WO2015060130A1 - 疎グラフ作成装置及び疎グラフ作成方法 - Google Patents
疎グラフ作成装置及び疎グラフ作成方法 Download PDFInfo
- Publication number
- WO2015060130A1 WO2015060130A1 PCT/JP2014/076910 JP2014076910W WO2015060130A1 WO 2015060130 A1 WO2015060130 A1 WO 2015060130A1 JP 2014076910 W JP2014076910 W JP 2014076910W WO 2015060130 A1 WO2015060130 A1 WO 2015060130A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- node
- sparse
- unit
- sparse graph
- graph
- Prior art date
Links
Images
Classifications
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/61—Aspects and characteristics of methods and arrangements for error correction or error detection, not provided for otherwise
- H03M13/615—Use of computational or mathematical techniques
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/03—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words
- H03M13/033—Theoretical methods to calculate these checking codes
- H03M13/036—Heuristic code construction methods, i.e. code construction or code search based on using trial-and-error
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/901—Indexing; Data structures therefor; Storage structures
- G06F16/9024—Graphs; Linked lists
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/03—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words
- H03M13/05—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words using block codes, i.e. a predetermined number of check bits joined to a predetermined number of information bits
- H03M13/11—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words using block codes, i.e. a predetermined number of check bits joined to a predetermined number of information bits using multiple parity bits
- H03M13/1102—Codes on graphs and decoding on graphs, e.g. low-density parity check [LDPC] codes
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/03—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words
- H03M13/05—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words using block codes, i.e. a predetermined number of check bits joined to a predetermined number of information bits
- H03M13/11—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words using block codes, i.e. a predetermined number of check bits joined to a predetermined number of information bits using multiple parity bits
- H03M13/1102—Codes on graphs and decoding on graphs, e.g. low-density parity check [LDPC] codes
- H03M13/1148—Structural properties of the code parity-check or generator matrix
- H03M13/118—Parity check matrix structured for simplifying encoding, e.g. by having a triangular or an approximate triangular structure
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/03—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words
- H03M13/05—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words using block codes, i.e. a predetermined number of check bits joined to a predetermined number of information bits
- H03M13/11—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words using block codes, i.e. a predetermined number of check bits joined to a predetermined number of information bits using multiple parity bits
- H03M13/1102—Codes on graphs and decoding on graphs, e.g. low-density parity check [LDPC] codes
- H03M13/1191—Codes on graphs other than LDPC codes
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/37—Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35
- H03M13/373—Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35 with erasure correction and erasure determination, e.g. for packet loss recovery or setting of erasures for the decoding of Reed-Solomon codes
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/37—Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35
- H03M13/3761—Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35 using code combining, i.e. using combining of codeword portions which may have been transmitted separately, e.g. Digital Fountain codes, Raptor codes or Luby Transform [LT] codes
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/63—Joint error correction and other techniques
- H03M13/635—Error control coding in combination with rate matching
- H03M13/6362—Error control coding in combination with rate matching by puncturing
- H03M13/6368—Error control coding in combination with rate matching by puncturing using rate compatible puncturing or complementary puncturing
- H03M13/6393—Rate compatible low-density parity check [LDPC] codes
Definitions
- the present invention relates to an apparatus and method for generating a sparse graph code, and more particularly, to an apparatus and method for efficiently enabling a sparse graph code by taking a large minimum loop formed between arbitrary nodes.
- error correction technology is widely used in various communication systems such as satellite digital broadcasting, communication on the Internet, and communication on mobile terminals.
- various communication systems such as satellite digital broadcasting, communication on the Internet, and communication on mobile terminals.
- video distribution services using the Internet are expected, and error correction technology for the Internet network has become important.
- the following is an example of the Internet network.
- the Internet network can be regarded as an erasure communication path (PEC: Packet Erasure Channel).
- PEC Packet Erasure Channel
- Fig. 1 binary erasure channel
- Fig. 1 a binary erasure channel grouped as a packet unit, and whether the channel output is output as correct information or as unknown due to some failure in the line.
- FEC Forward Error Correction
- ARQ Automatic Repeat reQuest
- FEC method a method that can correct errors only on the receiving side without requiring a feedback channel Is used. Since the FEC method does not perform retransmission control or the like, the FEC method is also used in a video conference system in which the delay is small and the real-time property is important. The following will discuss the FEC method.
- Reed-Solomon code As the FEC method, Reed-Solomon code (RS code) is widely used in digital broadcasting and the like. In Japanese digital broadcasting, it is defined as a code length of 204 bytes, and by adding a parity byte of about 10 [%] to the original data of 188 bytes, tolerance is given to errors. However, it is generally known that performance is improved if a code having a long code length is used as an error correction code. However, there is a known disadvantage that decoding becomes complicated and the amount of calculation becomes enormous as the code length increases. ing. For this reason, it is assumed that the RS code is handled with a code length of 256 bytes or less. In addition, when the RS code is applied to a packet called IP-based packet level FEC, it is necessary to handle 256 packets as one block for the above reason.
- IP-based packet level FEC it is necessary to handle 256 packets as one block for the above reason.
- a decoding method based on a probability propagation algorithm is known to have excellent decoding characteristics with a practical amount of computation when the code length is long, and is defined by a sparse graph
- a low-density parity check code (LDPC code) (for example, see Non-Patent Document 1), which is a linear code, is attracting attention as a practical error correction method that approaches the channel capacity defined by Shannon.
- LDPC code low-density parity check code
- erasure correction codes based on sparse graphs Digital Fountain's LT code (for example, see Non-Patent Document 2) and Raptor code (for example, Non-Patent Document 3) are known, and the coding efficiency is greatly degraded.
- Non-Patent Document 4 Asynchronous Layered Coding (for example, see Non-Patent Document 4)), which is an Internet multicast protocol. Therefore, it is widely used in layered multicast communication.
- the correspondence between the sparse matrix and the graph corresponds to an information node corresponding to a column of the matrix, a check node corresponding to a row, and one place of the sparse matrix corresponding to an edge connecting the information node and the check node.
- an irregular matrix (hereinafter referred to as an irregular matrix) with several column weights
- the PEG algorithm makes it difficult to find a node that will grow in a loop after a heavy column or row is entered. It becomes easy to happen to be a randomly generated sparse graph. This is because edges increase and loops are more likely to be born.
- the present invention has been made in view of the above points, and an object of the present invention is to efficiently create a sparse graph in a sparse graph code.
- the present invention relates to a method for generating a sparse matrix for an LDPC code, and in particular, an improved PEG (progressive edge-growth) algorithm between arbitrary node sets in a sparse matrix.
- PEG progressive edge-growth
- the efficiency of encoding according to the transmission rate is improved by selecting a node set that takes a long loop according to a sparse matrix request.
- the sparse graph creation device is: A sparse graph creation device for generating a sparse graph used for a sparse graph code, An inactivator that inactivates at least some of the nodes in the sparse graph; From the remaining sparse graphs made inactive by the inactive unit, a node that can be reached through the shortest number of edges from an active node connected to an arbitrary root node as a route when creating a local graph is searched.
- a search unit to A node selection unit that selects a node that satisfies a condition set from a search result of the search unit as a position for generating an edge;
- a root node edge connection unit that connects the node selected by the node selection unit and the root node; Is provided.
