CN101442390A - Equilibrium acceptance method and apparatus for Turbo of spatial correlation MIMO - Google Patents

Abstract

The invention relates to a Turbo equalization receiving method for spatial correlated MIMO and a device thereof. The invention provides an iterative equalization algorithm for carrying out combined detection on a sending antenna and provides a detection method for list spherical decoding (LSD) for reducing the complexity of realization, and belongs to the technical field of wireless communication. The method achieves the following aims: 1, iterative equalization is adopted to change a selective channel to a flat fading channel; the LSD decoding is used to reduce the complexity of detecting a system signal; and 2, a detection algorithm combining LSD and JAD-based SC/MMSE is adopted to reduce the complexity of an equalizer and the sensitivity of an MIMO signal detection algorithm on relevant properties of a spatial channel. The Turbo equalization receiving technology for carrying out combined detection on the sending antenna can be applied to a spatial correlated MIMO frequency selective channel and can effectively remove inter-antenna interference (IAI), inter-symbol interference (ISI) and multi-user inference.

Description

The Turbo equilibrium acceptance method and the device of spacing related MIMO

Technical field

The invention belongs to wireless communication technology field, particularly the balanced iteration reception technique of Turbo.

Background technology

Bit Interleave coded modulation (ST-BICM) is the simple and main flow scheme that multiple-input, multiple-output (MIMO) system uses in flat fading channel.Yet when it applied to broadband wireless communications, fading channel can present the serious frequency selectivity.On the other hand, after a plurality of transmitting antenna data process frequency selective fadings of mimo system, received signal is that multipath superposed signal and many antennas superposed signal are formed, and signal disturbs (IAI) to the signal that reception antenna receives between intersymbol interference (ISI) and antenna because the different pieces of information transmission of the multipath of channel and many antennas can form.In mimo system, it is by Maximum likelihood sequence detection (MLSE) that the performance of optimization obtains, but the complexity of MLSE is exponential increase with fate and multipath number.Recently, the couple candidate detection method that tabulation ball-type decoding (LSD) is mimo system under the smooth channel that falls, it can obtain the best compromise of performance and complexity.

Because it is in frequency selective fading channels, long with the search radius exponentially multiplication of LSD by the candidate symbol number of LSD.Recently, a kind of soft interference eliminated Minimum Mean Square Error of MIMO signal based on many antennas joint-detection (JAD) (SC/MMSE) iteration equalizing technology is used, it can change frequency-selective channel into equivalent flat fading channel group, therefore, the LSD technology can be applicable to the MIMO input under the frequency-selective channel, and its prerequisite is the SC/MMSE detection method that adopts based on JAD.

Summary of the invention

The present invention is directed to the balanced iteration reception technique of Turbo to the spatial coherence sensitive issue, proposed a kind of iteration equalizing algorithm that transmitting antenna is carried out joint-detection.Because the complexity of this algorithm is the index increase with the product of channel memory span, number of transmit antennas, the detection method that has therefore proposed tabulation ball-type decoding (LSD) is to reduce implementation complexity.The balanced reception technique of the Turbo of this transmitting antenna joint-detection can be applicable to can eliminate inter-antenna interference (IAI), intersymbol interference (ISI) and multi-user interference effectively in the MIMO frequency-selective channel of space correlation.

Innovation part of the present invention is: frequency-selective channel is become flat fading channel by iteration equalizing, reduce the complexity that system signal detects and can use LSD to decipher, and with LSD and the merging of JAD-based SC/MMSE detection algorithm, it not only can reduce the complexity of equalizer, can also reduce the sensitiveness of MIMO signal detection algorithm to the space channel correlation properties.Because existing is highstrung based on the independent SC/MMSE iteration equalizing technology that detects (ABA) of each transmitting antenna data stream to the space correlation channel, can be along with the enhancing of spatial correlation characteristic, the anxious great change of performance is poor.Because JAD-based SC/MMSE signal detecting method is regarded transmitting antenna data stream as signal phasor, by all possible transmission signal phasor being carried out the detection mode of nearly maximum likelihood, make the performance of mimo system be lower than SC/MMSE detection method based on ABA to the sensitiveness of spatial correlation characteristic.

1. transmitter and channel model

Space hour bit interlaced coded modulation (ST-BICM) is a kind of simple side signal transmission case when empty, method for transmitting signals when applying to the present invention as sky, and its transmitter structure is shown in explanation accompanying drawing 1.The serial u of information bit obtains coding data sequences c through encoder, obtains data sequence v through interweaving, and adopts mode along separate routes to obtain and the corresponding data subsequence of transmit antenna number x, shines upon the data symbol S that obtains on each transmit antenna through ovennodulation.Receiver structure is shown in explanation accompanying drawing 2.Suppose that desirable channel estimating and accurate received signal are synchronous.Receiver goes into softly to go out (SISO) and constitute by two-stage is soft, separates by interleaver and deinterleaver, and the external information of representing with log-likelihood ratio (LLR) exchanges between two-stage.

