WO2014082487A1 - 一种软输出固定复杂度球形译码检测方法和装置 - Google Patents

一种软输出固定复杂度球形译码检测方法和装置 Download PDF

Info

Publication number
WO2014082487A1
WO2014082487A1 PCT/CN2013/084017 CN2013084017W WO2014082487A1 WO 2014082487 A1 WO2014082487 A1 WO 2014082487A1 CN 2013084017 W CN2013084017 W CN 2013084017W WO 2014082487 A1 WO2014082487 A1 WO 2014082487A1
Authority
WO
WIPO (PCT)
Prior art keywords
layer
path
complement
nodes
retained
Prior art date
Application number
PCT/CN2013/084017
Other languages
English (en)
French (fr)
Inventor
邬钢
沈文水
Original Assignee
中兴通讯股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中兴通讯股份有限公司 filed Critical 中兴通讯股份有限公司
Priority to US14/648,089 priority Critical patent/US9356733B2/en
Priority to EP13858769.6A priority patent/EP2928084B1/en
Publication of WO2014082487A1 publication Critical patent/WO2014082487A1/zh

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
    • H04L25/03006—Arrangements for removing intersymbol interference
    • H04L25/03178—Arrangements involving sequence estimation techniques
    • H04L25/03203—Trellis search techniques
    • H04L25/03242—Methods involving sphere decoding
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L1/00—Arrangements for detecting or preventing errors in the information received
    • H04L1/004—Arrangements for detecting or preventing errors in the information received by using forward error control
    • H04L1/0045—Arrangements at the receiver end
    • H04L1/0054—Maximum-likelihood or sequential decoding, e.g. Viterbi, Fano, ZJ algorithms
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
    • H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
    • H04B7/0413—MIMO systems
    • H04B7/0456—Selection of precoding matrices or codebooks, e.g. using matrices antenna weighting
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
    • H04L25/03006—Arrangements for removing intersymbol interference
    • H04L25/03178—Arrangements involving sequence estimation techniques
    • H04L25/03312—Arrangements specific to the provision of output signals
    • H04L25/03318—Provision of soft decisions