- the node selection unit performs a loop with a small amount of computation by limiting the range to be searched next from the information related to the search obtained from the process of node generation so far. You may select the node taken large.
- the sparse graph creation device is: A sparse graph creation device for generating a sparse graph used for a sparse graph code, A search unit that searches for a node that can be reached through the shortest number of edges from an active node connected to an arbitrary root node that is a root when creating a local graph from a sparse graph; A node selection unit that selects a node that satisfies a condition set from a search result of the search unit as a position for generating an edge; A root node edge connection unit that connects the node selected by the node selection unit and the root node; A constrained interleaving unit that rearranges each node of the sparse graph in which an edge is generated at a position selected by the root node edge connection unit so that each weight of the sparse matrix is spatially uniformly arranged; Is provided.
- the constrained interleave unit may replace nodes in the sparse graph so as to maintain the graph structure of active nodes.
- the sparse graph creation apparatus may further include a node expansion unit that installs a new node so that the total number of edges of the sparse graph does not change and the minimum loop is not shortened.
- the sparse graph creation method is: A sparse graph creation method for generating a sparse graph used for a sparse graph code, An initiating procedure to inactivate at least some of the nodes in the sparse graph; From the remaining sparse graphs made inactive by the inactivation step, search for nodes that can be reached through the shortest number of edges from an active node connected to an arbitrary root node as a route for creating a local graph.
- a root node edge connection procedure for connecting the node selected in the node selection procedure and the root node; In order.
- a loop in the node selection procedure, a loop can be formed with a small amount of computation by limiting the range to be searched next from information related to the search obtained from the process of node generation so far. You may select the node taken large.
- the sparse graph creation method is: A sparse graph creation procedure for generating a sparse graph used for a sparse graph code, A search procedure for searching for a node that can be reached from an active node connected to an arbitrary root node as a route when creating a local graph through the shortest number of edges from a sparse graph, A node selection procedure for selecting a node that satisfies the conditions set from the search result of the search procedure as a position for generating an edge; A root node edge connection procedure for connecting the node selected in the node selection procedure and the root node; A constrained interleaving procedure for rearranging each node of the sparse graph in which an edge is generated at a position selected by the root node connection procedure so that each weight of the sparse matrix is spatially uniformly arranged; In order.
- the nodes in the sparse graph may be replaced so as to maintain the graph structure of the active nodes.
- a node expansion procedure for installing a new node is performed after the node selection procedure and the constraints so that the total number of edges of the sparse graph is not changed and the minimum loop is not shortened. You may also have before the attached interleaving procedure.
- a sparse graph in a sparse graph code can be created efficiently.
- An example of a packet loss communication path is shown.
- An example of the communication system which concerns on this embodiment is shown.
- generation apparatus which concerns on Embodiment 1 is shown.
- saved at the sparse matrix cache part 102 is shown.
- An example of tree expansion of the parity check matrix of an LDGM code is shown.
- An example of a loop of length 4 is shown.
- An example of a loop of length 6 is shown.
- An example of inactivity is shown.
- generation apparatus which concerns on Embodiment 2 is shown.
- An example of the matrix before interleaving is shown.
- An example of the matrix after interleaving is shown.
- An example of two matrices before interleaving is shown.
- FIG. 10 shows a second configuration example of a sparse matrix creation device according to Embodiment 3.
- FIG. 10 shows a second configuration example of a sparse matrix creation device according to Embodiment 3.
- the present invention can be implemented without depending on the type of communication path, but as an example, packet distribution using the UDP protocol used in multicast distribution will be described as an example.
- UDP protocol when used, even if the UDP packet is lost, unlike the TCP protocol, retransmission processing is not performed. However, as shown in FIG. 1, it is possible to determine whether or not the received packet is correct and the position of the lost packet is There is a feature that can be seen as an erasure channel that can be accurately identified.
- FIG. 2 shows an example of a communication system according to the present embodiment.
- the communication system according to this embodiment includes a channel encoding device 91, a packet transmission device 92, a packet base receiving device 93, and a channel decoding device 94.
- the communication path encoding device 91 encodes transmission data using a sparse matrix.
- the packet transmission device 92 transmits the encoded transmission data.
- the packet base receiving device 93 receives transmission data.
- the communication path decoding device 94 decodes received data using a sparse matrix.
- the encoding in the channel encoding device 91 and the decoding in the channel decoding device 94 use the sparse matrix created by the sparse matrix creating unit 10 that functions as the sparse matrix creating device according to the present embodiment.
- Transmission data such as video data and audio data is encoded by the communication path encoding device 91.
- the channel encoding device 91 normally performs fragmentation in consideration of the packet size of the packet transmission device 92 connected later.
- a linear code such as an LDPC code
- the channel encoding device 91 creates redundant data by the following encoding process.
- S is input data such as video data fragmented with a certain size
- G and T are sparse matrices corresponding to the sparse graph.
- T is a step matrix. T -1 can be replaced with an accumulator, and the use of an accumulator can speed up the encoding process.
- the sparse matrix creation unit 10 in the figure generates matrices G and T that improve the encoding efficiency and decoding efficiency with a small amount of computation. Since the amount of information of the created sparse matrix is large, generally the parameters of the sparse matrix are extracted and notified to the decoding side communication path decoding device 94. The parameters are, for example, generated parity data and FEC auxiliary information.
- the packet-based transmission device 92 transmits the extracted parameters together with video data and audio data in accordance with the FLUTE standard (RFC 3926), the MPEG media trans-port (ISO / IEC 23008-1) standard, and the like.
- the packet-side receiving device 93 on the receiving side receives the packet transmitted from the packet-based transmission device 92.
- the packet-based receiving device 93 attempts to recover the lost packet by the communication path decoding device 94 when packet loss has occurred in comparison with the FLUTE header format or the like.
- the communication path decoding device 94 corresponds to the encoding, and it is shared in advance from the auxiliary information that the following constraint equation holds.
- G and T are sparse matrices corresponding to the sparse graph, and use exactly the same information as the encoding side.
- the sparse matrix information has a large amount of data, it is generally shared by being notified as a parameter from the encoding side to the decoding side.
- the same sparse matrix as that on the encoding side is generated from the notified parameter by the sparse matrix creation unit 10.
- the lost packet is decoded using the above equation. Now, if ⁇ is an index set of lost packets, H ⁇ is a parity check matrix corresponding to lost packets, and x ⁇ is a lost packet set, the above equation can be rewritten as:
- ⁇ ′ on the right side is an index set of received packets.
- the lost packets can be recovered by maximum likelihood decoding.
- Various decoding methods have been proposed, including a method of recovery by a message passing algorithm and a method of recovery by a Gaussian elimination method.
- the message passing algorithm decoding utilizing the sparseness of the sparse graph is possible, so that decoding with O (N) is possible.
- O (N) is possible.
- preprocessing for obtaining an inverse matrix is performed with O (N 3 ), O (N 2 ) computation is required for computation processing that actually recovers the packet.
- a sparse matrix having high error correction capability and high coding efficiency can be created at high speed in a channel encoding device and a channel decoding device, and this matrix can be eliminated by a message passing algorithm or Gaussian elimination.
