CN1188975C - Space-time iterative multiuser detecting algorithm based on soft sensitive bit and space grouping - Google Patents

Abstract

The space-time iterative multi-user detection method based on soft-sensitive bits and space grouping is applied to the field of multi-user access technology in wireless communication. The outer group, and then use the simplified MA∏ iterative multiuser detection algorithm based on soft sensitive bits for the sub-multiusers in each group, that is, first distinguish the sensitive bits to obtain prior information, and then Perform MA∏ detection in the small subset, and then use the external information output by the iteratively processed channel MA∏ decoding of users in each group to realize soft interference cancellation outside the group. It can be used in the coded multi-user access system in broadband wireless communication, and can approach the single-user multi-antenna coded system with a small number of iterations when accessing 20 users at the same time in the AΩΓN channel and the frequency selective fading channel performance.

Description

Space-time iterative multi-user detection method based on soft sensitive bits and space grouping

Technical Field

Multi-user access system encoded in wireless communications: coded Direct Sequence-code Division Multiple Access (DS-CDMA), coded multi-carrier MC-CDMA, and coded Space Division Multiple Access (SDMA) systems.

Background

In a broadband wireless communication system, Multiple Access Interference (MAI) and Inter-Symbol Interference (ISI) constitute major obstacles affecting stable communication of the system. In order to better solve the problem of intersymbol interference and Multiple Access Interference (MAI) and improve the system capacity, the invention jointly adopts an algorithm of an intelligent antenna and a simplified iterative MAP multi-user Detection (Turbo MUD) technology, and a space-time iterative multi-user Detection method based on soft sensitive bits and space packets, and is applied to a coded multi-user access system.

The application of smart antenna technology in mobile communication is a research hotspot in the field of communication. The technology can greatly increase the capacity of the system and improve the efficiency of power and frequency spectrum under the condition of not increasing frequency spectrum resources. This is because the spatial filtering can effectively suppress multiple access interference in a direction different from the incident direction of the target user, while enhancing the target signal due to diversity combining of the target signal on the antenna array. In recent years, with the invention of Turbo Code with strong error correction capability, iterative (Turbo) processing technology has been increasingly emphasized in wireless communication systems. The Turbo code technology can obtain stable communication and the performance is close to the decoy theoretical value. When the MAP-based iterative (Turbo) multi-user detection technology is applied to a coded CDMA system, the performance of the algorithm can approach that of a single-user coded CDMA system even in a slightly low SNR range. Therefore, the performance of the system is further improved by combining the Turbo multi-user detection technology and the intelligent antenna technology. Recently, different Turbo multi-user detection techniques combined with antenna arrays have been proposed to enhance the performance of the system [1, 2 ]. In [1] and [2], the antenna array Turbo multi-user detection technology combined with the interference elimination method is used for DS-CDMA and MC-CDMA systems under a fading channel. The multi-user detection algorithm is similar to the Successive Interference Cancellation (SIC) method. Similar Turbo multi-user detection techniques are also proposed in [3, 4] except that Minimum Mean Square Error (MMSE) filtering is used to improve performance after interference cancellation. Although the complexity of these interference cancellation-based methods is linear with the number of users, the iterative multi-user detection algorithm requires more iterations to approach the optimal MAP. In [5], a MAP iterative multi-user detection algorithm combined with a smart antenna is proposed for a multi-beam (Multibeam) system, but the complexity of the algorithm is exponential to the number of users, and cannot be realized in practice.

In order to make MAP iterative multiuser detection technology possible to be applied to practical systems (dozens or even hundreds of users access in a sector at the same time), the invention combines intelligent antenna and iterative multiuser detection technology, and uses a space-time iterative (Turbo) multiuser detection algorithm for coding multiuser access systems.

[1]M.C.Reed and P.D.A.exander，“Iterative multiuser detection using antenna arrays and FEC on multipathchannels，”IEEE JSAC，Vol.17，No.12，pp.2082-89，Dec 1999.

[2]M.S.Akhter and J.Asenstorfer，“Iterative detection for MC-CDMA system with Base station antenna array forfading chanels，”IEEE GLOBECOM’98.Sydney，NSW，Australia，1998.

[3]X.D.Wang and H.V.Poor，“Iterative(Turbo)soft interference cancellation and decodig for coded CDMA，”IEEE Trans.Commun.，Vol.47，No.7，pp.1046-1061，July 1999.

[4]H.E.Gamal，and E.Geraniotis，“Iterative multiuser detection for coded CDMA signals in AWGN and Fadingchannels，”IEEE JSAC，Vol.18，No.1，pp.30-41，January 2000.

