CN103188037A - Depth-first search sphere decoding method and device - Google Patents

Abstract

The invention relates to a depth-first search sphere decoding method and device. According to the depth-first search sphere decoding method and device, depth-first search is carried out first to obtain K alternative emission symbolic vectors, wherein K is the size of a search list and is a positive integer; after a maximum route weight in the K alternative emission symbolic vectors is determined, the depth-first search is carried out continuously to search one or more new emission symbolic vectors; and then the decision that whether an alternative emission symbolic vector with the maximum route weight is replaced with the new emission symbolic vectors is made according to the fact that whether a route weight of the searched new emission symbolic vectors is smaller than the maximum route weight. In a searching process of the one or more new emission symbolic vectors, new nodes at the lowest level are searched, routes from a node at the top level of a search tree to the new nodes are taken as the new emission symbolic vectors, new nodes which are not at the lowest level of the search tree and route weights of which are larger than the maximum route weight are discarded, and subtrees below the search tree are discarded.

Description

Depth-first search globular decoding method and apparatus

Technical field

The present invention relates to the signal detecting method of multi-input multi-output-orthogonal frequency division multiplexing (MIMO-OFDM) system, especially relate to a kind of depth-first search globular decoding method and apparatus.

Background technology

The signal detection algorithm of multi-input multi-output-orthogonal frequency division multiplexing (MIMO-OFDM) system is divided linear detection algorithm and non-linear detection algorithm two classes.Linearity test algorithm commonly used mainly comprises ZF (ZF) and least mean-square error (MMSE), and its complexity is minimum, but performance is the poorest.Non-linear detection algorithm commonly used mainly comprises maximum likelihood (ML), disturb to eliminate (IC), decompose based on QR detection algorithm and globular decoding (SD).The Maximum Likelihood Detection algorithm is the Optimum Detection in the MIMO-OFDM system, but complexity is the highest, is unfavorable for real-time processing.Disturb and eliminate and calculate by simplifying based on the detection algorithm that QR decomposes, the realization suboptimum detects effect, obtains better balance between performance and complexity, but relative Maximum Likelihood Detection, still have bigger performance loss, and disturb the meeting of elimination the wrong phenomenon of propagating to occur.

(Sphere Decoding, SD) detection algorithm is proposed by Fincke and Pohst globular decoding the earliest, is used for research integer least square problem.Globular decoding comes down to MIMO-ML detection problem is configured to the problem of setting an optimal path of search a source signal constellation point, and constantly strengthens constraints in search procedure.The operation principle of globular decoding is: presetting one earlier in receiving signal space is the ball in the center of circle with the received signal points, again this ball is mapped as a ellipsoid in the space that transmits, and possible the transmitting a little of search in ellipsoid, in case finding a mapping point that transmits a little namely with this signaling point is that radius shrinks default ball with the distance that receives signal, thereby makes follow-up search be able to carry out in littler scope.

Nearest globular decoding technical research shows: the prerequisite that it can significantly reduce (becoming polynomial relation with number of transmit antennas) in complexity is issued to the error performance that approaches or be equal to Maximum Likelihood Detection.Therefore in the MIMO-OFDM of reality system, the performance that general receiving terminal adopts the SD algorithm to approach ML is used some simultaneously and can be reduced the strategy of complexity in the globular decoding detection algorithm.

The tree search plan of globular decoding generally has depth-first, breadth First and tolerance preferential.For hard sphere shape decoding, all be in order to find the ML solution to reach the purpose that reduces complexity and memory space as soon as possible, three class search plans have corresponding corrective measure.Concerning depth-first search, Schnorr Euchner (SE) ordering, QR ordering etc. are dynamically dwindled initial radium, are fixed each maximum access node of father node lower floor, adopt in common having.To BFS, common have that fixing every layer of maximum reserve section are counted, a Schnorr Euchner ordering, QR ordering etc.To measuring preferentially, also can use the thought of ordering and maximum surviving path.But these thoughts can only guarantee to declare firmly BER (Bit Error Rate, the error rate) performance, for LTE (Long Term Evolution) system that uses Turbo decoding, if BLER (Block Error Rate preferably will be arranged, Block Error Rate) performance, then need to guarantee the alternative emission symbolic vector of some, to produce soft information fully accurately, namely need to carry out soft globular decoding.

Soft globular decoding adopts the scheme that is called tabulation globular decoding (LSD) sometimes.This scheme is calculated soft bit LLR (b according to search listing (supposing that size is K) _{L, n}) (n bit of emission symbolic vector s l layer is designated as b _{L, n}, 1≤l≤N _T, 1≤n≤N _c, N _cRepresent the number of bits of single emission symbolic vector correspondence, relevant with modulation system).Yet this scheme need travel through whole search tree in the initial radium scope usually, and the node number of visit is more.And the size of search listing changes with the variation of initial radium.Some method is come the node number of limiting access by fixing maximum search list size, thereby can finish search in advance, but can to cause the symbolic vector that searches not be optimal value for this, and this can influence the accuracy of soft information.At this moment or choosing a quite bigger radius produces soft information fully accurately in addition, when tabulation when imperfect, some bit value can occur and be empty phenomenon,, this will cause very high complexity; Compose into maximum, perhaps weights or the current search radius approximate evaluation bit value according to symbolic vector in the tabulation is empty weights, and this will cause soft information inaccurate again.

