CN101621514A - Network data compressing method, network system and synthesis center equipment - Google Patents

Abstract

The invention provides a network data compressing method, a network system and synthesis center equipment. The method comprises the steps: sampling in a sensor network comprising J network nodes and obtaining node data of the J network nodes, wherein the node data of the J network nodes are relevant, and the node data of each network node have a pubic component and a special component with sparseness under a same base; carrying out compression processing on the node data of each network node in the sensor network by applying a random compression matrix, obtaining node compression data corresponding to the network node, wherein the random compression matrix is generated by a physical address of the network node; and sending the node compression data in each network node in the sensor network and the physical address corresponding to the node compression data to the synthesis center equipment. The invention can effectively and jointly compress the information of a plurality of nodes, avoid that node compunction takes up high bandwidth while calculating the relativity and satisfy the double requirements on efficiency and precision.

Description

The compression method of network data, network system and fusion center equipment

Technical field

The embodiment of the invention relates to the radio network technique field, relates in particular to a kind of compression method, network system and fusion center equipment of network data.

Background technology

Following monitoring and interception system will depend on distributed wireless sensor network miscellaneous, to guarantee still to obtain data reliably under extreme complexity and unsettled environment.Existing detection system adopts in network each node Information Monitoring to merge in destination node after transmitting and detects, and the monitoring system of this mode has been widely used in the military affairs and non-military fields such as remote sensing, navigation, meteorology, air traffic control and medical treatment that comprise control, communication, intellectual monitoring, tracking and identification.Owing to be subjected to the restriction of bandwidth and energy, active data compression and reconstruct become extremely important.Therefore, collaboration communication between multisensor and computation optimization are continuing to bring out.

The data compression method of present multinode network of the prior art mainly adopts distributed source coding, this method need communicate earlier between all different nodes, sharing of realization information, calculating the correlation between each node, thereby could effectively unite compression to the information of multinode; And adopt this method implementation procedure need take very high bandwidth, be difficult for realizing.

Compression sensing (Compressed Sensing is hereinafter to be referred as CS) is a kind of new sensing or Sampling techniques, has obtained very big difference with traditional data sampling.The CS theory shows that the collection quantity of signal can be far smaller than traditional sampled data (nyquist sampling theorem point out the minimum-rate of sampling must more than or equal to two times of signal peak frequency).The realization of CS is based on two criterions, sparse property and incoherence.Sparse property is meant that " information rate " of continuous time signal may be far smaller than its shared bandwidth, or the quantity of the discrete-time signal degree of freedom is far smaller than the length of signal.More precisely, CS is based on a fact, natural signal is sparse or compressible by after some conversion under suitable substrate is represented, for example discrete cosine transform (DCT), wavelet transform (DWT) and Fourier transform (FT) or the like.Incoherence has been represented a kind of antithesis characteristic, is similar to frequency and time relation.That is to say that the waveform of sampled signal under particular substrate is very intensive.

Generally, most signals can find at the bottom of the suitable transform-based and be expressed as sparse form, as voice signal, picture signal and some measurement data or the like.Under these substrates, the factor quantity of expression signal seldom, so CS can be good at realizing compressed encoding.To degree of rarefication is that K is (after the conversion under some substrate, have only K nonzero-divisor) signal, need not sample N time (supposing that N is a signal length), and only need c*K uncorrelated measuring-signal can realize compression, this has replaced original method to K sparse signal sampling N time, K＜＜N, c is constant (c 〉=2).Therefore, transducer can only send low volume data to destination node, and destination node still can also be handled these signals by these signals of reconstruct in any form.

Research for the data compression of single-sensor data is all to have obtained good effect in theory or in practice.But the data compression method of single-sensor can't be applied to comprise in the wireless distributed sensor network of a plurality of network nodes.

Summary of the invention

The embodiment of the invention provides a kind of compression method, network system and fusion center equipment of network data, need in the computing network correlation between each node in the prior art in order to solve, could unite the defective of compression to the information of multinode, realize a plurality of nodes are united compression effectively.

The embodiment of the invention provides a kind of compression method of network data, comprises:

In the sensor network that comprises J network node, sampling obtains the node data of a described J network node, have correlation between the node data of a wherein said J network node, and the node data of each network node has the special component that a component common and has sparse property under same substrate;

Use condensation matrix at random the node data of each network node in the described sensor network is compressed processing, obtain the node packed data corresponding with described network node, described condensation matrix at random is that the physical address by described network node generates;

The node packed data and the physical address corresponding of each network node in the described sensor network are sent to fusion center equipment.

The embodiment of the invention provides a kind of network system, comprising:

Processing module, be used at the sensor network that comprises J network node, sampling obtains the node data of a described J network node, have correlation between the node data of a wherein said J network node, and the node data of each network node has the special component that a component common and has sparse property under same substrate;

Compression module, be used for using condensation matrix at random the node data of each network node of described sensor network is compressed processing, obtain the node packed data corresponding with described network node, described condensation matrix at random is to be generated by the physical address of described network node;

Sending module is used for the node packed data and the physical address corresponding of described each network node of sensor network are sent to fusion center equipment.

The embodiment of the invention provides a kind of fusion center equipment, comprising:

Receiver module is used to receive the node packed data and the described physical address of a described J network node;

First acquisition module is used for each network node to described sensor network, obtains the described condensation matrix at random of described network node according to the physical address of the described network node that receives;

Second acquisition module is used for obtaining according to the node packed data of described J network node receiving the packed data of described sensor network; And obtain the condensation matrix at random of described sensor network according to the described condensation matrix at random of a described J network node;

Reconstructed module is used for the condensation matrix at random of the packed data according to described sensor network, described sensor network and the node packed data of described J network node receiving, the node data of described each network node of reconstruct.

The compression method of the network data of the embodiment of the invention, network system and fusion center equipment, do not need in the computing network correlation between each node, just can realize the information of multinode is united compression effectively, each node communication takies very high bandwidth when effectively avoiding calculating correlation, has satisfied the double requirements of efficient and accuracy.Correspond to actual needs.