By the transmitter and explanation accompanying drawing 2 receiver modules of explanation accompanying drawing 1, the sequence of the output sequence of encoder after interleaver interweaves is v=[v ₁, v ₂, v _B], wherein B is the interleaver frame length.According to number of transmit antennas n _TBe divided into sub-piece, in each height piece: $x^{\left(m\right)}={[x_1^{\left(m\right)},x_2^{\left(m\right)},\·\·\·x_P^{\left(m\right)}]}^T,$ M=1 ..., n _T, m is the transmit antenna order, and P is the bit number of each modulation symbol, and modulation symbol can be by s ^(m)=map (x ^{＜m 〉}) mapping function 2 ^PIndividual modulation symbol $S=\{{\mathrm{\α}}_0,{\mathrm{\α}}_1,\·\·\·,{\mathrm{\α}}_{2^P-1}\}$ In obtain.Again but each antenna transmission symbol is formed symbolic frame $\{s_1^{\left(m\right)},s_2^{\left(m\right)},\·\·\·s_{L_s}^{\left(m\right)}\}\∈S,$ Transmission symbol constitutes the transmission symbol matrix $s\left(l_s\right)=[s^{\left(1\right)}{\left(l_s\right)}^T,s^{\left(2\right)}{\left(l_s\right)}^T,\·\·\·,s^{\left(n_T\right)}{\left(l_s\right)}^T],$ Through arriving receiving terminal behind the frequency selectivity multidiameter fading channel.Receiver is through coherent demodulation, by n _RIndividual reception antenna samples time domain multipath composite signal, and signal y (k) can be expressed as when sampling instant k empty:

y(k)＝HS(k)+n(k) (1)

Wherein $y\left(k\right)\∈C^{{\mathrm{Ln}}_R\×1}$ Sampling received signal vector when being sky, L is the multipath number of frequency-selective channel, supposes the ideal communication channel estimation, the footpath number of every frame multipath signal and each footpath amplitude are known, receive the sampled signal vector and are produced in conjunction with received signal by formula (2):

y(k)＝[r ^T(k+L-1)，…，r ^T(k)] ^T (2)

$r\left(k\right)\∈C^{n_R\×1}$ For

$r\left(k\right)={[r_1\left(k\right),\·\·\·,r_{n_R}\left(k\right)]}^T---\left(3\right)$

r _n(k) the reception sampled signal on n antenna of definition, vector $n\left(k\right)\∈C^{n_R\×1}$ White Gaussian noise sampling when being sky, covariance is $E\left\{n\left(k\right)n^H\left(k\right)\right\}={\mathrm{\σ}}_n^{2}I.$ $H\∈C^{{\mathrm{Ln}}_R\×n_T(2L-1)}$ Be the known channel matrix,, can obtain, be expressed as by the separate multiple Gaussian process of each each antenna of footpath with distribute (i.i.d.) to irrelevant mimo channel:

Wherein

With the received signal k corresponding transmission symbol vector of sampling constantly $S\left(k\right)\∈C^{n_T(2L-1)\×1}$ For:

S(k)＝[s ^T(k+L-1)，…，s ^T(k)，…，s ^T(k-L+1)] ^T (5)

Wherein $s\left(k\right)=[s^{\left(1\right)}\left(k\right),\·\·\·,s^{\left(n_T\right)}\left(k\right)].$

2. iterative receiver

The present invention proposes a kind of receiver structure based on soft interference eliminated least mean-square error (JAD-SC/MMSE) algorithm that transmitting antenna is carried out joint-detection (JAD).To be that the inside of receiver is soft go into soft (SISO) level that goes out with the JAD-SC/MMSE detector, frequency-selective channel is transformed into an equivalent flat fading channel group, merge rake simultaneously, received signal is become the output vector of transmission data after decay and Gaussian noise are disturbed stack, each component is corresponding with the transmission data on the respective transmit antenna, but has experienced independently flat fading.With the output vector ball-type decoding (LSD) of tabulating, LSD will select and transmitting antenna sends the immediate candidate data symbol of data.Again the candidate symbol incoming symbol is arrived bit likelihood ratio calculator (MAP), calculate each encoded data bits LLR, the LLR sequence is sent into the SISO channel decoder that adopts bcjr algorithm through deinterleaving, obtain the LLR of coded sequence, LLR subtracts each other the coded-bit external information that obtains being used for next iteration with input.Fully after the iteration, the logarithm probability of decoder output information sequence obtains information bit through judgement.External information to the probability calculation of symbol, obtains being used for SC/MMSE and the symbol prior information to bit likelihood ratio transducer through bit.