Summary of the invention

Technical problem to be solved by this invention provides a kind of depth-first search globular decoding method and apparatus that reduces complexity.

The present invention is that to solve the problems of the technologies described above the technical scheme that adopts be to propose a kind of depth-first search globular decoding method, may further comprise the steps: the initialization search radius; Channel matrix is carried out QR decompose, obtain Q matrix and R matrix; Carry out depth-first search to obtain K alternative emission symbolic vector, wherein K is that search listing size and K are positive integer; Determine K the maximum path weights in the alternative emission symbolic vector; Continue to carry out depth-first search to seek one or more new emission symbolic vectors; And whether determine whether replacing the alternative emission symbolic vector that has the maximum path weights in described K the alternative emission symbolic vector with new emission symbolic vector less than described maximum path weights by the routine weight value of the new emission symbolic vector determining to find.

In one embodiment of this invention, continue carrying out depth-first search comprises with the step of seeking one or more new emission symbolic vectors: seek the new node that is positioned at the search bottom, with the path from the top-most node of search tree to this new node as new emission symbolic vector; And abandon not at the bottom of search tree and its routine weight value greater than the new node of described maximum path weights and the subtree of lower floor thereof.

In one embodiment of this invention, continue carrying out depth-first search also comprises with the step of seeking one or more new emission symbolic vectors: abandon not in the bottom of search tree and its routine weight value brotgher of node and the subtree thereof greater than the new node of described maximum path weights, directly return the last layer of search tree.

In one embodiment of this invention, replace the alternative emission symbolic vector that has the maximum path weights in described K the alternative emission symbolic vector with new emission symbolic vector, comprise that the routine weight value with described new emission symbolic vector upgrades described maximum path weights.

In one embodiment of this invention, whether determine whether that less than described maximum path weights upgrading the step that has the alternative emission symbolic vector of maximum path weights in described K the alternative emission symbolic vector with new emission symbolic vector also comprises afterwards at the routine weight value by definite new emission symbolic vector that finds: carry out soft bit and calculate.

In one embodiment of this invention, said method also comprises: according to the approximate weights of described search radius initialization one VB vacancy bit value; And replace with described new emission symbolic vector less than described maximum path weights at the routine weight value of the new emission symbolic vector that finds have the alternative emission symbolic vector of described maximum path weights before, give described VB vacancy bit value approximate weights described maximum path weights; In described soft bit calculates, give described VB vacancy bit value approximate weights to bit value for empty emission symbolic vector.

In one embodiment of this invention, when the routine weight value of the new emission symbolic vector that finds is not less than described maximum path weights, if described new emission symbolic vector is K+1 emission symbolic vector, then the routine weight value of described new emission symbolic vector is given described VB vacancy bit value approximate weights, otherwise have only when approximate weights of described VB vacancy bit value during greater than the routine weight value of described new emission symbolic vector, just give described VB vacancy bit value with the routine weight value of described new emission symbolic vector and be similar to weights.

In one embodiment of this invention, described depth-first search is the depth-first search of SE ordering.

In one embodiment of this invention, said method is to carry out in the multi-input multi-output-orthogonal frequency division multiplexing system.

The present invention proposes a kind of depth-first search globular decoding device in addition, comprising: the device that is used for the initialization search radius; Be used for that channel matrix is carried out QR and decompose, obtain the device of Q matrix and R matrix; Be used for carrying out depth-first search to obtain the device of K alternative emission symbolic vector, wherein K is that search listing size and K are positive integer; The device that is used for the maximum path weights of definite K alternative emission symbolic vector; Be used for to continue carry out depth-first search to seek the device of one or more new emission symbolic vectors; And be used for whether determining whether replacing the device that described K alternative emission symbolic vector has the alternative emission symbolic vector of maximum path weights with new emission symbolic vector less than described maximum path weights by the routine weight value of definite new emission symbolic vector that finds.

In one embodiment of this invention, described is to seek the new node that is positioned at the search bottom for continue carrying out depth-first search with the device of seeking one or more new emission symbolic vectors, path from the top-most node of search tree to this new node as new emission symbolic vector, and is abandoned not at the bottom of search tree and its routine weight value greater than the new node of described maximum path weights and the subtree of lower floor thereof.

In one embodiment of this invention, describedly also abandoning not in the bottom of search tree and its routine weight value brotgher of node and the subtree thereof greater than the new node of described maximum path weights with the device of seeking one or more new emission symbolic vectors for continue carrying out depth-first search, directly returning the last layer of search tree.

In one embodiment of this invention, said apparatus also comprises: be used for the device according to the approximate weights of described search radius initialization one VB vacancy bit value; Be used for routine weight value in the new emission symbolic vector that finds replace with described new emission symbolic vector less than described maximum path weights have the alternative emission symbolic vector of described maximum path weights before, described maximum path weights are given the device of the approximate weights of described VB vacancy bit value; And be used for calculating at soft bit, bit value is given the device of the approximate weights of described VB vacancy bit value for empty emission symbolic vector.

In one embodiment of this invention, be to be combined in the multi-input multi-output-orthogonal frequency division multiplexing system.