Description of drawings

Fig. 1 is the flow chart of compression method of the network data of the embodiment of the invention one;

Fig. 2 is the part flow chart of compression method of the network data of the embodiment of the invention two;

Fig. 3 is the schematic diagram of the network system of the embodiment of the invention three;

Fig. 4 is the schematic diagram of the fusion center equipment of the embodiment of the invention four.

Embodiment

Introduce the technical scheme of the embodiment of the invention in detail below in conjunction with the drawings and the specific embodiments.

The embodiment of the invention one provides a kind of compression method of network data, in order to solve by correlation between each node in the computing network, could unite the problem of compression to the information of multinode.

Fig. 1 is the flow chart of compression method of the network data of the embodiment of the invention one, and as shown in Figure 1, the compression method of the network data of present embodiment comprises following steps:

Step 100, in the sensor network that comprises J network node, sampling obtains the node data of a described J network node, have correlation between the node data of a wherein said J network node, and the node data of each network node has the special component that a component common and has sparse property under same substrate;

Particularly, get and include J transducer in the wireless sensor network, each transducer is equivalent to a network node, for each network node, obtain the node data of each network node for N time by sampling, the node data that obtains each network node is the vector of N * 1, have correlation between the node data of a described J network node, and the node data of each network node has a component common and the special component with sparse property under the identical substrate of a precognition; All node datas that also are the whole sensor network have a component common and J special component under substrate Ψ, the degree of rarefication of special component is K, and K the non-zero factor promptly arranged under substrate Ψ, and a last N-K element is zero; Component common is not done the hypothesis of sparse property, and then the node data of j network node in the sensor network is:

x _j＝z _c+z _j，j∈{1，2，...，J}，z _c＝Ψθ _c；z _j＝Ψθ _j，||θ _j|| ₀＝K _j

Wherein, z _cRepresent component common, z _j, j ∈ 1,2. ..., J} is the special component of the node data correspondence of j network node, for the component common z that is without loss of generality _cWith special component z _jAll be assumed to be and obey the independent identically distributed Gauss's vector of zero-mean; θ _cAnd θ _jBe the coefficient under substrate Ψ, while θ _jDegree of rarefication be K (K the non-vanishing factor), || θ || ₀Just represent degree of rarefication; For being without loss of generality, substrate Ψ is made as unit matrix promptly, Ψ=I _N

Step 101, use condensation matrix at random the node data of each network node in the described sensor network is compressed processing, obtain the node packed data corresponding with described network node, described condensation matrix at random is to be generated by the physical address of described network node;

Particularly, for each network node, be seed with the physical address of this network node, adopt a pseudo-random generator to generate one group of pseudo random number, form condensation matrix Φ at random by these pseudo random numbers _j, j ∈ [1,2 ..., J], Φ _jBe M _jThe condensation matrix at random on * N rank, M＜＜N, get M usually and be slightly larger than 2 * K, K is the degree of rarefication of special component; N is the sampling number that obtains the networking node data, also is the line number of the node data vector of network node, by Φ _jNode data to j network node compresses, and just obtains the node packed data y of j network node _j(also being referred to as observed quantity), i.e. y _j=Φ _jx _j=Φ _j(z _c+ z _j)=Φ _jz _c+ Φ _jz _j, the Φ of condensation matrix at random of this network node _jNode data vector x with the network node on N * 1 rank _jBecome the node packed data vector y of the network node on M * 1 rank _jWherein said Φ _jBe to utilize own physical address as the pseudo-random generator generation of seed by node j, σ ²Be Ф _jThe variance of middle all elements, generally, zero-mean, variance that the node packed data of each network node also includes this node place are σ _j ²Additive white Gaussian noise, the node packed data after therefore obtaining compressing is:

y _j＝Φ _jx _j+w _j＝Φ _jz _c+Φ _jz _j+w _j，j∈{1，2，……，J}；

w _jRepresentative is σ in zero-mean, the variance at each node place _j ²Additive white Gaussian noise.

The node packed data that also can consider this network node place respectively is y _j=y _Cj+ y _Uj+ w _j, y _Cj=Φ _jz _cThe compression result of component common and y _Uj=Φ _jz _jThe compression result of special component, w _jIdentical with above-mentioned physical significance.

Step 102, the node packed data and the physical address corresponding of each network node in the described sensor network is sent to fusion center equipment.

Particularly, the node packed data of each network node in the described sensor network is sent to fusion center equipment, because compress used condensation matrix at random and be physical address by the network node of correspondence and be seed, for the ease of the reconstructed network nodal information, therefore, the data after needing physical address with this network node with the compression number are sent to fusion center equipment.

The compression method of the network data that present embodiment provides, do not need in the calculating sensor network correlation between each node, just can realize the information of multinode is united compression effectively, each node communication takies very high bandwidth in the time of can avoiding calculating correlation, has satisfied the double requirements of efficient and accuracy.Correspond to actual needs.

The embodiment of the invention two provides a kind of compression method of network data, further includes the method that the node packed data to each network node in a kind of sensor network of the embodiment of the invention is reconstructed.

Fig. 2 is the part flow chart of compression method of the network data of the embodiment of the invention two, the compression method of the network data of present embodiment, after the step 102 of " the node packed data and the physical address corresponding of each network node in the described sensor network are sent to fusion center equipment " in the invention process one, also comprise the method for reconstructed network data as shown in Figure 2, specifically comprise following steps:

Step 200, described fusion center equipment receive the node packed data and the described physical address of a described J network node;

Particularly, at first at the node packed data of fusion center equipment according to J network node of certain communication protocol receiving sensor network, and described J network node physical address corresponding, the node packed data that described fusion center equipment can be told each network node correspondence automatically according to communications protocol with and physical address corresponding.

Step 201, described fusion center equipment obtain the described condensation matrix at random of described network node to each network node in the described sensor network according to the physical address of the described network node that receives;

Particularly, fusion center equipment need be to each network node in the sensor network, to receive the pairing physical address of this network node is seed, be used to generate the identical pseudo-random generator of pseudo-random generator that condensation matrix adopted at random when adopting, the identical condensation matrix at random of condensation matrix at random that adopts when generating with this node data of compression with this node data of compression.