Definitions

  • the present invention relates to multiple-input multiple-output (MIMO) detection technology in wireless communication, and more particularly to soft-output fixed-complexity Sphere Decoding (SFSD) Detection method and device.
  • MIMO multiple-input multiple-output
  • SFSD soft-output fixed-complexity Sphere Decoding
  • LTE Long Term Evolution
  • LTE-A Long Term Evolution-Advanced
  • SD detection can approximate the optimal performance in performance, that is, Maximum Likelihood (ML) performance, it is acceptable for optimal performance.
  • MIMO detection often uses SD detection.
  • SD detection is divided into fixed-complexity Sphere Decoding (FSD) and spherical decoding of non-fixed complexity characteristics. Since FSD is easy to implement in Very Large Scale Integration (VLSI), FSD is often used.
  • VLSI Very Large Scale Integration
  • soft-output MIMO detection can be combined with subsequent soft decoders to achieve better detection performance, so SFSD detection is often used.
  • the main processes of SFSD detection by a receiver in an LTE-A system include: pre-processing, sphere detection, and Likelihood Ratio (LRR) output, where the pre-processing includes equivalent channel matrix generation, ordering, and QR break down.
  • pre-processing includes equivalent channel matrix generation, ordering, and QR break down.
  • LRR Likelihood Ratio
  • the LTE-A system proposes a spectrum efficiency of 30 bps/Hz downstream and 15 bps/Hz uplink. To meet these metrics, the LTE-A system uses up to 8 layers of MIMO in the downlink and up to 4 layers of MIMO in the uplink, and supports Quadrature Phase Shift Keying (QPSK) /16 Quadrature Amplitude Modulation ( QAM) /64QAM modulation mode. Since the number of MIMO layers is greatly increased, the amount of calculation of the spherical detection becomes very large.
  • QPSK Quadrature Phase Shift Keying
  • QAM Quadrature Amplitude Modulation
  • the above method uses a bit-negating method to construct a complement path of ML, that is, only a measure of the path in which each bit of the ML path is reversed (for example) 64QAM has only 6 possible complement paths for 6 bits per symbol, which reduces the complexity but only calculates the bit. Excessive path loss information, detection performance of the ML performance is not optimal.
  • the main purpose of the embodiments of the present invention is to provide an SFSD detection and apparatus, which can obtain the detection performance of the ML performance for the MIMO of more than two layers, and can satisfy the currently accepted hardware implementation complexity.
  • An embodiment of the present invention provides an SFSD detection method, where the method includes:
  • QR decomposition of the channel response matrix to obtain a Q matrix and an R matrix
  • the conjugate transpose of the Q matrix is multiplied by the received signal to obtain an equalized signal of the received signal.
  • the ML path detection is performed on the equalized signal, and the layer-by-layer decreasing method is adopted for each layer of the reserved node.
  • likelihood ratio (LLR) information of each bit of each symbol of each layer is obtained.
  • the ML path detection is performed on the equalized signal, and the ML path is obtained by using a layer-by-layer decrementing manner of each layer, and the branches having the same number of iterations are retained as:
  • the ML complement path is detected on the branch, and all the nodes and other layer reserved nodes are used to obtain the ML complement path by using the complement layer to obtain the ML complement path:
  • the path is expanded from the reserved layer of the reserved branches to the bottom layer;
  • the branch with the smallest Euclidean distance is selected layer by layer from the complement layer to be used as the ML complement path of each layer.
  • an SFSD detecting apparatus which is applied to a receiver, and the apparatus includes:
  • a QR decomposition unit configured to perform QR decomposition on the channel response matrix to obtain a Q matrix and an R matrix
  • the equalization signal calculation unit is configured to multiply the conjugate transpose of the Q matrix by the received signal to obtain an equalized signal of the received signal;
  • the ML path detecting unit is configured to perform ML path detection on the equalized signal, and obtain an ML path by using a layer-by-layer decrementing manner of each layer, and retaining the same number of branches as the number of iterations;
  • the ML complement path detecting unit is configured to perform ML complementary path detection on the branch reserved by the ML path detecting unit, and obtain the ML complementary path by using the complement layer to retain all nodes and other layer retaining nodes to reduce the layer by layer;
  • the soft value information calculation unit is configured to obtain LLR information of each bit of each symbol of each layer according to the ML path and the ML complement path.
  • the ML path detecting unit is specifically configured to extend from the matrix R to the layer corresponding to only one non-zero element, that is, the top layer starts to expand to the bottom layer; All the nodes of the top layer are reserved, and the nodes retained by the layers below the top layer are decremented layer by layer to obtain the Euclidean distance of each branch;
  • the ML complement path detecting unit is specifically configured to start path expansion from the reserved complement layer of each reserved branch to the bottom layer;
  • the branch with the smallest Euclidean distance is selected layer by layer from the complement layer to be used as the ML complement path of each layer.
  • the ML complementary path detecting unit is configured to reserve all the nodes of the required complement layer, and the nodes retained by the layers below the complement layer are decremented layer by layer until the number of nodes in the bottom layer is 1, get the Euclidean distance of each branch.
  • the technical solution of the embodiment of the present invention includes: performing QR decomposition on the channel response matrix H to obtain a Q matrix and an R matrix; multiplying the conjugate transpose of the Q matrix by the received signal to obtain an equalized signal of the received signal.
  • the ML path detection is performed on the equalized signal, and the ML path is obtained by using the layer-by-layer decrementing manner of each layer, and the branch with the same number of iterations is retained; the ML complement path detection is performed on the branch, and the complement layer is reserved.
  • All the nodes and other layer-preserving nodes obtain the ML-complementing path in a layer-by-layer decrement manner; according to the ML path and the ML-complementing path, the LLR information of each bit of each symbol of each layer is obtained; thus, the embodiment of the present invention is greater than two Layer MIMO can also achieve near-ML performance detection performance while meeting the currently accepted hardware implementation complexity.
  • FIG. 1 is a flowchart of an implementation of a first embodiment of an SFSD detection method according to the present invention
  • FIG. 2 is a schematic structural diagram of an embodiment of an SFSD detection apparatus according to the present invention
  • Figure 3 is an ML path detection tree diagram
  • Figure 4 shows the ML path detection 4 pairs of diagrams
  • Figure 5 is a ML complement path detection tree diagram
  • FIG. 6 is a comparison diagram of throughput performance of the throughput (throughput) performance of the prior art 1, 2, and 3 and the 64QAM of the embodiment of the present invention
  • Figure 7 is a comparison of throughput performance simulation results of prior art 1, 2 and 3 and 16QAM of the embodiment of the present invention. detailed description
  • a first embodiment of the SFSD detection method provided by the present invention includes the following steps:
  • Step 101 Perform QR decomposition on the channel response matrix to obtain a Q matrix and an R matrix.
  • Step 102 Multiply the conjugate transpose of the Q matrix by the received signal to obtain an equalized signal of the received signal.
  • Step 103 Perform ML path detection on the equalized signal, obtain an ML path by using a layer-by-layer decrementing manner of each layer, and retain the same number of branches as the number of iterations;
  • Step 104 Perform ML complement path detection on the branch, and use the complement layer to retain all nodes and other layer reserved nodes to obtain the ML complement path by layer-by-layer decrement;
  • Step 105 Obtain LLR information of each bit of each symbol of each layer according to the ML path and the ML complement path.
  • the ML path detection is performed on the equalized signal, and the ML path is obtained by using a layer-by-layer decrementing manner of each layer, and the branch with the same number of iterations is retained, which may be: the matrix R has only one non-zero element corresponding to the branch. That layer, that is, the top layer starts the path expansion until the bottom layer;
  • All the nodes of the top layer are reserved, and the nodes retained by the layers below the top layer are decremented layer by layer to obtain the Euclidean distance of each branch; Select the branch with the smallest Euclidean distance as the ML path and keep the same number of branches as the number of iterations.
  • the ML complement path is detected on the branch, and all the nodes and other layer reserved nodes are used to obtain the ML complement path by using the complement layer to obtain the ML complement path, which may be: Find the complement layer to start the path expansion, until the bottom layer;
  • the branch with the smallest Euclidean distance is selected layer by layer from the complement layer to be used as the ML complement path of each layer.
  • the nodes retained by the layers below the complement layer are decremented layer by layer, and the nodes reserved by the layers below the required complement layer are decremented layer by layer until the number of nodes at the bottom layer is 1.
  • an SFSD detecting apparatus provided by the present invention, as shown in FIG. 2, the apparatus includes:
  • the QR decomposition unit 201 is configured to perform QR decomposition on the channel response matrix to obtain a Q matrix and an R matrix;
  • the equalization signal calculation unit 202 is configured to multiply the conjugate transpose of the Q matrix by the received signal to obtain an equalized signal of the received signal;
  • the ML path detecting unit 203 is configured to perform ML path detection on the equalized signal, and obtain an ML path by using a layer-by-layer decrementing manner of each layer, and retaining the same number of branches as the number of iterations;
  • the ML complement path detecting unit 204 is configured to perform ML complementary path detection on the branch reserved by the ML path detecting unit, and use the complement layer to reserve all nodes and other layer reserved nodes to obtain the ML complementary path in a layer-by-layer decreasing manner;
  • the soft value information calculation unit 205 is configured to obtain LLR information of each bit of each symbol of each layer according to the ML path and the ML complement path.
  • the ML path detecting unit 203 is specifically configured to have only one from the matrix R.
  • the layer corresponding to the non-zero element that is, the top layer starts the path expansion, up to the bottom layer; retains all the nodes of the top layer, and the nodes remaining in the lower layer of the top layer are decremented layer by layer to obtain the Euclidean distance of each branch;
  • the ML complement path detecting unit 204 is specifically configured to start path expansion from the required complement layer of each reserved branch to the bottom layer;
  • the branch with the smallest Euclidean distance is selected layer by layer from the complement layer to be used as the ML complement path of each layer.
  • the ML complement path detecting unit 204 is specifically configured to reserve all the nodes of the required complement layer, and the nodes retained by the layers below the complement layer are decremented layer by layer until the number of nodes in the bottom layer is 1, get the Euclidean distance of each branch.
  • the QR decomposition unit 201, the equalization signal calculation unit 202, the ML path detection unit 203, the ML complement path detection unit 204, and the soft value information calculation unit 205 may be configured by a central processing unit (CPU). , a digital signal processor (DSP, digital signal processor) or a programmable intermixed mutli bad 1 J (FPGA, Field - programmable Gate array) implementation.
  • CPU central processing unit
  • DSP digital signal processor
  • FPGA Field - programmable Gate array