The present invention is owing to adopt above technical scheme, and the globular decoding method based on deep search of existing relatively constraint search listing size can obtain soft information more accurately with identical search listing size; Relatively originally, do not retrain the globular decoding method based on deep search of search listing size, can significantly reduce the number of access node, finished search as soon as possible.

Description of drawings

For above-mentioned purpose of the present invention, feature and advantage can be become apparent, below in conjunction with accompanying drawing the specific embodiment of the present invention is elaborated, wherein:

Fig. 1 is the search tree schematic diagram that illustrates when using depth-first search.

Fig. 2 illustrates the globular decoding method flow diagram of the depth-first search of one embodiment of the invention.

Fig. 3 illustrates the globular decoding method flow diagram of the depth-first search of another embodiment of the present invention.

Fig. 4 illustrates the examplar search tree in the method for one embodiment of the invention.

Embodiment

The following embodiment of the present invention will describe the signal detecting method in multi-input multi-output-orthogonal frequency division multiplexing (MIMO-OFDM) system.More particularly, this signal detecting method adopts the globular decoding method based on depth-first search.

Suppose that MIMO-OFDM system signal model is:

y＝Hs+n，

Wherein, y is N _RThe reception vector of * 1 dimension, H is N _R* N _TThe channel matrix of dimension, s is N _TThe emission symbolic vector of * 1 dimension, n is that variance is σ ²White Gaussian noise.Here N _R, N _TRepresent reception antenna number and number of transmit antennas respectively.The maximal possibility estimation algorithm is whole

Seek optimal solution in the space, as shown in the formula:

${\hat{s}}_{\mathrm{ML}}=\underset{s\∈C^{N_T}}{\arg \min }{\left|\right|y-\mathrm{Hs}\left|\right|}^2$

Globular decoding has been quoted the thought of Maximum Likelihood Detection, but in the globular decoding algorithm, proposed to reduce the method for search volume, its core concept is exactly to select a suitable initial radium as the radius of the globular decoding region of search, seek all possible emission symbolic vector at a center for receiving in the ball that vector y, radius are r, namely satisfy inequality:

||y-Hs|| ²≤r ² (1)

Owing to be in the constraint radius r, rather than whole

Seek the emission symbolic vector in the space, thereby reduced amount of calculation.Here C represents the symbolic number that modulation constellation points comprises,

The space of representative emission symbolic vector.

Fig. 1 is the search tree schematic diagram that illustrates when using depth-first search.Search tree is to be divided into N in Fig. 1 _TIndividual layer, each layer have one or more symbols as the node of tree.Top layer be root node root, from root node down, be the brotgher of node with the node of one deck, node and lower level node are called father node and child node.Find the solution the emission symbolic vector in fact from N _TLayer (globular decoding top layer) beginning, setting is found out several modulation signal points of satisfying boundary condition and then according to known N according to radius _TThe signal of layer, and known N _TLayer present node and N _T-1 layer boundary condition is obtained N _T-1 layer of current point that satisfies condition.Carry out recursion like this and go down, obtain the point of the 1st layer (bottom of globular decoding) at last.Form a hypersphere body by the candidate point of these layers, the point of hypersphere body wherein is exactly the grid point of the required acquisition of globular decoding.

Traditional tabulation globular decoding method based on deep search will satisfy condition according to the order of depth-first

All emission symbolic vectors all find out, therefore need whole search tree in the traversal initial radium scope, calculate soft bit according to this tabulation that searches then.Retrain the size of search listing simply, for example search node is restricted to a fixing upper limit and no doubt can reduces complexity, but accuracy can reduce.

Embodiments of the invention propose to improve to the tabulation globular decoding method based on deep search, need the access node number in the search tree in the hope of reducing under the condition of not appreciable impact accuracy, finish search as soon as possible.According to an embodiment, dwindle the big or small K of search listing, after finding the alternative emission symbolic vector of K, only visit is hopeful to upgrade the node of K alternative emission symbolic vector, and guarantees to find K alternative emission symbolic vector to be The K of routine weight value minimum is individual in individual.That is to say, as long as the routine weight value of the alternative emission symbolic vector that assurance is found all is

Individual middle minimum, just can avoid in soft bit calculates, increasing K.

Fig. 2 illustrates the globular decoding method flow diagram of the depth-first search of one embodiment of the invention.With reference to this method of description shown in Figure 2.

In step 201, the initialization search radius is r.

If search radius is excessive usually, will comprise too much point so in the ball, make the calculating degree approach or reach the computation complexity of maximum likelihood algorithm.If search radius is too little, may in the field of search, there be the satisfactory point that will search for so, cause detecting failure.In an embodiment of the present invention, can use any known method to determine suitable initial search radius.

In step 202, channel matrix H (requiring the H alignment to have nothing to do) is carried out QR decomposing H=QR, obtain unitary matrice Q and upper triangular matrix R.

