CN101800624A - Generation grouped cross-random linear network coding method - Google Patents

Abstract

The invention discloses a generation grouped cross-random linear network coding method which comprises the following steps of: sequentially differentiating the generations of original data by an information source, sequentially grouping in each generation and forming a generation set by a generation and the next generation; coding in each generation set, wherein the coding process of each generation sets is independent, and confirming the information package structures of the generation sets by the information source; confirming local sparse coding coefficient vectors according to the number of coding-participating information packages by a node, then, generating a new coding package according to the local sparse coding coefficient vectors and the coding-participating information packages and sending the new coding package to downstream nodes; carrying out the linear correlation detection of the received new coding package by a node, using the new coding package for coding or decoding if the new coding package passes the detection or else discarding or retransmitting; and adopting a Gaussian column principal component elimination method to decode the original data by a destination option node after receiving enough coding packages passing the linear correlation detection. The invention has the advantages of lowering the network coding time and the space complexity and weakening the necessary conditions for correct coding, thereby being suitable for information transmission in information sharing.

Description

Generation grouped cross-random linear network coding method

Technical field

The invention belongs to communication technical field, relate to network code, specifically a kind of random linear network encoding method is used for the message transmission of information sharing process.

Background technology

2000, the network code theory that people such as Ashlswede propose was first overturned the route pattern that storage is transmitted in the conventional communication networks, and it allows the network service intermediate node that its information that receives is carried out data processing.The application of network code for existing network has brought plurality of advantages, comprises the throughput that improves network, improves the reliability of network, reduce communication energy consumption, improve bandwidth availability ratio, ensure information security, improve robustness and fault-tolerance etc.Existing file-sharing process adopts http protocol or similar agreement to realize transfer of data mostly, and these agreements all are according to network condition file elder generation piecemeal to be transmitted by piece then, and leaving of resource node usually can cause file transfer to be ended.So people begin one after another network code to be applied to the file-sharing process, with the throughput, bandwidth availability ratio and the fault-tolerance that improve communication network, ensure information security and system robustness is provided.

At the application problem of network code in the information sharing, a lot of in recent years experts and scholars have proposed many solutions.Philip A.Chou, people such as Yunnan Wu and Kamal Jain at first propose to divide for network coding method in the paper of delivering in 2003 " Practical Network Coding ".Though divide for network coding method and can solve the nothing branch for the excessive problem of amount of calculation, but thereby the encoding block combination that it still exists withdrawing from of node to make not had abundant linear independence in the network causes solving the problem of complete file, and switching between the generation of message transmission in addition also is to be difficult to the problem held.The Christos Gkantsidis of Microsoft Research and Pablo Rodriguez were published in 2005 in " Network Codingfor Large Scale Content Distribution " literary composition on " IEEE Conference on Information ", at first proposed not have to divide for network coding method.When the nothing branch is used for file-sharing for network coding method, though can improve the efficiency of transmission of network, but because cataloged procedure is to carry out between all piecemeals of a file, so amount of calculation and overhead are all very big especially when big file distributing, the necessary condition of its correct decoding is very harsh, and the linear independence encoding block number that requires to receive is equal to or greater than the original document block count.Huang Jiaqing, Wang Shuai and Chen Qingwen at first propose network coding method between generation in " application of network code in the P2P network " literary composition that was published on the general information of dimension in 2009.The advantage of network coding method was when this generation, certain concentrated generation did not have the encoding block of abundant linear independence between generation, can concentrate the linear independence encoding block in other generations fill up by this generation.But its shortcoming is successfully to solve separately the initial data of certain generation, and its computation complexity and space complexity are all too big.

Summary of the invention

The objective of the invention is to overcome the deficiency of above-mentioned prior art, a kind of generation grouped cross-random linear network coding method is proposed, to improve coding rate and decoding speed, reduce the space complexity of cataloged procedure, weaken the necessary condition that translates file, reduce node especially resource node leave influence to message transmission.