Step 202, described fusion center equipment obtain the packed data of described sensor network according to the node packed data of described J the network node that receives;

Particularly, fusion center equipment is with the node packed data y of each network node in the sensor network _jCombine the packed data that forms sensor network, i.e. Y=[y ₁, y ₂..., y _J], Y represents the packed data of sensor network, y _jBe the network node packed data of j network node, wherein j ∈ 1,2 ... J}.

Step 203, described fusion center equipment obtain the condensation matrix at random of described sensor network according to the described condensation matrix at random of a described J network node;

Particularly, fusion center equipment is with the node of each network node in the sensor network that recovers in the step 201 condensation matrix Φ at random _jCombine the condensation matrix at random that forms sensor network; Be Φ=[Φ ₁ ^T, Φ ₂ ^T..., Ф _J ^T] ^T, Φ represents the condensation matrix at random of sensor network, Φ _jBe the network node condensation matrix at random of j network node, wherein j ∈ 1,2 ... J}.

Step 204, described fusion center equipment are according to the packed data of described sensor network and the condensation matrix at random of described sensor network, the node data of described each network node of reconstruct.

Particularly, adopt the node data of the method reconstructed network node of iteration; Concrete special component two parts that also can be divided into reconstruct component common and each network node of reconstruct.

For the 1st iteration, the component common of reconstruct sensor network, because for any network node j, j ∈ 1,2 ... J}; The node initial data of network node is expressed as x _j=z _c+ z _j, z _cBe component common, z _jBe special component, and component common z _cWith special component z _jAll be assumed to be and obey the independent identically distributed Gauss's vector of zero-mean; If node number J is enough big, the mean value x by probability of the node data of all nodes equals satisfied in the whole sensor network $\stackrel{\‾}{x}\stackrel{p}{\→}{z}_c;$ P represents probability, the Φ of condensation matrix at random of the expression node j of arbitrary row among the Φ of condensation matrix at random of sensor network _jBe a M _j* N rank random matrix, wherein said Φ _jBe to utilize own MAC Address as the pseudo-random generator generation of seed by node j, σ ²Be Φ _jThe variance of middle all elements.So, utilize average notion to come the "ball-park" estimate component common, promptly all data are asked average; At first need the Φ of condensation matrix at random of sensor network is asked on average, specifically according to formula ${\widehat{\mathrm{\Φ}}}_j=\frac{1}{M_j{\mathrm{\σ}}_j^2}{\mathrm{\Φ}}_j,$ Each row to the Φ of condensation matrix at random of sensor network

Ask average.

According to central-limit theorem, adopt formula ${{\tilde{z}}_c}^1=\frac{1}{J}\underset{j=1}{\overset{J}{\mathrm{\Σ}}}{{\widehat{\mathrm{\Φ}}}_j}^T{y}_j,$ Obtain the component common of the node data of each network node in the 1st iteration sensor network

Y wherein _jPacked data Y=[y for the sensor network after integrating ₁, y ₂..., y _J] subvector, represent the network node packed data of j network node, wherein

Be the j row submatrix Φ of the Φ of condensation matrix at random of sensor network _jAverage data

Inverse matrix, wherein

According to formula ${\widehat{\mathrm{\Φ}}}_j=\frac{1}{M_j{\mathrm{\σ}}^2}{\mathrm{\Φ}}_j$ Calculate gained, y _jBe the j row of the packed data Y of sensor network, j ∈ 1,2 ... J}

Utilize the estimated value of component common in the node data of the 1st network node that iteration obtains

The special component estimated value of the node data correspondence of j network node in the sensor network of estimating the 1st time iteration obtaining

J ∈ 1,2 ... J};

Here the special component with j node of reconstruct is an example, carries out before the sub-iteration, at first according to the estimated value of the 1st component common that iteration has obtained

Utilize formula $Y_u=Y-\mathrm{\Φ}{{\tilde{z}}_c}^1,$ The component common of the node data of deletion all-network node from the packed data of sensor network, Y _uThe packed data of gained after the deletion component common in the packed data of expression sensor network;

The 1st second son iteration, the remaining matrix R of initialization ⁰=Y _u, initialization Y _u ⁰Full null matrix for M * J;

According to formula $n^1=\arg \underset{n=\mathrm{1,2},\·\·\·\·\·\·N}{\max }\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\frac{|\⟨r_j^0,{\mathrm{\φ}}_{j,n}\⟩|}{\left|\right|{\mathrm{\φ}}_{j,n}\left|\right|},$ Find out row mark, wherein r with maximal correlation degree _j ⁰Be remaining matrix R ⁰The j column vector, φ _{J, n}The Φ of the Φ of condensation matrix at random of representative sensor network _jThe n column vector of submatrix;

Set omega is set ¹=Ω ⁰∪ n ¹The matrix of forming M * 1 Described Ω ¹For according to calculating gained degree of correlation n ¹From the Φ of condensation matrix at random of sensor network, select Ω ¹Row; Carry out the 1st second son iteration, Ω because be ⁰Be empty set, the matrix of gained

Subclass for the Φ of condensation matrix at random of sensor network.

Utilize the least square estimation criterion

Estimate the special component of j network node of the 1st second son iteration

Adopt said method that all-network node in the sensor network is estimated respectively then, obtain j network node of the 1st second son iteration the node data correspondence the estimated value of special component

J ∈ 1,2 ... J}; Need to prove also needs to get here ${y_{u,j}}^1={\mathrm{\Φ}}_{j,{\mathrm{\Ω}}^1}{{\hat{z}}_j}^1,$ J ∈ { 1,2 ..., J}, and constitute matrix ${Y_u}^l={[y_{u,1}}^l,{y_{u,2}}^l,\·\·\·\·\·\·{y_{u,J}}^l],$ To calculate remaining matrix R ¹=Y _u-Y _u ¹For next second son iteration is prepared.