For example, carry out after QR decomposes, (1) formula can be equivalent to:

||Q ^Hy-Rs|| ²≤r ² (2)

Because R is that element is the particularity of arithmetic number on upper triangular matrix and the diagonal, (2) formula further abbreviation is:

${\left|\right|Q^{H}y-\mathrm{Rs}\left|\right|}^2={\left|\right|R\hat{s}-\mathrm{Rs}\left|\right|}^2\≤r^2$

$\⇒\underset{k=1}{\overset{N_T}{\mathrm{\Σ}}}{\left|R_{k,k}(s_k-({\hat{s}}_k-\underset{j=k+1}{\overset{N_T}{\mathrm{\Σ}}}(s_j-{\hat{s}}_j)R_{k,j}/R_{k,k}))\right|}^2\≤r^2---\left(3\right)$

$\⇔\underset{k=1}{\overset{N_T}{\mathrm{\Σ}}}{\left|R_{k,k}(s_k-{\mathrm{\ρ}}_k)\right|}^2\≤r^2$

Wherein, R _{K, j}Be the capable j column element of k of upper triangular matrix R, k=1,2 ..., N _T, represent the number of plies of transmitting antenna.

Step 203 adopts the depth-first search of Schnorr Euchner (SE) ordering, to find a plurality of alternative emission symbolic vectors.

Still be example with Fig. 1, describe for convenient, remember that the k layer is to N _TThe path of layer is

, this path has comprised from the k layer to N _TLayer all nodes or the symbol of process.

Routine weight value be:

${\mathrm{PED}}_k^2=\underset{i=k}{\overset{N_T}{\mathrm{\Σ}}}{R}_{i,i}^2{|s_i-{\mathrm{\ρ}}_i|}^2={\mathrm{PED}}_{k+1}^2+R_{k,k}^2{|s_k-{\mathrm{\ρ}}_k|}^2={\mathrm{PED}}_{k+1}^2+d_k^2.$

Based on (3) formula, can be from N _TIndividual element begins, and utilizes

Can obtain easily

(the alternative symbol that is marked as black in the corresponding diagram 1).The SE ordering is layer ordering, namely to all child nodes that search under the same father node

Sort from small to large by its corresponding routine weight value, the node that the priority access routine weight value is little is (as the child node priority access node under the root node among Fig. 1

Describe for convenient, be designated as

); In like manner, at N _TThe element of layer

Under the fixed situation, N _T-1 element Satisfy

Thereby find N _T-1 layer alternative symbol; Recurrence is to 1 layer of element s of reprocessing to the according to this ₁, can obtain desired satisfying condition

Alternative emission symbolic vector.Alternative emission symbolic vector is from top layer (N _TLayer) a certain alternative sign-on to another alternative symbol of the bottom (the 1st layer) the path of process.

When adopting list search, need the search listing size of determining that suitable needs keep.Size at this hypothesis search listing is K (K is positive integer), and namely the target number of alternative emission symbolic vector is K.That is to say, the total K bar alternative path from top-most node to bottom node, thus can be with number of paths as the foundation of judging whether search is finished.The corresponding weights of each alternative emission symbolic vector.Weights are weighed the distance from top-most node to bottom node.。

In step 204, if find the path (for example K paths) of destination number then enter step 205, continue deep search otherwise return step 203 through step 212.If do not find K paths (for example the path that under the current search radius, exists discontented K bar) but judge that in step 212 search finishes, and then enters step 213.

In step 205, determine the maximum path weights in the K paths.

For example the K paths is sorted according to the weights size, thereby find maximum weights.Maximum weights in this note K paths are

The position is Index _Max

At this, the K that the individual alternative emission symbolic vector of the K that finds may not be routine weight value minimum in the search radius is individual, thereby this K alternative emission symbolic vector may not be desired final result.

Therefore the flow process of present embodiment will be proceeded depth-first search, seek new emission symbolic vector.In the searching process, flow process can be sought the new node (it satisfies the condition of alternative symbol) that is positioned at the search tree bottom (the 1st layer), and with the path from the top-most node of search tree to the new node of the bottom as new emission symbolic vector, and abandon not at the 1st layer of search tree and its routine weight value greater than the new node of maximum path weights and the subtree of lower floor thereof.

Specifically, flow process begins to continue to seek new node search tree from step 206, when not finding new node, returns step 203.If find new node then further to judge new node whether the 1st layer (being the bottom) of search tree in step 207, if then the path from the top-most node of search tree to this new node is new emission symbolic vector.

If new node at the 1st layer of search tree, does not then continue deep search, up to the routine weight value of finding new node in step 210 do not satisfied less than

Condition, this moment return last layer in step 211, this means the subtree that abandons this node and lower floor thereof; At this, because the brotgher of node under the same father node all adopts SE ordering, then also discardable its all brotghers of node and subtree thereof, continuation is carried out the search of other nodes of last layer in step 203.

After finding new emission symbolic vector, (or replace) K alternative emission symbolic vector having found that flow process will judge further whether new emission symbolic vector may be upgraded.The standard of upgrading is that the routine weight value of new emission symbolic vector is less than the maximum path weights in K the alternative emission symbolic vector.