For achieving the above object, coding step of the present invention comprises as follows:

(1) element representation of the information that will be sent by the information source node handle in the finite field gf (q) obtains initial data, wherein q 〉=256; This initial data is divided into g generation in proper order, and G is used in g 〉=2 ⁱ(the expression of 1≤i≤g) i generation; Again each generation is divided into m group in order, P is used in m 〉=2 _j ⁱ(per generation and its next generation constitute for collecting, and use GE for 1≤i≤g, 1≤j≤m) represent the i j grouping in generation ⁱRepresent i for collection, 1≤i≤g wherein, last includes only last generation for collection;

(2) make i=1;

(3) determine its operation according to the attribute of present node,, change step (4) if present node is an information source node, otherwise, step (5) changeed;

(4) if 1≤i≤g then determines the packet infrastructure of i for collection by information source node, use GE ⁱStep (5) is changeed in expression, otherwise, finish;

(5) i in the present node is represented with k for the number of collection packets of information, if k 〉=2, after then present node is determined the local code coefficient vector of i for collection according to k, change step (6), otherwise, present node changes step (7) after an only encoded packets is sent to downstream node as new encoded packets;

(6) present node generates new encoded packets according to the local code coefficient vector and the i of the k in its internal memory of step (5) for the packets of information that collects, and sends to downstream node, changes step (7);

(7) downstream node is after receiving new encoded packets, it is carried out linear dependence detects, pass through if detect, then this new encoded packets is used for coding or decoding, otherwise, think that this new encoded packets can not provide useful information, abandon or please retransmit by upstream node,, change step (8) if present node is an information destination node, otherwise, change step (5);

(8) information destination node is being after receiving the encoded packets that k i detects by step (7) linear dependence for collection, draw the initial data of i that information source sends by decoding for collection after, change step (9), wherein, when 1≤i≤g-1, k=2m, when i=g, k=m;

(9) i is added 1, and change step (3).

The present invention has following advantage:

(1) the present invention has greatly improved coding rate owing to adopt sparse matrix as the local code coefficient matrix;

(2) the present invention has improved decoding speed owing to adopt Gauss's pivot in a column elimination to decipher;

(3) cataloged procedure of the present invention has reduced the time complexity and the space complexity of cataloged procedure owing to be to carry out in being collected by Ben Dai and the generation that constitutes of future generation thereof;

(4) decode procedure of the present invention is owing to be to carry out in being collected by Ben Dai and the generation that constitutes of future generation thereof, weakened the necessary condition that correctly translates file, as long as node is received k i and just can correctly be translated i for the collection initial data for the irrelevant encoded packets of line concentration, wherein, when 1≤i≤g-1, k=2m, when i=g, k=n;

(5) the present invention is because each encoded packets all comprises the information of all groupings in this generation collection, reduced node especially resource node leave influence to message transmission.

Description of drawings

Fig. 1 is a network code process schematic diagram of the present invention;

Fig. 2 is the communication network schematic diagram that the present invention uses;

Fig. 3 is a network code principle schematic of the present invention.

Embodiment

Followingly the present invention is described in further detail with reference to accompanying drawing.

One, applied mathematical theory of the present invention and technical term explanation

1, finite field sparse matrix full rank theorem:

If M=is (m _In) _{N * n}Be that a n * n on the finite field gf (q) ties up random matrix, the element m in the matrix _IjThe independence that obedience satisfies following probability evenly distributes:

$\mathrm{Pr}(m_{\mathrm{ij}}=m)=\left\{\begin{array}{cc}1-p& m=0\\ p/(q-1)& m\∈[\mathrm{1,2},\·\·\·,q-1]\end{array}\right.$

If p 〉=(logn+d)/n, then have:

Wherein, d is a non-negative constant.