  • the SFSD detecting apparatus in this embodiment can be applied to a receiver.
  • a second embodiment of the SFSD detection method provided by the present invention is described below.
  • a coded MIMO system has M r transmit antennas and M ⁇ M r receive antennas, and the encoded bit stream is mapped onto the constellation diagram and forms one transmit symbol, where 0 is a constellation point set.
  • Step 301 Sorting and QR decomposition of the channel response matrix H to obtain a Q matrix and an R matrix;
  • the Q matrix is an orthogonal matrix
  • the R matrix is an upper triangular array
  • Step 302 Transpose the conjugate of Q to the left to multiply the received signal y to obtain an equalized signal of the received signal.
  • Step 303 Start detecting from the layer of the R matrix that has only one non-zero element, that is, the top layer; here, as shown in FIG. 3, the top layer is the layer below the top root node.
  • Step 304 determining whether the number of branches of the root node is greater than a preset value, such as the number of constellation points; if less than or equal to the preset value, proceeds to step 305; if greater than the preset value, proceeds to step 308;
  • a preset value such as the number of constellation points
  • Step 305 Calculate the Euclidean distance of the top layer for each branch
  • Step 306 Perform path expansion from the top layer, and continue to calculate Euclidean distances of the remaining layers; retain each layer's optimal node and its Euclidean distance according to a preset value, and accumulate the Euclidean distance of the branch.
  • Step 307 it is determined whether the leaf node has been reached, if it has reached the leaf node, proceeds to step 308; if it does not reach the leaf node, returns to step 306;
  • the leaf node is an underlying node.
  • Step 308 Temporarily store the full path Euclidean distance obtained by the number of branches of the top root node and the constellation point number corresponding to each layer symbol; return to step 304 to start processing of the next branch;
  • Step 309 Sort the complete Euclidean distances of all the branches from small to large, select the branch with the smallest Euclidean distance as the ML path, and select the constellation point number corresponding to each layer symbol in the branch; and retain multiple optimal paths at the same time.
  • Each layer has a constellation point number, and the number of the optimal path is the same as the number of iterations, and is used for iteration of the subsequent ML complement path.
  • Step 310 Perform Euclidean distance and each layer symbol of each branch retained in the ML path detection.
  • the constellation point number is obtained, and the complement of each bit of each symbol in the top layer of the ML path is obtained.
  • the full complement of each bit of each symbol of each layer takes into account all possible situations from the top layer to the bottom layer.
  • Step 311 For the complement of the layers below the top layer, for example, the iRx layer, after multiple iterations, the path of the iteration of each layer is performed according to the Euclidean distance from small to sequential; in one iteration, the iRx layer is first sought. The Euclidean distance from each layer to the layer;
  • each bit of each symbol of the layer is traversed.
  • Step 312 Find the Euclidean distance of each layer below the iRx layer, and refer to step 306 to perform path extension node selection from the layer.
  • Step 313 Determine whether the leaf node is reached. If the leaf node has been reached, go to step 314. If it does not reach the leaf node, return to step 312.
  • Step 314 Sum the results of steps 311 and 312, and select the minimum value of each bit of each symbol of the iRx layer as the optimal path of the iteration for temporary storage.
  • Step 315 Determine whether the number of iterations has reached the preset value. If the preset value is not reached, proceed to step 316; if the preset value is reached, proceed to step 317.
  • Step 316 the number of iterations is increased by 1, and the next iteration is performed.
  • Step 317 For the layer that needs to be complemented, the optimal path temporarily stored in multiple iterations may be obtained, and the minimum value is selected from the paths as the final complement path of the bit of the symbol of the layer.
  • Step 318 determine whether all layers have completed the complement, if not complete the complement, go to step 319; if you have completed the complement, go to step 320;
  • the following is an example of the method of ML path detection.
  • the Euclidean distance can be regarded as the detection and selection of the tree nodes as shown in Fig. 4.
  • each parent node contains constellation points and sub-nodes, and so on from the root node to the leaf node.
  • SFSD detection in the ML path detection, a set of node-reserved numbers are used to select the node tree.
  • For each parent node only a few sub-nodes with the best Euclidean path are reserved, and the path expansion is performed from top to bottom according to the set of parameters, and finally the optimal ML path is selected among the remaining complete paths.
  • Figures 3 and 4 show the case of 4*4 MIMO, QPSK.
  • the embodiment of the present invention adopts a unified architecture in which the number of reserved nodes is decremented layer by layer, and the maximum supported branch metric when each layer is configured with the highest modulation in the maximum number of layers is bounded, and various combinations of various modulation modes and layer numbers are in the branch.
  • the metrics are configured below. Since the first detection layer has a fault propagation effect on the post-detection layer (from the top to the bottom of the tree), the number of top-layer reserved nodes is equal to the number of constellation points (for example, QPSK takes 4, 16QAM takes 16, 64QAM takes 64), and the middle layers are decremented. The last few layers are optional 1. Different modulation methods and different layers are followed by this rule.
  • the number of reserved nodes is configured to support (64 2 1 1) metric calculation of 128 branches, and detection of other various modulation layers. Branches are all in 128
  • Table 1 The current parameters are shown in Table 1:
  • the metric value can be found directly according to the constellation area, and no traversal is required for all constellation points. Therefore, for the most complex 64QAM, 4 layers ( 64 2 1 1 ), the cumulative only requires 128 paths (each path corresponds to 4 segments of metrics) independent operations.
  • the complement path uses the number of reserved nodes and the extended path consistent with the ML path to find the complement path corresponding to each bit in each layer node, which is embodied as follows:
  • Finite full complement Find the complement path of each layer node on the ML path. For a specific layer, you first need to determine the node that the layer uses to complement the path. In the embodiment of the present invention, a method of taking a finite full complement of each bit is adopted, and if the ML path is required to be a complement of the i-th bit of the node of the layer, according to the range indicated by the selected parameter (the number of iterations, the reserved node), The value at the bit needs to be inverted, and then the values at the remaining bits are freely combined to obtain a set of nodes, that is, the nodes of the complement path at that layer. Take QPSK as an example.
  • the embodiment of the present invention introduces the concept of limited full complement set. That is, directly traverse the constellation points in this layer instead of bit-by-bit to find the full complement;
  • the number of iterations is 1, and the solid path is the path reserved by the ML path detection, which is sorted according to the Euclidean distance. It is the ML path.
  • the layer 2 node is extracted from the seven candidate complement paths as '00. , , '01, the path (four in total), the Euclidean minimum distance in these paths is taken as the Euclidean distance of the ML complement path; the same can be used to find the Euclidean distance of the second bit of the ML complement path .
  • the top layer can reuse the result of the ML path, the following is the case of selecting the reserved node 1, which can be directly based on the constellation
  • Cyclic prefix (Normal) Cell ID (Cell ID) 0
  • Redundancy code sequence (Redundancy version for QPSK and 16QAM ⁇ 0,1,2,3 ⁇ coding sequence )
  • ACK/NACK feedback mode Cell reference signal (Cell-specific reference antenna port (Antenna port) 0, symbols) 1 Antenna port User reference signal (UE-specific reference)
  • Number of allocated resource blocks (Number of allocated physical resource blocks)
  • Propagation Condition Enhanced Walk Channel (EPA) 5 correlation matrix and antenna construction (Correlation 4x4 layer (layer), low correlation (Low Matrix and Antenna Configuration correlation)), 1 codeword (CW) corresponds to 2 layers ( Layer )
  • FIG. 6 is a comparison diagram of throughput performance simulation results of the prior art 1, 2, and 3 and the 64QAM of the embodiment of the present invention
  • FIG. 7 is a comparison of throughput performance simulation results of the prior art 1, 2, and 3 and the 16QAM of the embodiment of the present invention
  • Figure 6 and Figure 7 show a cross-sectional curve showing the throughput performance of the prior art 1, 2, a curve with a small circle indicating the throughput performance of the prior art 3, and a curve with a square indicating the implementation of the present invention.
  • the throughput performance of the example with a star-shaped curve representing the lower limit of throughput performance (minimum mean square error MMSE).
  • Lmax minimum mean square error
  • the curve with the cross is located below the curve with a star, and the signal-to-noise ratio of the prior art 1, 2 is higher in the case of the same throughput, which indicates the prior art 1, 2
  • the throughput performance does not reach the lower limit of throughput performance, which means that the performance is poor and the ML performance cannot be achieved.
  • the curve with a small circle is located above the curve with a square, and the signal-to-noise ratio of the embodiment of the present invention is slightly higher than that of the prior art 3 in the case of the same throughput.
  • the signal-to-noise ratio difference between the embodiment of the present invention and the prior art 3 is less than 0.3 decibels UB)
  • the present invention and the prior art The signal-to-noise ratio difference of 3 is less than 0.2 dB, and thus it can be seen that the embodiment of the present invention approximates the ML performance.
  • Table 3 is a complexity statistics table (64 QAM 2, 3, 4 layers, 16 QAM 2, 3, 4 layers) according to an embodiment of the present invention.
  • Table 4 is a complexity statistics table (16QAM, 4 layers) of the prior art 3
  • Table 5 is a complexity statistics table (64QAM, 4 layers) of the prior art 3 when the signal-to-noise ratio is 30 dB. It can be seen from Table 3 that the complexity of the embodiment of the present invention is fixed. It can be seen from Table 4 that the complexity of the prior art 3 varies with the signal to noise ratio. It can be seen from Table 5 that the complexity of the prior art 3 is very high at a high signal to noise ratio. high.
  • the addition of the embodiment of the present invention is 12794 times, and the multiplication is 84 times.
  • the addition of the prior art 3 has an average value of 22461 when Lmax is 1, and the multiplication Lmax is 1.
  • the time average is 14757; see Table 3 and Table 5, for example, for 64QAM, 4 layers, the addition of the embodiment of the present invention is 60370 times, the multiplication is 124 times, and the addition of the prior art 3 is Lmax 0.5, the signal to noise ratio When it is 30dB, it is added 292130 times and the multiplication is 40459 times.
  • 64QAM 4 layers 64QAM, 2 layers 64QAM
  • the embodiments of the present invention are applicable to a terminal receiver and also to a base station receiver.
  • the MIMO detection can obtain the performance close to the ML, and the complexity is greatly reduced, and the MIMO detection is achievable.
  • the iterative process used in the embodiment of the present invention has no feedback mechanism, and can be performed in parallel, and each branch adopts a unified processing unit and parameters. Therefore, the processing time delay of the embodiment of the present invention is small, and the buffer amount is small, which is suitable for the parallel pipeline structure. There are more advantages in the implementation of the receiver.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Power Engineering (AREA)
  • Artificial Intelligence (AREA)
  • Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
  • Radio Transmission System (AREA)