Specifically, in step 208, judge that whether the routine weight value of new node is less than current maximum path weights

If then replace K routine weight value the maximum in the alternative emission symbolic vector, substitute I ndex with new emission symbolic vector in step 209 _MaxAlternative emission symbolic vector and the weights thereof of position.That is to say that the weights of new emission symbolic vector will be replaced the maximum weights of former K alternative emission symbolic vector

Maximum weights in the individual alternative emission symbolic vector of K this moment To dwindle, this has optimized the result of deep search.Main purpose that it should be noted that step 209 is to replace K weights the maximum in the alternative emission symbolic vector with new emission symbolic vector, and the weights of new emission symbolic vector may not be real maximum weights.Although determine real maximum weights by extra ordering herein, because in the existing step of determining maximum weights of the step 205 of flow process, thereby need not determine real maximum weights herein.

In step 208, if the new node weights do not satisfy less than

Condition, then return step 203 and continue search.

After judging that in step 212 search finishes, enter step 213, carry out soft bit and calculate.

For example hypothesis emission symbol etc. is general, then has:

$\mathrm{LLR}\left(b_{l,n}\right)=\ln \frac{p(b_{l,n}=+1|y)}{p(b_{l,n}=-1|y)}=\ln \frac{\underset{s\∈S_{l,n,+1}}{\mathrm{\Σ}}p\left(y\right|s)}{\underset{s\∈S_{l,n,-1}}{\mathrm{\Σ}}p\left(y\right|s)}=\ln \frac{\underset{s\∈S_{l,n+1}}{\mathrm{\Σ}}{e}^{-\frac{{\left|\right|y-\mathrm{Hs}\left|\right|}^2}{2{\mathrm{\σ}}^2}}}{\underset{s\∈S_{l,n,-1}}{\mathrm{\Σ}}{e}^{-\frac{{\left|\right|y-\mathrm{Hs}\left|\right|}^2}{2{\mathrm{\σ}}^2}}}$

Use the approximate further abbreviation of MAX-LOG to be:

$\mathrm{LLR}\left(b_{l,n}\right)=\frac{1}{2{\mathrm{\σ}}^2}[-\underset{s\∈S_{l,n,+1}}{\min }{\left|\right|R\hat{s}-\mathrm{Rs}\left|\right|}^2+\underset{s\∈S_{l,n,-1}}{\min }{\left|\right|R\hat{s}-\mathrm{Rs}\left|\right|}^2]$

S _{L, n ,+1}={ s|b _{L, n}=+1} is set

In satisfy b _{L, n}=+1 and

The set of s, S _{L, n ,-1}={ s|b _{L, n}=-1} is set

In satisfy b _{L, n}=-1 and

The set of s.

Usually if search radius r is enough big, then have

If r is too little, certain bit b then can appear _{L, n}Value is at S _{L, n ,-1}∪ S _{L, n ,+1}In be empty.

In the present embodiment, can use any existent method to deal with bit value and be the situation of sky.For example bit value is chosen bigger search radius for empty solution comprises and is similar to.Approximate mode comprises, weights are composed into maximum, and perhaps weights or the current search radius approximate evaluation bit value according to emission symbolic vector in the tabulation is empty weights.

As described above, when bit value is sky, choose bigger search radius and can cause higher complexity.Therefore use to be similar to and to avoid higher complexity, but weights are composed into maximum or relatively poor according to the approximate accuracy of current search radius, improved approximate mode will be described in the following embodiments, wherein introduce the approximate weights of a VB vacancy bit value, be designated as PED_kong, and in search procedure, make the approximate weights of VB vacancy bit value constantly approach desired value.

Fig. 3 illustrates the globular decoding method flow diagram of the depth-first search of another embodiment of the present invention.Referring now to this method of description shown in Figure 3.

In step 301, the initialization search radius is r.Initialization PED_kong is r at the same time ²

If search radius is excessive usually, will comprise too much point so in the ball, make the calculating degree approach or reach the computation complexity of maximum likelihood algorithm.If search radius is too little, may in the field of search, there be the satisfactory point that will search for so, cause detecting failure.In an embodiment of the present invention, can use any known method to determine suitable initial search radius.

In step 302, channel matrix H (requiring the H alignment to have nothing to do) is carried out QR decomposing H=QR, obtain unitary matrice Q and upper triangular matrix R.

For example, carry out after QR decomposes, (1) formula can be equivalent to:

||Q ^Hy-Rs|| ²≤r ² (2)

Because R is that element is the particularity of arithmetic number on upper triangular matrix and the diagonal, (2) formula further abbreviation is:

${\left|\right|Q^{H}y-\mathrm{Rs}\left|\right|}^2={\left|\right|R\hat{s}-\mathrm{Rs}\left|\right|}^2\≤r^2$

$\⇒\underset{k=1}{\overset{N_T}{\mathrm{\Σ}}}{\left|R_{k,k}(s_k-({\hat{s}}_k-\underset{j=k+1}{\overset{N_T}{\mathrm{\Σ}}}(s_j-{\hat{s}}_j)R_{k,j}/R_{k,k}))\right|}^2\≤r^2---\left(3\right)$

$\⇔\underset{k=1}{\overset{N_T}{\mathrm{\Σ}}}{\left|R_{k,k}(s_k-{\mathrm{\ρ}}_k)\right|}^2\≤r^2$

Wherein, R _{K, j}Be the capable j column element of k of upper triangular matrix R, k=1,2 ..., N _T, represent the number of plies of transmitting antenna.

Step 303 adopts the depth-first search of Schnorr Euchner (SE) ordering, to find a plurality of alternative emission symbolic vectors.

