CN104202127A - Path metric value-based low complexity MIMO (multiple input multiple output) system sphere decoding signal detection method - Google Patents

Abstract

The invention provides a path metric value-based low complexity MIMO (multiple input multiple output) system sphere decoding signal detection method. The path metric value-based low complexity MIMO system sphere decoding signal detection method comprises the following steps: according to reversed order detection procedures of the sphere decoding, after the testing results from (i plus 1) layers to the m layer on a code tree are known, when the signal detection is conducted on the i layer, the traditional Euclidean distance square sum value is adopted to predict the symbolic value of the i layer; undetected code tree nodes from the first subsequent layer to the (i minus 1) layer are further utilized to calculate the path correction value and used for predicting the symbolic value of the i layer; then, the mutual combination of the Euclidean distance square sum value and the path correction value provided by the invention is utilized, so as to enable the predicated value of the i layer code tree node to be more accurate than two traditional detection algorithms of P o h s t and SE for the sphere decoding; during the detection process, as the search for path is accurate, the frequency of backspacing is reduced; the target path can be quickly and accurately found; the complexity in the signal detection is effectively lowered and the decoding time delay is obviously reduced; if the verification is carried out by combining a list and CRC, the performance of the path metric value-based low complexity MIMO system sphere decoding signal detection method is capable of exceeding the maximum likelihood detection. However, the complex rate of the path metric value-based low complexity MIMO system sphere decoding signal detection method is basically in line with the maximum likelihood detection method.

Description

Low-complexity MIMO system ball decoded signal detection method based on path metric value

Technical field

The present invention relates to a kind of low complex degree, high performance ball decoded signal detection method, exactly, relate to a kind of low-complexity MIMO system ball decoded signal detection method based on path metric value, for solving the decoding test problems after digital communication signal transmits by mimo system.Belong to the signal detection technique field of multi-antenna digital communication system.

Background technology

Ball decoding (Sphere Decoding) is a kind of signal detecting method for mimo system that Viterbo in 1999 and Bouros propose.Ball decoding detection algorithm principle is based on lattice multidimensional modulation theory, and its transmitted signal is equivalent to a lattice-site in multidimensional crystal lattice pattern, and after additive noise, received signal points can depart from original signal point.Under the condition that there is no prior information, adopting maximum-likelihood criterion decoding to detect is the optimum interpretation method that can realize.In ball decoding testing process, when receiving after signaling point, detect in order to reach maximum-likelihood decoding, in principle should traversal search crystal lattice pattern on all useful signal points, find the useful signal point nearest with received signal points Euclidean distance.But the complexity of this traversal search mode is exponential, after antenna amount and the rising of signal order of modulation, its complexity can increase sharply, and this algorithm does not have practical value in reality.The practical MIMO communication system ball interpretation method that Damen in 2003 etc. propose, has solved the problem of above-mentioned traversal search, and has had relatively low complexity.

The thinking of ball decoding detection method is: an initial radium is set before decoding starts, again taking received signal points as the centre of sphere, draw a spheroid taking the initial radium arranging as the radius of a ball, then on crystal lattice pattern, search out a useful signal point, if this signaling point is positioned at spheroid, the Euclidean distance taking this signaling point to received signal points, as radius and received signal points are the centre of sphere, draws a new spheroid.If this signaling point is not in spheroid, search signal point again, repeat according to this step, finally, the radius of decoding ball can constantly dwindle, until ball inside is not while thering is no other signaling points, the signaling point on sphere is now delivered to signal detector as testing result, reach thus Maximum Likelihood Detection.

Particularly, ball decode procedure is mainly used in solving every signal receiving antenna in mimo system all can receive the data that multiple transmitting antennas send simultaneously, causes original transmission data to be aliasing in together and the problem that cannot separate.In mimo system complex signal model, number of transmit antennas is M, and reception antenna quantity is N, and original transmitted signal is vectorial λ=(λ ₁, λ ₂..., λ _m) ^t, receiving signal is vectorial π=(π ₁, π ₂..., π _n) ^t.In formula, the each element in vectorial λ is respectively the transmission symbol on every transmitting antenna, and the each element in vectorial π is respectively the symbol receiving on every reception antenna, () ^tthe matrix transpose operation of representing matrix or vector.According to mimo system model, there is π=B λ+v, wherein, vectorial B is the channel response matrix of the capable M row of N, the multiple gaussian additive noise that v is receiving terminal.

In ball decoding testing process, first needing above-mentioned complex signal model conversion is real signal model of equal value, and reception signal is now vector $y=\left[\begin{array}{c}\mathrm{Re}\left(\mathrm{\π}\right)\\ \mathrm{Im}\left(\mathrm{\π}\right)\end{array}\right]={(y_1,y_2,...,y_n)}^T,$ And transmitted signal is vector $x=\left[\begin{array}{c}\mathrm{Re}\left(\mathrm{\λ}\right)\\ \mathrm{Im}\left(\mathrm{\λ}\right)\end{array}\right]={(x_1,x_2,...,x_m)}^T,$ Wherein, n=2N, m=2M.Additive noise $z=\left[\begin{array}{c}\mathrm{Re}\left(v\right)\\ \mathrm{Im}\left(v\right)\end{array}\right],$ Channel response matrix B in mimo system model is transformed to set $H=\left[\begin{array}{cc}\mathrm{Re}\left(B\right)& -\mathrm{Im}\left(B\right)\\ \mathrm{Im}\left(B\right)& \mathrm{Re}\left(B\right)\end{array}\right],$ Wherein, Re () and Im () represent respectively to get real part and imaginary part.So just original complex field signal model is converted to real signal model y=Hx+z of equal value.The ball decoding detection method of discussion of the present invention later all carries out under real signal model.