Abstract

本发明公开了一种软输出固定复杂度球形译码（SFSD）检测方法和装置，其中，所述方法包括：对信道响应矩阵H进行QR分解，得到Q矩阵和R矩阵；将Q矩阵的共轭转置QH与接收信号相乘，得到接收信号的均衡信号；对均衡信号进行最大似然（ML）路径检测，采用顶层保留全部节点和其他层保留节点逐层递减方式得到ML路径，并保留与迭代次数数量相同的分支；对所述分支进行ML补集路径检测，采用所求补集层保留全部节点和其他层保留节点逐层递减方式得到ML补集路径；根据所述ML路径和ML补集路径，得出各层各个符号各个比特的似然比（LLR）信息。本发明对于大于两层的MIMO也能获得逼近ML性能的检测性能，同时可以满足目前可接受的硬件实现复杂度。

Description

一种软输出固定复杂度球形译码检测方法和装置 技术领域
本发明 涉及无线通信中 的 多 输入多 输出 （ Multiple-Input Multiple-Out-put, MIMO )检测技术， 尤其涉及一种软输出固定复杂度球形 译码 ( Soft-output Fixed-complexity Sphere Decoding, SFSD )检测方法和装 置。 背景技术
第三代合作伙伴计划 （ 3rd Generation Partnership Project, 3GPP )长期 演进（ Long Term Evolution, LTE )技术的增强 （ LTE- Advanced, LTE-A ) 系统中的采用多层 MIMO发送来实现高频谱效率， 其对应的接收机需要对 多层 MIMO进行检测以得到解调数据。
MIMO检测有多种方法， 由于球形译码（Sphere Decoding, SD )检测 在性能上能逼近最优性能， 即最大似然（Maximum Likelihood, ML )性能， 因此为获得最优的性能， 在可接受的硬件实现复杂度情况下， MIMO检测 常采用 SD检测。 SD检测分为固定复杂度特性的球形译码（ Fixed-complexity Sphere Decoding, FSD )和非固定复杂度特性的球形译码， 由于 FSD易于 超大规模集成电路 ( Very Large Scale Integration, VLSI )实现， 因此常采用 FSD。 此外， 软输出 MIMO检测能够配合后续的软译码器使系统达到更优 的检测性能， 因此目前常采用 SFSD检测。
通常， LTE-A 系统中的接收机进行 SFSD检测的主要流程包括： 预处 理、 球形检测和似然比（Likelihood Ratio, LLR )输出， 其中所述预处理包 括等效信道矩阵生成、 排序和 QR分解。
由于国际电信联盟 ( International Telecommunications Union, ITU ) 的 高级国际移动通信 ( International Mobile Telecommunications-Advanced , IMT- Advanced )对于系统的高频谱效率要求， LTE-A 系统提出频谱效率要 达到下行 30bps/Hz以及上行 15bps/Hz。 为了满足这些指标， LTE-A系统采 用了下行最多 8层 MIMO， 上行最多 4层 MIMO， 同时支持正交相移键控 ( Quadrature Phase Shift Keying, QPSK ) /16 正交振幅调制 （Quadrature Amplitude Modulation, QAM ) /64QAM调制方式。 由于 MIMO层数大量增 加， 球形检测的运算量变得非常巨大。
名称为一种低复杂度 SFSD 检测方法 （A low-complexity soft mimo detector based on the fixed-complexity sphere decoder ) 的论文 (现有技术 1 ) (作者为 L. G. Barbero, T. Ratnarajah, and C. Cowan, 发表在 IEEE International Conference on acoustics, speech and signal processing (ICASSP，08), Las Vegas, USA, Mar./Apr. 2006 )， 以及名称为用于 MIMO检 测的固定复杂度特性的球形译码 ( Fixing the complexity of the sphere decoder for MIMO detection ) 的论文 （现有技术 2 ) (作者为 L.G.Barbero and J.Thompson, 发表在 IEEE Transaction, on Wireless Communications, vol. 7, no. 6, June 2008 )， 提出了降低 SFSD检测复杂度的方法， 上述方法采用比 特求反（bit-negating ) 的方法构造 ML 的补集路径， 也就是说只计算 ML 路径中每一位比特依次相反的那条路径的度量（比如 64QAM每个符号 6 个比特只有 6条可能的补集路径）， 这样虽然降低了复杂度， 但只计算比特 求反路径所损失的信息量过多， 检测性能不能达到最优的 ML性能。
名称为软输出球形译码： 算法与超大规模集成电路实现 （Soft-Output Sphere Decoding: Algorithms and VLSI Implementation )的论文 (现有技术 3 ) (作者为 Studer, C; Burg, A.; Bolcskei, Η·， 发表在 IEEE Journal on Selected Areas in Communications, vol. 26, no. 2, Feb. 2008 )， 提出了一种非固定复杂 度的球形译码方法， 不像固定复杂度的球形译码那样每层保留固定节点并 且每层检测完所有可能节点后再进入下一层检测， 而是每层检测完一个节 点后就进入下一层的一个节点一直持续到底层， 然后依据检测半径和一定 准则进行路径回溯， 直到找到最优路径。 这样虽然可以达到最优的 ML性 能， 但是复杂度不固定， 当信道随机变化到某种情况时复杂度极高， 不易 实现。 发明内容
有鉴于此， 本发明实施例的主要目的在于提供一种 SFSD检测和装置， 对于大于两层的 MIMO也能获得逼近 ML性能的检测性能， 同时可以满足 目前可接受的硬件实现复杂度。
为达到上述目的， 本发明实施例的技术方案是这样实现的：
本发明实施例提供了一种 SFSD检测方法， 所述方法包括：
对信道响应矩阵进行 QR分解， 得到 Q矩阵和 R矩阵；
将 Q矩阵的共轭转置与接收信号相乘， 得到接收信号的均衡信号； 对均衡信号进行 ML路径检测， 采用各层保留节点逐层递减方式得到
ML路径， 并保留与迭代次数数量相同的分支；
对所述分支进行 ML补集路径检测， 采用所求补集层保留全部节点和 其他层保留节点逐层递减方式得到 ML补集路径；
根据所述 ML路径和 ML补集路径， 得出各层各个符号各个比特的似 然比 （LLR )信息。
较佳地， 所述对均衡信号进行 ML路径检测， 采用各层保留节点逐层 递减方式得到 ML路径， 并保留与迭代次数数量相同的分支， 为：
从矩阵 R中只有一个非零元素对应的那一层， 即顶层开始路径扩展， 直至底层；
保留所述顶层的全部节点， 所述顶层以下各层保留的节点逐层递减， 得到各分支欧氏距离；
选取欧氏距离最小的分支作为 ML路径， 并保留与迭代次数数量相同 的分支。
较佳地， 所述对所述分支进行 ML补集路径检测， 采用所求补集层保 留全部节点和其他层保留节点逐层递减方式得到 ML补集路径， 为：
从保留的各分支的所求补集层开始路径扩展， 直至底层；
保留所述所求补集层的全部节点， 所述所求补集层以下各层保留的节 点逐层递减， 得到各分支欧氏距离；
从所求补集层开始逐层选取欧氏距离最小的分支作为每层的 ML补集 路径。
较佳地， 所述所求补集层以下各层保留的节点逐层递减， 为： 所述所求补集层以下各层保留的节点逐层递减直到底层的节点数为 1。 本发明实施例提供了一种 SFSD检测装置，应用于接收机，所述装置包 括：
QR分解单元， 配置为对信道响应矩阵进行 QR分解， 得到 Q矩阵和 R 矩阵；
均衡信号计算单元， 配置为将 Q矩阵的共轭转置与接收信号相乘， 得 到接收信号的均衡信号；
ML路径检测单元， 配置为对均衡信号进行 ML路径检测， 用各层保留 节点逐层递减方式得到 ML路径， 并保留与迭代次数数量相同的分支；
ML补集路径检测单元， 配置为对 ML路径检测单元保留的分支进行 ML补集路径检测，采用所求补集层保留全部节点和其他层保留节点逐层递 减方式得到 ML补集路径；
软值信息计算单元， 配置为根据 ML路径和 ML补集路径， 得出各层 各个符号各个比特的 LLR信息。
较佳地， 所述 ML路径检测单元， 具体配置为从矩阵 R中只有一个非 零元素对应的那一层， 即顶层开始路径扩展， 直至底层； 保留所述顶层的全部节点， 所述顶层以下各层保留的节点逐层递减， 得到各分支欧氏距离；
选取欧氏距离最小的分支作为 ML路径， 并保留与迭代次数数量相同 的分支。
较佳地， 所述 ML补集路径检测单元， 具体配置为从保留的各分支的 所求补集层开始路径扩展， 直至底层；
保留所述所求补集层的全部节点， 所述所求补集层以下各层保留的节 点逐层递减， 得到各分支欧氏距离；
从所求补集层开始逐层选取欧氏距离最小的分支作为每层的 ML补集 路径。
较佳地， 所述 ML补集路径检测单元， 具体配置为保留所述所求补集 层的全部节点， 所述所求补集层以下各层保留的节点逐层递减直到底层的 节点数为 1， 得到各分支欧氏距离。
由上可知，本发明实施例的技术方案包括：对信道响应矩阵 H进行 QR 分解， 得到 Q矩阵和 R矩阵； 将 Q矩阵的共轭转置^与接收信号相乘， 得到接收信号的均衡信号； 对均衡信号进行 ML路径检测， 采用各层保留 节点逐层递减方式得到 ML路径， 并保留与迭代次数数量相同的分支； 对 所述分支进行 ML补集路径检测， 采用所求补集层保留全部节点和其他层 保留节点逐层递减方式得到 ML补集路径； 根据所述 ML路径和 ML补集 路径， 得出各层各个符号各个比特的 LLR信息； 由此， 本发明实施例对于 大于两层的 MIMO也能获得逼近 ML性能的检测性能， 同时可以满足目前 可接受的硬件实现复杂度。 附图说明