Still be example with Fig. 1, describe for convenient, remember that the k layer is to N _TThe path of layer is

Routine weight value be:

${\mathrm{PED}}_k^2=\underset{i=k}{\overset{N_T}{\mathrm{\Σ}}}{R}_{i,i}^2{|s_i-{\mathrm{\ρ}}_i|}^2={\mathrm{PED}}_{k+1}^2+R_{k,k}^2{|s_k-{\mathrm{\ρ}}_k|}^2={\mathrm{PED}}_{k+1}^2+d_k^2.$

Based on (3) formula, can be from N _TIndividual element begins, and utilizes Can obtain easily

(the alternative symbol that is marked as black in the corresponding diagram 1).The SE ordering is layer ordering, namely to all child nodes that search under the same father node

Sort from small to large by its corresponding routine weight value, the node that the priority access routine weight value is little is (as the child node priority access node under the root node among Fig. 1

Describe for convenient, be designated as

); In like manner, at N _TThe element of layer

Under the fixed situation, N _T-1 element

Satisfy ${|s_{N_T-1}-{\mathrm{\ρ}}_{N_T-1}|}^2\≤\frac{r^2-R_{N_T,N_T}^2{|s_{N_T}-{\mathrm{\ρ}}_{N_T}|}^2}{R_{N_T-1,N_T-1}^2},$ Thereby find N _T-1 layer alternative symbol; Recurrence is to 1 layer of element s of reprocessing to the according to this ₁, can obtain desired satisfying condition Alternative emission symbolic vector.Alternative emission symbolic vector is from top layer (N _TLayer) a certain alternative sign-on to another alternative symbol of the bottom (the 1st layer) the path of process.

When adopting list search, need the search listing size of determining that suitable needs keep.Size at this hypothesis search listing is K (K is positive integer), and namely the target number of alternative emission symbolic vector is K.That is to say, the total K bar alternative path from top-most node to bottom node, thus can be with number of paths as the foundation of judging whether search is finished.The corresponding weights of each alternative emission symbolic vector.Weights are weighed the distance from top-most node to bottom node.。

In step 304, if find the path (for example K paths) of destination number then enter step 305, continue deep search otherwise return step 303 through step 316.If do not find K paths (for example the path that under the current search radius, exists discontented K bar) but judge that in step 316 search finishes, and then enters step 317.

In step 305, determine the maximum path weights in the K paths.

For example the K paths is sorted according to the weights size, thereby find maximum routine weight value.Maximum path weights in this note K paths are The position is Index _Max

At this, the K that the individual alternative emission symbolic vector of the K that finds may not be weights minimum in the search radius is individual, thereby this K alternative emission symbolic vector may not be desired final result.

Therefore the flow process of present embodiment will be proceeded depth-first search, seek new emission symbolic vector.In the searching process, flow process can be sought the new node (it satisfies the condition of alternative symbol) that is positioned at the search tree bottom (the 1st layer), and with the path from the top-most node of search tree to the new node of the bottom as new emission symbolic vector, and abandon not at the 1st layer of search tree and its routine weight value greater than the new node of maximum path weights and the subtree of lower floor thereof.Specifically, flow process begins to continue to seek new node search tree from step 306, when not finding new node, returns step 303.If find new node then further to judge new node whether the 1st layer (being the bottom) of search tree in step 307, if then the path from the top-most node of search tree to this new node is new emission symbolic vector.

If new node at the 1st layer of search tree, does not then continue deep search, up to find in step 311 the new node routine weight value do not satisfied less than

Condition, this moment return last layer in step 312, this means the subtree that abandons this node and lower floor thereof; At this, because the brotgher of node under the same father node all adopts SE ordering, then also discardable its all brotghers of node and subtree thereof, continuation is carried out the search of other nodes of last layer in step 303.

After finding new emission symbolic vector, (or replace) K alternative emission symbolic vector having found that flow process will judge further whether new emission symbolic vector may be upgraded.The standard of upgrading is that the weights of new emission symbolic vector are less than the maximum weights in K the alternative emission symbolic vector.

Specifically, in step 308, judge that whether the new node weights are less than current maximum weights

If then give new value at step 309 couple PED_kong

Replace K weights the maximum in the alternative emission symbolic vector, substitute I ndex with new emission symbolic vector in step 310 then _MaxAlternative vector and the weights thereof of position.That is to say that the weights of new emission symbolic vector will be replaced the maximum weights of former K alternative emission symbolic vector

Maximum weights in the individual alternative emission symbolic vector of K this moment

To dwindle, this has optimized the result of deep search.Main purpose that it should be noted that step 310 is replacing the weights the maximum in K the alternative emission symbolic vector with new emission symbolic vector, and the weights of new emission symbolic vector may not be real maximum weights.Although determine real maximum weights by extra ordering herein, because in the existing step of determining maximum weights of the step 305 of flow process, thereby need not determine real maximum weights herein.

In step 308, if the new node weights do not satisfy less than

Condition, namely this moment new node routine weight value Then make the following judgment:

If article one new route after step 313 is determined to find the K paths, namely the K+1 paths is then given PED_kong in step 314 with the weights of new node, namely Otherwise only exist

Situation under just upgrade PED_kong (step 315).

