is possible to improve the FEC decoding effect by using soft decision.

  The outputs of the demappers 15a to 15d are supplied to the P / S converters 16a, 16b, 16c, and 16d, respectively. The P / S converters 16a to 16d combine a number of channels equal to the number of OFDM subcarriers into one channel. Further, the P / S converter 17 combines the output data sequences of the P / S converters 16a to 16d into one received data sequence.

  The output of the P / S converter 17 is supplied to the FEC decoder 19 via a deinterleaver 18 that performs a deinterleave process corresponding to the interleave on the transmission side. The transmission bit sequence is decoded from the FEC code by the FEC decoder 19.

  When the transmission side is configured to perform interleaving and FEC encoding for each data sequence corresponding to the transmission antenna, the reception side also outputs each of the outputs of the P / S converters 16a to 16d. A deinterleaver is connected, an FEC decoder is connected to each deinterleaver, and the output of the FEC decoder is combined into one sequence by a P / S converter.

  Next, the symbol decoding unit 13 that characterizes the present invention will be described. In order to reduce the amount of processing delay, group detection (GD) (hereinafter referred to as GD as appropriate) is considered effective as an algorithm that enables parallel processing. In GD, a received signal is separated into a plurality of subgroups by a function called a projector, and detection by MLD, a hierarchical detection algorithm, or the like is performed for each group. By separating the received signals into groups, GD loses diversity gain compared to MLD that is not grouped, but it is possible to reduce the amount of computation instead. Furthermore, since detection for each group can be performed in parallel, it is advantageous in terms of processing delay.

  FIG. 4 shows a configuration example of the symbol decoding unit 13 in one embodiment of the present invention. Received symbols are supplied to the multiplier 21. An output from the QR decomposition unit 22 is supplied to the multiplication unit 21, and the received symbol is multiplied by the output of the QR decomposition unit 22. A plurality of multipliers 21 and QR decomposition units 22 are provided corresponding to the number of subgroups. The multiplier 22 multiplies the conjugate transpose matrix and the received symbol, thereby orthogonalizing the received symbol. The permutation channel matrix from the adaptive grouping scheme (AGS) unit 23 is supplied to the QR decomposition unit 22. The AGS unit 23 is supplied with the channel response matrix from the channel estimation unit 14. A permutation of the channel response matrix for rows or columns is referred to as a permutation channel matrix. The multiplication unit 21, the QR decomposition unit 22, and the AGS unit 23 constitute a projector.

  Received symbols divided into a plurality of subgroups from the multiplier 21 and orthogonalized are supplied to the local detector 24. The local detector 24 performs signal detection by a hierarchical algorithm (for example, M algorithm) using the fact that each subgroup is orthogonalized.

One embodiment of the present invention described above, in particular, the processing of the symbol decoding unit 13 will be described in more detail. For simplicity, the case where the AGS unit 23 does not perform adaptive group control will be described. First, the received signal model will be described. Let L be the number of transmitting antennas and M be the number of receiving antennas. Between time t, subcarrier k, symbol transmitted from transmission antenna l, s _l (t, k), symbol received by reception antenna m, r _m (t, k), between transmission antenna l and reception antenna m The received symbol vector r (t, k) can be expressed by the following equation (1), where h _{m, l} (t, k) is the channel response and n _m (t, k) is the noise. The channel response matrix is estimated by the channel estimation unit 14 and supplied to the symbol decoding unit 13.

  In the following description, t and k are omitted. The signal detection process is performed every symbol time and every subcarrier.

  The received symbols are divided into Gr subgroups by the projector (multiplication unit 21, QR decomposition unit 22, and AGS unit 23). In the following description, for example, Gr = 2 and the number of transmission symbols belonging to one subgroup is L / Gr. Thus, the processing delay can be shortened most when the number of symbols belonging to each subgroup is made equal.

The channel response matrix H is rewritten as a replacement channel matrix H ₁ . As shown in Expression (2), H ₁ can be decomposed into a unitary matrix Q ₁ and an upper triangular matrix R ₁ by QR decomposition.

As shown in Equation (3), the received symbols can be orthogonalized by multiplying r by the conjugate transpose matrix of Q ₁ .