2, Gauss's pivot in a column elimination

Gauss's pivot in a column elimination approach is a kind of method of separating system of linear equations, it is at first to the coefficient matrix of linear equations unit that disappears successively, before the unit that at every turn disappears, earlier the big element of absolute value is exchanged on the position of leading diagonal, unit again disappears, change into upper triangular matrix up to coefficient matrix, then solution of equations is obtained in back substitution.

3, relational language

Fig. 2 is a directed acyclic graph, represents a communication network, with G=(V, E) expression, V is the node set in this communication network, V={S, R1, R2, R3, R4, R5, T}, wherein, S is an information source node, and R1, R2, R3, R4 and R5 are intermediate node, and T is an information destination node, E is the link set in this communication network, E={S → R1, S → R3, S → R4, R1 → R2 ..., R5 → T} describes relational language according to Fig. 2 below:

(1) information source node S is meant the supplier of information.

(2) information destination node T is meant the requestor of information.

(3) intermediate node R1, R2, R3, R4, R5, be information from the information source node to the information destination node the node of process, they are only encoded to information and transmit, and do not decipher.

(4) all communication links all are oriented channels, and the top of oriented channel is its terminal upstream node, and the end of oriented channel is the downstream node at its top, and as in communication link R1 → R2, R1 is the upstream node of R2, and R2 is the downstream node of R1.

Two, cataloged procedure of the present invention

With reference to Fig. 1, concrete steps of the present invention are:

Step 1, the information that will send is carried out branch generation grouping by information source node.

With reference to Fig. 3, information source node to the branch of information for grouping process is:

Be the information translation that will send the interior element of finite field gf (q) 1a), obtain initial data, wherein q 〉=256;

1b) this initial data is divided in proper order g equal big or small generation, G is used in g 〉=2 ⁱ(in the expression of 1≤i≤g) i generation, when the size in last generation during less than other generations, make the size in all generations all equate with 0 polishing;

1c) each generation is divided into m group in order, P is used in m 〉=2 _j ⁱ(size of all groups all equates for 1≤i≤g, 1≤j≤m) represent i j the grouping in generation;

1d) per generation and its next generation constitute the generation collection, use GE ⁱRepresent i for collection, 1≤i≤g wherein, when 1≤i≤g-1, i comprises i generation and i+1 generation for collection, when i=g, i is exactly g generation for collection.

Step 2, give parameter i initialize, make i=1.

Step 3, determine its operation,, change step 4 if present node is an information source node according to the attribute of present node, otherwise, step 5 changeed.

If step 4 1≤i≤g then determines the packet infrastructure of i for collection by information source node, use GE ⁱStep 5 is changeed in expression, otherwise, finish.

Described i is for the packet infrastructure GE of collection ⁱBe divided into two kinds of situations:

First kind of situation is: when 1≤i≤g-1, i for the packet infrastructure of collection is:

Wherein, D ⁱBe the matrix of i in the present node for the data formation of collection packets of information,

GC ⁱBe the matrix of i in the present node for the overall code coefficient formation of collection packets of information,

SD ⁱBe the matrix of initial data i for collection grouping formation,

E _{2m * 2m}Be the unit matrix of the capable 2m row of 2m,

SD _j ⁱBe the vector of initial data i for j grouping formation in the collection, and 1≤j≤2m,

GC _j ⁱBe the overall code coefficient vector of i for j grouping in the collection, and 1≤j≤2m,

x _Jl ⁱBe l the data of initial data i for j grouping in the collection, and 1≤j≤2m, 1≤l≤n,

GE _j ⁱBe i for j packets of information in the collection, and 1≤j≤2m;

Second kind of situation is: when i=g, i for the packet infrastructure of collection is:

Wherein, E _{M * m}Be the unit matrix of the capable m row of m,

SD _j ⁱBe the vector of initial data i for j grouping formation in the collection, and 1≤j≤m,

GC _j ⁱBe the overall code coefficient vector of i for j grouping in the collection, and 1≤j≤m,

x _Jl ^gBe l the element of initial data g for j grouping in the collection, and 1≤j≤m, 1≤l≤n,

GE _j ^gBe g for j packets of information in the collection, and 1≤j≤m.