- the method can also correct errors efficiently. Furthermore, by combining with a variable rate LDPC code, the matrix does not have to be recreated from 0 each time, so even when one sparse matrix is applied to various rates, error correction with high coding efficiency can be performed. it can.
- the selective PEG algorithm generates a sparse matrix while keeping row weight / column weight at an arbitrary multilevel, and in the process, an arbitrary edge is made inactive so that the matrix space is narrow. Improve coding efficiency.
- the sparse graph creation method includes an inactivation procedure, a search procedure, a node selection procedure, and a root node edge connection procedure in this order.
- the inactivation procedure at least some of the nodes in the sparse graph are made inactive.
- the search procedure a node that can be reached through the shortest number of edges from an active node connected to an arbitrary root node as a root when creating a local graph is searched from the remaining sparse graphs that have been made inactive.
- the node selection procedure a node satisfying the set condition is selected from the search result.
- the root node edge connection procedure the node selected in the node selection procedure is connected to the root node.
- FIG. 3 shows a basic configuration diagram of a sparse matrix creation device according to this embodiment.
- the sparse matrix creation device 100 includes an algebraic structure generation unit 101, a sparse matrix cache unit 102, a random number generation unit 103, a root node edge connection unit 104, a node inactivation unit 105, an inactive A control unit 106, a search depth control unit 107, a search unit 108, a node selection unit 109, and an interleaving unit 110.
- the sparse matrix creation unit 100 and the sparse graph creation method according to the present embodiment realize a variable rate error correction code that achieves multilevel rate control with high efficiency.
- the algebraic structure generation unit 101 is considered when the sparse graph has a special structure.
- the sparse matrix cache unit 102 stores a sparse matrix to be generated.
- the random number generation unit 103 constructs a code probabilistically.
- the root node edge connection unit 104 selects an edge to be connected to the root node as a reference when a large loop in the local graph is taken.
- the node inactivating unit 105 selects nodes that are not considered when taking a large loop in the local graph.
- the inactivation control unit 106 determines the conditions in the processing of the node inactivation unit 105.
- the search depth control unit 107 controls the depth of a node to be searched when taking a large loop in the local graph.
- the search unit 108 searches for a node that can be reached from the active node connected to the root node through the shortest number of edges based on the control information of the search depth control unit 107.
- the node selection unit 109 selects a node that satisfies the condition from the search result of the search unit 108.
- the condition is, for example, a check node that cannot be reached from the root node through any five edges.
- the interleave unit 110 is a constrained interleave unit that replaces the created sparse matrix by column replacement or the like. Each component will be described below.
- This system is a communication path decoding system that efficiently recovers erasures that occur on erasure communication paths.
- IP packet transmission is performed on the Internet, which is a typical communication path of an erasure communication path, and LDPC configured by a sparse graph -An example that assumes a Staircase code will be described.
- the input data is input to the sparse matrix creation device 100 and creation of the sparse matrix is started.
- information necessary for generating a matrix such as the size of the matrix related to the coding rate, information about the weight of the matrix, and the initial value of the search depth, are input.
- the matrix size is, for example, a vertical size and a horizontal size.
- the column weight 3 is the number of columns and the column weight 7 is the number of columns.
- the algebraic structure generation unit 101 assumes the LDPC-Staircase code this time, a step matrix is generated as an algebraic structure.
- the generated staircase matrix is stored in the sparse matrix cache unit 102.
- the root node when creating the local graph at the root node edge connection unit 104 and one node are connected through the edge.
- the local graph is a graph composed of candidate nodes connected from the root node in the sparse graph generation process.
- the root node is a node that is a target to which a new edge is added, and is a node arranged at the top when represented by a tree.
- the node inactivation unit 105 designates nodes that are not included in the search when creating a local graph based on information of the inactivation control unit 106. However, when there is no special request such as extremely reducing the search time in generating the matrix, there is no merit of making the node inactive. Normally, you do not specify the node to be inactive.
- the search unit 108 searches for an active node that has not been made inactive based on the information of the search depth control unit 107.
- the search operation will be described with reference to FIGS.
- FIG. 4 shows an example of a matrix stored in the sparse matrix cache unit 102
- FIG. 5 shows a connection relationship between a root node and a check node connected with the minimum number of edges.
- the search unit 108 performs a one-step search.
- the one-step search is a search for a node in which a root node and an active check node are connected through three edges (connected by a length of 3). In the example shown in FIG.
- the node selection unit 109 receives the search result of the search unit 108 and determines the node position to be added. For example, a node that cannot be reached from the root node through any three edges.
- a method such as selecting from a candidate node set at random, filling from the upper part of the matrix, or filling from the lowest row weight is used.
- a random number generated by the random number generation unit 103 is used.
- the process returns to the root node edge connection unit 104, and this time, an edge with the selected node is added to the node selection unit 109 and the same processing as 105, 108, 109 is performed. Do.
- an arbitrary number of nodes eg, three nodes, etc.
- the selected node is stored in the sparse matrix cache unit 102.
- the search depth control unit 107 changes the search depth to narrow the search range.
- the inactivation control unit 106 selects a node to be inactive, and performs a process of reducing active nodes to be searched. These processes can be performed one by one or simultaneously.
- the search range control unit 107 changes the depth of search, and there is no node that can be selected even in one-step search. In this case, it is effective to change the setting of the inactivation control unit 106, make the node generated so far inactive, increase the search depth of the search range control unit 107 again, and continue matrix generation.
- the processing of the node inactivating unit 105 will be described with reference to FIG.
- the matrix on the left in FIG. 8 is in the middle of the process of generating the matrix.
- the node in the ninth column is to be created, and the first check node is selected as the root node.
- the second, third, fourth, fifth, and all other check nodes are connected with an edge of length 3, and no matter which node is selected, the length This is a situation where a loop of 4 occurs. Therefore, the inactivation control unit 106 determines that half of the already created nodes are inactive, and the node inactivation unit 105 inactivates the nodes from the first column to the fourth column (right in FIG. 8). .
- any of the third, fourth, and fifth check nodes can be selected.
- the sparse matrix creation device 100 can achieve excellent coding efficiency even when a part of the generated matrix is used, and can change the node to be made inactive by the column weight, etc. By doing so, it becomes possible to create a sparse matrix with good coding efficiency quickly and efficiently.
- the node to be inactivated is arbitrarily changed by a variable parameter or the like. For example, in this example, the node corresponding to the staircase matrix has a high degree of demand, so it is desirable that the node is always inactive without being made inactive.
- the generated sparse matrix is sent to the interleave unit 110.
- the interleave unit 110 rearranges the columns so that the loop becomes wide and the matrix weights are averaged even when cut out in an arbitrary range, and sparse matrix information is output as output data.
- the selective PEG algorithm generates a sparse matrix while keeping row weights / column weights at an arbitrary multilevel, and performs encoding in the case where the matrix space is narrow by performing constrained interleaving. Improve efficiency.
- the sparse graph creation method includes a search procedure, a node selection procedure, a root node edge connection procedure, and a constrained interleaving procedure in this order.
- search procedure a node that can be reached through the shortest number of edges from an active node connected to an arbitrary root node serving as a route for creating a local graph is searched from the sparse graph.
- node selection procedure a node satisfying the set condition is selected from the search result.