During l second son iteration, according to formula $n^l=\arg \underset{n=\mathrm{1,2},\·\·\·\·\·\·N}{\max }\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\frac{|\⟨r_j^{l-1},{\mathrm{\φ}}_{j,n}\⟩|}{\left|\right|{\mathrm{\φ}}_{j,n}\left|\right|},$ The degree of correlation of computing network data; R wherein _j ^L-1Be remaining matrix R ^L-1The j column vector, φ _{J, n}The Φ of the Φ of condensation matrix at random of representative sensor network _jThe n column vector of submatrix, wherein l＞1;

Set omega is set ^l=Ω ^L-1∪ n ^lThe matrix of forming M * l

Described Ω ^lFor from the Φ of condensation matrix at random of sensor network, selecting Ω according to calculating the gained degree of correlation ^lRow;

Utilize the least square estimation criterion

Estimate the special component estimated value of the node data correspondence of l second son iteration j network node

Can obtain the estimated value of special component of resultant node data correspondence of the l second son iteration of other all-network nodes in the sensor network then equally, and get ${y_{u,j}}^l={\mathrm{\Φ}}_{j,{\mathrm{\Ω}}^1}{{\hat{z}}_j}^l,$ J ∈ { 1,2 ..., J}, and by integrating the formation matrix ${{Y_u}^l=[y_{u,1}}^l,{y_{u,2}}^l,\·\·\·\·\·\·{y_{u,J}}^l],$ Calculate remaining matrix $R^l=Y_u-{Y_u}^l;$

Judge whether sub-iterations l equals the degree of rarefication K of network node; If equal, then current sub-iteration obtains the special component estimated value of the node data correspondence of described network node Be the special component estimated value of the node data correspondence of j network node in the sensor network that the 1st time iteration obtains

If current sub-iterations l is not equal to the degree of rarefication K of network node, then proceed sub-iteration, equal the coefficient degree K of the special component of this network node until sub-iterations l, when obtaining carrying out the 1st iteration, the estimated value of the special component of j network node in the sensing network J ∈ { 1,2 ..., J}.

According to formula ${{\tilde{x}}_j}^1={{\tilde{z}}_c}^1+{{\tilde{z}}_j}^1$ Calculate the estimated value of node data of j network node of the 1st iteration J ∈ { 1,2 ..., J} just can obtain the estimated value of the node data of all J network node according to said method.

And then proceed iteration, during the t time iteration, following method is adopted for the estimation of the component common of node data: the special component that utilizes the last iteration estimation in t 〉=2

In real number field

(H _jRepresent dimension) one group of base of last structure $B_j=[{{\mathrm{\Φ}}_{j,{\widehat{\mathrm{\Ω}}}_j}}^t,{Q_j}^t],$ Set omega _j ^tFor the N of the node data of the resulting described J network node of last iteration ties up special component

The special component of the sensor network that integration obtains

The sequence number set of middle N dimension nonzero element, ${{\mathrm{\Ω}}_j}^t\⋐\{\mathrm{1,2},\·\·\·\·\·\·N\},$

Be Φ by the Φ of condensation matrix at random of sensor network _jThe q column vector of submatrix The subspace of structure, wherein q ∈ Ω _j, Q _j ^tFor

Orthogonal intersection space; Here adopt orthogonal intersection space to leave out special component,, following relational expression arranged for arbitrary node j: ${\tilde{y}}_j=Q_j^{\mathrm{tT}}{y}_j=Q_j^{\mathrm{tT}}{{\mathrm{\Φ}}_j}^t(z_c+z_j)=Q_j^{\mathrm{tT}}{{\mathrm{\Φ}}_j}^t{z}_c={{\widetilde{\mathrm{\Φ}}}_j}^t{z}_c;$

Promptly ${{\widetilde{\mathrm{\Φ}}}_j}^t=Q_j^{\mathrm{tT}}{{\mathrm{\Φ}}_j}^t,$ Just can obtain z by finding the solution variance _c

By integrating $\tilde{Y}={[{{\tilde{y}}_1}^T,{{\tilde{y}}_2}^T,\·\·\·\·\·\·,{{\tilde{y}}_J}^T]}^T$ With $\widetilde{\mathrm{\Φ}}={[{{\widetilde{\mathrm{\Φ}}}_1}^T,{{\widetilde{\mathrm{\Φ}}}_2}^T,\·\·\·\·\·\·,{{\widetilde{\mathrm{\Φ}}}_J}^T]}^T,$ Can obtain: the component common of estimating during the t time iteration is:

${{\tilde{z}}_c}^t={\widetilde{\mathrm{\Φ}}}^{+t}\tilde{Y}$

Q _j ^tBe the subspace Orthogonal intersection space, in order to delete special component; Have ${\tilde{y}}_j=Q_j^{\mathrm{tT}}{y_j}^t=Q_j^{\mathrm{tT}}{{\mathrm{\Φ}}_j}^t(z_c+z_j)=Q_j^{\mathrm{tT}}{{\mathrm{\Φ}}_j}^t{z}_c={{\widetilde{\mathrm{\Φ}}}_j}^t{z}_c,$ Wherein ${{\widetilde{\mathrm{\Φ}}}_j}^t=Q_j^{\mathrm{tT}}{{\mathrm{\Φ}}_j}^t;$ So according to formula ${{\tilde{z}}_c}^t={\widetilde{\mathrm{\Φ}}}^{t+}\tilde{Y}$ Obtain the component common of the estimation of this iteration of t

Wherein ${\widetilde{\mathrm{\Φ}}}^{t+}={\left({\widetilde{\mathrm{\Φ}}}^{\mathrm{tT}}\widetilde{{\mathrm{\Φ}}^t}\right)}^{-1}{\widetilde{\mathrm{\Φ}}}^{\mathrm{tT}}$ Represent matrix

Generalized inverse matrix.