Step 5, with i in the present node for the collection packets of information number represent with k, if k 〉=2, then present node according to k determine i for the collection the local code coefficient vector after, change step 6, otherwise present node changes step 7 after an only encoded packets is sent to downstream node as new encoded packets.

Present node is determined the local code coefficient vector of i for collection according to following formula:

${\mathrm{LC}}^i=\left[\begin{array}{cccc}{\mathrm{lc}}_1^i& {\mathrm{lc}}_2^i& \·\·\·& {\mathrm{lc}}_k^i\end{array}\right]$

Wherein, LC ⁱBe the local code coefficient vector of i for collection, lc _j ⁱBe the interior random number of finite field gf (q) that generates according to certain rule, 1≤j≤k, it obeys independent same distribution, and its probability-distribution function is:

$\left\{\begin{array}{c}\mathrm{Pr}({\mathrm{lc}}_j^i=x)=\left\{\begin{array}{cc}1-p& x=0\\ p/(q-1)& x\∈[\mathrm{1,2},\·\·\·,q-1]\end{array}\right.\\ p=(\log N+10)/k\end{array}\right.$

Wherein, N is initial data i for the number of grouping in the collection, when 1≤i≤g-1, and N=2m, when i=g, N=m; K is the number of the i that has of present node for the collection packets of information, if present node is an information source node, and k=N then.

Step 6, present node generate new encoded packets, and send to downstream node.

New encoded packets is the local code coefficient vector and k in its internal memory individual i encoded packets for the packets of information generation of collection of present node according to step 5, and this encoded packets comprises new information data and new overall code coefficient, and their generative process is:

6a) produce new information data by following formula:

$D_0^i={\mathrm{LC}}^i\⊗D^i$

Wherein,

Be the multiplying in the finite field, D ₀ ⁱBe new information data, it is the row vector of a n dimension, if present node is information source node, then D ⁱ=SD ⁱ

6b) produce new overall code coefficient by following formula:

${\mathrm{GC}}_0^i={\mathrm{LC}}^i\⊗{\mathrm{GC}}^i$

Wherein, GC ₀ ⁱBe new overall code coefficient, if present node is information source node, then GC ⁱ=E, GC ₀ ⁱ=LC ⁱ

Step 7, downstream node carry out linear dependence to it and detect after receiving new encoded packets.

7a) by the overall code coefficient matrix GC of following formula structure present node i for collection ⁱ:

${\mathrm{GC}}^i={\left[\begin{array}{ccccc}{\mathrm{GC}}_1^i& {\mathrm{GC}}_2^i& \·\·\·& {\mathrm{GC}}_k^i& {\mathrm{GC}}_0^i\end{array}\right]}^T$

Wherein, GC ₁ ⁱ, GC ₂ ⁱ... GC _k ⁱBe the overall code coefficient of the existing k of present node i for the collection packets of information;

7b) obtain matrix GC by the determinant variation ⁱOrder, with rank (GC ⁱ) expression;

If 7c) rank (GC ⁱ)=k+1, then this new encoded packets is used for coding or decoding to this new encoded packets by detecting, otherwise, think that this new encoded packets can not provide useful information, abandon or please retransmit by upstream node;

If 7d) present node is an information destination node, change step 8, otherwise, change step 5.

Step 8, information destination node are deciphered after receiving the encoded packets that k i detects by the step 7 linear dependence for collection, wherein, when 1≤i≤g-1, k=2m, when i=g, k=m.