In the formula (3), the second half elements z ₁ orthogonalization received vector _{z 1, L / Gr + 1} ··· z i, when focusing _L, and they transmit symbol _{s L / Gr + 1 ··· s} L Consists of only. That is, by extracting these elements, it is possible to separate received symbols having the number L of transmitted symbols into L / Gr groups.

Next, for a column of the previous channel response matrix H, a permutation channel matrix in which the first half and the second half are exchanged is defined as H ₂ (Formula (5)).

The matrix H ₂ is processed in the same manner as H ₁ to obtain the orthogonalized reception vector z ₂ shown in Expression (6).

As for the vector z ₂ , similarly to z ₁ , when the latter half elements z _{2, L / Gr + 1} ... Z _{2, L} are extracted, as shown in the equation (7), they are transmitted symbols s _1.・ It is composed only of s _{L / Gr} .

Thus, the received symbol of the transmission symbol number L is divided into two sub-groups each containing transmission symbol _{s L / Gr + 1 ··· s} L or s ₁ ··· s L / _Gr.

  In the local detector 24, detection by the hierarchical algorithm is performed by making use of the fact that Gr (for example, 2) subgroups from the multiplication unit 21 are orthogonalized. Typical algorithms for layered algorithms include the M algorithm and sphere decoding. Here, a signal detection method based on the M algorithm will be schematically described. For simplicity, only the first subgroup will be described.

The M algorithm is a method for reducing the metric calculation amount as a whole by leaving only S _{1 in the} order of the metrics calculated in the first stage to the next stage. FIG. 5 shows the operation of the M algorithm in the case of QPSK (Quadrature Phase Shift Keying). The local detector 24 uses the square Euclidean distance as a metric. However, a metric other than the square Euclidean distance can also be used. In the first stage, a metric μ _{1, u} (1 ≦ μ ≦ S, S is the number of symbol points for each modulation) shown in Expression (8) is calculated.

Here, s ^ _l is a replica of the symbol transmitted from the transmission antenna l. The replica is a symbol generated by a process similar to that on the transmission side with respect to a symbol decoded from the received signal. From this metric, S ₁ is the cumulative metric Λ _l, ν (1 ≦ ν ≦ S ₁ ) to the next stage in ascending order. Thereafter, the metric μ _{1, u} (1 ≦ μ ≦ SS _l−1 ) in the l-th stage (1 <l ≦ L / Gr) is calculated by the following equation.

Here, s ₁ is a transmission symbol replica used to calculate the cumulative metric Λ _1, ν. When calculation of a metric is completed at all stages, a log-likelihood ratio (LLR) is calculated from the metric.

In the above description, when a subgroup is created, a condition such as a difference in received power for each transmission symbol is not considered. The AGS (adaptive group control) unit 23 performs adaptive group control based on a predetermined condition. In the following, adaptive group control based on a received signal-to-noise power ratio (SNR) is shown as an example. First, define a receiving SNRganma _l for the transmission symbols s _l by a formula (10) below. || a || in Eq. (10) indicates the Euclidean norm of the vector a.

Ordered in descending order of reception SNRganma _l obtained, replacing it in a variable called rank (l) (1 ≦ l ≦ L). rank (1) corresponds to the maximum SNR, and rank (L) corresponds to the minimum SNR. According to this ranking, the columns of the permutation channel matrices H ₁ and H ₂ are controlled before QR decomposition. As an example, when L = M = 4 and the number of symbols for each subgroup is the same, one of the following four patterns AGS1, AGS2, AGS3, and AGS4 is selected.

AGS1 is a method of dividing s _{rank (1)} and s _{rank (4)} into one group, and s _{rank (2)} and s _{rank (3)} into another group. The permutation channel matrix in this case is as shown below.

AGS2 is a method of dividing s _{rank (1)} and s _{rank (2)} into one group, and s _{rank (3)} and s _{rank (4)} into another group. The permutation channel matrix in this case is as shown below.

AGS3 is a method of dividing s _{rank (1)} and s _{rank (3)} into one group, and s _{rank (2)} and s _{rank (4)} into another group. The permutation channel matrix in this case is as shown below.

  AGS4 does not consider ranking. The permutation channel matrix in this case is as shown below.

  An embodiment of the present invention will be further described. FIG. 6 shows the flow of processing according to an embodiment. In the first step ST1, the channel estimator 14 calculates the channel quality and obtains a channel response matrix. In the next step ST2, adaptive group control is performed. That is, Gr subgroups are formed by any of AGS1 to AGS4 as described above.