图 1为本发明提供的一种 SFSD检测方法的第一实施例的实现流程图； 图 2为本发明提供的一种 SFSD检测装置的实施例的结构示意图； 图 3为 ML路径检测树图；
图 4为 ML路径检测 4对图；
图 5为 ML补集路径检测树图；
图 6为现有技术 1、 2和 3与本发明实施例 64QAM的吞吐量（ Throughput ) 性能仿真结果对比图；
图 7为现有技术 1、 2和 3与本发明实施例 16QAM的吞吐量性能仿真结 果对比图。 具体实施方式
本发明提供的一种 SFSD检测方法的第一实施例，如图 1所示， 包括以 下步骤：
步骤 101、 对信道响应矩阵进行 QR分解， 得到 Q矩阵和 R矩阵； 步骤 102、 将 Q矩阵的共轭转置与接收信号相乘， 得到接收信号的均 衡信号；
步骤 103、对均衡信号进行 ML路径检测， 采用各层保留节点逐层递减 方式得到 ML路径， 并保留与迭代次数数量相同的分支；
步骤 104、对所述分支进行 ML补集路径检测， 采用所求补集层保留全 部节点和其他层保留节点逐层递减方式得到 ML补集路径；
步骤 105、根据所述 ML路径和 ML补集路径，得出各层各个符号各个 比特的 LLR信息。
优选地， 所述对均衡信号进行 ML路径检测， 采用各层保留节点逐层 递减方式得到 ML路径， 并保留与迭代次数数量相同的分支， 可以为： 从矩阵 R中只有一个非零元素对应的那一层， 即顶层开始路径扩展， 直至底层；
保留所述顶层的全部节点， 所述顶层以下各层保留的节点逐层递减， 得到各分支欧氏距离； 选取欧氏距离最小的分支作为 ML路径， 并保留与迭代次数数量相同 的分支。
优选地， 所述对所述分支进行 ML补集路径检测， 采用所求补集层保 留全部节点和其他层保留节点逐层递减方式得到 ML补集路径， 可以为： 从保留的各分支的所求补集层开始路径扩展， 直至底层；
保留所述所求补集层的全部节点， 所述所求补集层以下各层保留的节 点逐层递减， 得到各分支欧氏距离；
从所求补集层开始逐层选取欧氏距离最小的分支作为每层的 ML补集 路径。
优选地， 所述所求补集层以下各层保留的节点逐层递减， 为： 所述所 求补集层以下各层保留的节点逐层递减直到底层的节点数为 1。
本发明提供的一种 SFSD检测装置的实施例，如图 2所示，所述装置包 括：
QR分解单元 201， 配置为对信道响应矩阵进行 QR分解， 得到 Q矩阵 和 R矩阵；
均衡信号计算单元 202， 配置为将 Q矩阵的共轭转置与接收信号相乘， 得到接收信号的均衡信号；
ML路径检测单元 203， 配置为对均衡信号进行 ML路径检测， 用各层 保留节点逐层递减方式得到 ML路径， 并保留与迭代次数数量相同的分支；
ML补集路径检测单元 204， 配置为对 ML路径检测单元保留的分支进 行 ML补集路径检测， 采用所求补集层保留全部节点和其他层保留节点逐 层递减方式得到 ML补集路径；
软值信息计算单元 205， 配置为根据 ML路径和 ML补集路径，得出各 层各个符号各个比特的 LLR信息。
优选地， 所述 ML路径检测单元 203， 具体配置为从矩阵 R中只有一 个非零元素对应的那一层， 即顶层开始路径扩展， 直至底层； 保留所述顶层的全部节点， 所述顶层以下各层保留的节点逐层递减， 得到各分支欧氏距离；
选取欧氏距离最小的分支作为 ML路径， 并保留与迭代次数数量相同 的分支。
优选地， 所述 ML补集路径检测单元 204， 具体配置为从保留的各分支 的所求补集层开始路径扩展， 直至底层；
保留所述所求补集层的全部节点， 所述所求补集层以下各层保留的节 点逐层递减， 得到各分支欧氏距离；
从所求补集层开始逐层选取欧氏距离最小的分支作为每层的 ML补集 路径。
优选地， 所述 ML补集路径检测单元 204， 具体配置为保留所述所求补 集层的全部节点， 所述所求补集层以下各层保留的节点逐层递减直到底层 的节点数为 1， 得到各分支欧氏距离。
在实际应用中， 所述 QR分解单元 201、 均衡信号计算单元 202、 ML 路径检测单元 203、 ML补集路径检测单元 204、 软值信息计算单元 205可 以由中央处理器 （CPU， Central Processing Unit ), 数字信号处理器 （DSP， Digital Signal Processor )或可编程遝辑阵歹1 J ( FPGA, Field - Programmable Gate Array ) 实现。
本实施例中的 SFSD检测装置可以应用于接收机。
下面对本发明提供的一种 SFSD检测方法的第二实施例进行介绍。 本 实施例中， 假设一个带编码的 MIMO系统有 Mr个发射天线和 M ≥Mr个接 收天线， 编码后的比特流映射到星座图上并形成^个发射符号 其 中0是星座点集合。 则 UE的接收信号可以表示为： = + 其中 H表 示 * 信道响应矩阵， n为噪声。 步骤 301、 对信道响应矩阵 H进行排序和 QR分解， 得到 Q矩阵和 R 矩阵；
这里， 所述 Q矩阵为正交阵， R矩阵为上三角阵。
步骤 302、 将 Q的共轭转置^左乘接收信号 y， 得到接收信号的均衡 信号。
步骤 303、 从 R矩阵中只有一个非零元素的那一层， 即顶层开始检测； 这里， 参见图 3所示， 所述顶层为顶部根节点向下的那一层。
步骤 304、判断根节点向下的分支数是否大于预先设定的值， 比如星座 点数目； 如果小于等于预先设定的值， 进入步骤 305; 如果大于预先设定的 值， 进入步骤 308;
步骤 305、 对于每一条分支， 计算顶层的欧氏距离；
步骤 306、 由顶层向下进行路径扩展， 继续计算其余各层的欧氏距离； 根据预设值保留每层最优的节点及其欧氏距离， 同时累加到该分支的 欧氏距离中。
步骤 307、 判断是否已经到叶子节点， 如果已经到叶子节点， 进入步骤 308; 如果没有到叶子节点， 返回步骤 306;
这里， 参见图 3所示， 所述叶子节点为底层节点。
步骤 308、将顶层根节点向下的一条分支数所得到的完整路径欧氏距离 及对应各层符号对应的星座点序号进行暂存； 返回步骤 304， 开始下一条分 支的处理；
步骤 309、将所有分支的完整欧氏距离按从小到大进行排序， 选取欧氏 距离最小的分支作为 ML路径， 并选取分支中各层符号对应的星座点序号； 同时保留多条最优路径的各层星座点序号， 所述最优路径的条数与迭代次 数相同， 用以后续 ML补集路径时的迭代。
步骤 310、根据 ML路径检测中保留下来的各个分支欧氏距离及各层符 号星座点序号， 得到 ML路径的顶层中每个符号每个比特的补集 这里， 每层每个符号每个比特的全补集要考虑由顶层到底层的所有可 能情况。
步骤 311、对于顶层以下各层的补集， 比如第 iRx层，要经过多次迭代， 每层的迭代依据的路径遵照欧氏距离由小到依次进行； 在一次迭代中， 先 求 iRx层以上各层到该层的欧氏距离；
这里， 要对该层每个符号每个比特进行遍历。
步骤 312、 再求 iRx层以下各层的欧氏距离， 参照步骤 306， 从该层往 下进行路径扩展节点取舍。
步骤 313、判断是否到叶子结点，如果已经到叶子节点，进入步骤 314; 如果没有到叶子节点， 返回步骤 312。
步骤 314、 将步骤 311和 312的结果求和， 选取 iRx层各符号各比特全 补集中的最小值作为该次迭代的最优路径加以暂存。
步骤 315、 判断迭代次数是否已达到预设值， 如果没有达到预设值， 进 入步骤 316; 如果达到预设值， 进入步骤 317。
步骤 316、 迭代次数加 1， 进行下一次迭代。
步骤 317、对于需要求补集的层，可以得到多次迭代所暂存的最优路径， 从这些路径中再选取最小值作为该层该符号该比特的最终补集路径。
步骤 318、 判断是否所有层都已求完补集， 如果没有求完补集， 进入步 骤 319; 如果都已求完补集， 进入步骤 320;
步骤 319、 层数减 1， 进入下一层， 依此类推求得各个层各个符号各个 比特的补集 ； 步骤 320、最终由 ML路径检测得到的 和 ML补集路径搜得到的 ^， 根据公式（2 )求得各个层各个符号各个比特的 LLR信息； ( - )， if ^ = l
( 2 )
其中， 表示第 i个符号第 k个比特位置的值， N。表示噪声。
下面举例说明 ML路径检测的方法。
经过对信道矩阵 H进行 QR分解之后， 可以将路径欧氏距离的求值看 成如图 4所示的树形节点的检测、 取舍的问题。 对于 ML检测， 每个父节 点下面均包含星座点数个子节点， 从根节点到叶节点依此类推； 对于 SFSD 检测， 在 ML路径检测时， 采用一组各层节点保留数对节点树进行取舍， 对于每个父节点， 只保留部分欧氏路径最优的几个子节点， 同时依据该组 参数从上到下进行路径扩展， 最终在保留下来的几条完整路径中选最优作 为 ML路径。 图 3、 4所示为 4*4 MIMO, QPSK的情况， 图 3中所示为各 层保留节点数 二^，" """ (其中 表示父节点下面保留的子结点的个 数)取 (4， 1， 1， 1)，从图 3中可见最终幸存下来的完整路径数为 =Π "'· =4， 即图 3 中的虚线路径和箭头线路径。 最终这 4条路径中的最优路径， 参见 图 4中的箭头线路径， 即为 ML路径。
本发明实施例采用保留节点数逐层递减的统一架构， 以最大层数下每 层配置最高调制时的最大支持分支度量为界， 其他各种多种调制方式和层 数的组合都在该分支度量以下进行配置。 由于先检测层对后检测层 （从树 顶层到底层）有错误传播作用， 顶层保留节点数就等于星座点个数（比如 QPSK取 4， 16QAM取 16， 64QAM取 64 )， 中间各层递减， 最后几层均可 选 1。 不同调制方式和不同层数都遵循这种规则。 比如本发明实施例考虑最 大 4层，每层都配置最大调制方式为 64Q AM时，保留节点数配置为支持 （64 2 1 1)共 128个分支的度量计算， 其他各种调制层数的检测分支都在 128之 目前采用参数如表所 1示:
Figure imgf000014_0001
表 1
对于顶层以下保留节点选 2或 1 的情况， 可以直接根据星座图区域找 到度量值， 不需进行对所有星座点遍历。 因此， 对于最复杂的 64QAM, 4 层（ 64 2 1 1 )， 累计只需要 128条路径 (每条路径对应 4段度量 )独立运算。
下面举例说明 ML补集路径检测的方法。