2.1 Turbo MIMO equalizer

Turbo equalizer of the present invention is based on the expansion of single antenna receiving algorithm, and expands to PSK and qam constellation mode.The soft symbol of channel decoder feedback is estimated

K can be expressed as in the time of reception:

${\tilde{s}}^{\left(m\right)}\left(k\right)=E\left\{s^{\left(m\right)}\left(k\right)\right\}=\underset{{\mathrm{\α}}_j\∈S}{\mathrm{\Σ}}{\mathrm{\α}}_j{P}_a(s^{\left(m\right)}\left(k\right)={\mathrm{\α}}_j)---\left(6\right)$

With $E\left\{{\left|{\tilde{s}}^{\left(m\right)}\left(k\right)\right|}^2\right\}=\underset{{\mathrm{\α}}_j\∈S}{\mathrm{\Σ}}{\left|{\mathrm{\α}}_j\right|}^2{p}_a(s^{\left(m\right)}\left(k\right)={\mathrm{\α}}_j)---\left(7\right)$

P wherein _a(s ^(m)(k)=α _j) the is-symbol prior probability, and hypothesis feedback log likelihood ratio

Independent based on bit, a transmission symbol s ^(m)(k) form by P coded-bit mapping, be expressed as

p _a(s ^(m)(k)=α _j) can be expressed as:

$P_a(s^{\left(m\right)}\left(k\right)={\mathrm{\α}}_j)={\left(\frac{1}{2}\right)}^p\underset{p=1}{\overset{P}{\mathrm{\Π}}}(1-x_{k,p}^{\left(m\right)}\tanh (L_a\left(x_{k,p}^{\left(m\right)}\right)/2))---\left(8\right)$

The symbol variance

For:

$\mathrm{var}\left({\tilde{s}}^{\left(m\right)}\left(k\right)\right)=E\left\{{\left|{\tilde{s}}^{\left(m\right)}\left(k\right)\right|}^2\right\}-{\left|E\left\{{\tilde{s}}^{\left(m\right)}\left(k\right)\right\}\right|}^2.---\left(9\right)$

Expression formula relevant with the modulation constellation mode, if | α _j|=const, ${\∀\mathrm{\α}}_j\∈S$ (as P-PSK),

Be constant and equal the transmission symbol energy

The calculating of variance does not need the prior information of modulation symbol.For other constellation (i.e.2 ^P-QAM, P〉2), average and variance can be calculated by formula (9):

$\hat{S}\left(k\right)=\hat{S}\left(k\right)-\tilde{S}\left(k\right)\⊗e_k---\left(10\right)$

Wherein $e_k=[0_{1\×(L-1)n_T},1_{1\×n_T},0_{1\×(L-1)\×n_T}],$

Definition is based on each vector product that multiplies each other of vector.

Obtain by each transmission symbol item in the soft information replacement formula (5).

Transmission symbol

M=1 ..., n _TBy soft interference eliminated received signal

Filtering is carried out joint-detection and is obtained:

$\hat{y}\left(k\right)=y\left(k\right)-H\hat{s}\left(k\right),k=0,\·\·\·,L_s-1.---\left(11\right)$

With linear least mean-square poor (MMSE) filter filtering, weighting matrix W (k) satisfies criterion formula (11):

$[W\left(k\right),A\left(k\right)]=\arg \underset{W,A}{\min }{\left|\right|W^H\hat{y}\left(k\right)-A^H\mathrm{\β}\left(k\right)\left|\right|}^2---\left(12\right)$

Matrix $W\left(k\right)\∈C^{{\mathrm{Ln}}_R\×n_T},$ $W\left(k\right)=[w^{\left(1\right)}\left(k\right),\·\·\·,w^{\left(n_T\right)}\left(k\right)],$ Vector $\mathrm{\β}\left(k\right)\∈C^{n_T\×1}$ By $\mathrm{\β}\left(k\right)={[s^{\left(1\right)}\left(k\right),\·\·\·,s^{\left(n_T\right)}\left(k\right)]}^T$ It is given, $A\left(k\right)\∈C^{n_T\×n_T}$ Be diagonal matrix, satisfy $a_{\mathrm{1,1}}\left(k\right)=\·\·\·a_{n_T,n_T}\left(k\right)=1$ Separate [W (k), A (k)]=[0,0] to avoid virtual.

Optimize the column vector of weighting matrix W (k) $w^{\left(m\right)}\left(k\right)\∈C^{{\mathrm{Ln}}_R\×1},$ M is the transmitting antenna order, w ^(m)(k)=M (k) ^-1h ^(m), wherein:

$M\left(k\right)=\mathrm{H\Λ}\left(k\right)H^H+{\mathrm{\σ}}_n^{2}I-\underset{m=1}{\overset{n_T}{\mathrm{\Σ}}}{h}^{\left(m\right)}{h}^{{\left(m\right)}^H}$ (13)

$=R_{\mathrm{cov}}-\underset{m=1}{\overset{n_T}{\mathrm{\Σ}}}{h}^{\left(m\right)}{h}^{{\left(m\right)}^H}$

h ^(m)Be [the Ln of matrix H _T+ m]-the th column vector, $R_{\mathrm{cov}}=\mathrm{H\Λ}\left(k\right)H+{\mathrm{\σ}}_n^{2}I.$

$\mathrm{\Λ}\left(k\right)=\mathrm{diag}\{v^{\left(1\right)}(k+L-1),\·\·\·v^{\left(n_T\right)}(k+L-1),\·\·\·,v^{\left(1\right)}(L-1),\·\·\·,v^{\left(n_T\right)}(L-1),$ ?（14)