[5]Michael L.Moher，“Multiuser decoding for multibeam systems，”IEEE Trans.Vehicular Technology，Vol.49，No.4，pp.1226-34，July 2000

Disclosure of Invention

The soft sensitive bit and space packet space-time iteration (Turbo) multi-user detection algorithm based on the intelligent antenna and the simplified MAP iteration multi-user detection technology can be applied to a multi-user access system of coding in broadband wireless communication. For example, as shown in fig. 1, 2 and 3, a space-time iterative multi-user detection algorithm is applied to a multi-carrier CDMA system to significantly improve the capacity and performance of the system. The spatio-temporal iterative multi-user detection algorithm is described as follows: the spatio-temporal iterative multi-user receiver classifies all users into groups and corresponding "outer groups" according to spatial correlation, with all users classified into different groups. After grouping, a multi-user detection algorithm based on MAP iteration is applied to each group for reducing the number of effective users. Although the number of valid users in each group is greatly reduced relative to the number of all users in a sector, the conventional MAP iterative multi-user detection algorithm is not feasible when the number of users in a group is close to or exceeds 10. In order to reduce the complexity of the MAP multi-user detection algorithm in each group, we propose a simplified iterative MAP multi-user detection algorithm based on soft sensitive bits as a sub multi-user detection algorithm in each group. The basic idea of the MAP iterative multi-user detection algorithm based on sensitive bits is that we first distinguish the sensitive bits, which can obtain "coarse" prior information, which gives information whether each coded bit is likely to estimate a pair or an error. With this a priori information, we can perform MAP detection in a small subset of corresponding resolved sensitive bits. In addition, because the spatial filtering method is limited to inhibit the inter-group user interference, before MAP iterative multi-user detection is performed in each group, the multi-access interference (MAI) of the 'outer group' interfering users is eliminated, and a soft interference elimination method is adopted in the algorithm. Due to Turbo (iterative) processing within a group, extrinsic information output by channel MAP decoding of users within each group can be used to achieve soft interference cancellation outside the group to improve algorithm performance. Because if a hard interference cancellation method is used, errors in hard interference cancellation can result when the hard decision estimate is incorrect. By doing so, the space-time iterative multi-user detection algorithm based on soft sensitive bits and spatial grouping can significantly reduce the complexity of the algorithm and can achieve better performance than the traditional multi-user detection combining soft interference cancellation of the antenna array. In the coded MC-CDMA system, the space-time iterative multi-user detection algorithm can approach the performance of a single-user multi-antenna coding system under a few iteration times under an AWGN channel and a frequency-selective fading channel when twenty users are accessed simultaneously. The algorithm complexity of the space-time iterative multi-user detection algorithm is in a linear relation with the number of users, and the MAP multi-user detection algorithm is possible to be applied in practice.

Therefore, the present invention aims to provide a spatio-temporal iterative multi-user detection method based on soft sensitive bits and spatial grouping.

The invention is characterized in that: classifying all users into a plurality of groups and corresponding groups except the groups, namely 'outer groups', according to spatial correlation, and then using a simplified MAP iterative multi-user detection algorithm based on soft sensitive bits for sub-multi-users in each group, namely firstly distinguishing the sensitive bits to obtain prior information, then carrying out MAP detection in a small subset corresponding to the distinguished sensitive bits, and then using outer information output by channel MAP decoding of users in each group after iterative processing to realize soft interference elimination outside the groups; it comprises the following steps in sequence:

1) receiving multiple users by using a multi-antenna matrix, setting M user access signals, and performing spatial filtering and frequency domain matching filtering according to the incident direction of user signals; according to the correlation of spatial filtering weight coefficients among users, all M users are classified and divided into G groups and corresponding 'outer groups';

namely: obtaining a spatial filtering weight vector of each user according to the following formulaAll M users are classified into G groups and corresponding 'outer groups' by grouping criteria:

${\stackrel{\→}{W}}_m=\frac{{\stackrel{\→}{a}}_m}{\left|\right|{\stackrel{\→}{a}}_m\left|\right|}=k\·{\stackrel{\→}{a}}_m,$ 1≤m≤M $k=\frac{1}{\sqrt{Q}}$

wherein: q is the number of antenna elements;

${\stackrel{\→}{a}}_m={[1,e^{-\mathrm{j\π}\sin Q_m},.......,e^{-j(Q-1)\mathrm{\π}\sin {\mathrm{\θ}}_m}]}^T$ for the mth user at an incident angle of Q_mA lower antenna array response;

the grouping criterion is as follows:

1.1) when beta_μ，ν≥β₁And beta is_v，μ≥β₂When the users are distributed to the same group omega_gI.e. { μ, ν, v }. epsilon. OMEGA_g；

Wherein, mu is 1, 1., M, v is 1, 2., v-1,

β₁，β₂is a set threshold, and₁＞β₂，

β_μ，vfor spatial filtering vectors W between users mu, v_μ， Such as:

${\mathrm{\β}}_{\mathrm{\μ},v}=\frac{\left|\right|{{\stackrel{\→}{W}}_{\mathrm{\μ}}}^H\·{\stackrel{\→}{W}}_v\left|\right|}{\left|\right|W_{\mathrm{\μ}}\left|\right|\left|\right|W_v\left|\right|}$

superscript (·)^HFor conjugate transpose, "·" denotes vector dot product;

1.2) when beta_μ，v≥β₁And beta is_ν，v≤β₂And then distributing the users mu and v to be the same group omega_gThe number of members of (a), i.e. { mu.,ν}∈Ω_gand is $\left\{v\right\}\∉{\mathrm{\Ω}}_g,$

Thus, the total number of all users assigned to the G groups satisfies $\underset{g=0}{\overset{G=1}{\mathrm{\Σ}}}{k}_g=M,$ Wherein k is_gThe total number of users in the g-th group, and the 'outer group' is the g-th group omega_gFor other users, using Ω_g ^IIndicates the number of users $k_g^I=M-k_g,k_g^I,$ Represents omega_g ^IThe total number of users;

2) calculating initial hard and soft estimates of the coded bits for all M users and assigning truth values to G groups and corresponding "outer groups";

after the receiving end is processed by beam forming and traditional matched filtering, the initial hard estimation and soft estimation of the coded bits of all M users are obtained, and the initial hard estimation and the soft estimation sequentially and respectively comprise:

${\hat{g}}_{\stackrel{\‾}{g},t}^{\left(R\right)}=\mathrm{sign}\left(\mathrm{real}\left(y_{m(k,\stackrel{\‾}{g})}\right)\right),$

${\tilde{d}}_{\stackrel{\‾}{g},t}^{\left(k\right)}=\mathrm{real}\left(y_{m(k,\stackrel{\‾}{g})}\right);$

is the received signal Matched Filtered (MF) output of the kth user classified as the g "outer group" by the mth user;

3) defining the maximum sensitive bit number and the maximum iteration number of each group;

3.1) according to the following inequalityTo resolve estimated multi-user coded bit vectors

Likelihood metric of (d) and the original individual user code bits of a conventional single-user Matched Filter (MF) estimateThe upper limit of the likelihood metric difference value of (c),

the larger the value of (d), the more likely the corresponding adjusted bit is to estimate an error, i.e., a sensitive bit;

3.2) then k_gAnMiddle searchf(f＜k_g) F is the number of sensitive bits in the g group;

4) a simplified iterative MAP multi-user detection algorithm based on soft-sensitive bits is performed in the g-th group, which in turn has the following steps:

4.1) realizing soft interference elimination according to the 'outer group' soft estimation;

according to "outer group" omega_g ^ICoded bits for inter-user transmission

To achieve soft interference cancellation:

the above soft estimation

${\tilde{d}}_{\stackrel{\‾}{g},k}^{\left(k\right)}=\mathrm{real}\left(y_{m(k,\stackrel{\‾}{g})}\right);$

After the interference is eliminated, there is ${{\stackrel{\→}{X}}_g}^n=H_g{G}_{g,t}{\stackrel{\→}{d}}_{g,t}+{\stackrel{\→}{Z}}_{g,}$ WhileIs the autocorrelation matrix R_g，

Is composed of

$E\left[{\stackrel{\&RightArrow;}{Z}}_g{{\stackrel{\&RightArrow;}{Z}}_g}^H\right]=H_{\stackrel{\&OverBar;}{g}}\&CenterDot;E[\mathrm{\&Delta;}{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g}}\&CenterDot;\mathrm{\&Delta;}{{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g}}}^H]\&CenterDot;{H_{\stackrel{\&OverBar;}{g}}}^H+\frac{1}{\sqrt{Q}}{H}_g{{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="C0312082000212.png,C0312082000222.png"\; state="\{\{state\}\}">n</figure-callout>}}}^2;$

Wherein H_gIs the g-th group omega_gThe inner users and the correlation matrix are,is group omega_gInner users and corresponding "outer groups" omega_g ^ICorrelation matrix between inner users, G_g，tIs an internal omega of the group_gThe average power of the users is then calculated,

is a group omega_gThe vector of coded bits transmitted by the users,is an outer group omega_g ^ICoding ratio of inner user transmissionSpecially for treating diabetesThe soft-state estimation of (a) is performed,is an "outer group" omega_g ^IEstimation error, σ, of MAP channel decoded output of a user_n ²Is the variance of additive white gaussian noise, AWGN;

4.2) finding out f sensitive bits by using a sensitive bit algorithm, and calculating the output extrinsic information of each user in the g group by using a simplified MAP multi-user detection algorithm according to the sensitive bits;

when the prior probability of the sensitive bits is set to be equal probability distribution and the prior probability of the non-sensitive bits is set to be 1 in the first iteration, the MAP multi-user detection algorithm can only consider 2 corresponding to f sensitive bits^fA vector of coded bits

4.2.1) computing extrinsic information λ of the kth user in the g-th group_1e ^k；