Again the matrix H in real signal model is carried out to QR decomposition, obtains: $H=\left[\begin{array}{cc}Q& U\end{array}\right]\left[\begin{array}{c}R\\ 0\end{array}\right],$ Wherein, R is the upper triangular matrix of the capable m row of m.Also by unitary matrice Q ^h(being the transposed matrix of Q) premultiplication above formula, can obtain y=Rx+z, and wherein, x and y are respectively transmission signal vector and received signal vector, the additive noise vector that z is receiving terminal.Above formula y=Rx+z can expand into following equation:

By after the matrix in this equation and vector operation expansion, obtain expression formula again: in formula, the interference of (a-1) individual symbol before a symbol has been subject to, and m symbol is only subject to the impact of noise.

In the time of the MIMO input of decomposing based on QR, in x, each symbol is to carry out backward in the mode of counteracting serial interference according to sequence number to detect, and from m sign-on, is then m-1, m-2 ..., until last the 1st.The testing process of ball decoded signal is equivalent at one taking m symbol as root node, in the code tree of leaf node, searches out a paths taking the 1st symbol.

Referring to Fig. 1, in maximum-likelihood decoding testing process, require the decode results and the Euclidean distance minimum that receives signal of the path representative that in code tree, search obtains.Damen has proposed two kinds of practical ball decoded signal detection methods, be referred to as Pohst algorithm and SE algorithm, but the two respectively has its pros and cons.Below the basic operational steps of these two kinds of algorithms is briefly introduced:

(1) Pohst algorithm: in Pohst ball decoding testing process, after the decoding of i layer finishes, the Euclidean distance of a record square accumulated value is: wherein, y _abe the data that a root reception antenna receives, i.e. a the element of vectorial y in real signal model, r _a,bfor the capable b column element of a in matrix R, it is the symbol sebolic addressing having detected in b element.When in forecourt decoding testing process, the decoding radius of a ball is C, if T _iexceed C ², can not continue to detect forward decoding.Detect rule according to ball decoding backward, after the detection of i+1 layer finishes, while detecting i layer, can utilize T _i+1to the x of i layer _ivalue is made prediction, and this prediction is according to T _i+1calculate and make i layer testing result be no more than C ²'s span X _i, this X _iinterval be [A _i, B _i].

Under complex signal model, it is q that transmitted signal adopts order of modulation ²two-dimensional modulation mode, the one-dimensional modulation that is q for order of modulation under corresponding real signal model, the value set of useful signal point is S, works as X _iwith the common factor of S be empty set, when (wherein ∩ represents two intersection of sets collection), just do not need to carry out i layer and detect.Now directly return to front one deck at X _i+1in ∩ S, choose another according to order from small to large before replacing value; If at X _i+1in ∩ S, not new value is desirable, continues rollback.If at X _i+1in ∩ S, successfully obtain new utilize upgrade T _i+1, again calculate span X _i.And continue sequentially to carry out according to this process.If successfully found find an active path, namely found a useful signal point, utilized T ₁upgrade the decoding radius of a ball then at X ₁in ∩ S, choose new continue to carry out corresponding operating according to above-mentioned steps, until m layer while also needing rollback, detects and finishes, now searched maximum likelihood path.

Pohst algorithm has been avoided the defect of traversal search signaling point, and its computation complexity is obviously to have reduced with respect to traversal search.But, when this algorithm is searched at every one deck of code tree, be all from forecast interval [A _i, B _i] lower bound A _itoward upper bound B _ivalue successively, such search is with certain blindness.Therefore on its computation complexity, also has obvious redundancy.But Pohst algorithm can ensure to realize Maximum Likelihood Detection.

(2) SE algorithm: its basic ideas are consistent with Pohst algorithm, just at every turn at X _iwhen interior value, all from [A _i, B _i] mid point start to choose, if value not within the scope of S, scatter gradually to this interval the right and left from interval mid point, until while falling in S set, record now calculate again T _i.If T _i>C ², direct rollback, and no longer need to have searched for X _iother useful signal points in ∩ S, until m layer is while also needing rollback, just detection of end.

Owing to not searched X at every turn _iother useful signal points in ∩ S, therefore SE algorithm there will be the phenomenon of undetected signaling point, finally do not reach Maximum Likelihood Detection thereby cause, and its performance is obviously poor than Pohst algorithm.But it directly starts search from mid point, has not searched X in decode procedure at every turn _iother useful signal points in ∩ S, this makes SE algorithm complex is obviously to have reduced with respect to Pohst algorithm.

In a word, the existing detection technique shortcoming of above-mentioned two kinds of practical MIMO balls decoding is: decoding testing process is just utilized T _ipredict the value of next layer data, cause (i-1) layer decoding value can only travel through and take out successively from small to large, or take out in the mode of reducing by half; These two kinds of methods are not optimum detection methods.The traversal value mode of Pohst algorithm representative can ensure Maximum Likelihood Detection, but because value is with blindness, detection complexity obviously improves.The mode value by half of SE algorithm has had certain improvement, but its algorithm structure makes not searched X _iother useful signal points in ∩ S.This just makes after detection computations reduced complexity, and the obvious variation of its performance, cannot reach Maximum Likelihood Detection.

Sum up above-mentioned two kinds of algorithms, its inferior position is mainly reflected in for each layer signal point value there is no to predict its value order in the set of useful signal point most scientificly.