${\mathrm{\δ}}_s^2\×1_{1\×n_T},v^{\left(1\right)}\left(1\right),\·\·\·,v^{\left(n_T\right)}\left(1\right),\·\·\·,v^{\left(1\right)}(k-L+1),\·\·\·,v^{\left(n_T\right)}(k-L+1)\}$

V wherein ^(m)(.) is soft estimate symbol

Variance.Suppose the output of MMSE filter $z\left(k\right)\∈C^{n_T\×1}$ Can be approximately equivalent Gaussian noise channel, can be written as:

$z\left(k\right)=W^H\left(k\right)\hat{y}\left(k\right)=H_e\left(k\right)\mathrm{\β}\left(k\right)+{\mathrm{\ψ}}_e\left(k\right)---\left(15\right)$

Matrix wherein $H_e\left(k\right)\∈C^{n_R\×n_T}$ Be equivalent flat fading channel gain, it is found the solution and can be obtained by formula (15):

H _e(k)＝E{z(k)β ^H(k)}＝W ^H(k)H _ML，(16)

$H_{\mathrm{ML}}=[h^{\left(1\right)},\·\·\·,h^{\left(n_T\right)}],$ Vector ${\mathrm{\ψ}}_e\left(k\right)\∈C^{n_T\×1}$ Be equivalent Gaussian noise covariance matrix:

$R_e\left(k\right)=E\left\{{\mathrm{\ψ}}_e\left(k\right){\mathrm{\ψ}}_e^H\left(k\right)\right\}=W^H\left(k\right)R_{\mathrm{cov}}W\left(k\right)-H_e\left(k\right)H_e^H\left(k\right)---\left(17\right)$

Therefore, after received signal is by the JAD-SC/MMSE equalizer processes, the result who can equivalence obtains after through a flat channel for transmission symbol.But more for antenna number, when order of modulation was higher, possible transmission symbol number was 2 ⁿ _T ^{* P}, very complicated according to the bit log-likelihood ratio (LLR) of MAP algorithm computation transmission symbol.The ball-type decoding (SD) that document [8] proposes is chosen one near the subclass of transmission symbol in possible selection symbol, but can only be applied to the frequency-flat fading channel.And the JAD-SC/MMSE equalizer that this paper proposes can become frequency-selective channel frequency-flat channel, can list of application ball-type decoding LSD algorithm.

Has the equivalent channel Parameter H _e(k) and R _e(k) vector z (k) obtains most probable candidate's transmission symbol collection through LSD decoding, obtains the external information log-likelihood probability of each bit by symbol to the MAP algorithm of bit mapping, and send softly to go into softly to go out (SISO) channel decoder and decipher.Can be expressed as equalizer Gauss equivalence output vector:

z(k)＝H _eβ(k)+Ψ _e(k) (18)

Possible transmission subclass β (k) calculates its Maximum Likelihood Detection probability by formula (19):

$\underset{\mathrm{\β}\∈D^{\mathrm{nT}}}{\max }{p}_{z|S}\left(z\left(k\right)\right|\mathrm{\β}\left(k\right))---\left(19\right)$

The search volume

Be n _TThe dimension integer lattice, its conditional probability distribution is:

$p\left(z\left(k\right)\right|\mathrm{\β}\left(k\right))=\frac{\exp [-\frac{1}{{2R}_e\left(k\right)}{\left|\right|z\left(k\right)-H_e\mathrm{\β}\left(k\right)\left|\right|}^2]}{{\left(2\mathrm{\π}{R}_e\left(k\right)\right)}^{n_R}}---\left(20\right)$

Formula (19) can equivalence be the optimization aim function of finding the solution formula (20):

$\underset{\mathrm{\β}\∈D^{\mathrm{nT}}}{\min }{\left|\right|z\left(k\right)-H_e\mathrm{\β}\left(k\right)\left|\right|}^2---\left(21\right)$

The problem of finding the solution of formula (21) just becomes the integer lattice least mean square problems, and the selection of " recently " lattice point is realized by LSD.

2.2 comparison based on the SC/MMSE of JAD and ABA

JAD-SC/MMSE is that the defeated signal of multi-antenna transmitting carries out joint-detection, and formula (12) is its optimal design criterion.ABA-SC/MMSE with $w_{\mathrm{ABA}}^{\left(m\right)}=\arg \underset{w_{\mathrm{ABA}}^{\left(m\right)}}{\min }\left|\right|w_{\mathrm{ABA}}^{\left(m\right)}\hat{y}{\left(k\right)}^{\left(m\right)}-s_k^{\left(m\right)}\left|\right|$ Be the basis, based on each transmitting antenna detection signal, the LLR of coded-bit is directly calculated by equalizer output, and the soft information of feedback is also calculated based on each antenna.Yet, JAD-SC/MMSE exports a symbolic vector, calculating the LLR of each bit with all transmission symbols and all composite symbol vectors of transmitting on all transmit antennas, adopt LSD to reduce implementation complexity, serves as the feedback that its soft information is realized on the basis with all antennas simultaneously.When the mimo channel space was irrelevant, the systematic function of two kinds of algorithms did not have difference; But to the space correlation channel, when data symbol arrives receiving terminal, certain relevant ambiguity is arranged, be difficult to separate transmission data on line every day by the ABA-SC/MMSE equalization methods that line is handled based on every day; JAD-SC/MMSE equalization algorithm based on the transmitting antenna joint-detection, do not adopt the mode of separating each transmitting antenna transmission data, and employing is combined with the immediate transmission data of transmitting antenna by the received signal search and the method acquisition candidate of LSD is transmitted the data vector collection, carry out bit LLR and calculate, reduced sensitivity the space channel correlation.

