ESA.tar.gz_ESA_MATLAB esa_compress sensing_压缩感知_压缩感知 信号

function [xnew,MemoryNode_new,MemoryCluster_new,powerUO,LinkMat,varargout]= ... EsaR_iter(tri_actif,nontri,xinit,y,V,MemoryNode,MemoryCluster,... Dist,LinkMatOld,alpha, proba_indep, type_fail,varargin) % % usage [xnew,MemoryNode_new,MemoryCluster_new,powerUO,LinkMat,varargout]=... % EsaR_iter(tri_actif,nontri,xinit,y,V,MemoryNode,MemoryCluster,... % Dist,LinkMatOld,alpha,etat, proba_indep, type_fail,varargin) % % Input : tri_actif : ntrix3 matrix of nodes that are in the embedded triangles % nontri : 1:n \ tri_actif : nodes that are in the network but % not in the embedded trianlges % xinit : current posterior mean % y : observations/sigma (normalised observations) % V : matrix to be inverted % MemoryNode : memory that each node has of its neighbors. This % is used in the robust update % MemoryCluster : inside an embedded triangle, each sensor i keeps % in memory the value \hat{y_k}^m for k = the two % other nodes that in the same triangle as i % % Dist : matrix with Distance between nodes in the network, % when these nodes are linked % LinkMatOld : Matrix giving the state of each link. % LinkMatOld(i,j) = 1 if link is active, =1 if it is % the first iteration where link is inactive =-2 if % it is the second step, etc,... % LinkMatOld(i,j) = 0 if there are no link between i % and j % alpha : exponent in computation of the Energy as a fct of Dist % proba_indep = [p1 M], with p1 probability of link failure at each iteration % and M maximum number of iteratinons where a link or % node can be off. % type_fail = 'node_fail' or 'link_fail' % varargin{1} = nodeOff = indices of nodes that are sleeping (for % type_fail = node_fail) % % Output : xnew : posterior mean at next iteration % MemoryNode_new: updated MemoryNode % MemoryCluster_new : updated MemoryCluster % powerUO : power unidirectional and omni consumed at this % iteration % LinkMat : new matrix indicating links that are ON or OFF (and % if OFF, at which stage) % varargout = updated nodeOFF : indices of the new nodes that % are sleeping % % called by Esa_robust.m % Adjacency matrix dag = (LinkMatOld ~= 0); n=size(V,1); taille = size(tri_actif); nbtri=taille(1); % number of disjoint embedded triangles nbnodes = taille(1)*taille(2); DD=spdiags(sparse(V),0); J=sparse(diag(DD)); for k=1:nbtri trik = tri_actif(k,:); J(trik,trik) = V(trik,trik); end; K = J-V; x=xinit; MemoryCluster_new = MemoryCluster; elemtri =sort(reshape(tri_actif, taille(1)*taille(2),1)); switch type_fail case 'link_fail' [LinkMat] = matfailureN_L(LinkMatOld, proba_indep,type_fail); case 'node_fail' nodeOff=varargin{1}; [LinkMat, ind_fail] = matfailureN_L(LinkMatOld, proba_indep,... type_fail,nodeOff); varargout(1)={ind_fail}; otherwise error('type_fail in EsaR_iter should be link_fail or node_fail') end % product Ktri_x = Ktri*x but now with link failure and/or node down % new value transmitted : % NOTE : linkmat(i,j) gives the transmission node from i to j % here what we need is transmission from x_atnode to x_trans % hence we need to transpose _linkmat_ % from LinkMat build a matrix of 0 and 1 : p1 = proba_indep(1); linkmat = (LinkMat> 0); % x_trans = x transmitted : values that were transmitted x_trans = (K.*linkmat')*x; Kfailed = K.*(1 - linkmat'); % x_nontrans = x non transmitted : values that were NOT transmitted x_nontrans = sum(Kfailed.*MemoryNode,2); Kx = x_trans + x_nontrans; for k=1:nbtri % For all embedded triangles trik=tri_actif(k,:); Jtrik=J(trik,trik); % UPDATE STEP : either ynewk is computed from x_transmitted and % x_nontransmitted. ynewk = y(trik) + Kx(trik); for indtri = 1:3 memk = MemoryCluster{indtri}(k,:); one = indtri; switch type_fail case 'link_fail' oneOK =1; otherwise if length(ind_fail)>0 oneOK = ~sum(ind_fail == trik(one)); else oneOK=1; end end % end switch type_fail switch oneOK case 0 % node 'one' not solved x(trik(one)) = xinit(trik(one)); case 1 twothreeind = setdiff(1:3,indtri); two= (twothreeind(1)); three = (twothreeind(2)); % If the link between two and one or between three and one % fails, we use the value in the memory if ~linkmat(trik(two), trik(one)) ynewk(two) = memk(1); end if ~linkmat(trik(three), trik(one)) ynewk(three) = memk(2); end % SOLVE step when transmission was successful MemoryCluster_new{indtri}(k,:) = [ynewk(two),ynewk(three)]; xtmp = Jtrik\ynewk; x(trik(one))= xtmp(one); end% end switch oneOK end % end for indtri end % for k % If a node if off, it will not receive its new value, even if % its neighbors try to communicate with him if strcmp(type_fail,'node_fail') % node who fails do not update their value if length(ind_fail) >0 x(ind_fail) = xinit(ind_fail); end end % Update for other nodes: % Gauss-seidel type: we use the newly computed values for x % Ktri_x = Ktri*x; x_trans = (K.*linkmat')*x; Kfailed = K.*(ones(size(linkmat)) - linkmat'); x_nontrans = sum(Kfailed.*MemoryNode,2); Kx = x_trans + x_nontrans; D = spdiags(V(nontri,nontri),0); if length(nontri) >0 x(nontri,1) = (y(nontri,1) + Kx(nontri,1))./D; end % Again, if some node in nontri are OFF, their value is not updated %---- Now this part is treated above ... if ~strcmp(type_fail,'link_fail') % node who are sleeping do not update their value if length(ind_fail) >0 x(ind_fail) = xinit(ind_fail); end end xnew = x; % Total energy spent by the sensors % Here you take into account if the sleeping modes of the sensors % are synchronised or not synchronised if strcmp(type_fail,'node_fail') if length(ind_fail) > 0 powerUO = TrianPowerHeadR(tri_actif,Dist,J,K,alpha,ind_fail); else powerUO = TrianPowerHead(tri_actif,Dist,J,K,alpha); end else % type_fail = 'link_fail' powerUO = TrianPowerHead(tri_actif,Dist,J,K,alpha); end %x_atnode is the value transmitted at this iteration: hence % if the transmission has been successful it must be recorded MemoryNode_new = newMemory(linkmat, dag, x, MemoryNode); %------------------------------------------------------------------