Processing ST3 ₁ , ST3 ₂ ,..., ST3 _Gr is performed for each subgroup. That is, for each subgroup, QR decomposition processing, processing for multiplying the received transposed matrix by the conjugate transpose matrix Q ^H, and local detection processing for the orthogonalized vector are performed in order. The local detection process may be MLD other than hierarchical detection.

As shown in FIG. 7, the process of multiplying the received symbol by Q ^H does not need to be calculated for each subgroup from the matrix structure, and can be performed by one multiplication as a whole. .

  An example of the M algorithm and sphere decoding (sphere decoding) will be described for the hierarchical detection algorithm. As described above, in one embodiment, hierarchical detection is performed for each subgroup, but here, in order to simplify the description of the algorithm itself, a case where subgroup division is not performed will be described. Let L be the number of transmission antennas.

The QR decomposition process and the process of multiplying the conjugate transpose matrix Q ^H by the received symbol are common regardless of the algorithm. The received symbol vector z after Q ^H multiplication is as follows.

When attention is paid only to z _{L in the} equation (11), it can be seen that the following equation is obtained.

That is, it consists of only one transmission symbol s _L. Therefore, the metric is calculated for this z _{L using the} transmission symbol replica s _L ^ (the number of transmission symbol replicas is S. For example, if QPSK, S = 4, and 16QAM, S = 16). This metric μ _{L, u} is calculated by the following equation.

  However, there is no problem even if the right side of equation (13) is not squared.

Here, S _L (≦ S) items are selected in order from the smallest value of this metric μ _{L, u} , and these are set as cumulative metrics Λ _L, ν (1 ≦ ν ≦ S _L ).

Next, paying attention to z _L-1 in the equation (11), it can be seen that the following equation is obtained.

Here, for s _L , the metric is calculated by the equation (13). Therefore, here, the metric μ _L−1u (1 ≦ u ≦ SS _L ) for s _L−1 is calculated by the following equation. .

Here, S _L-1 items are selected in order from the smallest value of this metric μ _{L-1, u} , and these are selected as cumulative metrics Λ _L-1, ν (1 ≦ ν ≦ S _L-1 ). To do.

In the M algorithm, the same calculation is repeated up to z _l . The metric μ _{l, u} (1 ≦ u ≦ SS _L-1 ) at z _l in the middle is as shown in the following equation.

Here, s _{p, v} ^ is a transmission symbol replica from the transmission antenna p used for calculating Λ _{l + 1,} ν.

When the calculation of the metric is completed up to z _l , the final hard decision value or soft decision value of the transmission symbol is output according to the metric.

FIG. 8 shows a basic flowchart of the processing of the M algorithm described above. In step ST11, the variable l is set to L. In step ST12, the metric μ _{L, u} is calculated. In step ST13, S _L (≦ S) items are selected in order from the smallest value of the metrics μ _{L, u} , and these are set as cumulative metrics.

  In step ST14, it is determined whether l = 1. Otherwise, in step ST15, the value of l is decremented by -1 (decrement), and the process returns to step ST12 (metric calculation). If it is determined in step ST14 that l = 1, the metric calculation ends. In step ST16, a hard decision value or a soft decision value for the calculated metric is output.

  Next, sphere decoding will be described. The sphere decoding is an algorithm for reducing the number of transmission symbol candidate points in MLD using a sphere with a certain radius C. Metrics are calculated only for transmission symbol candidates within the sphere C from the reception symbol point.

  If the transmission symbol candidate vector is s ^, the presence of a symbol candidate in the sphere C satisfies the following formula (17).

  When the expression (17) is expanded, the following expression (18) is obtained.

Paying attention to s _L ^, in order to establish the equation (18), the following equation (19) must be established.

Therefore, s _L ^ that satisfies the following equation (20) is selected.

  Furthermore, in order to establish the equation (18), the following equation (21) must also be established.

Here, for s _L ^, since a candidate has already been selected in Expression (20), the candidate is selected from Expression (21) by paying attention to s _L-1 ^. That is, s _L-1 ^ satisfying the following formula (22) is selected.