3. improved tabulation ball-type decoding

In the 2.1st joint, ML detection problem is become a least mean square problems, at first utilize the real number equivalents of formula (21), in the present invention, only consider the decoding of real number ball-type.Definition M=2n _T, N=2n _R, and vector S _r(k), Z _r(k), H _rBe from β (k), Z (k), H _eThe real vector that obtains.

S _r(k)＝[Re(β(k)) ^T?Im(β(k)) ^T] ^T，Z _r(k)＝[Re(z(k)) ^T?Im(z(k)) ^T] ^T；(22)

H wherein _r∈ R ^{N * M}Can be by being defined as

$H_r=\left[\begin{array}{cc}\mathrm{Re}\left(H_e\right)& \mathrm{Im}\left(H_e\right)\\ -\mathrm{Im}\left(H_e\right)& \mathrm{Re}\left(H_e\right)\end{array}\right]---\left(23\right)$

Formula (21) can be written as:

$\underset{S_r\&Element;{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="A200710050539E00121.png"\; state="\{\{state\}\}">D</figure-callout>}}^{2\mathrm{nT}}}{\min }{\left|\right|Z_r\left(k\right)-H_r{S}_r\left(k\right)\left|\right|}^2---\left(24\right)$

To Γ=H _eH _e ^TMatrix carries out Cholesky and decomposes, and can obtain Γ=AA ^T, its lower triangular matrix is A={a _Ij.Yet, in this patent, need with the tabulation ball-type decipher after the real number candidate symbol, use formula (22) to obtain plural transmission symbol again..

Existing ball-type decoding only is applicable to integer lattice, and modulation symbol is not an integer under normal conditions, needs a non-integer symbol increase or deduct a side-play amount to become integer lattice; Another kind method is to enumerate a little to force constellation point to form integer lattice an integer range, and this lattice scheme adopts in this paper.Tabulation ball-type decoding (LSD) is that with the difference of ball-type decoding (SD) LSD just carries out the candidate symbol search in a given radius, and SD selects best distance radius point, and therefore, the performance of SD algorithm is better than LSD.The LSD algorithm flow of this The thesis is shown in explanation accompanying drawing 3.

Be the candidate symbol table,, can obtain the tabulation lattice point under radius C, selected symbol is used for the calculating of maximum a posteriori probability in the lump, as the prior information of input and output (SISO) decoder according to the LSD algorithm.

4. the calculating of logarithm bit likelihood ratio

The bit number that each equivalent output vector z (k) is carried is p _c=P * n _TIndividual coded-bit is corresponding to the coded-bit that receives sampling instant k $X\left(k\right)=[x^{\left(1\right)}\left(k\right),x^{\left(2\right)}\left(k\right),\&CenterDot;\&CenterDot;\&CenterDot;,x^{\left(n_T\right)}\left(k\right)],$ Wherein $x^{\left(m\right)}\left(k\right)=[x_{k,1}^{\left(m\right)},x_{k,2}^{\left(m\right)},\&CenterDot;\&CenterDot;\&CenterDot;x_{k,P}^{\left(m\right)}]$ 。The symbol vector $\mathrm{\&beta;}\left(k\right)=\{s^{\left(1\right)}\left(k\right),\&CenterDot;\&CenterDot;\&CenterDot;s^{\left(n_T\right)}\left(k\right)\}$ The priori symbol probability be

The LLR of this coded-bit is

Can calculate generation by the candidate symbol vector collection that LSD obtains, computational process is as follows:

$L\left(x_{k,p}^{\left(m\right)}\right|z\left(k\right))=\log \frac{p[x_{k,p}^{\left(m\right)}=+1|z\left(k\right)]}{p[x_{k,p}^{\left(m\right)}=-1|z\left(k\right)]},---\left(25\right)$

Wherein, suppose binary character information 0, and 1} be mapped to 1 ,+1}, formula (25) can be changed into:

$L\left(x_{k,p}^{\left(m\right)}\right|z\left(k\right))=\log \frac{p[x_{k,p}^{\left(m\right)}=+1|z\left(k\right)]}{p[x_{k,p}^{\left(m\right)}=-1|z\left(k\right)]}=\log \frac{p[z\left(k\right),x_{k,p}^{\left(m\right)}=+1]}{p[z\left(k\right),x_{\mathrm{pk}}^{\left(m\right)}=-1]}$ (26)