4.2.2) calculating the posterior LOG Likelihood Ratio (LLR) of the kth user by using MAP algorithm and using Lambda₂Indicating that the user decodes the output extrinsic information into ${\mathrm{\&lambda;}}_{2e}^k={\mathrm{\&Lambda;}}_2-{\mathrm{\&lambda;}}_{1e}^k;$

4.2.3) treatment of Lambda_2e ^kFeeding back to MAP multi-user detection module, calculating posterior LOG likelihood of information bit when iteration is finished, so as to decode received bit;

4.2.4) obtaining improved prior information of the coding bits according to the external information fed back by each user channel decoding, namely: ${\mathrm{\&lambda;}}_{1\mathrm{\&sigma;}}^k={\mathrm{\&lambda;}}_{2e}^k,$ thereby obtaining more accurate hard and soft estimates of the user coded bits;

4.3) judging whether the iteration is finished:

if not, each user MAP channel decoder calculates the extrinsic information of the coded bit and returns to the step (2);

if the calculation is finished, namely G groups are calculated, the MAP channel decoder of each user calculates the extrinsic information of the information bit to be output as a multi-user detection signal;

3) and (4) obtaining improved hard estimation and soft estimation of all M users, and returning to the step (3).

The application proves that the invention makes the application of the MAP iterative multi-user detection technology possible in a practical system.

Drawings

FIG. 1: the multi-carrier CDMA system model of Turbo space-time multi-user detection.

FIG. 2: turbo space-time multi-user receiver.

FIG. 3: and g, a Turbo multi-user detection algorithm block diagram based on a soft sensitive bit algorithm in the group.

FIG. 4: simplified Turbo multi-user detection algorithm performance for single antenna reception of AWGN channel (M10, L15).

FIG. 5: performance of Turbo space-time multi-user detection algorithm under AWGN channel (M20, L7), 4 groups, and 15 degrees of group interval.

FIG. 6: turbo space-time multi-user detection algorithm performance under AWGN channel (M20), ${\mathrm{\&rho;}}_{{m,m}^{\&prime;}}^s=0.4.$

FIG. 7: the performance of Turbo space-time multi-user detection algorithm under frequency selective fading channel (M20, L7).

FIG. 8: the program flow of the method is schematically shown.

Detailed Description

In particular, when applied to a coded multi-carrier CDMA system, the algorithm is implemented as follows.

To facilitate understanding of the formula, we first define the meaning of the symbols in the formula: variable table vector of upper arrow; an uppercase blackbody variable table matrix; lower horizontal line variable table time series. Dot-table products in the formula; the notation  defines the Kronecker product. Superscript (·)^TDefining a transpose; superscript (·)^HA coreference transpose is defined.

Fig. 1 shows a coded MC-CDMA system with M users. The users are randomly distributed within a 120 sector. Since the mobile terminal is not easy to use the antenna array in the actual system, only the base station is provided with the antenna array of Q elements. In the system, the information bit sequence b (m) of the mth user passes through a convolutional coder (or Turbo coder) and then is interleaved to obtain a coding bit sequence d_t ^(m)To avoid burst bit errors caused by deep fading. Mth user, t time coded bit sequence d_t ^(m)After spreading with a pseudo-random (PN) sequence, the spread spectrum is transmitted using MC-CDMA techniques, where the number of subcarriers N is equal to the PN sequence length L. Here we assume that the sub-channels are flat fading and that the channel response between sub-channels is independent, which can be achieved by frequency domain interleaving. The frequency domain channel response of the ith subchannel of the mth user is $H_{t,l}^m={\mathrm{\&rho;}}_{m.l}\exp \left({\mathrm{\&theta;}}_{m,l}\right),$ Where ρ is_m，lAnd theta_m，lAmplitude and phase respectively. At the receiving end, the system achieves complete frame synchronization, which can be achieved by various existing time synchronization techniques. In addition, we also assume the antenna array response a of each user_m(0. ltoreq. m.ltoreq.M-1) has been accurately estimated. After beamforming, the signals for each user are despread in the frequency domain and maximal ratio combined. Definition of r_tReceiving signals for the array antenna of the r-th time interval

${\stackrel{\&OverBar;}{r}}_t=\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}{d}_t^{\left(m\right)}\&CenterDot;{\stackrel{\&OverBar;}{a}}_m\&CenterDot;\underset{l=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{\mathrm{\&rho;}}_{m,l}\&CenterDot;c_m\left[l\right]\&CenterDot;\exp \{{\mathrm{\&omega;}}_{l}t+{\mathrm{\&theta;}}_{m,l}\}+{\stackrel{\&RightArrow;}{n}}_t---\left(1\right)$