Summary of the invention

In view of this, the object of this invention is to provide a kind of low-complexity MIMO system ball decoded signal detection method based on path metric value, the inventive method has very high practical value for the input of mimo system: its error-correcting performance can reach maximum-likelihood decoding, and the computation complexity that decoding detects obviously reduces than the maximum likelihood ball decoding algorithm of conventional art.Because the method combines the advantage of Pohst and two kinds of traditional decoding detection algorithms of SE, if again in conjunction with list and two operating procedures of CRC check, the performance of the inventive method can exceed Maximum Likelihood Detection, but its complexity is basic and maximum likelihood detection method maintains an equal level.

In order to achieve the above object, the invention provides a kind of low-complexity MIMO system ball decoded signal detection method based on path metric value, it is characterized in that: the flow process detecting according to ball decoding backward, in known code tree, (i+1) layer is after the testing result of m layer, in the time that i layer signal is detected, not only square accumulated value of the Euclidean distance by traditional ball decoding is predicted the value of i layer symbol, also utilize follow-up the 1st layer of code tree node calculating path correction value not detecting to (i-1) layer, for predicting the value of i layer symbol; Utilize mutually combining of two kinds of metrics, make the predicted value of i layer code tree node more accurate than Pohst and two kinds of traditional ball decoding detection algorithms of SE; And in testing process, because route searching is accurately reduced to rollback number of times, fast, accurately find destination path, effectively reduce input complexity and obviously reduce decoding delay; The method comprises following operating procedure:

Step 1, according to transmission conditions and decode procedure needs, initialization arranges following parameter:

Current decoding radius of a ball C=C is set ₀, under real signal model, mimo system transmitting antenna total quantity is m, in formula, and C ₀for the decoding ball initial radium arranging; Natural number i is the current antenna sequence number detecting, and the maximum of sequence number i is m; And MIMO testing process is according to antenna sequence number order from big to small, detects successively the symbol sending on each antenna, therefore while starting to detect, current detection antenna sequence number i is set to maximum antenna sequence number m, i.e. i=m; The vector that initialization length is m is set again with be respectively used to the symbol sebolic addressing temporarily obtaining in store M IMO testing process and the sequence number sequence finally obtaining; in each element represent the symbol that on i root antenna, detection obtains, finish the rear symbol sebolic addressing finally obtaining for recording MIMO detection; The vector T that initialization length is m is set, the each element T in T _ifor recording when the symbol of i root transmitting antenna is detected, the Euclidean distance between the symbol obtaining after testing on i+1 to m root antenna and reception signal square accumulated value; Transmission total number of symbols amount in transmission assemble of symbol and this set χ that χ and q represent that respectively under real signal model, every transmitting antenna of mimo system is corresponding is set; It is the vectorial β of m that initialization length is set, the each element β in β _ibe used to indicate when the symbol of i root transmitting antenna is detected the sequence number of selected symbol correspondence in alternative assemble of symbol; It is capable that initialization arranges m, matrix S and the matrix P of q row, wherein, and when q the capable paired i root of the element set transmitting antenna of i in matrix S detects, corresponding alternative assemble of symbol; The vector that the 1st capable row of i in matrix P are total to q element composition to q row is P _{i, 1:q}, for recording the transmission assemble of symbol χ that i root transmitting antenna is corresponding, the likelihood probability metric of each element under log-domain;

Step 2, detects ball decoded signal: the regulation that receiving terminal detects according to ball decoding backward, put in order according to the size of antenna sequence number, detect one by one the transmission symbol on every antenna; Final output for final detection result;

Step 3, utilize the CRC check bit in initial data to improve detection performance:

It is capable that initialization arranges τ, matrix L and the list storage position indicator pointer of m row wherein, τ is the byte total capacity for storing each row symbol alternative path, initial setting up in many code trees;

In the time that step 2 finishes each detection operation, will be demodulated into after corresponding bit sequence, this bit sequence is carried out to CRC check: if verification is passed through, will as final detection result output, and stop detecting; Otherwise, will deposit this list in, and check whether this list fills up: if list is filled up, the symbol sebolic addressing that deposits first list in is exported as final detection result, and stopped detecting; Otherwise, will from code tree, delete, continue to find new return and carry out above-mentioned steps 1, until stop detecting.

The low-complexity MIMO system ball decoded signal detection side who the present invention is based on path metric value compares with the signal detection technique of Pohst and two kinds of traditional ball decodings of SE, and its innovation advantage is:

The present invention utilizes square accumulated value of Euclidean distance and path modification value to combine the Search Results of code tree node is predicted, thereby make the route searching process in code tree more accurate, quick, effectively reduce the computation complexity of input and detect time delay.Moreover in conjunction with list, the inventive method can be in the situation that computation complexity be almost fair with maximum likelihood detection method, combination property exceedes Maximum Likelihood Detection.

Because the testing process of the inventive method is compute sign likelihood probability metric, therefore it can also provide the information of the log-likelihood ratio in bit-level, can combine with various channel coding technologies easily like this.With respect to traditional mimo system ball decoded signal detection technique, the present invention is more suitable for being applied in practical communication system, has good popularizing application prospect.

Brief description of the drawings

Fig. 1 is the schematic diagram of ball decoded signal testing process searching route in code tree.

Fig. 2 is the mimo system ball decoded signal detection method operating procedure flow chart that the present invention is based on path metric value.

Fig. 3 is that the inventive method is at the improvement algorithm operating flow chart adding after list and CRC check.

Fig. 4 is in the mimo system of employing 16QAM modulation, 8 transmitting antennas and 8 reception antennas, the embodiment of the present invention and the performance comparison schematic diagram of traditional ball decoded signal detection algorithm in bit error rate (BER).

Fig. 5 is in the mimo system of employing 16QAM modulation, 8 transmitting antennas and 8 reception antennas, the embodiment of the present invention and the performance comparison of traditional ball decoded signal detection algorithm in floating-point operation amount (Flops), i.e. computation complexity contrast schematic diagram.