Decode procedure is found the solution separating of system of linear equations for adopting Gauss's pivot in a column elimination, and concrete steps are:

8a) k that information destination node is received i is for data division composition data matrix D in the collection encoded packets ⁱ, it is the matrix of a k * n dimension;

8b) k that information destination node is received i partly constitutes overall code coefficient matrix GC for overall code coefficient in the collection encoded packets ⁱ, it is the matrix of a k * N dimension;

8c) can get according to coding principle:

$D^i={\mathrm{GC}}^i\⊗{\mathrm{SD}}^i$

To coefficient matrix GC ⁱThe unit that disappears successively before the unit that at every turn disappears, exchanges to the big element of absolute value on the position of leading diagonal earlier, and the unit that disappears again is up to GC ⁱChange into upper triangular matrix, then raw data matrix SD is obtained in back substitution ⁱ, it is exactly the information that information source sends that initial data is converted to after the binary format.

Step 9, i is added 1, and change step 3, carry out the message transmission of collection of future generation.

Claims (7)

1. a generation grouped cross-random linear network coding method comprises the steps:

(1) element representation of the information that will be sent by the information source node handle in the finite field gf (q) obtains initial data, wherein q 〉=256; This initial data is divided into g generation in proper order, and G is used in g 〉=2 ⁱ(the expression of 1≤i≤g) i generation; Again each generation is divided into m group in order, P is used in m 〉=2 ⁱ _j(per generation and its next generation constitute for collecting, and use GE for 1≤i≤g, 1≤j≤m) represent the i j grouping in generation ⁱRepresent i for collection, 1≤i≤g wherein, last includes only last generation for collection;

(2) make i=1;

(3) determine its operation according to the attribute of present node,, change step (4) if present node is an information source node, otherwise, step (5) changeed;

(4) if 1≤i≤g then determines the packet infrastructure of i for collection by information source node, use GE ⁱStep (5) is changeed in expression, otherwise, finish;

(5) i in the present node is represented with k for the number of collection packets of information, if k 〉=2, after then present node is determined the local code coefficient vector of i for collection according to k, change step (6), otherwise, present node changes step (7) after an only encoded packets is sent to downstream node as new encoded packets;

(6) present node generates new encoded packets according to the local code coefficient vector and the i of the k in its internal memory of step (5) for the packets of information that collects, and sends to downstream node, changes step (7);

(7) downstream node is after receiving new encoded packets, it is carried out linear dependence detects, pass through if detect, then this new encoded packets is used for coding or decoding, otherwise, think that this new encoded packets can not provide useful information, abandon or please retransmit by upstream node,, change step (8) if present node is an information destination node, otherwise, change step (5);

(8) information destination node is being after receiving the encoded packets that k i detects by step (7) linear dependence for collection, draw the initial data of i that information source sends by decoding for collection after, change step (9), wherein, when 1≤i≤g-1, k=2m, when i=g, k=m;

(9) i is added 1, and change step (3).

2. generation grouped cross-random linear network coding method according to claim 1, wherein the described i of step (4) is divided into two kinds of situations for the packet infrastructure of collection:

First kind of situation is: when 1≤i≤g-1, i for the packet infrastructure of collection is:

Wherein, D ⁱBe the matrix of i in the present node for the data formation of collection packets of information,

GC ⁱBe the matrix of i in the present node for the overall code coefficient formation of collection packets of information,

SD ⁱBe the matrix of initial data i for collection grouping formation,

E _{2m * 2m}Be the unit matrix of the capable 2m row of 2m,

SD _j ⁱBe the vector of initial data i for j grouping formation in the collection, and 1≤j≤2m,

GC _j ⁱBe the overall code coefficient vector of i for j grouping in the collection, and 1≤j≤2m,

x _Jl ⁱBe l the data of initial data i for j grouping in the collection, and 1≤j≤2m, 1≤l≤n,

GE _j ⁱBe i for j packets of information in the collection, and 1≤j≤2m;

Second kind of situation is: when i=g, i for the packet infrastructure of collection is:

Wherein, E _{M * m}Be the unit matrix of the capable m row of m,

SD _j ⁱBe the vector of initial data i for j grouping formation in the collection, and 1≤j≤m,

GC _j ⁱBe the overall code coefficient vector of i for j grouping in the collection, and 1≤j≤m,

x _Jl ^gBe l the element of initial data g for j grouping in the collection, and 1≤j≤m, 1≤l≤n,

GE _j ^gBe g for j packets of information in the collection, and 1≤j≤m.