Estimated after the component common of node data of the t time iterative network node, utilized the estimated value of component common in the node data of the t time iterative network node

Estimate the special component estimated value of the node data correspondence of j network node in the sensor network that the t time iteration obtain ${{\tilde{z}}_j}^t,j\∈\{\mathrm{1,2},\·\·\·\·\·\·J\}.$

Estimation for the special component of each network node is finished in a sub-iteration, is example with the special component of j node of reconstruct still here, carries out before the sub-iteration, according to the component common that the 1st time iteration has been estimated, utilizes formula $Y_u=Y-\mathrm{\Φ}{{\tilde{z}}_c}^t,$ Delete the component common of all node datas in the t time iteration sensor network, Y _uThe packed data of the later gained of component common of all node data correspondences is deleted in expression from the packed data of sensor network;

The 1st second son iteration in the t time iteration, the remaining matrix R of initialization ⁰=Y _u, initialization Y _u ⁰Full null matrix for M * J;

According to formula $n^1=\arg \underset{n=\mathrm{1,2},\·\·\·\·\·\·N}{\max }\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\frac{|\⟨r_j^0,{\mathrm{\φ}}_{j,n}\⟩|}{\left|\right|{\mathrm{\φ}}_{j,n}\left|\right|},$ Find row mark, wherein r with maximal correlation degree _j ⁰Be remaining matrix R ⁰The j column vector, φ _{J, n}The Φ of the Φ of condensation matrix at random of representative sensor network _jThe n column vector of submatrix;

Set omega is set ¹=Ω ⁰∪ n ¹The matrix of forming M * 1

Described Ω ¹For according to calculating gained degree of correlation n ¹From the Φ of condensation matrix at random of sensor network, select Ω ¹Row; Carry out the 1st second son iteration, Ω because be ⁰Be empty set, the matrix of gained

Subclass for the Φ of condensation matrix at random of sensor network.

Utilize the least square estimation criterion

Estimate the special component estimated value of the node data correspondence of the 1st second son iteration j network node J ∈ 1,2 ... J}; Adopt the estimated value of the special component of the node data correspondence of all-network node in the said method estimation sensor network then, and get ${y_{u,j}}^1={\mathrm{\Φ}}_{j,{\mathrm{\Ω}}^1}{{\hat{z}}_j}^1,$ J ∈ { 1,2 ..., J}, and integrate the formation matrix ${{Y_u}^l=[y_{u,1}}^l,{y_{u,2}}^l,\·\·\·\·\·\·{y_{u,J}}^l],$ To calculate remaining matrix $R^1=Y_u-{Y_u}^1;$ For next second son iteration is prepared.

During l second son iteration in the t time iteration, according to formula

Calculate the degree of correlation; R wherein _j ^L-1Be remaining matrix R ^L-1The j column vector, φ _{J, n}The j column vector of the Φ of condensation matrix at random of representative sensor network, wherein l＞1;

Set omega is set ^l=Ω ^L-1∪ n ^lThe matrix of forming M * l

Described Ω ^lFor from the Φ of condensation matrix at random of sensor network, selecting Ω according to calculating the gained degree of correlation ^lRow;

Utilize the least square estimation criterion

Estimate the special component of l second son iterative network node Need the special component of the estimation of the l second son iteration of J network node in the estimation sensor network then equally, and get ${y_{u,j}}^l={\mathrm{\Φ}}_{j,{\mathrm{\Ω}}^l}{{\hat{z}}_j}^l,$ J ∈ { 1,2 ..., J}, and constitute matrix ${{Y_u}^l=[y_{u,1}}^l,{y_{u,2}}^l,\·\·\·\·\·\·{y_{u,J}}^l],$ Calculate remaining matrix $R^l=Y_u-{Y_u}^l;$

Judge that the degree of rarefication whether sub-iterations l equals the special component of described network node is K; Then proceed sub-iteration if be not equal to; If equal, then current sub-iteration obtains the special component estimated value of the node data correspondence of described network node

Be the special component estimated value of the node data correspondence of j network node in the sensor network that the t time iteration obtain

J ∈ { 1,2 ..., J}

According to formula ${{\tilde{x}}_j}^t={{\tilde{z}}_c}^t+{{\tilde{z}}_j}^t$ Calculate the estimated value of node data of j network node of the t time iteration

J ∈ { 1,2 ..., J} just can calculate the estimated value of the node data of all J of this iteration network node.

Estimated value with the node data of j network node of the t time iteration gained

Estimated value with the node data of j network node of the t-1 time iteration gained Compare and take absolute value, obtain a comparison value;

If described comparison value greater than predetermined value, is proceeded iteration; If described comparison value is less than predetermined value, current iteration gained then

Be the node data of the j network node of reconstruct gained, the predetermined value here is arbitrarily small constant, promptly before and after the data of the network node that obtains of twice iteration enough approach.Stop iteration, final obtaining

Be the node data of j the network node that reconstruct obtains, can reconstruct obtain the node data of J network node in the sensor network according to the method.

The compression method of the network data that present embodiment provides, further comprised the process that the node data after the compression is reconstructed, in the restructuring procedure, the component common of at first shooting off is estimated special component, the special component reconstruct component common accurately of skimming is again carried out the data convergence of iteration until reconstruct then, obtains the reconstruct data of degree of precision, the present embodiment method has satisfied the double requirements of efficient and accuracy, corresponds to actual needs.

Through the above description of the embodiments, those skilled in the art can be well understood to each execution mode and can realize by the mode that software adds essential general hardware platform, can certainly pass through hardware.Based on such understanding, the part that technique scheme contributes to prior art in essence in other words can embody with the form of software product, this computer software product can be stored in the computer-readable recording medium, as ROM/RAM, magnetic disc, CD etc., comprise that some instructions are with so that a computer equipment (can be a personal computer, server, perhaps network equipment etc.) carry out the described method of some part of each embodiment or embodiment.

One of ordinary skill in the art will appreciate that: all or part of step that realizes said method embodiment can be finished by the relevant hardware of program command, aforesaid program can be stored in the computer read/write memory medium, this program is carried out the step that comprises said method embodiment when carrying out; And aforesaid storage medium comprises: various media that can be program code stored such as ROM, RAM, magnetic disc or CD.

The embodiment of the invention three provides a kind of network system, Fig. 3 is the schematic diagram of the network system of the embodiment of the invention three, as shown in Figure 3, the network system of present embodiment comprises, processing module 11, compression module 12 and sending module 13, processing module 11 is used at the sensor network that comprises J network node, sampling obtains the node data of a described J network node, have correlation between the node data of a wherein said J network node, and the node data of each network node has the special component that a component common and has sparse property under same substrate; Compression module 12 is used for using condensation matrix at random the node data of each network node of described sensor network is compressed processing, obtain the node packed data corresponding with described network node, described condensation matrix at random is to be generated by the physical address of described network node; Sending module 13 is used for the node packed data of described each network node of sensor network and physical address corresponding are sent to fusion center equipment.