在 ML路径确定后， 补集路径采用与求 ML路径一致的保留节点数和 扩展路径， 来求各层节点中各比特所对应的补集路径， 具体体现为：
有限全补集： 求 ML路径上各层节点的补集路径， 对于特定的某层， 首先需要确定该层用于补集路径的节点。 本发明实施例采用各比特取有限 全补集的方法， 根据所选参数（迭代次数、 保留节点）指示的范围内， 如 果需要求 ML路径在该层的节点的第 i个比特的补集，需要将该比特处的值 取反， 然后其余比特处的值自由组合得到的节点的集合， 即为补集路径在 该层的节点。 以 QPSK为例， 假如该层 ML路径的节点为 '01，， 对于第一 个比特， 其所对应的该层的补集节点应该是 '10，、 '11' 两个节点； 同样， 对于第二个比特， 对应补集节点为 '00，、 '10，。 可以发现， 两个比特所对 应的补集节点已经包含了 QPSK的四个星座点， 同时 '10， 这个节点为两 者共有， 为避免重复运算， 本发明实施例引入有限全补集的概念， 即在该 层直接遍历星座点， 来替代逐比特求全补集；
路径扩展： 对于位于求补集的层以上的各层节点， 可以复用路径检测 中幸存路径的节点； 而对于该层下面的点， 采用和求 ML路径一致的节点 取舍原理， 根据表 1参数 " "" "" "J 进行路径扩展直至叶子结点。 由于 采用有限全补集方式， 最终保留下来星座点数 (或者其倍数， 视" 2、 参数 而定)条完整路径； 迭代： 对求 ML路径时幸存下来的 =riJ "'条路径， 除去 ML路径外 的路径按欧氏距离从小到大进行排序（比如图 4中的虚线路径）， 根据迭代 次数选取存留路径。 上述有限全补集和路径扩展都是在幸存的 ML路径上 进行的， 迭代就是将这两个步骤依次移植到 ML路径外的其他存留路径中 进行求值。
下面以 4*4MIMO、 QPSK为例， 如图 5所示为迭代次数为 1的情况， 实线路径为 ML 路径检测保留下来的路径， 根据欧氏距离排序有
Figure imgf000015_0001
即为 ML路径。
在求 ML补集路径时， 对于顶层（图 5中 i=4 )由于在算 ML路径时已 经遍历，各比特的补集就直接从实线路径中取。 由于 ML路径节点为 '01，， 因此第一比特的补集在 '10，、 '11， 两个节点所在的完整路径即 、 中取 小， 为 ; 同理， 第二比特的补集为 最终将 、 s 这三个值通过公 式 2组合成该层两比特的 LLR信息。
对于顶层以下 3层的每一层 （1≤'·≤3 )， 依据有限全补集、 路径扩展、 迭代这三个步骤来求补集。 假设迭代次数设为 2。 对于 i=3， 先依托" ^路径
(就是 4巴 路径第一层的节点 '01， 当成父节点）进行有限全补集节点扩充， 如图 5框中所示。 而后对层 2扩充出来的节点往下面几层进行路径扩展直 至叶子结点。对于图 5所示迭代次数为 1的情况， 还要再依托 路径， 同样 进行有限全补集节点扩充及路径扩展。 最终对于层 2来说， 有七条候选的 补集路径 ( 层 2分支旁的三条曲线、 层 2分支旁的三条曲线及 路径）。 由于 ML路径（箭头线所示）层 2的节点为 ' 10，， 第一比特对应全补集节 点为 '00，、 '01，， 因此从七条候选补集路径中抽取层 2节点为 '00，、 '01， 的路径（共 4条）， 在这几条路径中取欧氏最小距离作为 ML补集路径的欧 氏距离； 同理可求第二比特的 ML补集路径的欧氏距离。下面几层（i=2、 1 ) 的 ML补集路径搜索也依此类推。
根据表 1参数 = ("4，"3，"2，"l) 进行ML补集路径的计算，顶层可以复用 ML路径的结果， 以下都是选保留节点 1的情况， 可以直接根据星座图区域 找到度量值， 不需进行对所有星座点遍历。 因此， 对于最复杂的 64QAM、 4层、 迭代次数为 2， i=3层需要 128 (每条路径对应 3段度量 )条路径独立 运算， i=2层需要 128 (每条路径对应 2段度量）条路径独立运算， i=l层 需要 128 (每条路径对应 1段度量 )条路径独立运算。 并且所有层的运算是 可以独立进行的。
下面通过对 LTE-A系统（版本（Release ) 10 )接收机的仿真来说明本 发明实施例的有益效果。 具体仿真条件参照 3GPP标准（Release 8 )， 主要 仿真参数如表 2所示。
参数 ( Parameter ) 单位（Unit ) 数值（Value )
上 下行构造 ( Uplink downlink
1
configuration )
特殊子帧构造 ( Special subframe
4
configuration )
循环前缀（ Cyclic prefix ) 常规（Normal ) 小区识别号 （Cell ID ) 0
传 输 时 间 间 隔 （ Inter-TTI
1
Distance )
混合自动重传请求 （HARQ )进程数 进 程
7
( Number of HARQ processes ) ( Processes )
混合自动重传请求 （HARQ ) 最大重
传数 ( Maximum number of HARQ 4
transmission )
冗余编码序列 ( Redundancy version 对于 QPSK和 16QAM {0,1,2,3} coding sequence ) 对于 64QAM {0,0,1,2} 下行物理控制信道（PDCCH )的正交 OFDM符号
频分复用 （OFDM )符号数（Number 数 ( OFDM 2
of OFDM symbols for PDCCH ) symbols )
命令正确应答 /命令不正确应答反馈
复用 ( Multiplexing ) 模式（ ACK/NACK feedback mode ) 小区参考信号 ( Cell-specific reference 天线端口（Antenna port ) 0， symbols ) 1 天线端 口 （ Antenna port ) 用户参考信号 ( UE-specific reference
7,8,9,10
symbols )
分配资源块数目 （ Number of allocated 物理资源块
50
resource blocks ) ( PRB )
64 QAM, 3/4码率； 16 QAM, 调制编码方式（MCS )
1/2码率；
传输信道条件 ( Propagation Condition ) 增强步行信道（EPA ) 5 相关矩阵和天线构造 ( Correlation 4x4 层 （layer ), 低相关 ( Low Matrix and Antenna Configuration ) correlation ), 1个码字（CW )对 应 2层（ layer )
表 2
图 6为现有技术 1、 2和 3与本发明实施例 64QAM的吞吐量性能仿真结 果对比图， 图 7为现有技术 1、 2和 3与本发明实施例 16QAM的吞吐量性能 仿真结果对比图； 图 6、 图 7中带有十字的曲线表示现有技术 1、 2的吞吐 量性能， 带有小圓圈的曲线表示现有技术 3 的吞吐量性能， 带有正方形的 曲线表示本发明实施例的吞吐量性能， 带有星形的曲线表示吞吐量性能的 下限（最小均方误差 MMSE )。 当修剪等级参数 Lmax>0.2时可以认为逼近 ML性能， 图 6中现有技术 3的曲线为 Lmax取 0.5时得到的曲线， 图 7中 现有技术 3的曲线为 Lmax取 1时得到的曲线。
参见图 6所示， 带有十字的曲线位于带有星形的曲线之下， 相同吞吐 量的情况下， 现有技术 1、 2的信噪比更高， 这表明现有技术 1、 2的吞吐 量性能没有达到吞吐量性能的下限， 也就是说性能很差不能达到 ML性能。
参见图 6、 7所示， 带有小圓圈的曲线位于带有正方形的曲线之上， 相 同吞吐量的情况下， 本发明实施例的信噪比略高于现有技术 3 的信噪比， 图 6中在吞吐量为 70%时， 本发明实施例和现有技术 3的信噪比差小于 0.3 分贝 UB )， 图 7中在吞吐量为 70%时， 本发明实施例和现有技术 3的信噪 比差小于 0.2dB， 由此可知本发明实施例逼近 ML性能。
但是， 本发明实施例的复杂度远小于现有技术 3的复杂度， 表 3为本 发明实施例的复杂度统计表（64 QAM 2、 3、 4层、 16QAM 2、 3、 4层）， 表 4为现有技术 3的复杂度统计表 ( 16QAM, 4层），表 5为信噪比为 30dB 时现有技术 3的复杂度统计表 ( 64QAM, 4层）。 由表 3可知本发明实施例 的复杂度固定， 由表 4可知现有技术 3的复杂度随信噪比的不同而变化， 由表 5可知高信噪比时现有技术 3的复杂度很高。
参见表 3和表 4所示，例如对于 16QAM, 4层，本发明实施例的加法为 12794次， 乘法为 84次， 现有技术 3的加法在 Lmax为 1时均值为 22461， 乘法 Lmax为 1时均值为 14757; 参见表 3和表 5所示， 例如对于 64QAM, 4层， 本发明实施例的加法为 60370次， 乘法为 124次， 现有技术 3的加法 在 Lmax为 0.5、 信噪比为 30dB时加法 292130次， 乘法为 40459次。 通过 对比可知， 本发明实施例的加法和乘法的次数均比现有技术 3 的加法和乘 法次数大幅减少， 其中乘法的次数减少尤其显著， 这表明本发明实施例的 复杂度较之现有技术 3大大降低。
64QAM, 3 16QAM, 4 16QAM, 3 16QAM, 2
64QAM, 4层 64QAM, 2层
层 层 层
加法 60370 27833 4472 12794 5952 1032 乘法 124 75 42 84 51 20
表 3
Figure imgf000019_0001
表 4
Figure imgf000019_0002
表 5
本发明实施例适用于终端接收机， 也适用于基站接收机。 采用本发明 实施例可使 MIMO检测获得逼近 ML的性能， 同时大大降低了复杂度， 使 MIMO检测具有可实现性。本发明实施例所采用的迭代处理没有反馈机制， 可以并行进行， 并且各个分支采用统一的处理单元和参数， 因此本发明实 施例的处理时延短， 緩存量小， 适合并行流水结构， 在终端接收机的实现 中更有优势。
以上所述， 仅为本发明的较佳实施例而已， 并非用于限定本发明的保 护范围。