3. generation grouped cross-random linear network coding method according to claim 1, wherein the described i of step (5) is for the local code coefficient vector of collection, by producing with following formula:

${\mathrm{LC}}^i=\left[\begin{array}{cccc}{\mathrm{lc}}_1^i& {\mathrm{lc}}_2^i& \·\·\·& {\mathrm{lc}}_k^i\end{array}\right]$

Wherein, LC ⁱBe the local code coefficient vector of i for collection, lc _j ⁱBeing i for the j number in the local code coefficient vector of collection, is the random number in the finite field gf (q), 1≤j≤k wherein, if present node is an information source node, and then when 1≤i≤g-1, k=2m, when i=g, k=m.

4. generation grouped cross-random linear network coding method according to claim 3, wherein i is for the j number lc of collection local code coefficient vector _j ⁱBe the interior random number of finite field gf (q) that generates according to certain rule, it obeys independent same distribution, and its probability-distribution function is:

$\left\{\begin{array}{c}\mathrm{Pr}({\mathrm{lc}}_j^i=x)=\left\{\begin{array}{cc}1-p& x=0\\ p/(q-1)& x\∈[\mathrm{1,2},\·\·\·,q-1]\end{array}\right.\\ p=(\log N+10)/k\end{array}\right.$

Wherein, N is initial data i for the number of grouping in the collection, when 1≤i≤g-1, and N=2m, when i=g, N=m.

5. generation grouped cross-random linear network coding method according to claim 1, the described new encoded packets of step (6) wherein, comprise the new information data and the new overall code coefficient that generate for the collection packets of information according to k in local code coefficient vector and the present node i, they produce by following two formulas respectively:

$D_0^i={\mathrm{LC}}^i\⊗D^i$

${\mathrm{GC}}_0^i={\mathrm{LC}}^i\⊗{\mathrm{GC}}^i$

Wherein, Be the multiplying in the finite field, D ₀ ⁱBe new information data, GC ₀ ⁱBe new overall code coefficient, if present node is information source node, then D ⁱ=SD ⁱ,

6. generation grouped cross-random linear network coding method according to claim 1, wherein the described downstream node of step (7) carries out linear dependence to it and detects after receiving new encoded packets, carries out as follows:

(6a), represent with following formula with behind the overall code coefficient matrix of i for collection in the overall code coefficient adding present node of new encoded packets:

${\mathrm{GC}}^i={\left[\begin{array}{ccccc}{\mathrm{GC}}_1^i& {\mathrm{GC}}_2^i& \·\·\·& {\mathrm{GC}}_k^i& {\mathrm{GC}}_0^i\end{array}\right]}^T$

Wherein, GC ₁ ⁱ, GC ₂ ⁱ... GC _k ⁱBe the overall code coefficient of k in the present node i for the collection packets of information;

(6b) obtain overall code coefficient matrix GC by the determinant conversion ⁱOrder, with rank (GC ⁱ) expression, if rank is (GC ⁱ)=k+1, then this new encoded packets is passed through to detect, otherwise, think that this new encoded packets can not provide useful information, can not be by detecting.

7. generation grouped cross-random linear network coding method according to claim 1, the wherein described decoding of step (8), adopt Gauss's pivot in a column elimination to obtain separating of system of linear equations:

$D^i={\mathrm{GC}}^i\⊗{\mathrm{SD}}^i$

It is initial data i for the collection SD ⁱ

Images