Wherein,

is a block vector of dimension Q x 1

${\stackrel{\&RightArrow;}{a}}_m={[1,e^{-\mathrm{j\&pi;}\sin {\mathrm{\&theta;}}_m},\&CenterDot;\&CenterDot;\&CenterDot;,e^{-j(Q-1)\mathrm{\&pi;}\sin {\mathrm{\&theta;}}_m}]}^T$ Defined as the mth user at an incident angle theta_mThe antenna array response in time. n is_tIs an Additive White Gaussian Noise (AWGN) vector over the antenna array and considers the noise on each element of the antenna array to be independent. c. C_m[l]Is a random sequencec ^m＝{c_m[l]}_{l＝1，…，L}The ith chip of (1). Since the signal bandwidth is much smaller than the frequency of the radio frequency, the antenna array response on all sub-carriers of each user can be approximated

Are the same. Taking into account the received signal on the qth antenna in the t-th time interval, a matrix-form representation of the signal

${{\stackrel{\&RightArrow;}{r}}_t}^q=\mathrm{IFFT}\left\{A_t^q{G}_t{\stackrel{\&OverBar;}{d}}_t\right\}+{\stackrel{\&RightArrow;}{n}}_t---\left(2\right)$

Wherein A is_t ^qIs an L × M dimensional matrix, A_t ^qContains the channel response and PN sequence information on all subcarriers of the mth user.

Is a vector of coded bits transmitted by M users. And G_tDefined as the average power matrix of all users.

In particular to

$A_t^q=[{\mathrm{\&alpha;}}_{1,t}^q{\stackrel{\&RightArrow;}{s}}_{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="C0312082000201.png,C0312082000202.png"\; state="\{\{state\}\}">t</figure-callout>}}^1,{\mathrm{\&alpha;}}_{2,t}^q{\stackrel{\&RightArrow;}{s}}_{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="C0312082000212.png,C0312082000222.png"\; state="\{\{state\}\}">t</figure-callout>}}^2,\&CenterDot;\&CenterDot;\&CenterDot;,{\mathrm{\&alpha;}}_{M,t}^q{\stackrel{\&RightArrow;}{s}}_t^M],$ g＝1，…，Q

Wherein,

${\stackrel{\&RightArrow;}{s}}_t^m={{[s}_{t,1}^m,s_{t,2}^m,\&CenterDot;\&CenterDot;\&CenterDot;,s_{t,L}^m]}^T,$ m＝1，…，M，

$s_{t,l}^m=H_{t,l}^m\&CenterDot;c_m\left[l\right],$ i＝1，…L，

${\mathrm{\&alpha;}}_m^q=\exp \left\{\mathrm{j\&pi;}(q-1)\sin \left({\mathrm{\&theta;}}_m\right)\right\}$

and

$G_t=\mathrm{Diag}\{\sqrt{{\mathrm{<figure-callout\; id="1"\; label="p\; t"\; filenames="C0312082000201.png,C0312082000202.png"\; state="\{\{state\}\}">p</figure-callout>}}_t^{}},\sqrt{{\mathrm{<figure-callout\; id="2"\; label="p\; t"\; filenames="C0312082000212.png,C0312082000222.png"\; state="\{\{state\}\}">p</figure-callout>}}_t^{}},\&CenterDot;\&CenterDot;\&CenterDot;,\sqrt{p_t^M}\},$ ${\stackrel{\&RightArrow;}{d}}_t={[d_t^{\left(1\right)},d_t^{\left(2\right)},\&CenterDot;\&CenterDot;\&CenterDot;{,d}_t^{\left(M\right)}]}^T$

wherein H_t，i ^mIs the frequency domain channel response, p, on the t-th user, the ith subcarrier_t ^mIs the received power of the mth user at time t. Thus, the frequency domain received signal at the q-th antenna element can be expressed as

${\stackrel{\&RightArrow;}{R}}_t^q=\mathrm{FFT}\&CenterDot;\mathrm{IFFT}\{A_t^q\&CenterDot;G_t\&CenterDot;{\stackrel{\&RightArrow;}{d}}_t\}+\mathrm{FFT}\left\{{\stackrel{\&RightArrow;}{n}}_t^q\right\}.---\left(3\right)$

We define

Spatial filter vector for mth user, which can be based on antenna array response of mth user

Given by wiener algorithm as

${\stackrel{\&RightArrow;}{W}}_m=\frac{R_{\mathrm{uu}}^{-1}{\stackrel{\&RightArrow;}{a}}_m}{{\stackrel{\&RightArrow;}{a}}_m^H{R}_{\mathrm{uu}}^{-1}{\stackrel{\&RightArrow;}{a}}_m}=\mathrm{\&kappa;}\&CenterDot;{\stackrel{\&RightArrow;}{a}}_m(\mathrm{\&kappa;}:\mathrm{cons}\tan t)---\left(4\right)$

Wherein R is_uuIs the covariance matrix of interference and noise for the mth user. Then the m-th user's beamforming and frequency domain matched filtering outputs are