Embodiment

For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing, the present invention is described in further detail.

The present invention is based on the low-complexity MIMO system ball decoded signal detection method of path metric value, it is the flow process detecting according to ball decoding backward, in known code tree, (i+1) layer is after the testing result of m layer, in the time that i layer signal is detected, not only square accumulated value of the Euclidean distance by traditional ball decoding is predicted the value of i layer symbol, also utilize follow-up the 1st layer of code tree node calculating path correction value not detecting to (i-1) layer, for predicting the value of i layer symbol.Then utilize mutually combining of two kinds of metrics, make the predicted value of i layer code tree node more accurate than Pohst and two kinds of traditional ball decoding detection algorithms of SE; And in testing process, because route searching is accurately reduced to rollback number of times, fast, accurately find destination path, effectively reduce input complexity and obviously reduce decoding delay.

In ball decoded signal testing process, prediction step can also be set, to obviously reduce the amount of calculation of path modification value.In addition, testing process can also utilize the CRC cyclic redundancy check (CRC) position in original transmission Bit data further to improve detection performance, testing process utilizes a fixed-size list to remove to store the alternative path in many code trees, after each detection, not directly the path of finding out to be exported as final result, but corresponding this paths bit sequence is carried out to CRC check: if verification is passed through, this path is exported as final detection result, and detection of end; If verification is not passed through, the path of having found out is deposited into list, more whether the path of storing in detection list is filled up list.If fill up, export as final result in the path that Article 1 is deposited in to list, and detection of end; If do not fill up, the path that just deposits list in is left out from code tree, repeat above-mentioned steps and find new path, finish until detect.

Referring to Fig. 2, finish the concrete operation step of mimo system ball decoded signal detection method of the present invention:

Step 1, according to transmission conditions and decode procedure needs, initialization arranges following parameter:

Current decoding radius of a ball C=C is set ₀, under real signal model, mimo system transmitting antenna total quantity is m, in formula, and C ₀for the decoding ball initial radium arranging; Natural number i is the current antenna sequence number detecting, and the maximum of sequence number i is m; And because MIMO testing process is according to antenna sequence number order from big to small, detect successively the symbol sending on each antenna, therefore while starting to detect, current detection antenna sequence number i is set to maximum antenna sequence number m, i.e. i=m; The vector that initialization byte length is m is set again with be respectively used to the symbol sebolic addressing temporarily obtaining in store M IMO testing process and the sequence number sequence finally obtaining; in each element represent the symbol that on i root antenna, detection obtains, finish the rear symbol sebolic addressing finally obtaining for recording MIMO detection; The vector T that initialization byte length is m is set, the each element T in T _ifor recording when the symbol of i root transmitting antenna is detected, the Euclidean distance between the symbol obtaining after testing on i+1 to m root antenna and reception signal square accumulated value; Transmission total number of symbols amount in transmission assemble of symbol and this set χ that χ and q represent that respectively under real signal model, every transmitting antenna of mimo system is corresponding is set; It is the vectorial β of m that initialization length is set, the each element β in β _ibe used to indicate when the symbol of i root transmitting antenna is detected the sequence number of selected symbol correspondence in alternative assemble of symbol; It is capable that initialization arranges m, matrix S and the matrix P of q row, wherein, and when q the capable paired i root of the element set transmitting antenna of i in matrix S detects, corresponding alternative assemble of symbol; The vector that the 1st capable row of i in matrix P are total to q element composition to q row is P _{i, 1:q}, for recording the transmission assemble of symbol χ that i root transmitting antenna is corresponding, the likelihood probability metric of each element under log-domain.

Step 2, detects ball decoded signal: the regulation that receiving terminal detects according to ball decoding backward, put in order according to the size of antenna sequence number, detect one by one the transmission symbol on every antenna; Final output for final detection result.This step 2 comprises following content of operation:

(21) initialization arranges i element β in vectorial β _i=1.

(22) judge β _iwhether >q sets up, if set up, i is added to 1, redirect execution step (26); Otherwise, record vector in i element in formula, for the capable β of i in matrix S _icolumn element.

(23) by β _iafter adding 1; According to formula calculate and upgrade i element T of vector T _ivalue, wherein, y _abe the receiving symbol on a root reception antenna, in mimo system real signal model, receive a the element of vectorial y, r _a,bfor the capable b column element of a in matrix R, for the symbol sebolic addressing obtaining after testing in b element; Then, judge T _i>C ²whether set up, if set up, after the antenna sequence number i detecting being added to 1, carry out subsequent step (24); If be false, redirect execution step (25).

(24) judge whether i>m sets up, if set up, will as final detection result output, finish all to detect operation; If be false, return to execution step (22).

(25) judge whether i equals 1, if so, carry out subsequent step (26); Otherwise, after i is subtracted to 1, the capable j column element of the i P of compute matrix P _i,jcorresponding likelihood probability metric, wherein, natural number j is the corresponding sequence number of element in assemble of symbol χ, its maximum is q; Obtain so the capable vectorial P of q element composition altogether of i in this matrix P _{i, 1:q}numerical value, P _{i, 1:q}for the capable vector of q element composition altogether of i of matrix P.Then, the capable j column element of the i P of matrix P _i,jas j the corresponding likelihood probability metric of symbol in transmission assemble of symbol χ corresponding to i root transmitting antenna, then to the capable all elements P of i in this matrix P _{i, 1:q}according to carrying out from small to large after ascending order arrangement, obtain vector and show that thus sequence is rear vectorial in each element at former vectorial P _{i, 1:q}the original sequence number of middle correspondence is η _ψ, wherein, ψ is vector in element sequence number, its maximum is q, for vector in ψ element; Again according to η _ψ, by all elements S of capable i of matrix S ψ row _{i, ψ}be updated to respectively the η in set χ _ψindividual element