- the node selected in the node selection procedure is connected to the root node.
- the constrained interleaving procedure the nodes of the sparse graph in which nodes are generated at the selected positions are rearranged so that the weights of the sparse matrix are spatially uniformly arranged so as to maintain the local graph relationship.
- FIG. 9 shows a basic configuration diagram of a sparse matrix creation device according to this embodiment.
- the sparse matrix creation device 200 includes an algebraic structure generation unit 201, a sparse matrix cache unit 202, a random number generation unit 203, a root node edge connection unit 204, a search unit 208, and a search depth control unit. 207, a node selection unit 209, and a constrained interleaving unit 210.
- the sparse matrix creation apparatus 200 according to the present embodiment realizes a variable rate error correction code that achieves multilevel rate control with high efficiency.
- the algebraic structure generation unit 201 is considered when the sparse graph has a special structure.
- the sparse matrix cache unit 202 stores a sparse matrix to be generated.
- the random number generation unit 203 constructs a code stochastically.
- the root node edge connection unit 204 selects an edge to be connected to the root node as a reference when a large loop in the local graph is taken.
- the search unit 208 searches for a node that can be reached through the shortest number of edges based on the control information of the search depth control unit 207.
- the search depth control unit 207 controls the depth searched by the search unit 208.
- the node selection unit 209 selects a node that satisfies the condition from the search result.
- the constrained interleave unit 210 replaces the created sparse matrix by column replacement or the like. Each component will be described below.
- This system is a communication path decoding system that efficiently recovers erasures that occur on erasure communication paths.
- IP on the Internet which is a typical communication path of the lost communication path.
- An example in which a packet transmission is assumed and an LDPC-Staircase code including a sparse graph is assumed will be described.
- the input data is input to the sparse matrix creation device 200 and creation of the sparse matrix is started.
- the input data includes the matrix size (eg, vertical size, horizontal size) related to the coding rate, information about the matrix weight (eg, column weight 3 and column weight 7). The number of columns), the initial value of the search depth, and other information necessary for matrix generation are input.
- the algebraic structure generation unit 201 assumes an LDPC-Staircase code this time, a step matrix is generated as an algebraic structure.
- the generated staircase matrix is stored in the sparse matrix cache unit 202.
- the root node edge connection unit 204 connects nodes connected from a node serving as a route when creating a local graph by edges.
- the edge connected to the check node from the root node is selected for the first time in the column, there are various methods for selecting the edge. For example, the row weight is calculated and the check node set having the lightest weight is selected. A random selection from among them is performed. When this restriction is imposed, a regular sparse matrix is generated in the row direction. In this process, check nodes are selected by random numbers generated by the random number generation unit 203 and connected by edges.
- the search unit 208 When an edge is connected from the root node by the root node edge connection unit 204, the search unit 208 performs a search based on the information of the search depth control unit 207. Since the operation of the search unit 208 is the same as that of the search unit 108 described in the first embodiment, the details of the search operation are omitted here.
- the node selection unit 209 receives a search result from the search unit 208 and determines a node to be added. When there are a plurality of selection candidates, a method such as selecting from a candidate node set at random, filling from the upper part of the matrix, or filling from the lowest row weight is used. In addition, when selecting at random, a random number generated by the random number generation unit 203 is used.
- the process returns to the root node edge connection unit 204, and this time, the selected edge is connected to the root node, and the same processing as the search unit 208 and the node selection unit 209 is performed. Repeat until an arbitrary number of nodes (eg, 3 nodes, etc.) are created and sequence processing is completed. Thereafter, the selected node is stored in the sparse matrix cache unit 202.
- the search depth control unit 207 changes the search depth to narrow the search range. Is the same as that of the first embodiment. However, there may be a case where a node that can be selected by the node selection unit 204 does not appear even when the search range is narrowed. In that case, the search depth control unit 207 performs control so that the search is not performed, and the search unit 208 is skipped to generate a sparse matrix having an arbitrary weight.
- a column having a light weight (eg, a column having a column weight of 3) is generated first, and then a column having a heavy weight (e.g., a column having a column weight of 7) can be generated to generate a matrix with good coding efficiency.
- FIG. 10 and FIG. 11 show examples of sparse matrices stored in the sparse matrix cache unit 202 after node selection of all sparse matrices is performed.
- the sparse matrix has a staircase matrix created by the algebraic structure generator 201 from the left, followed by a sparse matrix with column weight 3 and finally a sparse matrix with column weight 7. .
- the constrained interleaving unit 210 rearranges the sparse matrix shown in FIG. 10 into the matrix shown in FIG.
- the staircase matrix indicated by “T” in FIG. 10 is located on the right side of the parity check matrix.
- the mixed sparse matrix “G” shown in FIG. 11 is created as the sparse matrix indicated by “G1” with column weight 3 and the sparse matrix indicated by “G2” with column weight 7.
- the mixed sparse matrix “G” is mixed so that the sparse matrix with the column weight 3 and the sparse matrix with the column weight 7 take a spatial average and become uniform.
- the mixed sparse matrix “G” maintains the order relationship in the sparse matrix with the column weight 3 and the sparse matrix with the column weight 7. That is, the columns N 31 , N 32 , N 33 , N 34 with column weight 3 and the columns N 71 , N 72 , N 73 , N 74 with column weights are represented by N 31 , N 71 in the mixed sparse matrix “G”. , N 32 , N 72 , N 33 , N 73 , N 34 , and N 74 are arranged in this order.
- the constrained interleave unit 210 may generate one sparse matrix by combining two sparse matrices.
- FIG. 12 shows two sparse matrices with different column weights generated by the sparse matrix creation apparatus according to this embodiment.
- the constrained interleave unit 210 mixes the sparse matrix with column weight 3 and the sparse matrix with column weight 7 as a mixed sparse matrix “G” with column weight 3 and column weight 7 as shown in FIG.
- the staircase matrix is rearranged so that it is on the right side.
- a function similar to that of the node inactivation unit 105 of the first embodiment can be provided by creating a plurality of sparse matrices.
- FIG. 13 shows a basic configuration diagram of a sparse matrix creation device according to the present embodiment.
- the sparse matrix creation device 300 according to the present embodiment realizes a variable rate error correction code that achieves multilevel rate control with high efficiency.
- the sparse matrix creation apparatus 300 according to the present embodiment includes an algebraic structure generation unit 301, a sparse matrix cache unit 302, a random number generation unit 303, a root node edge connection unit 304, a node inactivation unit 305, an inactive A control unit 306, a search depth control unit 307, a search unit 308, a node selection unit 309, an extended edge selection unit 310, a node expansion unit 311, and a constrained interleaving unit 312.
- the algebraic structure generation unit 301 is considered when the sparse graph has a special structure.
- the sparse matrix cache unit 302 stores a sparse matrix to be generated.
- the random number generation unit 303 constructs a code stochastically.
- the root node edge connection unit 304 connects nodes connected from a node serving as a route when creating a local graph by edges.
- the node inactivator 305 selects nodes that are not considered when taking a large loop in the local graph.
- the inactivation control unit 306 determines the conditions in the processing of the node inactivation unit 305.
- the search depth control unit 307 controls the search depth when taking a large loop in the local graph.