$y_m={\stackrel{\&RightArrow;}{s}}_t^{m^T}\&CenterDot;\{{\stackrel{\&RightArrow;}{W}}_m\&CenterDot;{[{\stackrel{\&RightArrow;}{R}}_{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="C0312082000201.png,C0312082000202.png"\; state="\{\{state\}\}">t</figure-callout>}}^1,\&CenterDot;\&CenterDot;\&CenterDot;,{\stackrel{\&RightArrow;}{R}}_t^q,\&CenterDot;\&CenterDot;\&CenterDot;,{\stackrel{\&RightArrow;}{R}}_t^Q]}^T\}$

$={\stackrel{\&RightArrow;}{s}}_t^{m^T}\&CenterDot;\{{\stackrel{\&RightArrow;}{W}}_m\&CenterDot;(\left[\begin{array}{c}{A}_{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="C0312082000201.png,C0312082000202.png"\; state="\{\{state\}\}">t</figure-callout>}}^1\\ .\\ .\\ .\\ A_t^Q\end{array}\right]\&CircleTimes;G_t\&CenterDot;{\stackrel{\&RightArrow;}{d}}_t+\left[\begin{array}{c}{\stackrel{\&RightArrow;}{\mathrm{\&eta;}}}_{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="C0312082000201.png,C0312082000202.png"\; state="\{\{state\}\}">t</figure-callout>}}^1\\ .\\ .\\ .\\ {\stackrel{\&RightArrow;}{\mathrm{\&eta;}}}_t^Q\end{array}\right]\left)\right\}$

Wherein the spatial and random sequence correlation coefficients are respectively

${\mathrm{\&rho;}}_{m,m^{\&prime;}}^a={\stackrel{\&RightArrow;}{W}}_m\&CenterDot;{\stackrel{\&RightArrow;}{a}}_{m^{\&prime;}},$ ${\mathrm{\&rho;}}_{m,m^{\&prime;}}^s={\stackrel{\&RightArrow;}{s}}_t^{m^T}\&CenterDot;{\stackrel{\&RightArrow;}{s}}_t^{m^{\&prime;}}$ (1≤m≤M：1≤m′≤M)

${\stackrel{\&RightArrow;}{\mathrm{\&eta;}}}_t^q=\mathrm{FFT}\left\{{\stackrel{\&RightArrow;}{n}}_t^q\right\}$ Is another complex white Gaussian noise random process, and the variance thereof satisfies ${\mathrm{\&sigma;}}_{\mathrm{\&eta;}}^2={\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="C0312082000212.png,C0312082000222.png"\; state="\{\{state\}\}">n</figure-callout>}}^2.$ σ_n ²Is Gaussian noise n_t ^q(Q ═ 1, …, Q). Therefore, we can model the signal of the whole system as

Or be recorded as

$\stackrel{\&RightArrow;}{y}=H{G}_t{\stackrel{\&RightArrow;}{d}}_t+\stackrel{\&RightArrow;}{N}---\left(6\right)$

Where H is the correlation matrix between all users.Is colored complex Gaussian noise with a mean of zero and a variance of

$E(\stackrel{\&RightArrow;}{N}\&CenterDot;{\stackrel{\&RightArrow;}{N}}^H)=\frac{1}{\sqrt{Q}}H{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="C0312082000212.png,C0312082000222.png"\; state="\{\{state\}\}">n</figure-callout>}}^2.$

As can be seen from equation (6), when tens or even hundreds of users access in a sector, an iterative multi-user detection algorithm with an optimal MAP is not feasible at the base station.

As shown in FIG. 2, we present a block diagram of a Turbo space-time multi-user detection algorithm combining smart antennas and MAP iterative multi-user detection. The Turbo space-time multi-user receiver classifies all users into a plurality of groups and corresponding 'outer groups' according to spatial correlation, and before MAP iterative multi-user detection is carried out in each group, MAI of interference users of the 'outer groups' is eliminated. In the algorithm, a soft interference elimination method is adopted. Due to Turbo (iterative) processing within a group, extrinsic information output by channel MAP decoding of users within each group can be used to achieve soft interference cancellation outside the group to improve algorithm performance. Because if a hard interference cancellation method is used, errors in hard interference cancellation can result when the hard decision estimate is incorrect. In addition, in order to reduce the complexity of the intra-group MAP multi-user detection algorithm, we apply a simplified MAP multi-user detection algorithm based on soft sensitive bits as a sub-multi-user detection algorithm within the group. Specifically, the Turbo space-time multi-user detection algorithm is described as follows:

it is assumed that the base station can correctly estimate the antenna array response for each user. For simplifying the algorithm we take the spatial filtering weight vector as