In this step (25), calculate the likelihood probability metric P that each symbol is corresponding _i,joperation comprises following content:

(251) arrange in formula, r _a,wfor the capable w column element of a in matrix R, ρ _wbe certain element in the transmission assemble of symbol χ that w root transmitting antenna is corresponding, i.e. ρ _w∈ χ, natural number a is to remain the antenna sequence number not detecting, and its maximum is i-1, and natural number w calculates γ _aintermediate variable in Distribution Statistics process, for the antenna sequence number that represents that computational process uses, its maximum is (i-1).Again to all w ∈ 1,2 ..., i-1}, the each element ρ in traversal set χ _winstitute's value likely, obtain γ _adistribution Statistics after, calculate γ _aaverage μ _aand variance

(252) to γ _acarry out non-uniform quantizing: first by γ _abe quantified as N _qplant value, make the quantized signal point set G after quantizing _ain comprise altogether N _qindividual element, obtains the quantized signal point set G that a root transmitting antenna is corresponding thus _ain k value g _a,k, wherein, g _a,k∈ G _a, a=1,2 ..., i, k=1,2 ..., N _q.

(253) according to formula: $P_{i,j}=-\underset{a=i}{\overset{m}{\mathrm{\Σ}}}{(y_a-\underset{b=a}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b)}^2+\underset{a=1}{\overset{i-1}{\mathrm{\Σ}}}\ln \underset{k=1}{\overset{N_Q}{\mathrm{\Σ}}}\exp [-{(y_a-\underset{b=i}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b-g_{a,k})}^2]$ Or formula of reduction $P_{i,j}=\underset{a=i}{\overset{m}{\mathrm{\Σ}}}{(y_a-\underset{b=a}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b)}^2+\underset{a=1}{\overset{i-1}{\mathrm{\Σ}}}\underset{g_{a,k}}{\min }{[y_a-\underset{b=i}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b-g_{a,k}]}^2$ Calculate P _i,jvalue; Wherein, y _abe the signal receiving on a root reception antenna, i.e. a the element of vectorial y in real signal model, r _a,bfor the capable b column element of a in matrix R, with j the element χ sending in assemble of symbol χ _jnumerical value is equal, for the symbol sebolic addressing obtaining after testing in b element, and subscript b ∈ i+1, i+2 ..., m}, g _a,kk value in quantized signal point set corresponding to a root transmitting antenna that obtains for above-mentioned steps; Ln () and exp () represent respectively to get with natural logrithm e=2.71828 ... for logarithm and the index of the truth of a matter; represent traversal searching g _a,kall values so that the result of calculation minimum in bracket; Because P _i,jbe worth littlely, the reliability of its corresponding symbol is higher.

It should be noted that: for above-mentioned three operating procedures (251)～(253), the computational process of path modification value can arrange prediction step δ and adjusts accordingly and improve, to reduce computation complexity and to improve performance.

Above-mentioned improved three operating procedures the contents are as follows described (referring to the content of operation shown in Fig. 3 dotted line):

(25A) arrange in formula, r _a,wfor the capable w column element of a in matrix R, ρ _wbe certain element in the transmission assemble of symbol χ that w root transmitting antenna is corresponding, i.e. ρ _w∈ χ; Natural number a is to remain the antenna sequence number not detecting, and its maximum is i-1, and a ∈ { max{1, i-δ }, 2 ... i-1}, max{, represent get two number in higher value; Natural number w calculates γ _aintermediate variable in Distribution Statistics process, for the antenna sequence number that represents that computational process uses, its maximum is (i-1); Again according to w ∈ 1,2 ..., i-1}, the each element ρ in traversal set χ _winstitute's value likely, obtain γ _adistribution Statistics after, calculate γ _aaverage μ _aand variance

(25B) to γ _acarry out non-uniform quantizing: first by γ _abe quantified as N _qplant value, make the quantized signal point set G after quantizing _ain comprise altogether N _qindividual element, obtains the quantized signal point set G that a root transmitting antenna is corresponding thus _ain k value g _a,k, wherein g _a,k∈ G _a, a=max{1, i-δ }, 2 ..., i, k=1,2 ..., N _q;

(25C) according to formula: $P_{i,j}=-\underset{a=i}{\overset{m}{\mathrm{\Σ}}}{(y_a-\underset{b=a}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b)}^2+\underset{a=\max \{1,i-\mathrm{\δ}\}}{\overset{i-1}{\mathrm{\Σ}}}\ln \underset{k=1}{\overset{N_Q}{\mathrm{\Σ}}}\exp [-{(y_a-\underset{b=i}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b-g_{a,k})}^2]$ Or formula of reduction $P_{i,j}=\underset{a=i}{\overset{m}{\mathrm{\Σ}}}{(y_a-\underset{b=a}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b)}^2+\underset{a=\max \{1,i-\mathrm{\δ}\}}{\overset{i-1}{\mathrm{\Σ}}}\underset{g_{a,k}}{\min }{[y_a-\underset{b=i}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b-g_{a,k}]}^2$ Calculate P _i,jvalue; In formula, y _abe the signal receiving on a root reception antenna, i.e. a the element of vectorial y in real signal model, r _a,bfor the capable b column element of a in matrix R, with j the element χ sending in assemble of symbol χ _jnumerical value is equal, for the symbol sebolic addressing obtaining after testing in b element, and subscript b ∈ i+1, i+2 ..., m}; g _a,kk value in quantized signal point set corresponding to a root transmitting antenna that obtains for above-mentioned steps; Ln () and exp () represent respectively to get with natural logrithm e=2.71828 ... for logarithm and the index of the truth of a matter; represent traversal searching g _a,kall values so that the result of calculation minimum in bracket; Because P _i,jbe worth littlely, the reliability of its corresponding symbol is higher.