- the search unit 308 searches for a node that can be reached from the active node connected to the root node through the shortest number of edges based on the control information of the search depth control unit 307.
- the node selection unit 309 selects a node that satisfies the condition from the search result of the search unit 308.
- the extended edge selection unit 310 selects an edge to be transferred to the check node to be extended.
- the node expansion unit 311 increases the number of check nodes while maintaining the number of edges in the sparse graph.
- the constrained interleaving unit 312 replaces the created sparse matrix by column replacement or the like. Each component will be described below.
- This system is a communication path decoding system that efficiently recovers erasures that occur on erasure communication paths.
- IP packet transmission is performed on the Internet, which is a typical communication path of the lost communication path, and the code is configured in a sparse graph.
- An example assuming the LDPC-Staircase code will be described. Further, in the present embodiment, a description will be given using an example specialized for a variable rate type in which the redundancy can be greatly changed.
- the input data is input to the sparse matrix creation device 300 and creation of the sparse matrix is started.
- the input data includes the matrix size (eg, vertical size, horizontal size) related to the coding rate, information about the matrix weight (eg, column weight 3 and column weight 7). The number of columns), the initial value of the search depth, and other information necessary for matrix generation are input.
- information such as the minimum matrix size, the maximum matrix size, and the redundancy rate of each is input as a parameter corresponding to the variable rate.
- the processing from the algebraic structure generation unit 301 to the processing of the node selection unit 309 is basically the same as the processing from 101 to 109 in the first embodiment, and the details are omitted.
- nodes are selected so that they fall within the minimum matrix size range in the node selection by the root node edge connection unit 304 and the node selection unit 309, and are connected by edges. That is, as shown in FIG. 14, a small sparse matrix is generated by limiting the area for selecting a node, and is stored in the sparse matrix cache unit 302.
- G1 in FIGS. 14 to 17 is a sparse matrix having an arbitrary weight, and a regular matrix and an irregular matrix can be created.
- FIG. 18 shows an outline of the processing.
- the left side of FIG. 18 shows a state before G1 in FIG. 14 is expanded.
- the extended edge selection unit 310 selects an extended edge. That is, in FIG. 18, the edge to be expanded is selected from the edges connected to NC 1 and NC 2 .
- the node extension unit 311 adds check nodes NC 3 and NC 4 and replaces the edge selected by the extension edge selection unit 310.
- the check nodes NC 3 and NC 4 increase as shown on the right side of FIG.
- the extended edge selection unit 310 generates a selection of which edge is connected to the extended node (or remains in the check node connected so far) by the random number generation unit 303. Randomly selected by random number. In this process, the extended edges are selected so that the number of edges connected to any check node is approximately the same.
- the node expansion unit 311 expands the check node. As a result of this expansion, G1 in FIG. 14 becomes G2 in FIG. 15, and has a sparse graph structure with more check nodes.
- the expanded sparse matrix is stored in the sparse matrix cache unit 302.
- the procedure is the same as the process so far, and a sequence creation procedure from the root node edge connection unit 304 to the node selection unit 309 is executed.
- the root node edge connection unit 304 and the node selection unit 309 do not provide the spatial restriction that was performed in the previous procedure.
- a partial sparse matrix that is not limited to a region is generated as shown by G3 in FIG.
- the generated sparse matrix is sent to the constrained interleaving unit 312, and even if it is cut out in an arbitrary range, the loop becomes wide and the matrix weight becomes average. Rearrangement is performed, and sparse matrix information is output as output data.
- the sparse matrices G2 and G3 are rearranged so that the column weights are averaged to become sparse matrices G2 'and G3'.
- the sparse matrices G2 'and G3' are interleaved with a constraint that the columns are not replaced if the column weights are the same as in the constrained interleaving shown in FIGS. With this configuration, a matrix with high coding efficiency can be generated even when the rate is varied by padding with 0 from the right side of G3 'or when G2' is varied by puncturing.
- the configuration of the second embodiment may further include an extended edge selection unit 310 and a node expansion unit 311.
- the invention according to the present embodiment further includes the constrained interleaving function described in the second embodiment in the configuration of the first embodiment.
- the invention of Embodiment 1 and the invention of Embodiment 2 can be combined. This combination can further improve the encoding efficiency.
- the present invention provides a sparse matrix that realizes a variable-rate error correcting code that achieves multi-level rate control with high efficiency in creating a sparse matrix of a sparse graph code. Can be created.