The above calculation is repeated until s _l ^. When attention is paid in the middle of the s _l ^, it will select the s ₁ ^ to establish the equation (23) below. It should be noted that the sphere radius can be changed for each repetition when selecting a repetition candidate. That is, C is replaced with Cl. When the processing up to s ₁ ^ is completed, a hard decision value or a soft decision value is output from the selected transmission symbol candidate.

  FIG. 9 shows a basic flowchart of the sphere decoding process described above. In step ST21, the variable l is set to L. In step ST22, the radius C of the sphere is set. In step ST23, a transmission symbol candidate in the sphere is selected.

  In step ST24, it is determined whether l = 1. Otherwise, in step ST25, the value of l is decremented by -1 (decrement), and the process returns to step ST22 (setting of the radius of the sphere). If it is determined in step ST24 that l = 1, the calculation of the metric ends. In step ST26, a hard decision value or a soft decision value for the calculated metric is output.

  Table 1 shows the results (the number of real multiplications required per packet) of evaluating the amount of computation of the symbol decoding unit of the embodiment of the present invention described above.

Table 1 shows calculation amounts for a plurality of signal detection algorithms in the case of QPSK and the case of 16QAM. The numbers of multiplications of ZF (Zero-Focing), MLD, QRM-MLD, and Multi-QRD-GD (one embodiment) are shown. Focusing on QRM-MLD and Multi-QRD-GD, in the case of 16QAM, Multi-QRD-GD with S ₁ = 14 is QRM- with [S ₁ , S ₂ , S ₃ ] = [12, 12, 12]. The calculation amount of MLD is reduced by about 18%. As will be described later, these BER (Bit Error Rate) characteristics are substantially the same, and Multi-QRD-GD can effectively reduce the amount of calculation and the amount of delay time. Further, QRM-MLD with [S ₁ , S ₂ , S ₃ ] = [10, ₁₀ , 10] has a larger calculation amount than Multi-QRD-GD with S ₁ = 14, but the BER characteristics will be described later. Deteriorates greatly. That is, it can be said that QRM-MLD has a large influence of BER degradation due to reduction of S ₁ .

  The BER characteristic of one embodiment (Multi-QRD-GD) of the present invention and the conventional algorithm is evaluated by computer simulation. Table 2 below shows simulation specifications.

  In Table 2, L is the number of transmitting antennas, M is the number of receiving antennas, and (L = M = 4). QPSK and 16QAM are used for subcarrier modulation. The number of divided groups Gr is 2. For channel estimation, an estimation method using CI (Carrier Interferometory) pilot symbols is used. It is assumed that the number of pilot symbols is 2, and that 2 pilot antenna pilots are included per symbol. The ideal state is assumed for the synchronous system. The channel model is exponentially attenuated 24-wave Rayleigh fading, path interval 62.5 nsec, maximum delay time 1437.5 nsec, and delay spread 360 nsec. The maximum Doppler frequency is 10 Hz and a quasi-static environment is assumed.

FIG. 10 shows the effect of AGS in one embodiment of the present invention. In the case of QPSK, AGS 1 and 3 show an Eb / No improvement of about 0.7 dB at BER = 10 ⁻⁵ compared to the others. This is because it is possible to improve the detection accuracy of symbols having a low SNR by making a symbol having a large received SNR and a symbol having a small received SNR into the same group. Also in the case of 16QAM, AGS1 shows the best BER.

FIG. 11 shows the effect of the number of surviving paths S ₁ for each stage in the M algorithm. In the case of QPSK, since the number of signal points S is as small as 4 in the first place, it can be seen that if the number of paths is reduced even by one, the BER is greatly deteriorated. On the other hand, in the case of 16QAM, the deterioration of BER due to the reduction of the number of surviving paths is not as great as that of QPSK. In particular, the BER of S ₁ = 16 and S ₁ = 14 has almost similar characteristics.

FIG. 12 compares the BER for QPSK. In the case of QPSK, Multi-QRD-GD (one embodiment) has an Eb / No degradation of about 0.5 dB at BER = 10 ⁻⁵ compared to QRM-MLD. In the case of QPSK, error propagation of the M algorithm in QRM-MLD is small, and it has a diversity gain effect, so that it has good characteristics. However, as described above, the calculation amount of Multi-QRD-GD is smaller by about 10%.