Particularly, processing module 11 is used for the network node that each transducer of sensor network that comprises J transducer forms is carried out sampling processing, obtain the node data of each network node, have correlation between the node data of a described J network node, and the node data of each network node has a component common and the special component with sparse property under the identical substrate of a precognition; All node datas that also are the whole sensor network have a component common and J special component under substrate Ψ, the degree of rarefication of special component is K, K the non-zero factor promptly arranged under substrate Ψ, under the situation of the hypothesis of component common not being done sparse property, then the node data of j network node in the sensor network is:

x _j＝z _c+z _j，j∈{1，2，...，J}，z _c＝Ψθ _c；z _j＝Ψθ _j，||θ _j|| ₀＝K _j

Compression module 12 is realized the node data of each network node is compressed, particularly, for each network node, be seed with the physical address of this network node, adopt a pseudo-random generator to generate one group of pseudo random number, form condensation matrix Φ at random by these pseudo random numbers _j, j ∈ [1,2 ..., J], Φ _jBe M _jThe condensation matrix at random on * N rank, M＜＜N, get M=2 * K usually, K is the degree of rarefication of special component; N is the sampling number that obtains the networking node data, also is the line number of the node data vector of network node, by Φ _jNode data to j network node compresses, and just obtains the node packed data y of j network node _j(also being referred to as observed quantity), i.e. y _j=Φ _jx _j=Φ _j(z _c+ z _j)=Φ _jz _c+ Φ _jz _j, the Φ of condensation matrix at random of this network node _jNode data vector x with the network node on N * 1 rank _jBecome the node packed data vector y of the network node on M * 1 rank _jWherein said Φ _jBe to be that the pseudo-random variable of 1/N produces by zero-mean, variance, generally, zero-mean, variance that the node packed data of each network node also includes this node place are σ _j ²Additive white Gaussian noise, the node packed data after therefore obtaining compressing is:

y _j＝Φ _jx _j+w _j＝Φ _jz _c+Φ _jz _j+w _j，j∈{1，2，……，J}；

w _jRepresentative is σ in zero-mean, the variance at each node place _j ²Additive white Gaussian noise.

The transmission that sending module 13 is realized the node packed data of the network node after the compression, node packed data with each network node in the described sensor network sends to fusion center equipment particularly, because compress used condensation matrix at random and be physical address by the network node of correspondence and be seed, for the ease of the reconstructed network nodal information, therefore, the data after needing physical address with this network node with the compression number are sent to fusion center equipment.

The compression process that realizes network data between each module of the network system of present embodiment is identical with the compression process of embodiment one network data.

The network system that present embodiment provides, do not need in the calculating sensor network correlation between each node, just can realize the information of multinode is united compression effectively, each node communication takies very high bandwidth in the time of can avoiding calculating correlation, has satisfied the double requirements of efficient and accuracy.Correspond to actual needs.

The embodiment of the invention four provides a kind of fusion center equipment, Fig. 4 is the schematic diagram of the fusion center equipment of the embodiment of the invention four, as shown in Figure 4, the fusion center equipment of present embodiment comprises, receiver module 21, first acquisition module 22, second acquisition module 23, reconstructed module 24, receiver module 21 is used to receive the node packed data and the described physical address of a described J network node; First acquisition module 22 is used for each network node to described sensor network, obtains the described condensation matrix at random of described network node according to the physical address of the described network node that receives; Second acquisition module 23 is used for obtaining according to the node packed data of described J the network node that receives the packed data of described sensor network; And obtain the condensation matrix at random of described sensor network according to the described condensation matrix at random of a described J network node; Reconstructed module 24 is used for the condensation matrix at random of the packed data according to described sensor network, described sensor network and the node packed data of described J network node receiving, the node data of described each network node of reconstruct.

Particularly, fusion center equipment receiver module 21 is according to the node packed data of J network node of certain communication protocol receiving sensor network, and described J network node physical address corresponding, the node packed data that described fusion center equipment can be told each network node correspondence automatically according to communications protocol with and physical address corresponding.Each network node in 22 pairs of sensor networks of first acquisition module of fusion center equipment, to receive the pairing physical address of this network node is seed, be used to generate the identical pseudo-random generator of pseudo-random generator that condensation matrix adopted at random when adopting, the identical condensation matrix at random of condensation matrix at random that adopts when generating with this node data of compression with this node data of compression.The node of each network node condensation matrix Φ at random in the sensor network that second acquisition module 23 of fusion center equipment obtains first acquisition module 22 _iCombine the condensation matrix at random that forms sensor network according to row; Promptly $\mathrm{\Φ}={[{{\mathrm{\Φ}}_1}^T,{{\mathrm{\Φ}}_2}^T,\·\·\·\·\·\·,{{\mathrm{\Φ}}_J}^T]}^T,$ Φ represents the condensation matrix at random of sensor network, Φ _jBe the network node condensation matrix at random of j network node, wherein j ∈ 1,2 ... J}; The node packed data y of each network node in the sensor network that also receiver module 21 is received simultaneously _jCombine the packed data that forms sensor network, i.e. Y=[y according to row ₁, y ₂..., y _J], Y represents the packed data of sensor network, y _jBe the network node packed data of j network node, wherein j ∈ 1,2 ... J}.Reconstructed module 24 realizes the reconstruct of the node data of each network node in the sensor network, and described restructuring procedure is identical with the embodiment of the invention two, does not repeat them here.

The fusion center equipment of present embodiment has been realized the double requirements of efficient and accuracy has been satisfied in the accurate reconstruct of node data after the compression.Correspond to actual needs.

Device embodiment described above only is schematic, wherein said unit as the separating component explanation can or can not be physically to separate also, the parts that show as the unit can be or can not be physical locations also, promptly can be positioned at a place, perhaps also can be distributed on a plurality of network element.Can select wherein some or all of module to realize the purpose of present embodiment scheme according to the actual needs.Those of ordinary skills promptly can understand and implement under the situation of not paying performing creative labour.