- the present invention can be applied to the information and communication industry.
- Sparse matrix creation unit 91 Channel encoding device 92: Packet-based transmission device 93: Packet-based reception device 94: Channel decoding devices 100, 200, 300, 400: Sparse matrix creation devices 101, 201, 310, 401 : Algebraic structure generation units 102, 202, 302, 402: Sparse matrix cache units 103, 203, 303, 403: Random number generation units 104, 204, 304, 404: Root node edge connection units 105, 305: Node inactive Units 106, 306: inactivation control units 107, 207, 307, 406: search depth control units 108, 208, 308, 405: search units 109, 209, 309, 407: node selection unit 110: interleaving units 210, 312, 410: Constrained interleave unit 310, 408: Extended edge selection unit 3 1,409: node extensions 911: channel-coding section 921: packet transmission unit 931: received data analysis unit 9
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Probability & Statistics with Applications (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Algebra (AREA)
- Computational Mathematics (AREA)
- Computing Systems (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Error Detection And Correction (AREA)
- Complex Calculations (AREA)
Abstract
Description
疎グラフ符号に用いられる疎グラフを生成する疎グラフ作成装置であって、
疎グラフ中の少なくとも一部のノードをイナクティブにするイナクティブ化部と、
前記イナクティブ化部によってイナクティブにされた残りの疎グラフのなかから、ローカルグラフを作成する際のルートとなる任意のルートノードと接続されているアクティブなノードから最短のエッジ数を通じて到達できるノードを探索する探索部と、
前記探索部の探索結果から設定された条件を満たすノードを、エッジを生成する位置に選択するノード選択部と、
前記ノード選択部にて選択されたノードとルートノードを接続するルートノードエッジ接続部と、
を備える。
疎グラフ符号に用いられる疎グラフを生成する疎グラフ作成装置であって、
疎グラフのなかから、ローカルグラフを作成する際のルートとなる任意のルートノードと接続されているアクティブなノードから最短のエッジ数を通じて到達できるノードを探索する探索部と、
前記探索部の探索結果から設定された条件を満たすノードを、エッジを生成する位置に選択するノード選択部と、
前記ノード選択部にて選択されたノードとルートノードを接続するルートノードエッジ接続部と、
前記ルートノードエッジ接続部によって選択された位置にエッジが生成された疎グラフの各ノードを、疎行列の各重みが空間的に均一に配置されるように並び替える制約付きインターリーブ部と、
を備える。
疎グラフ符号に用いられる疎グラフを生成する疎グラフ作成方法であって、
疎グラフ中の少なくとも一部のノードをイナクティブにするイナクティブ化手順と、
前記イナクティブ化手順によってイナクティブにされた残りの疎グラフのなかから、ローカルグラフを作成する際のルートとなる任意のルートノードと接続されているアクティブなノードから最短のエッジ数を通じて到達できるノードを探索する探索手順と、
前記探索手順の探索結果から設定された条件を満たすノードを、エッジを生成する位置に選択するノード選択手順と、
前記ノード選択手順にて選択されたノードとルートノードを接続するルートノードエッジ接続手順と、
を順に有する。
疎グラフ符号に用いられる疎グラフを生成する疎グラフ作成手順であって、
疎グラフのなかから、ローカルグラフを作成する際のルートとなる任意のルートノードと接続されているアクティブなノードから最短のエッジ数を通じて到達できるノードを探索する探索手順と、
前記探索手順の探索結果から設定された条件を満たすノードを、エッジを生成する位置に選択するノード選択手順と、
前記ノード選択手順にて選択されたノードとルートノードを接続するルートノードエッジ接続手順と、
前記ルートノード接続手順によって選択された位置にエッジが生成された疎グラフの各ノードを、疎行列の各重みが空間的に均一に配置されるように並び替える制約付きインターリーブ手順と、
を順に有する。
本実施形態では、選択的PEGアルゴリズムは行重み/列重みを任意のマルチレベルで保ちながら疎行列を生成していき、その過程で、任意のエッジをinactiveにすることにより行列空間が狭い場合の符号化効率を改善する。
イナクティブ化手順では、疎グラフ中の少なくとも一部のノードをイナクティブにする。
探索手順では、イナクティブにされた残りの疎グラフのなかから、ローカルグラフを作成する際のルートとなる任意のルートノードと接続されているアクティブなノードから最短のエッジ数を通じて到達できるノードを探索する。
ノード選択手順では、探索結果から設定された条件を満たすノードを選択する。
ルートノードエッジ接続手順では、ノード選択手順にて選択されたノードとルートノードを接続する。
次に、ルートノードエッジ接続部104にてローカルグラフを作成する際のルートノードと1つのノードがエッジを通じて接続される。ここで、ローカルグラフとは、疎グラフの生成過程においてルートノードから接続される候補となるノードから構成されるグラフである。ルートノードとは、新しくエッジを追加していくターゲットとしているノードであり、ツリーで表現したときに一番上に配置されるノードである。
本実施形態では、選択的PEGアルゴリズムは行重み/列重みを任意のマルチレベルで保ちながら疎行列を生成していき、その過程で、制約付きインターリーブを行うことにより行列空間が狭い場合の符号化効率を改善する。
探索手順では、疎グラフのなかから、ローカルグラフを作成する際のルートとなる任意のルートノードと接続されているアクティブなノードから最短のエッジ数を通じて到達できるノードを探索する。
ノード選択手順では、探索結果から設定された条件を満たすノードを選択する。
ルートノードエッジ接続手順では、ノード選択手順にて選択されたノードとルートノードを接続する。
制約付きインターリーブ手順では、選択された位置にノードが生成された疎グラフの各ノードを、ローカルグラフの関係を保つように疎行列の各重みが空間的に均一に配置されるように並び替える。
次に、ルートノードエッジ接続部204は、ローカルグラフを作成する際のルートとなるノードから接続されるノードをエッジで接続する。ルートノードからチェックノードに接続されるエッジがその列で初めて選択される場合、エッジの選択方法は様々な方法が考えられるが、例えば、行重みを計算して一番重みの軽いチェックノード集合の中からランダムに選択するなどが実施される。この制約を課した場合、行方向にレギュラーな疎行列が生成されることとなる。この処理の際、乱数生成部203により生成される乱数によりチェックノードを選択しエッジで接続する。
図13に、本実施形態に係る疎行列作成装置の基本構成図を示す。本実施形態に係る疎行列作成装置300は、マルチレベルのレート制御を高効率に達成するレート可変型の誤り訂正符号を実現する。本実施形態に係る疎行列作成装置300は、代数的構造生成部301と、疎行列キャッシュ部302と、乱数生成部303と、ルートノードエッジ接続部304と、ノードイナクティブ化部305と、イナクティブ化制御部306と、探索深度制御部307と、探索部308と、ノード選択部309と、拡張エッジ選択部310と、ノード拡張部311と、制約付きインターリーブ部312と、を備える。
91:通信路符号化装置
92:パケットベース伝送装置
93:パケットベース受信装置
94:通信路復号装置
100、200、300、400:疎行列作成装置
101、201,310、401:代数的構造生成部
102、202、302、402:疎行列キャッシュ部
103,203、303、403:乱数生成部
104、204、304、404:ルートノードエッジ接続部
105、305:ノードイナクティブ化部
106、306:イナクティブ化制御部
107、207、307、406:探索深度制御部
108、208、308、405:探索部
109、209、309、407:ノード選択部
110:インターリーブ部
210、312、410:制約付きインターリーブ部
310、408:拡張エッジ選択部
311、409:ノード拡張部
911:通信路符号化部
921:パケット伝送部
931:受信データ解析部
941:通信路復号部
Claims (10)
- 疎グラフ符号に用いられる疎グラフを生成する疎グラフ作成装置であって、
疎グラフ中の少なくとも一部のノードをイナクティブにするイナクティブ化部と、
前記イナクティブ化部によってイナクティブにされた残りの疎グラフのなかから、ローカルグラフを作成する際のルートとなる任意のルートノードと接続されているアクティブなノードから最短のエッジ数を通じて到達できるノードを探索する探索部と、
前記探索部の探索結果から設定された条件を満たすノードを、エッジを生成する位置に選択するノード選択部と、
前記ノード選択部にて選択されたノードとルートノードを接続するルートノードエッジ接続部と、
を備える疎グラフ作成装置。 - 前記ノード選択部は、それまでのノード生成の過程から得られた探索に関する情報から次に探索する範囲を限定して探索することで少ない演算量でループを大きくとるノードを選択する
ことを特徴とする請求項1に記載の疎グラフ作成装置。 - 疎グラフ符号に用いられる疎グラフを生成する疎グラフ作成装置であって、
疎グラフのなかから、ローカルグラフを作成する際のルートとなる任意のルートノードと接続されているアクティブなノードから最短のエッジ数を通じて到達できるノードを探索する探索部と、
前記探索部の探索結果から設定された条件を満たすノードを、エッジを生成する位置に選択するノード選択部と、
前記ノード選択部にて選択されたノードとルートノードを接続するルートノードエッジ接続部と、
前記ルートノードエッジ接続部によって選択された位置にエッジが生成された疎グラフの各ノードを、疎行列の各重みが空間的に均一に配置されるように並び替える制約付きインターリーブ部と、
を備える疎グラフ作成装置。 - 前記制約付きインターリーブ部は、アクティブなノードのグラフ構造を保つように、疎グラフ中のノードを入れ替える
ことを特徴とする請求項3に記載の疎グラフ作成装置。 - 疎グラフのエッジ総数が変わらずかつ最小のループが短くならないように、新たなノードを設置するノード拡張部をさらに備えることを特徴とする請求項1から4のいずれかに記載の疎グラフ作成装置。
- 疎グラフ符号に用いられる疎グラフを生成する疎グラフ作成方法であって、
疎グラフ中の少なくとも一部のノードをイナクティブにするイナクティブ化手順と、
前記イナクティブ化手順によってイナクティブにされた残りの疎グラフのなかから、ローカルグラフを作成する際のルートとなる任意のルートノードと接続されているアクティブなノードから最短のエッジ数を通じて到達できるノードを探索する探索手順と、
前記探索手順の探索結果から設定された条件を満たすノードを、エッジを生成する位置に選択するノード選択手順と、