(26) upgrade the current decoding radius of a ball then, by vector value be copied to vector return to again execution step (22).

Step 3, utilize the CRC check bit in initial data to improve detection performance:

It is capable that initialization arranges τ, the matrix L of m row and each row-column list memory location pointer of this matrix L wherein, τ is the byte total capacity for storing each row symbol alternative path, initial setting up in many code trees;

In the time that step 2 finishes each detection operation, will be demodulated into after corresponding bit sequence, this bit sequence is carried out to CRC check: if verification is passed through, will as final detection result output, and stop detecting; Otherwise, will deposit this list in, and check whether this list fills up: if list is filled up, the symbol sebolic addressing that deposits first list in is exported as final detection result, and stopped detecting; Otherwise, will from code tree, delete, continue to find new return and carry out above-mentioned steps 1, until stop detecting.

This step 3 comprises following content of operation:

(31) judge whether i>m sets up, if set up, carry out subsequent step (32); If be false, return to execution step (22);

(32) will be demodulated into corresponding bit sequence, then this bit sequence is carried out to CRC check; If verification is passed through, will as final detection result output, and detection of end decoding all operations were; If verification is not passed through, carry out subsequent step (33);

(33) for each κ, respectively by vector in κ element assignment is in matrix L row κ column element wherein, κ is the vector element sequence number using in instruction assignment procedure, and its maximum is m; Then by list storage position indicator pointer after adding 1, judge now whether be greater than τ; If so, the 1st row of matrix L is exported as final detection result, and detection of end decoding all operations were; If not, carry out subsequent step (34);

(34) will the symbol sebolic addressing of representative is deleted from code tree, and decoding radius of a ball C=+ ∞ is set, and turns to step (22).

Because the testing process of the inventive method is compute sign likelihood probability metric, therefore the likelihood ratio information in bit-level can also be provided, can be combined with various channel coding technologies easily like this, improve systematic function.

The present invention has carried out the experiment of Multi simulation running embodiment and the use of simulated scenario, and with regard to the result of the test of emulation embodiment, performance and complexity to signal detecting method of the present invention are made a concrete analysis of below:

(1) performance comparison of the inventive method and traditional ball decoded signal detection algorithm:

Shown in Figure 4, in 88 mimo systems of receiving, adopt 16QAM modulation system, can obviously find out SE algorithm in the traditional ball decoded signal detection technique Maximum Likelihood Detection with respect to Pohst algorithm, the gap on error bit ability is approximately 2dB.This is to have the phenomenon of undetected signaling point due to SE algorithm, although reduced computation complexity, has lost performance.And the ball decoded signal detection algorithm that the present invention is based on path metric value has reached Maximum Likelihood Detection level in performance.In addition, adding after list and CRC check, improvement algorithm of the present invention is 8 o'clock in list length, with respect to Maximum Likelihood Detection, in performance, can also improve about 3dB, and in the time that list length increases, performance can further promote.

(2) computation complexity of the inventive method and traditional ball decoded signal detection algorithm contrast:

Shown in Figure 5, in 88 mimo systems of receiving, adopt 16QAM modulation system, can obviously find out the SE algorithm in traditional ball decoded signal detection technique, obviously reduce with respect to the computation complexity of Pohst algorithm, but the poor-performing of SE algorithm.And the ball decoded signal detection algorithm that the present invention is based on path metric value is in ensureing Maximum Likelihood Detection, computation complexity has reduced again a magnitude with respect to Pohst algorithm, adding after list and CRC check, improvement algorithm of the present invention is substantially constant in complexity, but is greatly improved in performance.

Can see by above contrast, the MIMO detection algorithm that the present invention proposes has double dominant with respect to traditional ball decoded signal detection algorithm on performance and computation complexity, and when aerial array scale becomes large, when signal modulation order uprises, its advantage can be more and more obvious.

In addition, utilize the inventive method that the information of the log-likelihood ratio in bit-level can also be provided, can be combined with various channel coding technologies easily like this.With respect to traditional ball decoding detection technique, the present invention is more suitable for, in applying, having good popularizing application prospect in practical communication system.

The foregoing is only preferred embodiment of the present invention, in order to limit the present invention, within the spirit and principles in the present invention not all, any amendment of making, be equal to replacement, improvement etc., within all should being included in the scope of protection of the invention.

Claims (5)

1. the low-complexity MIMO system ball decoded signal detection method based on path metric value, it is characterized in that: the flow process detecting according to ball decoding backward, in known code tree, (i+1) layer is after the testing result of m layer, in the time that i layer signal is detected, not only square accumulated value of the Euclidean distance by traditional ball decoding is predicted the value of i layer symbol, also utilize follow-up the 1st layer of code tree node calculating path correction value not detecting to (i-1) layer, for predicting the value of i layer symbol; Utilize mutually combining of two kinds of metrics, make the predicted value of i layer code tree node more accurate than Pohst and two kinds of traditional ball decoding detection algorithms of SE; And in testing process, because route searching is accurately reduced to rollback number of times, fast, accurately find destination path, effectively reduce input complexity and obviously reduce decoding delay; The method comprises following operating procedure:

Step 1, according to transmission conditions and decode procedure needs, initialization arranges following parameter:

Current decoding radius of a ball C=C is set ₀, under real signal model, mimo system transmitting antenna total quantity is m, in formula, and C ₀for the decoding ball initial radium arranging; Natural number i is the current antenna sequence number detecting, and the maximum of sequence number i is m; And MIMO testing process is according to antenna sequence number order from big to small, detects successively the symbol sending on each antenna, therefore while starting to detect, current detection antenna sequence number i is set to maximum antenna sequence number m, i.e. i=m; The vector that initialization byte length is m is set again with be respectively used to the symbol sebolic addressing temporarily obtaining in store M IMO testing process and the sequence number sequence finally obtaining; in each element represent the symbol that on i root antenna, detection obtains, finish the rear symbol sebolic addressing finally obtaining for recording MIMO detection; The vector T that initialization byte length is m is set, the each element T in T _ifor recording when the symbol of i root transmitting antenna is detected, the Euclidean distance between the symbol obtaining after testing on i+1 to m root antenna and reception signal square accumulated value; Transmission total number of symbols amount in transmission assemble of symbol and this set χ that χ and q represent that respectively under real signal model, every transmitting antenna of mimo system is corresponding is set; It is the vectorial β of m that initialization byte length is set, the each element β in β _ibe used to indicate when the symbol of i root transmitting antenna is detected the sequence number of selected symbol correspondence in alternative assemble of symbol; It is capable that initialization arranges m, matrix S and the matrix P of q row, wherein, and when q the capable paired i root of the element set transmitting antenna of i in matrix S detects, corresponding alternative assemble of symbol; The vector that the 1st capable row of i in matrix P are total to q element composition to q row is P _{i, 1:q}, for recording the transmission assemble of symbol χ that i root transmitting antenna is corresponding, the likelihood probability metric of each element under log-domain;

Step 2, detects ball decoded signal: the regulation that receiving terminal detects according to ball decoding backward, put in order according to the size of antenna sequence number, detect one by one the transmission symbol on every antenna; Final output for final detection result;

Step 3, utilize the CRC check bit in initial data to improve detection performance:

It is capable that initialization arranges τ, the matrix L of m row and each row symbol memory location pointer of this matrix L wherein, τ is the byte total capacity for storing each row symbol alternative path, initial setting up in many code trees;

In the time that step 2 finishes each detection operation, will be demodulated into after corresponding bit sequence, this bit sequence is carried out to CRC check: if verification is passed through, will as final detection result output, and stop detecting; Otherwise, will deposit this list in, and check whether this list fills up: if list is filled up, the symbol sebolic addressing that deposits first list in is exported as final detection result, and stopped detecting; Otherwise, will from code tree, delete, continue to find new return and carry out above-mentioned steps 1, until stop detecting.

2. method according to claim 1, is characterized in that: described step 2 comprises following content of operation:

(21) initialization arranges i element β in vectorial β _i=1;

(22) judge β _iwhether >q sets up, if set up, i is added to 1, redirect execution step (26); Otherwise, record vector in i element in formula, for the i in matrix S is capable, β _icolumn element;

(23) by β _iafter adding 1; According to formula calculate and upgrade i element T of vector T _ivalue, wherein, y _abe the receiving symbol on a root reception antenna, in mimo system real signal model, receive a the element of vectorial y, r _a,bfor the capable b column element of a in matrix R, for the symbol sebolic addressing obtaining after testing in b element; Then, judge T _i>C ²whether set up, if set up, after the antenna sequence number i detecting being added to 1, carry out subsequent step (24); If be false, redirect execution step (25);

(24) judge whether i>m sets up, if set up, will as final detection result output, finish all to detect operation; If be false, return to execution step (22);

(25) judge whether i equals 1, if so, carry out subsequent step (26); Otherwise, after i is subtracted to 1, the capable j column element of the i P of compute matrix P _i,jcorresponding likelihood probability metric, wherein, natural number j is the corresponding sequence number of element in assemble of symbol χ, its maximum is q; Obtain so the capable vectorial P of q element composition altogether of i in this matrix P _{i, 1:q}numerical value, wherein, P _{i, 1:q}for the capable vector of q element composition altogether of i of matrix P; Then, the capable j column element of the i P of matrix P _i,jas j the corresponding likelihood probability metric of symbol in transmission assemble of symbol χ corresponding to i root transmitting antenna, then to the capable all elements P of i in this matrix P _{i, 1:q}according to carrying out from small to large after ascending order arrangement, obtain vector and show that thus sequence is rear vectorial in each element at former vectorial P _{i, 1:q}the original sequence number of middle correspondence is η _ψ, wherein, ψ is vector in element sequence number, its maximum is q, for vector in ψ element; Again according to η _ψ, by all elements S of capable i of matrix S ψ row _{i, ψ}be updated to respectively the η in set χ _ψindividual element

(26) upgrade the current decoding radius of a ball then, by vector value be copied to vector return to again execution step (22).

3. method according to claim 2, is characterized in that: in described step (25), calculate the likelihood probability metric P that each symbol is corresponding _i,jcomprise following content of operation:

(251) arrange in formula, r _a,wfor the capable w column element of a in matrix R, ρ _wbe certain element in the transmission assemble of symbol χ that w root transmitting antenna is corresponding, i.e. ρ _w∈ χ, natural number a is to remain the antenna sequence number not detecting, and its maximum is i-1, and natural number w calculates γ _aintermediate variable in Distribution Statistics process, for the antenna sequence number that represents that computational process uses, its maximum is (i-1); Again to all w ∈ 1,2 ..., i-1}, the each element ρ in traversal set χ _winstitute's value likely, obtain γ _adistribution Statistics after, calculate γ _aaverage μ _aand variance

(252) to γ _acarry out non-uniform quantizing: first by γ _abe quantified as N _qplant value, make the quantized signal point set G after quantizing _ain comprise altogether N _qindividual element, obtains the quantized signal point set G that a root transmitting antenna is corresponding thus _ain k value g _a,k, wherein, g _a,k∈ G _a, a=1,2 ..., i, k=1,2 ..., N _q;