${\stackrel{\&RightArrow;}{W}}_m=\frac{{\stackrel{\&RightArrow;}{a}}_m}{\left|\right|{\stackrel{\&RightArrow;}{a}}_m\left|\right|}=\mathrm{\&kappa;}\&CenterDot;{\stackrel{\&RightArrow;}{a}}_m,1\&le;m\&le;M---\left(7\right)$

Where "|" is defined as the modulus of the vector, $\mathrm{\&kappa;}=1/\sqrt{Q}.$ according to inter-user spatial filtering weightsCorrelation of coefficients, all M users are categorized into G groups. We define inter-user spatial filtering vectors

Has a correlation coefficient of

${\mathrm{\&beta;}}_{\mathrm{\&mu;},v}=\frac{\left|\right|{\stackrel{\&RightArrow;}{W}}_u^H\&CenterDot;{\stackrel{\&RightArrow;}{W}}_v\left|\right|}{\left|\right|{\stackrel{\&RightArrow;}{W}}_u\left|\right|\left|\right|{\stackrel{\&RightArrow;}{W}}_v\left|\right|}u=1,\&CenterDot;\&CenterDot;\&CenterDot;,M,v=1,\&CenterDot;\&CenterDot;\&CenterDot;,u-1.---\left(8\right)$

Where ". cndot.is defined as the dot product of two vectors, defining β₁And beta₂Is a threshold and has beta₁＞β₂. From the correlation coefficients, we group with the following classification criteria:

1. when beta is_μ，ν≥β₁And beta_ν，υ≥β₂When the user u, the user v and the user upsilon are distributed to be the same group omega_gAnd satisfies { u, v, upsilon }, e.g.. omega_g

2. When beta is_μ，ν≥β₁And beta_ν，υ≤β₂When we allocate user u and user v as group Ω_gA member of (1). Then { u, ν, upsilon }, belongs to omega_g，and $\left\{\mathrm{\&nu;}\right\}\&NotElement;{\mathrm{\&Omega;}}_g$ Where u-1 …, M, ν -1, … u, ν -2, …, ν -1, and define K_gIs the number of users of the g-th group. The total number of all users allocated to the G groups satisfies

$\underset{g=0}{\overset{G-1}{\mathrm{\&Sigma;}}}{K}_g=M.---\left(9\right)$

Due to the limited spatial filtering interference suppression, the out-of-group users can still generate severe interference to the in-group users. Therefore, we set the g-th group Ω_gThe outer users are classified as corresponding "outer group" omega_g ^INumber of users K_g ^ISatisfy the requirement of

$K_g^I=M-K_g.---\left(10\right)$

According to the above criteria, all users in a sector are assigned to several groups and corresponding "outer groups" according to their angle of incidence (DOA).

In order to reduce the influence of the inter-group multiple access interference MAI, a soft interference elimination method is adopted in the inter-group interference elimination method. If a group is considered to be a user,the method can be considered as a soft Parallel Interference Cancellation (PIC) method. Without loss of generality, we consider K as shown in FIG. 3_gThe g group omega of an active user_gAnd correspondingly have K_g ^I"outer group" omega of individual interfering users_g ^I. K of the g-th group according to equation (6)_gA Matched Filter (MF) output of

${\stackrel{\&RightArrow;}{y}}_g=H_g{G}_{g,t}{\stackrel{\&RightArrow;}{d}}_{g,t}+H_{\stackrel{\&OverBar;}{g}}{G}_{\stackrel{\&OverBar;}{g},t}{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g},t}+{\stackrel{\&RightArrow;}{N}}_g---\left(11\right)$

Wherein H_gGroup g Ω_gThe correlation matrix of the inner users, and

is group omega_gInner users and corresponding "outer groups" omega_g ^ICorrelation matrix between inner users.

And

are respectively defined as a group omega_gAnd outer group Ω_g ^IA vector of coded bits transmitted by a user. In addition, G_g，tAndare respectively defined as omega_gAnd Ω_g ^IAverage power of the users. In particular, we define

And the noise vector is

It is zero-mean colored complex Gaussian noise with variance of

$E({\stackrel{\&RightArrow;}{N}}_g\&CenterDot;{\stackrel{\&RightArrow;}{N}}_g^H)=\frac{1}{\sqrt{Q}}{H}_g{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="C0312082000212.png,C0312082000222.png"\; state="\{\{state\}\}">n</figure-callout>}}^2.---\left(15\right)$

The first term in formula (11) is Ω_gTarget signal of the user and the second term is from Ω_g ^IThe interference MAI. Thus, the interference cancellation operation is based on the "outer group" Ω_g ^ICoded bits for inter-user transmissionTo achieve interference cancellation, then interference cancellation is performedAfter disturbance have

${\stackrel{\&RightArrow;}{x}}_g^n=H_g{G}_{g,t}{\stackrel{\&RightArrow;}{d}}_{g,t}+H_{\stackrel{\&OverBar;}{g}}{G}_{\stackrel{\&OverBar;}{g},t}({\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g},t}-{\stackrel{\widetilde{\&OverBar;}}{d}}_{\stackrel{\&OverBar;}{g},t}^n)+{\stackrel{\&RightArrow;}{N}}_g$