前記ノード選択手順にて選択されたノードとルートノードを接続するルートノードエッジ接続手順と、
を順に有する疎グラフ作成方法。 - 前記ノード選択手順において、それまでのノード生成の過程から得られた探索に関する情報から次に探索する範囲を限定して探索することで少ない演算量でループを大きくとるノードを選択する
ことを特徴とする請求項6に記載の疎グラフ作成方法。 - 疎グラフ符号に用いられる疎グラフを生成する疎グラフ作成手順であって、
疎グラフのなかから、ローカルグラフを作成する際のルートとなる任意のルートノードと接続されているアクティブなノードから最短のエッジ数を通じて到達できるノードを探索する探索手順と、
前記探索手順の探索結果から設定された条件を満たすノードを、エッジを生成する位置に選択するノード選択手順と、
前記ノード選択手順にて選択されたノードとルートノードを接続するルートノードエッジ接続手順と、
前記ルートノード接続手順によって選択された位置にエッジが生成された疎グラフの各ノードを、疎行列の各重みが空間的に均一に配置されるように並び替える制約付きインターリーブ手順と、
を順に有する疎グラフ作成方法。 - 前記制約付きインターリーブ手順において、アクティブなノードのグラフ構造を保つように、疎グラフ中のノードを入れ替える
ことを特徴とする請求項8に記載の疎グラフ作成方法。 - 疎グラフのエッジ総数が変わらずかつ最小のループが短くならないように、新たなノードを設置するノード拡張手順を、前記ノード選択手順の後でありかつ前記制約付きインターリーブ手順の前にさらに有することを特徴とする請求項6から9のいずれかに記載の疎グラフ作成方法。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP17181109.4A EP3264612B1 (en) | 2013-10-22 | 2014-10-08 | Ldpc code construction using a modified progressive edge-growth (peg) algorithm |
EP14855171.6A EP3062445A4 (en) | 2013-10-22 | 2014-10-08 | Sparse graph creation device and sparse graph creation method |
US15/027,456 US11303305B2 (en) | 2013-10-22 | 2014-10-08 | Sparse graph creation device and sparse graph creation method |
BR112016007458A BR112016007458A2 (pt) | 2013-10-22 | 2014-10-08 | dispositivo de criação de gráfico esparso que cria um gráfico esparso utilizado para um código de gráfico esparso, e, método de criação de gráfico esparso que cria um gráfico esparso utilizado para um código de gráfico esparso |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2013-218933 | 2013-10-22 | ||
JP2013218933A JP5792256B2 (ja) | 2013-10-22 | 2013-10-22 | 疎グラフ作成装置及び疎グラフ作成方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2015060130A1 true WO2015060130A1 (ja) | 2015-04-30 |
Family
ID=52992732
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2014/076910 WO2015060130A1 (ja) | 2013-10-22 | 2014-10-08 | 疎グラフ作成装置及び疎グラフ作成方法 |
Country Status (5)
Country | Link |
---|---|
US (1) | US11303305B2 (ja) |
EP (2) | EP3062445A4 (ja) |
JP (1) | JP5792256B2 (ja) |
BR (1) | BR112016007458A2 (ja) |
WO (1) | WO2015060130A1 (ja) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10540398B2 (en) * | 2017-04-24 | 2020-01-21 | Oracle International Corporation | Multi-source breadth-first search (MS-BFS) technique and graph processing system that applies it |
CN109165322B (zh) * | 2018-08-01 | 2022-04-19 | 成都数联铭品科技有限公司 | 基于路径关系的网络特征提取系统及方法 |
US20190340511A1 (en) * | 2019-06-20 | 2019-11-07 | Intel Corporation | Sparsity control based on hardware for deep-neural networks |
CN113012555B (zh) * | 2019-12-20 | 2022-06-24 | 百度在线网络技术(北京)有限公司 | 地图显示方法、装置、电子设备和存储介质 |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2006027897A1 (ja) * | 2004-09-03 | 2006-03-16 | Mitsubishi Denki Kabushiki Kaisha | Ldpc符号生成方法、通信装置および符号列生成方法 |
JP2011109228A (ja) * | 2009-11-13 | 2011-06-02 | Mitsubishi Electric Corp | 復号装置及び方法 |
Family Cites Families (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5327561A (en) * | 1991-09-20 | 1994-07-05 | International Business Machines Corporation | System and method for solving monotone information propagation problems |
FI121431B (fi) | 2004-09-13 | 2010-11-15 | Tamfelt Pmc Oy | Paperikoneessa käytettävä kudosrakenne ja menetelmä sen valmistamiseksi |
WO2006039801A1 (en) * | 2004-10-12 | 2006-04-20 | Nortel Networks Limited | System and method for low density parity check encoding of data |
US8117523B2 (en) * | 2007-05-23 | 2012-02-14 | California Institute Of Technology | Rate-compatible protograph LDPC code families with linear minimum distance |
US7747633B2 (en) * | 2007-07-23 | 2010-06-29 | Microsoft Corporation | Incremental parsing of hierarchical files |
EP2134018A1 (en) * | 2008-05-23 | 2009-12-16 | Deutsches Zentrum für Luft- und Raumfahrt e.V. | Method for recovery of lost and/ or corrupted data |
US20100146007A1 (en) * | 2008-12-09 | 2010-06-10 | Alex Kononov | Database For Managing Repertory Grids |
EP2503698B1 (en) * | 2009-11-17 | 2018-02-14 | Mitsubishi Electric Corporation | Error correction method and device, and communication system using the same |
US10394778B2 (en) * | 2010-09-03 | 2019-08-27 | Robert Lewis Jackson, JR. | Minimal representation of connecting walks |
US20120095957A1 (en) * | 2010-10-18 | 2012-04-19 | Tata Consultancy Services Limited | Component Based Approach to Building Data Integration Tools |
CN102467709B (zh) * | 2010-11-17 | 2017-03-01 | 阿里巴巴集团控股有限公司 | 一种发送商品信息的方法和装置 |
WO2012174565A2 (en) * | 2011-06-16 | 2012-12-20 | Radiumone, Inc. | Building a social graph with sharing activity between users of the open web |
US8583659B1 (en) * | 2012-07-09 | 2013-11-12 | Facebook, Inc. | Labeling samples in a similarity graph |
US8929390B2 (en) * | 2012-09-24 | 2015-01-06 | Alcatel Lucent | Methods and apparatuses for channel estimation in wireless networks |
EP2918047A4 (en) * | 2012-11-06 | 2016-04-20 | Hewlett Packard Development Co | IMPROVED GRAPHIDE RUNNING |
JPWO2014087590A1 (ja) * | 2012-12-05 | 2017-01-05 | 日本電気株式会社 | 最適化装置、最適化方法および最適化プログラム |
US20140208297A1 (en) * | 2013-01-20 | 2014-07-24 | International Business Machines Corporation | Validation of revised computer programs |
BR112015027153B1 (pt) * | 2013-05-02 | 2021-12-14 | Sony Corp | Dispositivo e método de processamento de dados |
US9501503B2 (en) * | 2013-05-09 | 2016-11-22 | Microsoft Technology Licensing, Llc | Inferring entity attribute values |
EP2816767A1 (en) * | 2013-06-17 | 2014-12-24 | Ericsson Modems SA | Equalization in the receiver of a multiple input multiple output system |
CN104518801A (zh) * | 2013-09-29 | 2015-04-15 | Lsi公司 | 非二进制的分层低密度奇偶校验解码器 |