After judging that in step 316 search finishes, enter step 317, carry out soft bit and calculate.

For example hypothesis emission symbol etc. is general, then has:

$\mathrm{LLR}\left(b_{l,n}\right)=\ln \frac{p(b_{l,n}=+1|y)}{p(b_{l,n}=-1|y)}=\ln \frac{\underset{s\∈S_{l,n,+1}}{\mathrm{\Σ}}p\left(y\right|s)}{\underset{s\∈S_{l,n,-1}}{\mathrm{\Σ}}p\left(y\right|s)}=\ln \frac{\underset{s\∈S_{l,n+1}}{\mathrm{\Σ}}{e}^{-\frac{{\left|\right|y-\mathrm{Hs}\left|\right|}^2}{2{\mathrm{\σ}}^2}}}{\underset{s\∈S_{l,n,-1}}{\mathrm{\Σ}}{e}^{-\frac{{\left|\right|y-\mathrm{Hs}\left|\right|}^2}{2{\mathrm{\σ}}^2}}}$

Use the approximate further abbreviation of MAX-LOG to be:

$\mathrm{LLR}\left(b_{l,n}\right)=\frac{1}{2{\mathrm{\σ}}^2}[-\underset{s\∈S_{l,n,+1}}{\min }{\left|\right|R\hat{s}-\mathrm{Rs}\left|\right|}^2+\underset{s\∈S_{l,n,-1}}{\min }{\left|\right|R\hat{s}-\mathrm{Rs}\left|\right|}^2]$

S _{L, n ,+1}={ s|b _{L, n}=+1} is set

In satisfy b _{L, n}=+1 and

The set of s, S _{L, n ,-1}={ s|b _{L, n}=-1} is set

In satisfy b _{L, n}=-1 and

The set of s.

Usually if search radius r is enough big, then have

If r is too little, certain bit b then can appear _{L, n}Value is at S _{L, n ,-1}∪ S _{L, n ,+1}In be empty.Can use this moment the approximate VB vacancy bit value of weights PED_kong of VB vacancy bit to be similar to.

In bit value calculated, if the alternative emission symbolic vector number＜K that searches, then Ci Shi PED_kong still was initial search radius; If the alternative emission symbolic vector number 〉=K that searches then uses K+1 individual less

Approximate.

In the embodiment of Fig. 2 and Fig. 3, in search tree, seek the littler new node of weights that can upgrade alternative emission symbolic vector.In the embodiment of replaceability, after step 204 and 304 finds K alternative emission symbolic vector, can also visit several child nodes at most (as only visiting weights by each father node of control

Minimum that) come the quantity of limiting access node.Even the worst situation takes place like this, namely

Minimum alternative emission symbol is last tree, does not also need to travel through whole tree.

In addition, if wish to obtain soft information more accurately, can increase memory space seldom in an alternative embodiment, the leaf node beyond K the minimal path weights that search is also preserved, this moment, PED_kong was approximate with about beam radius.

Globular decoding method of the present invention has the advantage of complexity or accuracy with respect to existing the whole bag of tricks.With shown in Figure 4 at initial radium r (r ²〉=1.5) search tree is example in the scope, and existing tabulation globular decoding based on the degree of depth can travel through whole search tree, with v ₁₀～v ₁₆All search for out, complexity is very high undoubtedly.If according to the size of existing method constraint search listing, as get K=3, then search node v ₁₂In time, can finish to search for, but obvious v ₁₀, v ₁₁, v ₁₂Three of the weights minimum nodes in the not all bottom layer node, and this moment PED_kong=r ²〉=1.5 to be used as the bit value also not accurate enough for empty weights.

If according to embodiments of the invention, only need access node v ₀, v ₁, v ₄, v ₅, v ₂, v ₆, reduced complexity.The existing relatively conventional method that does not retrain search listing has reduced search complexity and memory space widely, and is little to performance impact simultaneously; Though and the conventional method of existing relatively constraint search listing has been visited node v more ₂, v ₆, but can guarantee K node v keeping ₁₀, v ₁₂, v ₁₃Routine weight value be minimum in all 7 leaf nodes, and PED_kong=1.3, soft information is more accurate.

Based on above-mentioned, embodiment proposed by the invention has following advantage with respect to the globular decoding method of existing depth-first search:

1, the globular decoding method based on deep search of existing relatively constraint search listing size, embodiments of the invention can be obtained soft information more accurately with identical search listing size; Relatively originally, do not retrain the globular decoding method based on deep search of search listing size, can significantly reduce the number of access node, finished search as soon as possible;

It is with maximum or the approximate scheme of about beam radius this moment, in the time finding more than K alternative emission symbolic vector, individual less with K+1 at embodiments of the invention when 2, existing relatively bit value is empty

Value is approximate, can make the soft information of partial bit more accurate;

3, single relatively tree search, embodiments of the invention can earlier finish search and little to performance impact; Embodiments of the invention can increase memory space seldom simultaneously, and the leaf node beyond K the minimal path weights that search is also preserved, and can further approach single tree search performance of exhaustive search.

Though the present invention discloses as above with preferred embodiment; right its is not in order to limiting the present invention, any those skilled in the art, without departing from the spirit and scope of the present invention; when can doing a little modification and perfect, so protection scope of the present invention is when with being as the criterion that claims were defined.