  FIG. 13 compares the BER for 16QAM. In the case of 16QAM, although the amount of calculation is smaller than Multi-QRD-GD, the BER characteristic is equal to that of QRM-MLD or Multi-QRD-GD is superior. This is because, in the case of QRM-MLD, there is an error propagation in the M algorithm, and thus the influence of the propagation cannot be covered even with a diversity gain. On the other hand, although Multi-QRD-GD has a decrease in diversity gain due to group division, the influence of error propagation in the M algorithm is small, so BER degradation is small.

  Although the embodiment of the present invention has been specifically described above, the present invention is not limited to the above-described embodiment, and various modifications based on the technical idea of the present invention are possible. For example, the number of subgroups and the number of symbols for each subgroup are not limited to the example of the embodiment described above. A subgroup division method will be described, where the number of transmission antennas is L, the number of subgroups to be divided is Gr, and the number of transmission symbols in subgroup g is Ng.

The relationship between the number of subgroups Gr and the number of transmitting antennas L is Gr ≦ L. The number of QR decompositions coincides with the number of subgroups Gr. The symbol s _l transmitted from an arbitrary transmission antenna l can belong to two or more subgroups. Therefore, not only ΣNg = L but ΣNg ≧ L is also possible. Σ means taking the sum from g = 1 to Gr. In this case, since the output for sl is output from a plurality of local detectors, these outputs are selected or synthesized. Although this method increases the amount of computation, a selection / combination diversity effect can be obtained for the output of the local detector.

  In addition, the present invention is not limited to a multi-carrier transmission scheme such as OFDM, but is applied to a communication system combining a narrowband single carrier transmission scheme and SDM, a wideband single carrier transmission scheme using frequency domain equalization, and a communication system combining SDM, and the like. Is also possible.

DESCRIPTION OF SYMBOLS 13 ... Symbol decoding part 14 ... Channel estimation part 21 ... Multiplication part 22 ... QR decomposition part 23 ... Adaptive group control part 24 ... Local detector

Claims (6)

A signal input unit that emits L transmission signals independent of each other obtained from a predetermined information signal from L transmission antennas, and receives M mixed signals obtained by mixing the L transmission signals;
A channel estimation unit that performs channel estimation and estimates a channel response matrix of M rows and L columns;
A symbol decoding unit that receives a received symbol and the channel response matrix and detects a transmission signal;
The symbol decoding unit
A processing unit that divides the channel response matrix from the channel estimation unit into a plurality of different permutation channel matrices to form subgroups, and performs parallel processing for each subgroup,
The processing unit is supplied with a QR decomposition processing unit for the permutation channel matrix, a multiplication unit that multiplies a reception symbol by a conjugate transpose matrix and orthogonalizes the reception symbol, and an output of the multiplication unit. It consists of a local detection unit that performs signal detection every time,
A signal decoding apparatus that, in the multiplication unit, collectively multiplies the received symbols of the subgroups with the conjugate transpose matrix. The signal decoding device according to claim 1,
A signal decoding apparatus that adaptively forms the permutation channel matrix according to a predetermined condition. The signal decoding device according to claim 1,
The local detection unit is a signal decoding device which is a hierarchical detection algorithm. A signal input step in which L transmission signals independent from each other obtained from a predetermined information signal are emitted from L transmission antennas, and M mixed signals obtained by mixing the L transmission signals are input.
A channel estimation step of performing channel estimation and estimating a channel response matrix of M rows and L columns;
A symbol decoding step for receiving a received symbol and the channel response matrix and detecting a transmission signal;
The symbol decoding step includes:
Dividing the channel response matrix into a plurality of different permutation channel matrices to form subgroups, and performing parallel processing for each subgroup,
The processing step is provided with a QR decomposition processing step for the permutation channel matrix, a multiplication step for multiplying a received symbol by a conjugate transpose matrix and orthogonalizing the received symbol, and an output of the multiplication step, It consists of a local detection step that performs signal detection every time,
A signal decoding method in which, in the multiplication step, a plurality of received symbols of the subgroup are collectively multiplied by the conjugate transpose matrix. The signal decoding method according to claim 4, wherein
A signal decoding method for adaptively forming the permutation channel matrix according to a predetermined condition. The signal decoding method according to claim 1,
The local detection step is a signal decoding method which is a hierarchical detection algorithm.

Images