WO2015047423A1 (en) * | 2013-09-30 | 2015-04-02 | Mindjet Llc | Scoring members of a set dependent on eliciting preference data amongst subsets selected according to a height-balanced tree |
US9396164B2 (en) * | 2013-10-21 | 2016-07-19 | International Business Machines Corporation | Sparsity-driven matrix representation to optimize operational and storage efficiency |
-
2013
- 2013-10-22 JP JP2013218933A patent/JP5792256B2/ja active Active
-
2014
- 2014-10-08 WO PCT/JP2014/076910 patent/WO2015060130A1/ja active Application Filing
- 2014-10-08 US US15/027,456 patent/US11303305B2/en active Active
- 2014-10-08 BR BR112016007458A patent/BR112016007458A2/pt active Search and Examination
- 2014-10-08 EP EP14855171.6A patent/EP3062445A4/en not_active Ceased
- 2014-10-08 EP EP17181109.4A patent/EP3264612B1/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2006027897A1 (ja) * | 2004-09-03 | 2006-03-16 | Mitsubishi Denki Kabushiki Kaisha | Ldpc符号生成方法、通信装置および符号列生成方法 |
JP2011109228A (ja) * | 2009-11-13 | 2011-06-02 | Mitsubishi Electric Corp | 復号装置及び方法 |
Non-Patent Citations (10)
Title |
---|
"Asynchronou layered coding protocol instantiation", IETF RFC 3450, December 2002 (2002-12-01) |
KEN'YA SUGIHARA ET AL.: "Coding Rate Adjustment for LDPC Codes by Row-Splitting", THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS SOGO TAIKAI KOEN RONBUNSHU, 2010 NEN-KISO-KYOKAI, 2 March 2010 (2010-03-02), pages 123, XP008171580 * |
KEN'YA SUGIHARA ET AL.: "Constructing Algorithms of Large-Girth LDPC Codes Based on the Shortest Paths in Tanner Graphs", IEICE TECHNICAL REPORT, ISEC, INFORMATION SECURITY, vol. 110, no. 443, 24 February 2011 (2011-02-24), pages 69 - 74, XP008183653 * |
M. LUBY: "LT Codes", THE 43RD ANNUAL IEEE SYMPOSIUM ON FOUNDATIONS OF COMPUTER SCIENCE, 2002 |
R. G. GALLAGER: "Research Monograph series. Cambridge", 1963, MIT PRESS, article "Low density parity check codes" |
See also references of EP3062445A4 |
SHIVA K. PLANJERY; T. AARON GULLIVER: "Design of Rate-Compatible Punctured Repeat-Accumulate Codes", IEEE GLOBAL TELECOMMUNICATIONS CONFERENCE, 2007. GLOBECOM'07 |
SHOKROLLAHI, A: "Raptor codes", INFORMATION THEORY, IEEE TRANSACTIONS, vol. 52, no. 6, 2006 |
XIAO-YU HU ET AL.: "Progressive edge-growth Tanner graphs", GLOBAL TELECOMMUNICATIONS CONFERENCE, 2001. GLOBECOM '01., vol. 2, 29 November 2001 (2001-11-29), pages 995 - 1001, XP055119661 * |
XIAO-YU HU; EVANGELOS ELEFTHERIOU; DIETER-MICHAEL ARNOLD: "Progressive Edge-Growth Tanner Graphs", IEEE GLOBAL TELECOMMUNICATIONS CONFERENCE, 2001. GLOBECOM'01 |
Also Published As
Publication number | Publication date |
---|---|
US20160248448A1 (en) | 2016-08-25 |
JP2015082712A (ja) | 2015-04-27 |
EP3062445A1 (en) | 2016-08-31 |
EP3264612B1 (en) | 2022-08-17 |
US11303305B2 (en) | 2022-04-12 |
EP3062445A4 (en) | 2017-10-25 |
JP5792256B2 (ja) | 2015-10-07 |
BR112016007458A2 (pt) | 2017-08-01 |
EP3264612A1 (en) | 2018-01-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US9270414B2 (en) | Multiple-field based code generator and decoder for communications systems | |
CN107026709B (zh) | 一种数据包编码处理方法及装置、基站及用户设备 | |
Shokrollahi | Raptor codes | |
US7418651B2 (en) | File download and streaming system | |
US7720174B2 (en) | Multi-stage code generator and decoder for communication systems | |
EP2228909A1 (en) | Coding method, coding device, decoding method and decoding device for low density generator matrix code | |
JP4005084B2 (ja) | 検査行列生成方法および検査行列生成装置 | |
JP5792256B2 (ja) | 疎グラフ作成装置及び疎グラフ作成方法 | |
Roca et al. | Rs+ ldpc-staircase codes for the erasure channel: Standards, usage and performance | |
Joshi et al. | Fountain codes | |
Mirrezaei et al. | Towards fountain codes | |
KR101145673B1 (ko) | 특정 구조를 가지는 lt코드를 이용한 부호화 방법 및 이를 기반으로 하는 랩터 코드를 이용한 부호화 방법 | |
JP5952971B2 (ja) | 通信路復号方法及び通信路復号装置 | |
Andreadou et al. | Quasi-Cyclic Low-Density Parity-Check (QC-LDPC) codes for deep space and high data rate applications | |
Sy et al. | Extended non-binary low-density parity-check codes over erasure channels | |
Baldi et al. | A simple scheme for belief propagation decoding of BCH and RS codes in multimedia transmissions | |
JP5238060B2 (ja) | 符号化装置及び方法及び符号化・復号化システム及び復号化方法 | |
Mattoussi | Design and optimization of al-fec codes: the gldpc-staircase codes | |
JP4972128B2 (ja) | 符号化・復号システム及び符号化・復号方法 | |
JP2007329828A (ja) | 復号方法、復号器および通信装置 | |
Majeed | LDPC error floor improvement | |
Balasubramanian | ITERATIVE DECODING OF SPATIALLY-COUPLED PRODUCT CODES USING GUESSING RANDOM ADDITIVE NOISE DECODING ALGORITHMS | |
Casu | Optimization of protection techniques based on FEC codes for the transmission of multimedia packetized streams | |
Poulliat et al. | Conception et optimisation de codes AL-FEC: les codes GLDPC-Staircase | |
El-Gohary et al. | Study the performance of fountain codes in wireless communication systems |
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: 14855171 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 15027456 Country of ref document: US |
|
REG | Reference to national code |
Ref country code: BR Ref legal event code: B01A Ref document number: 112016007458 Country of ref document: BR |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
REEP | Request for entry into the european phase |
Ref document number: 2014855171 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2014855171 Country of ref document: EP |
|
ENP | Entry into the national phase |
Ref document number: 112016007458 Country of ref document: BR Kind code of ref document: A2 Effective date: 20160404 |