$+\log \frac{\underset{X\left(k\right):x_{k,p}^{\left(m\right)}=+1}{\mathrm{\&Sigma;}}p\left[z\left(k\right)\right|x_{k,p}^{\left(m\right)}][p\left(X\left(k\right)\right]}{\underset{X\left(k\right):x_{k,p}^{\left(m\right)}=-1}{\mathrm{\&Sigma;}}p\left[z\left(k\right)\right|x_{k,p}^{\left(m\right)}]p\left[X\left(k\right)\right]}$

Suppose that the bit among the X (k) is separate, formula (26) becomes:

$L\left(x_{k,p}^{\left(m\right)}\right|z\left(k\right))=\log \frac{p[x_{k,p}^{\left(m\right)}=+1]}{p[x_{k,p}^{\left(m\right)}=-1]}+\log \frac{\underset{X\left(k\right):x_{k,p}^{\left(m\right)}=+1}{\mathrm{\&Sigma;}}p\left[z\left(k\right)\right|X\left(k\right)]\underset{j,j\&NotEqual;p}{\mathrm{\&Pi;}}p\left[x_{k,j}^{\left(m\right)}\right]}{\underset{X\left(k\right):x_{k,p}^{\left(m\right)}=-1}{\mathrm{\&Sigma;}}p\left[z\left(k\right)\right|X\left(k\right)]\underset{j,j\&NotEqual;p}{\mathrm{\&Pi;}}p\left[x_{k,j}^{\left(m\right)}\right]}$

First of formula (27) is priori logarithm probability Second is called the outside logarithm probability of coded-bit

Bit block x ^(m)Unique symbolic vector β of expression ^(m)Under Gauss's equivalent channel, can be expressed as:

$L\left(x_{k,p}^{\left(m\right)}\right|z\left(k\right))=\log \frac{\underset{\mathrm{\&beta;}\left(k\right):x_{k,p}^{\left(m\right)}=+1}{\mathrm{\&Sigma;}}p\left[z\left(k\right)\right|\mathrm{\&beta;}\left(k\right)]\underset{m}{\mathrm{\&Pi;}}p\left[s_k^{\left(m\right)}\right]}{\underset{\mathrm{\&beta;}\left(k\right):x_{k,p}^{\left(m\right)}=-1}{\mathrm{\&Sigma;}}p\left[z\left(k\right)\right|\mathrm{\&beta;}\left(k\right)]\underset{m}{\mathrm{\&Pi;}}p\left[s_k^{\left(m\right)}\right]}=\log \frac{\underset{\mathrm{\&beta;}\left(k\right):x_{k,p}^{\left(m\right)}=+1}{\mathrm{\&Sigma;}}{e}^{-\frac{1}{{\mathrm{\&sigma;}}^2}\left|\right|z\left(k\right)-H_e\mathrm{\&beta;}\left(k\right)\left|\right|+{\mathrm{\&Sigma;}}_m\log p\left[s_k^{\left(m\right)}\right]}}{\underset{\mathrm{\&beta;}\left(k\right):x_{k,p}^{\left(m\right)}=-1}{\mathrm{\&Sigma;}}{e}^{-\frac{1}{{\mathrm{\&sigma;}}^2}\left|\right|z\left(k\right)-H_e\mathrm{\&beta;}\left(k\right)\left|\right|+{\mathrm{\&Sigma;}}_j\log p\left[s_k^{\left(m\right)}\right]}}---\left(28\right)$

If calculating formula (28) is that whole symbol space is carried out, its complexity is very high, but for some candidates' transmission symbol, complexity can reach reasonable range.

Description of drawings

The conveyer structure chart of Figure 1B CIM scheme

Fig. 2 receives structure based on the Turbo equalizer of LSD and SC/MMSE

The tabulate algorithm flow chart of globular decoding of Fig. 3.

Claims (6)

1. detection method and device based on spatial correlation characteristic multiple-input, multiple-output (MIMO) multi-antenna wireless communication signal is characterized in that, for the frequency selectivity mimo channel, by a kind of transmission symbol associating vector detection algorithm.Its core is: many antenna transmission symbol is sent vector as one, by the time domain equalization that carries out to received signal, the frequency property selected fading channel etc. can be imitated be flat fading channel, then, utilize the candidate transmitted signal vector to calculate the soft information of output that sends symbol and detect the transmission signal.The wireless device that this method makes the MIMO signal supervisory instrument can be applied to space correlation channel and space non-correlation channel makes the use border.

2. require described detection algorithm as right 1, when base is characterised in that from receive vector detection of transmitted signals, all possible candidate who sends signal need be sent vector as the estimated vector that sends the soft information of symbol.