(253) according to formula: $P_{i,j}=-\underset{a=i}{\overset{m}{\mathrm{\Σ}}}{(y_a-\underset{b=a}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b)}^2+\underset{a=1}{\overset{i-1}{\mathrm{\Σ}}}\ln \underset{k=1}{\overset{N_Q}{\mathrm{\Σ}}}\exp [-{(y_a-\underset{b=i}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b-g_{a,k})}^2]$ Or formula of reduction $P_{i,j}=\underset{a=i}{\overset{m}{\mathrm{\Σ}}}{(y_a-\underset{b=a}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b)}^2+\underset{a=1}{\overset{i-1}{\mathrm{\Σ}}}\underset{g_{a,k}}{\min }{[y_a-\underset{b=i}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b-g_{a,k}]}^2$ Calculate P _i,jvalue; Wherein, y _abe the signal receiving on a root reception antenna, i.e. a the element of vectorial y in real signal model, r _a,bfor the capable b column element of a in matrix R, with j the element χ sending in assemble of symbol χ _jnumerical value is equal, for the symbol sebolic addressing obtaining after testing in b element, and subscript b ∈ i+1, i+2 ..., m}, g _a,kk value in quantized signal point set corresponding to a root transmitting antenna that obtains for above-mentioned steps; Ln () and exp () represent respectively to get logarithm and the index taking natural logrithm e as the truth of a matter; represent traversal searching g _a,kall values so that the result of calculation minimum in bracket; Because P _i,jbe worth littlely, the reliability of its corresponding symbol is higher.

4. method according to claim 3, it is characterized in that: in order to reduce computation complexity and to improve performance, prediction step δ is set, for operating procedure (251)～(253) to step 2, the computational process of path modification value is adjusted and improved, the content of operation after improvement is as follows:

(25A) arrange in formula, r _a,wfor the capable w column element of a in matrix R, ρ _wbe certain element in the transmission assemble of symbol χ that w root transmitting antenna is corresponding, i.e. ρ _w∈ χ; Natural number a is to remain the antenna sequence number not detecting, and its maximum is i-1, and a ∈ { max{1, i-δ }, 2 ... i-1}, max{, represent get two number in higher value; Natural number w calculates γ _aintermediate variable in Distribution Statistics process, for the antenna sequence number that represents that computational process uses, its maximum is (i-1); Again according to w ∈ 1,2 ..., i-1}, the each element ρ in traversal set χ _winstitute's value likely, obtain γ _adistribution Statistics after, calculate γ _aaverage μ _aand variance

(25B) to γ _acarry out non-uniform quantizing: first by γ _abe quantified as N _qplant value, make the quantized signal point set G after quantizing _ain comprise altogether N _qindividual element, obtains the quantized signal point set G that a root transmitting antenna is corresponding thus _ain k value g _a,k, wherein g _a,k∈ G _a, a=max{1, i-δ }, 2 ..., i, k=1,2 ..., N _q;

(25C) according to formula: $P_{i,j}=-\underset{a=i}{\overset{m}{\mathrm{\Σ}}}{(y_a-\underset{b=a}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b)}^2+\underset{a=\max \{1,i-\mathrm{\δ}\}}{\overset{i-1}{\mathrm{\Σ}}}\ln \underset{k=1}{\overset{N_Q}{\mathrm{\Σ}}}\exp [-{(y_a-\underset{b=i}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b-g_{a,k})}^2]$ Or formula of reduction $P_{i,j}=\underset{a=i}{\overset{m}{\mathrm{\Σ}}}{(y_a-\underset{b=a}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b)}^2+\underset{a=\max \{1,i-\mathrm{\δ}\}}{\overset{i-1}{\mathrm{\Σ}}}\underset{g_{a,k}}{\min }{[y_a-\underset{b=i}{\overset{m}{\mathrm{\Σ}}}{r}_{a,b}{\hat{d}}_b-g_{a,k}]}^2$ Calculate P _i,jvalue; In formula, y _abe the signal receiving on a root reception antenna, i.e. a the element of vectorial y in real signal model, r _a,bfor the capable b column element of a in matrix R, with j the element χ sending in assemble of symbol χ _jnumerical value is equal, for the symbol sebolic addressing obtaining after testing in b element, and subscript b ∈ i+1, i+2 ..., m}; g _a,kk value in quantized signal point set corresponding to a root transmitting antenna that obtains for above-mentioned steps; Ln () and exp () represent respectively to get logarithm and the index taking natural logrithm e as the truth of a matter; represent traversal searching g _a,kall values so that the result of calculation minimum in bracket; Because P _i,jbe worth littlely, the reliability of its corresponding symbol is higher.

5. method according to claim 1, is characterized in that: described step 3 comprises following content of operation:

(31) judge whether i>m sets up, if set up, carry out subsequent step (32); If be false, return to execution step (22);

(32) will be demodulated into corresponding bit sequence, then this bit sequence is carried out to CRC check; If verification is passed through, will as final detection result output, and detection of end decoding all operations were; If verification is not passed through, carry out subsequent step (33);

(33) for each κ, respectively by vector in κ element assignment is in matrix L row κ column element wherein, κ is the vector element sequence number using in instruction assignment procedure, and its maximum is m; Then by list storage position indicator pointer after adding 1, judge now whether be greater than τ; If so, the 1st row of matrix L is exported as final detection result, and detection of end decoding all operations were; If not, carry out subsequent step (34);

(34) will the symbol sebolic addressing of representative is deleted from code tree, and decoding radius of a ball C=+ ∞ is set, and turns to step (22).