It should be noted that at last: above embodiment only in order to technical scheme of the present invention to be described, is not intended to limit; Although with reference to previous embodiment the present invention is had been described in detail, those of ordinary skill in the art is to be understood that: it still can be made amendment to the technical scheme that aforementioned each embodiment put down in writing, and perhaps part technical characterictic wherein is equal to replacement; And these modifications or replacement do not make the essence of appropriate technical solution break away from the spirit and scope of various embodiments of the present invention technical scheme.

Claims (10)

1, a kind of compression method of network data is characterized in that, comprises:

In the sensor network that comprises J network node, sampling obtains the node data of a described J network node, have correlation between the node data of a wherein said J network node, and the node data of each network node has the special component that a component common and has sparse property under same substrate;

Use condensation matrix at random the node data of each network node in the described sensor network is compressed processing, obtain the node packed data corresponding with described network node, described condensation matrix at random is that the physical address by described network node generates;

The node packed data and the physical address corresponding of each network node in the described sensor network are sent to fusion center equipment.

2, the compression method of network data according to claim 1, it is characterized in that, described application condensation matrix is at random compressed processing to the node data of each network node in the described sensor network, obtains the node packed data corresponding with described network node and is specially:

According to formula: y _j=Φ _jx _j=Φ _j(z _c+ z _j)=Φ _jz _c+ Φ _jz _jObtain the node packed data of described network node correspondence, wherein subscript j represents j network node in the sensor network, j ∈ 1,2 ..., J}, x _jThe node data of representing j network node, described node data x _jBe the N dimension; z _cThe expression component common, z _jThe special component of representing the node data correspondence of j network node; y _jThe node packed data of representing j network node, described node packed data y _jBe the M dimension; Φ _jThe physical address of representing j network node is the condensation matrix at random that seed utilizes pseudo-random generator to generate, condensation matrix Φ at random _jFor M is capable, N row, wherein M＜＜N.

3, the compression method of network data according to claim 2 is characterized in that, the additive white Gaussian noise that the node packed data of described network node correspondence also includes described network node is y _j=Φ _jx _j+ w _j, w _jIt is the additive white Gaussian noise at j network node place.

4, the compression method of network data according to claim 1 is characterized in that, also comprises:

Described fusion center equipment receives the node packed data and the described physical address of a described J network node;

Described fusion center equipment obtains the described condensation matrix at random of described network node to each network node in the described sensor network according to the physical address of the described network node that receives;

Described fusion center equipment obtains the packed data of described sensor network according to the node packed data of described J the network node that receives;

Described fusion center equipment obtains the condensation matrix at random of described sensor network according to the described condensation matrix at random of a described J network node;

Described fusion center equipment is according to the packed data of described sensor network and the condensation matrix at random of described sensor network, the node data of described each network node of reconstruct.

5, the compression method of network data according to claim 4 is characterized in that, the described condensation matrix at random that the physical address of the described network node that described basis receives obtains described network node is specially:

Physical address with described network node is a seed, adopts to generate the condensation matrix at random that the used pseudo-random generator of described condensation matrix at random generates described network node.

6, the compression method of network data according to claim 4 is characterized in that, the packed data that the node packed data of described J the network node that described fusion center equipment basis receives obtains described sensor network is specially:

Application of formula Y=[y ₁, y ₂..., y _J] the node packed data of J network node is combined the packed data that forms sensor network, y ₁, y ₂..., y _JBe column vector, be respectively the packed data of J network node; Wherein Y represents the packed data of described whole sensor network, y _jBe the node packed data of j network node, wherein j ∈ 1,2 ... J}; Subscript T is the transposition symbol of vector;

Described fusion center equipment is specially according to the condensation matrix at random that the described condensation matrix at random of a described J network node obtains described sensor network:

Application of formula Φ=[Φ ₁ ^T, Φ ₂ ^T..., Φ _J ^T] ^TThe described condensation matrix at random of J network node is combined the condensation matrix at random that forms sensor network, and wherein Φ represents the condensation matrix at random of sensor network, Φ _jBe the network node condensation matrix at random of j network node, wherein j ∈ 1,2 ... J}; Subscript T is the transposition symbol of vector.

7, according to the compression method of the arbitrary described network data of claim 4 to 6, it is characterized in that, described fusion center equipment is according to the packed data of described sensor network and the condensation matrix at random of described sensor network, the node data of described each network node of reconstruct, the concrete alternative manner that adopts may further comprise the steps:

During the 1st iteration, according to central-limit theorem, according to formula ${{\tilde{z}}_c}^1=\frac{1}{J}\underset{j=1}{\overset{J}{\mathrm{\Σ}}}{{\widehat{\mathrm{\Φ}}}_j}^T{y}_j,$ Obtain the estimated value of component common in the node data of described network node, wherein

The estimated value of representing component common in the node data of the 1st network node that iteration obtains;

Be j the submatrix Φ of the Φ of condensation matrix at random of sensor network _jAverage data

Inverse matrix, wherein

According to formula ${\widehat{\mathrm{\Φ}}}_j=\frac{1}{M_j{\mathrm{\σ}}^2}{\mathrm{\Φ}}_j$ Calculate gained, M _jBe submatrix Φ _jLine number, σ ²Be submatrix Φ _jThe variance of middle all elements, y _jBe the j row of the packed data Y of sensor network, j ∈ 1,2 ... J};

Utilize the estimated value of component common in the node data of the 1st network node that iteration obtains The special component estimated value of the node data correspondence of j network node in the sensor network of estimating the 1st time iteration obtaining ${{\tilde{z}}_j}^1,j\∈\{\mathrm{1,2}......J\};$

According to formula ${{\tilde{x}}_j}^1={{\tilde{z}}_c}^1+{{\tilde{z}}_j}^1$ Calculate the estimated value of the node data of the 1st j the network node that iteration obtains ${{\tilde{x}}_j}^1,j\∈\{\mathrm{1,2},......J\};$

During the t time iteration, t 〉=2; Set omega is set _j ^tFor the N of the node data of the resulting described J network node of last iteration ties up special component Middle K _jThe sequence number set of individual nonzero element, ${\mathrm{\Ω}}_j^t\⋐\{\mathrm{1,2},......N\},$ By sensor network with loom condensation matrix Φ _jColumn vector