3. require described detection computations method as right 2, base is characterised in that the soft information that sends signal is estimated and need be met the following conditions:

The soft symbol of channel decoder feedback is estimated

K can be expressed as in the time of reception:

${\tilde{s}}^{\left(m\right)}\left(k\right)=E\left\{s^{\left(m\right)}\left(k\right)\right\}=\underset{{\mathrm{\&alpha;}}_j\&Element;S}{\mathrm{\&Sigma;}}{\mathrm{\&alpha;}}_j{p}_a(s^{\left(m\right)}\left(k\right)={\mathrm{\&alpha;}}_j)---\left(1\right)$

With

$E\left\{{\left|{\tilde{s}}^m\left(k\right)\right|}^2\right\}=\underset{{\mathrm{\&alpha;}}_i\&Element;S}{\mathrm{\&Sigma;}}{\left|{\mathrm{\&alpha;}}_j\right|}^2{p}_a(s^{\left(m\right)}\left(k\right)={\mathrm{\&alpha;}}_j)---\left(2\right)$

P wherein _a(s ^(m)(k)=α _j) the is-symbol prior probability, and hypothesis feedback log likelihood ratio

Independent based on bit, a defeated symbol s ^(m)(k) form by P coded-bit mapping, be expressed as

p _a(s ^(m)(k)=α _j) can be expressed as:

$p_a(s^{\left(m\right)}\left(k\right)={\mathrm{\&alpha;}}_j)={\left(\frac{1}{2}\right)}^p\underset{p=1}{\overset{P}{\mathrm{\&Pi;}}}(1-x_{k,p}^{\left(m\right)}\tanh (L_a\left(x_{k,p}^{\left(m\right)}\right)/2))---\left(3\right)$

The symbol variance For:

$\mathrm{var}\left({\tilde{s}}^{\left(m\right)}\left(k\right)\right)=E\left\{{\left|{\tilde{s}}^{\left(m\right)}\left(k\right)\right|}^2\right\}-{\left|E\left\{{\tilde{s}}^{\left(m\right)}\left(k\right)\right\}\right|}^2.---\left(4\right)$

Expression formula relevant with the modulation constellation mode, if | α _j|=const, $\&ForAll;{\mathrm{\&alpha;}}_j\&Element;S$ (as P-PSK),

Be constant and equal the transmission symbol energy

The calculating of variance does not need the prior information of modulation symbol.For other constellation (i.e.2 ^P-QAM, P〉2), average and variance can be calculated by formula (4):

$\hat{S}\left(k\right)=\hat{S}\left(k\right)-\hat{S}\left(k\right)\&CircleTimes;e_k---\left(5\right)$

Wherein $e_k=[0_{1\&times;(L-1)n_T},1_{1\&times;n_T},0_{1\&times;(L-1)\&times;n_T}],$ Definition is based on each vector product that multiplies each other of vector. Replace by soft information

S(k)＝[s ^T(k+L-1)，…，s ^T(k)，…，s ^T(k-L+1)] ^T

In each transmission symbol item obtain.

4. the soft breath letter of transmission signal as claimed in claim 3 detects, and base is characterised in that, needs the tap coefficient of calculating filter, and its computational methods are as follows:

Transmission symbol M=1 ..., n _TBy soft interference eliminated received signal

Filtering is carried out joint-detection and is obtained:

$\hat{y}\left(k\right)=y\left(k\right)-H\hat{s}\left(k\right),k=0,\&CenterDot;\&CenterDot;\&CenterDot;,L_s-1.---\left(6\right)$

With linear least mean-square poor (MMSE) filter filtering, weighting matrix W (k) satisfies criterion formula (6):

$[W\left(k\right),A\left(k\right)]=\arg \underset{W,A}{\min }{\left|\right|W^H\hat{y}\left(k\right)-A^H\mathrm{\&beta;}\left(k\right)\left|\right|}^2---\left(7\right)$

Matrix $W\left(k\right)\&Element;C^{{\mathrm{Ln}}_R\&times;n_T},$ $W\left(k\right)=[w^{\left(1\right)}\left(k\right),\&CenterDot;\&CenterDot;\&CenterDot;,w^{\left(n_T\right)}\left(k\right)],$ Vector $\mathrm{\&beta;}\left(k\right)\&Element;C^{n_T\&times;1}$ By $\mathrm{\&beta;}\left(k\right)={[s^{\left(1\right)}\left(k\right),\&CenterDot;\&CenterDot;\&CenterDot;s^{\left(n_T\right)}\left(k\right)]}^T$ It is given, $A\left(k\right)\&Element;C^{n_T\&times;n_T}$ Be diagonal matrix, satisfy $a_{\mathrm{1,1}}\left(k\right)=\&CenterDot;\&CenterDot;\&CenterDot;a_{n_T,n_T}\left(k\right)=1$ Separate [W (k), A (k)]=[0,0] to avoid virtual.