Claims

权利要求书
1、一种软输出固定复杂度球形译码（SFSD )检测方法，所述方法包括： 对信道响应矩阵进行 QR分解， 得到 Q矩阵和 R矩阵；
将 Q矩阵的共轭转置与接收信号相乘， 得到接收信号的均衡信号； 对均衡信号进行最大似然（ML )路径检测， 采用各层保留节点逐层递 减方式得到 ML路径， 并保留与迭代次数数量相同的分支；
对所述分支进行 ML补集路径检测， 采用所求补集层保留全部节点和 其他层保留节点逐层递减方式得到 ML补集路径；
根据所述 ML路径和 ML补集路径， 得出各层各个符号各个比特的似 然比 （LLR )信息。
2、 根据权利要求 1所述的方法， 其中， 所述对均衡信号进行 ML路径 检测， 采用各层保留节点逐层递减方式得到 ML路径， 并保留与迭代次数 数量相同的分支， 为：
从矩阵 R中只有一个非零元素对应的那一层， 即顶层开始路径扩展， 直至底层；
保留所述顶层的全部节点， 所述顶层以下各层保留的节点逐层递减， 得到各分支欧氏距离；
选取欧氏距离最小的分支作为 ML路径， 并保留与迭代次数数量相同 的分支。
3、 根据权利要求 2所述的方法， 其中， 所述对所述分支进行 ML补集 路径检测， 采用所求补集层保留全部节点和其他层保留节点逐层递减方式 得到 ML补集路径， 为：
从保留的各分支的所求补集层开始路径扩展， 直至底层；
保留所述所求补集层的全部节点， 所述所求补集层以下各层保留的节 点逐层递减， 得到各分支欧氏距离；
从所求补集层开始逐层选取欧氏距离最小的分支作为每层的 ML补集 路径。
4、 根据权利要求 3所述的方法， 其中， 所述所求补集层以下各层保留 的节点逐层递减， 为：
所述所求补集层以下各层保留的节点逐层递减直到底层的节点数为 1。
5、一种软输出固定复杂度球形译码（SFSD )检测装置，所述装置包括： QR分解单元， 配置为对信道响应矩阵进行 QR分解， 得到 Q矩阵和 R 矩阵；
均衡信号计算单元， 配置为将 Q矩阵的共轭转置与接收信号相乘， 得 到接收信号的均衡信号；
ML路径检测单元， 配置为对均衡信号进行 ML路径检测， 用各层保留 节点逐层递减方式得到 ML路径， 并保留与迭代次数数量相同的分支；
ML补集路径检测单元， 配置为对 ML路径检测单元保留的分支进行 ML补集路径检测，采用所求补集层保留全部节点和其他层保留节点逐层递 减方式得到 ML补集路径；
软值信息计算单元， 配置为根据 ML路径和 ML补集路径， 得出各层 各个符号各个比特的 LLR信息。
6、 根据权利要求 5所述的装置， 其中， 所述 ML路径检测单元， 配置 为从矩阵 R中只有一个非零元素对应的那一层， 即顶层开始路径扩展， 直 至底层；
保留所述顶层的全部节点， 所述顶层以下各层保留的节点逐层递减， 得到各分支欧氏距离；
选取欧氏距离最小的分支作为 ML路径， 并保留与迭代次数数量相同 的分支。
7、 根据权利要求 6所述的装置， 其中， 所述 ML补集路径检测单元， 配置为从保留的各分支的所求补集层开始路径扩展， 直至底层；
保留所述所求补集层的全部节点， 所述所求补集层以下各层保留的节 点逐层递减， 得到各分支欧氏距离；
从所求补集层开始逐层选取欧氏距离最小的分支作为每层的 ML补集 路径。
8、 根据权利要求 7所述的装置， 其中， 所述 ML补集路径检测单元， 配置为保留所述所求补集层的全部节点， 所述所求补集层以下各层保留的 节点逐层递减直到底层的节点数为 1， 得到各分支欧氏距离。
PCT/CN2013/084017 2012-11-29 2013-09-23 一种软输出固定复杂度球形译码检测方法和装置 WO2014082487A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US14/648,089 US9356733B2 (en) 2012-11-29 2013-09-23 Method and apparatus for soft output fixed complexity sphere decoding detection
EP13858769.6A EP2928084B1 (en) 2012-11-29 2013-09-23 Method and apparatus for soft output fixed complexity sphere decoding detection