The constructor space

Wherein, φ _{J, q}Represent the Φ of the Φ of condensation matrix at random of described sensor network _jThe q row of submatrix, Q _j ^tFor

Orthogonal intersection space, in order to estimate component common, remove the interference of special component; By ${\tilde{y}}_j={Q_j^t}^T{y}_j={Q_j^t}^T{\mathrm{\Φ}}_j^t(z_c+z_j)={Q_j^t}^T{\mathrm{\Φ}}_j^t{z}_c={\widetilde{\mathrm{\Φ}}}_j^t{z}_c,$ Wherein ${\tilde{Q}}_j^t={Q_j^t}^T{\mathrm{\Φ}}_j^t;$ Draw formula ${{\tilde{z}}_c}^t={\widetilde{\mathrm{\Φ}}}^{t+}\tilde{Y},$ And $\tilde{Y}={[{{\tilde{y}}_1}^T,{{\tilde{y}}_2}^T,......,{{\tilde{y}}_J}^T]}^T,$ $\widetilde{\mathrm{\Φ}}={[{{\widetilde{\mathrm{\Φ}}}_1}^T,{{\widetilde{\mathrm{\Φ}}}_2}^T,......,{{\widetilde{\mathrm{\Φ}}}_J}^T]}^T$ Calculate the estimated value of component common in the node data of the t time iterative network node

Wherein

Matrix when representing the t time iteration

Generalized inverse matrix; Subscript T is the transposition symbol of vector;

Utilize the estimated value of component common in the node data of the t time iterative network node

Estimate the special component estimated value of the node data correspondence of j network node in the sensor network that the t time iteration obtain ${{\tilde{z}}_j}^t,j\∈\{\mathrm{1,2},......J\};$

According to formula ${{\tilde{x}}_j}^t={{\tilde{z}}_c}^t+{{\tilde{z}}_j}^t$ Calculate the estimated value of node data of j network node of the t time iteration ${{\tilde{x}}_j}^t,j\∈\{\mathrm{1,2},......J\};$

Estimated value with the node data of j network node of the t time iteration gained

Estimated value with the node data of j network node of the t-1 time iteration gained

Subtract each other and take absolute value, obtain a comparison value;

If described comparison value greater than predetermined value, is proceeded iteration; If described comparison value is less than predetermined value, the current iteration gained

Be the node data of the j network node of reconstruct gained.

8, the compression method of network data according to claim 7 is characterized in that, the estimated value of component common in the described node data that utilizes the t time iterative network node Estimate the special component estimated value of the node data correspondence of j network node in the sensor network that the t time iteration obtain

Specifically may further comprise the steps:

According to formula $Y_u=Y-\mathrm{\Φ}{{\tilde{z}}_c}^t,$ Component common among the packed data Y of deletion sensor network in the node data of all-network node;

The remaining matrix of initialization $R^0=Y_u=[y_1-{\mathrm{\Φ}}_1{{\tilde{z}}_c}^t,y_2-{\mathrm{\Φ}}_2{{\tilde{z}}_c}^t,...,y_J-{\mathrm{\Φ}}_J{{\tilde{z}}_c}^t],$ M * J dimension, initialization Y _u ⁰Full null matrix for M * J;

During l second son iteration, according to formula $n^l=\arg \underset{n=\mathrm{1,2},......N}{\max }\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\frac{|<r_j^{l-1},{\mathrm{\&phi;}}_{j,n}>|}{\left|\right|{\mathrm{\&phi;}}_{j,n}\left|\right|},$ Find row mark with maximal correlation degree; R wherein _j ^L-1Be remaining matrix R ^L-1The j column vector, φ _{J, n}Represent the Φ of the Φ of condensation matrix at random of described sensor network _jThe n row of submatrix, wherein l 〉=1; Wherein＜be the inner product compute sign,

Set omega is set ^l=Ω ^L-1∪ n ^lThe matrix of forming M * l

Described Ω ^lFor from the Φ of condensation matrix at random of described sensor network, selecting Ω according to calculating the gained degree of correlation ^lRow;

Utilize the least square estimation criterion

Estimate the special component of the node data correspondence of j network node of l second son iteration

Get ${y_{u,j}}^l={\mathrm{\&Phi;}}_{j,{\mathrm{\&Omega;}}^l}{\tilde{z}}_j^l,j\&Element;\{\mathrm{1,2}......,J\},$ And formation matrix $Y_u^l=[{y_{u,1}}^l,{y_{u,2}}^l,......{y_{u,J}}^l],$ Calculate remaining matrix $R^l=Y_u-Y_u^l;$

Judge that the degree of rarefication whether sub-iterations l equals the special component of described network node is K; Then proceed sub-iteration if be not equal to; If equal, then current sub-iteration obtains the special component estimated value of the node data correspondence of described network node

Be the special component estimated value of the node data correspondence of j network node in the sensor network that the t time iteration obtain

9, a kind of network system is characterized in that, comprising:

Processing module, be used at the sensor network that comprises J network node, sampling obtains the node data of a described J network node, have correlation between the node data of a wherein said J network node, and the node data of each network node has the special component that a component common and has sparse property under same substrate;

Compression module, be used for using condensation matrix at random the node data of each network node of described sensor network is compressed processing, obtain the node packed data corresponding with described network node, described condensation matrix at random is to be generated by the physical address of described network node;

Sending module is used for the node packed data and the physical address corresponding of described each network node of sensor network are sent to fusion center equipment.

10, a kind of fusion center equipment is characterized in that, comprising:

Receiver module is used to receive the node packed data and the described physical address of a described J network node;

First acquisition module is used for each network node to described sensor network, obtains the described condensation matrix at random of described network node according to the physical address of the described network node that receives;

Second acquisition module is used for obtaining according to the node packed data of described J network node receiving the packed data of described sensor network; And obtain the condensation matrix at random of described sensor network according to the described condensation matrix at random of a described J network node;

Reconstructed module is used for the condensation matrix at random of the packed data according to described sensor network, described sensor network and the node packed data of described J network node receiving, the node data of described each network node of reconstruct.

Images