Claims (14)

1. depth-first search globular decoding method comprises:

The initialization search radius;

Channel matrix is carried out QR decompose, obtain Q matrix and R matrix;

Carry out depth-first search to obtain K alternative emission symbolic vector, wherein K is that search listing size and K are positive integer;

Determine K the maximum path weights in the alternative emission symbolic vector;

Continue to carry out depth-first search to seek one or more new emission symbolic vectors; And

Whether the routine weight value by definite new emission symbolic vector that finds determines whether replacing the alternative emission symbolic vector that has the maximum path weights in described K the alternative emission symbolic vector with new emission symbolic vector less than described maximum path weights.

2. the method for claim 1 is characterized in that, continue to carry out depth-first search and comprises with the step of seeking one or more new emission symbolic vectors:

Searching is positioned at the new node of the search bottom, with the path from the top-most node of search tree to this new node as new emission symbolic vector;

Abandon not at the bottom of search tree and its routine weight value greater than the new node of described maximum path weights and the subtree of lower floor thereof.

3. method as claimed in claim 2 is characterized in that, continue to carry out depth-first search and also comprises with the step of seeking one or more new emission symbolic vectors:

Abandon not in the bottom of search tree and its routine weight value brotgher of node and the subtree thereof greater than the new node of described maximum path weights, directly return the last layer of search tree.

4. the method for claim 1, it is characterized in that, replace the alternative emission symbolic vector that has the maximum path weights in described K the alternative emission symbolic vector with new emission symbolic vector, comprise that the routine weight value with described new emission symbolic vector upgrades described maximum path weights.

5. the method for claim 1, it is characterized in that whether determine whether that less than described maximum path weights upgrading the step that has the alternative emission symbolic vector of maximum path weights in described K the alternative emission symbolic vector with new emission symbolic vector also comprises afterwards at the routine weight value by definite new emission symbolic vector that finds: carry out soft bit and calculate.

6. method as claimed in claim 5 is characterized in that, also comprises:

According to the approximate weights of described search radius initialization one VB vacancy bit value;

And replace with described new emission symbolic vector less than described maximum path weights at the routine weight value of the new emission symbolic vector that finds have the alternative emission symbolic vector of described maximum path weights before, give described VB vacancy bit value approximate weights described maximum path weights;

In described soft bit calculates, give described VB vacancy bit value approximate weights to bit value for empty emission symbolic vector.

7. method as claimed in claim 6, it is characterized in that, when the routine weight value of the new emission symbolic vector that finds is not less than described maximum path weights, if described new emission symbolic vector is K+1 emission symbolic vector, then the routine weight value of described new emission symbolic vector is given described VB vacancy bit value approximate weights, otherwise have only when approximate weights of described VB vacancy bit value during greater than the routine weight value of described new emission symbolic vector, just give described VB vacancy bit value with the routine weight value of described new emission symbolic vector and be similar to weights.

8. the method for claim 1 is characterized in that, described depth-first search is the depth-first search of SE ordering.

9. the method for claim 1 is characterized in that, is to carry out in the multi-input multi-output-orthogonal frequency division multiplexing system.

10. depth-first search globular decoding device comprises:

The device that is used for the initialization search radius;

Be used for that channel matrix is carried out QR and decompose, obtain the device of Q matrix and R matrix;

Be used for carrying out depth-first search to obtain the device of K alternative emission symbolic vector, wherein K is that search listing size and K are positive integer;

The device that is used for the maximum path weights of definite K alternative emission symbolic vector;

Be used for to continue carry out depth-first search to seek the device of one or more new emission symbolic vectors; And

Be used for whether determining whether replacing the device that described K alternative emission symbolic vector has the alternative emission symbolic vector of maximum path weights with new emission symbolic vector less than described maximum path weights by the routine weight value of definite new emission symbolic vector that finds.

11. device as claimed in claim 10, it is characterized in that, described is to seek the new node that is positioned at the search bottom for continue carrying out depth-first search with the device of seeking one or more new emission symbolic vectors, path from the top-most node of search tree to this new node as new emission symbolic vector, and is abandoned not at the bottom of search tree and its routine weight value greater than the new node of described maximum path weights and the subtree of lower floor thereof.

12. device as claimed in claim 10, it is characterized in that, describedly also abandoning not in the bottom of search tree and its routine weight value brotgher of node and the subtree thereof greater than the new node of described maximum path weights with the device of seeking one or more new emission symbolic vectors for continue carrying out depth-first search, directly returning the last layer of search tree.

13. device as claimed in claim 10 is characterized in that, also comprises:

Be used for the device according to the approximate weights of described search radius initialization one VB vacancy bit value;

Be used for routine weight value in the new emission symbolic vector that finds replace with described new emission symbolic vector less than described maximum path weights have the alternative emission symbolic vector of described maximum path weights before, described maximum path weights are given the device of the approximate weights of described VB vacancy bit value;

Be used for calculating at soft bit, bit value given the device of the approximate weights of described VB vacancy bit value for empty emission symbolic vector.

14. device as claimed in claim 10 is characterized in that, is to be combined in the multi-input multi-output-orthogonal frequency division multiplexing system.

Images