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201210499242.0 2012-11-29
CN201210499242.0A CN103856254B (zh) 2012-11-29 2012-11-29 一种软输出固定复杂度球形译码检测方法和装置

Publications (1)

Publication Number Publication Date
WO2014082487A1 true WO2014082487A1 (zh) 2014-06-05

Family

ID=50827155

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2013/084017 WO2014082487A1 (zh) 2012-11-29 2013-09-23 一种软输出固定复杂度球形译码检测方法和装置

Country Status (4)

Country Link
US (1) US9356733B2 (zh)
EP (1) EP2928084B1 (zh)
CN (1) CN103856254B (zh)
WO (1) WO2014082487A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10171207B2 (en) * 2017-04-26 2019-01-01 Cavium, Llc Methods and apparatus for control bit detection
US10523480B1 (en) * 2018-11-08 2019-12-31 Nxp B.V. K-bit enumerative sphere shaping of multidimensional constellations
US10904131B1 (en) * 2020-07-08 2021-01-26 Eci Telecom Ltd. Systems and methods for configuring a communications network
US10903904B1 (en) 2020-07-08 2021-01-26 Eci Telecom Ltd. Systems and methods for configuring a communications network
CN115051734B (zh) * 2022-05-24 2023-09-19 西安电子科技大学 一种固定复杂度球形译码的改进方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101373975A (zh) * 2007-08-22 2009-02-25 株式会社Ntt都科摩 一种球形译码方法
CN101557269A (zh) * 2009-05-18 2009-10-14 北京天碁科技有限公司 一种基于超大规模集成电路的球形译码检测方法
CN101582750A (zh) * 2009-06-02 2009-11-18 北京天碁科技有限公司 一种基于宽度优先搜索的球形译码检测方法
WO2010000075A1 (en) * 2008-07-03 2010-01-07 Eth Zurich Computation of extrinsic information in a branch-and-bound detector
US20100014606A1 (en) * 2008-07-16 2010-01-21 Industrial Technology Research Institute Symbol detector and sphere decoding method

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4387282B2 (ja) * 2004-10-20 2009-12-16 株式会社エヌ・ティ・ティ・ドコモ 信号分離装置及び信号分離方法
JP4776692B2 (ja) * 2006-09-25 2011-09-21 パナソニック株式会社 信号分離装置及び信号分離方法
CN101022325A (zh) * 2007-03-23 2007-08-22 东南大学 多输入多输出天线系统空分复用的最大似然简化检测方法
EP2341676A1 (en) * 2009-12-30 2011-07-06 ST-Ericsson SA Branch processing of search tree in a sphere decoder
CN102594467B (zh) * 2011-08-15 2014-07-02 上海交通大学 无线多输入多输出系统的接收机检测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101373975A (zh) * 2007-08-22 2009-02-25 株式会社Ntt都科摩 一种球形译码方法
WO2010000075A1 (en) * 2008-07-03 2010-01-07 Eth Zurich Computation of extrinsic information in a branch-and-bound detector
US20100014606A1 (en) * 2008-07-16 2010-01-21 Industrial Technology Research Institute Symbol detector and sphere decoding method
CN101557269A (zh) * 2009-05-18 2009-10-14 北京天碁科技有限公司 一种基于超大规模集成电路的球形译码检测方法
CN101582750A (zh) * 2009-06-02 2009-11-18 北京天碁科技有限公司 一种基于宽度优先搜索的球形译码检测方法

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
L. G. BARBERO; T. RATNARAJAH; C. COWAN: "A low-complexity soft MIMO detector based on the fixed-complexity sphere decoder", IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP'08, March 2006 (2006-03-01)
L.G.BARBERO; J.THOMPSON: "Fixing the complexity of the sphere decoder for MIMO detection", IEEE TRANSACTION. ON WIRELESS COMMUNICATIONS, vol. 7, no. 6, June 2008 (2008-06-01), XP011216351
STUDER, C.; BURG, A.; BOLCSKEI, H.: "Soft-Output Sphere Decoding: Algorithms and VLSI Implementation", IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, vol. 26, no. 2, February 2008 (2008-02-01), XP011200796, DOI: doi:10.1109/JSAC.2008.080206

Also Published As

Publication number Publication date
CN103856254A (zh) 2014-06-11
US20150304069A1 (en) 2015-10-22
CN103856254B (zh) 2017-09-12
EP2928084B1 (en) 2019-02-06
EP2928084A4 (en) 2015-12-09
US9356733B2 (en) 2016-05-31
EP2928084A1 (en) 2015-10-07

Similar Documents

Publication Publication Date Title
KR101301189B1 (ko) 연속하는 간섭 소거를 이용하여 ｍｉｍｏ 시스템에서 디코딩 순서를 결정하기 위한 방법들
US10491245B2 (en) Method and apparatus for grid mapping in a wireless communication system
US20220060365A1 (en) Mimo-ofdm wireless signal detection method and system capable of channel matrix pre-processing during detection
US8391429B2 (en) Methods for determining reconstruction weights in a MIMO system with successive interference cancellation
US8306165B2 (en) Traffic-to-pilot ratio estimation for MIMO-OFDM system
JP5185434B2 (ja) バッファされた復調機能およびデマップ機能
TW200304307A (en) Data transmission with non-uniform distribution of data rates for a multiple-input multiple-output (mimo) system
WO2014082487A1 (zh) 一种软输出固定复杂度球形译码检测方法和装置
TW200917698A (en) Near soft-output maximum-likelihood detection for multiple-input multiple-output systems
US20110122004A1 (en) Method of decoding a signal implementing a progressive construction of a decoding tree, corresponding computer program and decoding device
CN107094063B (zh) 半穷举迭代块解码方法和设备
US8111770B1 (en) High-bandwidth over-the-air signal processing
US9118446B2 (en) Methods and user equipment in a wireless communication network
US10084526B2 (en) Path detection method and device, and sphere decoding detection device
JP5487090B2 (ja) 無線信号処理方法および無線通信装置
US11791880B2 (en) MIMO-OFDM wireless signal detection method and system capable of channel matrix pre-processing before detection
US9066247B2 (en) Communication devices and methods for signal detection
JP2012060407A (ja) 受信装置、通信システム、受信装置の制御プログラムおよび集積回路
TW201218701A (en) A method and apparatus for detecting a plurality of symbol blocks using a decoder
JP2014204254A (ja) 端末装置、基地局装置、無線通信システムおよび通信方法
KR20160024375A (ko) 무선 통신 네트워크들을 위한 개선된 수신기
CN116708094A (zh) 多输入多输出系统的检测方法和装置、设备、介质
KR20200094611A (ko) 5g 통신 시스템을 위한 nsc 데이터 검출 방법
WO2021171069A1 (en) Multiple-input and multiple-output detection based on transmitted signal streams permutation
Knievel et al. Evaluation and extension of a multi-dimensional graph-based receiver concept for MIMO-OFDM

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 13858769

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

WWE Wipo information: entry into national phase

Ref document number: 2013858769

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 14648089

Country of ref document: US