5. as requirement as described in the right 3,4, base is characterised in that, needs to calculate the soft information that sends symbol, can be obtained by following algorithm: the column vector of matrix W (k) $w^{\left(m\right)}\left(k\right)\&Element;C^{{\mathrm{Ln}}_R\&times;1},$ M is the transmitting antenna order, w ^(m)(k)=M (k) ^-1H (m), wherein:

$M\left(k\right)=\mathrm{H\&Lambda;}\left(k\right)H^H+{\mathrm{\&sigma;}}_n^{2}I-\underset{m=1}{\overset{n_T}{\mathrm{\&Sigma;}}}{h}^{\left(m\right)}{h}^{{\left(m\right)}^H}$ (8)

$=R_{\mathrm{cov}}-\underset{m=1}{\overset{n_T}{\mathrm{\&Sigma;}}}{h}^{\left(m\right)}{h}^{{\left(m\right)}^H}$

h ^(m)Be [the Ln of matrix H _T+ m]-the th column vector, $R_{\mathrm{cov}}=\mathrm{H\&Lambda;}\left(k\right)H+{\mathrm{\&sigma;}}_n^{2}I.$

$\mathrm{\&Lambda;}\left(k\right)=\mathrm{diag}\{v^{\left(1\right)}(k+L-1),\&CenterDot;\&CenterDot;\&CenterDot;v^{\left(n_T\right)}(k+L-1),\&CenterDot;\&CenterDot;\&CenterDot;,v^{\left(1\right)}(L-1),\&CenterDot;\&CenterDot;\&CenterDot;,v^{\left(n_T\right)}(L-1),$

(9)

${\mathrm{\&delta;}}_s^2\&times;1_{1\&times;n_T},v^{\left(1\right)}\left(1\right),\&CenterDot;\&CenterDot;\&CenterDot;,v^{\left(n_T\right)}\left(1\right),\&CenterDot;\&CenterDot;\&CenterDot;,v^{\left(1\right)}(k-L+1),\&CenterDot;\&CenterDot;\&CenterDot;,v^{\left(n_T\right)}(k-L+1)\}$

V wherein ^(m)(.) is soft estimate symbol

Variance.Suppose the output of MMSE filter $z\left(k\right)\&Element;C^{n_T\&times;1}$ Can be approximately equivalent Gaussian noise channel, can be written as:

$z\left(k\right)=W^H\left(k\right)\hat{y}\left(k\right)=H_e\left(k\right)\mathrm{\&beta;}\left(k\right)+{\mathrm{\&psi;}}_e\left(k\right)---\left(10\right)$

Matrix wherein $H_e\left(k\right)\&Element;C^{n_R\&times;n_T}$ Be equivalent flat fading channel gain, it is found the solution and can be obtained by formula (10):

H _e(k)＝E{z(k)β ^H(k)}＝W ^H(k)H _ML， (11)

$H_{\mathrm{ML}}=[h^{\left(1\right)},\&CenterDot;\&CenterDot;\&CenterDot;,h^{\left(n_T\right)}],$ Vector ${\mathrm{\&psi;}}_e\left(k\right)\&Element;C^{n_T\&times;1}$ Be equivalent Gaussian noise covariance matrix:

$R_e\left(k\right)=E\left\{{\mathrm{\&psi;}}_e\left(k\right){\mathrm{\&psi;}}_e^H\left(k\right)\right\}=W^H\left(k\right)R_{\mathrm{cov}}W\left(k\right)-H_e\left(k\right)H_e^H\left(k\right)---\left(12\right)$

Therefore, after received signal is by the JAD-SC/MMSE equalizer processes, the result who can equivalence obtains after through a flat channel for transmission symbol.

6. select based on right 1,5 described transmission glossary of symbols, it is characterized in that, can adopt the tabulation globular decoding to simplify, its simplification process is:

Has the equivalent channel Parameter H _e(k) and R _e(k) vector z (k) obtains most probable candidate's transmission symbol collection through LSD decoding, obtains the external information log-likelihood probability of each bit by symbol to the MAP algorithm of bit mapping, and send softly to go into softly to go out (SISO) channel decoder and decipher.Can be expressed as equalizer Gauss equivalence output vector:

z(k)＝H _eβ(k)+Ψ _e(k) (13)

Possible transmission subclass β (k) calculates its Maximum Likelihood Detection probability by formula (14):

$\underset{\mathrm{\&beta;}\&Element;D^{\mathrm{nT}}}{\max }{p}_{z|S}\left(z\left(k\right)\right|\mathrm{\&beta;}\left(k\right))---\left(14\right)$

The search volume Be n _TThe dimension integer lattice, its conditional probability distribution is:

$p\left(z\left(k\right)\right|\mathrm{\&beta;}\left(k\right))=\frac{\exp [-\frac{1}{{2R}_e\left(k\right)}{\left|\right|z\left(k\right)-H_e\mathrm{\&beta;}\left(k\right)\left|\right|}^2]}{{\left(2\mathrm{\&pi;}{R}_e\left(k\right)\right)}^{n_R}}---\left(15\right)$

Formula (14) can equivalence be the optimization aim function of finding the solution formula (15):

$\underset{\mathrm{\&beta;}\&Element;D^{\mathrm{nT}}}{\min }{\left|\right|z\left(k\right)-H_e\mathrm{\&beta;}\left(k\right)\left|\right|}^2---\left(16\right)$

The problem of finding the solution of formula (16) just becomes the integer lattice least mean square problems, and the selection of " recently " lattice point is realized by LSD.

Images