$=H_g{G}_{g,t}{\stackrel{\&RightArrow;}{d}}_{g,t}+H_{\stackrel{\&OverBar;}{g}}{G}_{\stackrel{\&OverBar;}{g},t}\mathrm{\&Delta;}{\stackrel{\&RightArrow;}{d}}_{g,t}+{\stackrel{\&RightArrow;}{N}}_g$

$=H_g{G}_{g,t}{\stackrel{\&RightArrow;}{d}}_{g,t}+{\stackrel{\&OverBar;}{Z}}_g---\left(16\right)$

Wherein,is that

Is obtained from the feedback soft information given by the MAP channel decoding of each user for the nth iteration.Is an "outer group" omega_g ^IThe estimated error of the user's MAP channel decoding output. Assuming that the estimation error output by MAP channel decoding of each user is white Gaussian noise, the co-correlation matrix of the estimation error vector is

$E[\mathrm{\&Delta;}{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g}}\&CenterDot;\mathrm{\&Delta;}{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g}}^H]=\mathrm{diag}\left(\right[{\mathrm{\&sigma;}}_{e,1,\stackrel{\&OverBar;}{g}}^2,{\mathrm{\&sigma;}}_{e,2,\stackrel{\&OverBar;}{g}}^2,\&CenterDot;\&CenterDot;\&CenterDot;,{\mathrm{\&sigma;}}_{e,K_g^l,\stackrel{\&OverBar;}{g}}^2\left]\right).---\left(17\right)$

Thus, the total noise vector in equation (16)Is also Gaussian and has a covariance matrix

$E\left[{\stackrel{\&RightArrow;}{Z}}_g{\stackrel{\&RightArrow;}{Z}}_g^H\right]=H_{\stackrel{\&OverBar;}{g}}\&CenterDot;E[\mathrm{\&Delta;}{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g}}\&CenterDot;\mathrm{\&Delta;}{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g}}^H]\&CenterDot;H_{\stackrel{\&OverBar;}{g}}^H+\frac{1}{\sqrt{Q}}{H}_g{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="C0312082000212.png,C0312082000222.png"\; state="\{\{state\}\}">n</figure-callout>}}^2=R_{g,\stackrel{\&RightArrow;}{Z}\stackrel{\&RightArrow;}{Z}}---\left(18\right)$

Wherein, the total noise covariance matrix comprises two parts of residual interference and noise. It can be seen that if we can calculate per iteration Ω_g ^IError variance of channel decoding of kth user in the block

And variance σ of AWGN noise_n ²Then, the traditional optimal MAP iterative multi-user detection algorithm can be used as the sub multi-user detection algorithm in each group. We define the error variance of the channel decoded output as

${\mathrm{\&sigma;}}_{e,k,\stackrel{\&OverBar;}{g}}^2=E\left[{(d_{\stackrel{\&OverBar;}{g},t}^{\left(k\right)}-{\tilde{d}}_{\stackrel{\&OverBar;}{g},t}^{\left(k\right)})}^2\right]---\left(19\right)$

Wherein,is omega_g ^IThe coded bits actually transmitted by the k-th user, andis a soft estimate of the coded bits. In the case of an actual receiver, it is, is not possible to know. Therefore, we give an approximate error variance estimate

Wherein,

is a coded bitHard decision estimation of. WhileIs a soft estimate of the coded bits. The prior probability of the coded bit can be obtained according to the external information fed back by MAP channel decoding of each user $p(d_{\stackrel{\&OverBar;}{g},t}^{\left(k\right)}=\&PlusMinus;1).$ Then the process of the first step is carried out,

is composed ofExpected value of

It should be noted that the coded bits in the first iteration are without prior probability information, and therefore, let us

${\tilde{d}}_{\stackrel{\&OverBar;}{g},t}^{\left(k\right)}=\mathrm{real}\left(y_{m(k,\stackrel{\&OverBar;}{g})}\right)$ And

wherein, y_m(k，g)Is the received signal Matched Filtered (MF) output in equation (6), and m (k, g) indicates that the mth user is classified as the kth user in the gth "outer group".

Although the optimal MAP iterative multi-user detection algorithm can be applied as a sub multi-user detection algorithm in each group, the complexity of the algorithm is exponential to the number of users in the group. Therefore, when the number of users in the group is large, such as greater than 10 users, the optimal MAP iterative multi-user detection algorithm is not feasible. In the following, we will apply the simplified MAP multi-user detection algorithm based on soft sensitive bits as the sub-multi-user detection algorithm for each group.

The MAP iterative multi-user detection algorithm within each group is shown in fig. 3. Without loss of generality, we consider MAP iterative multi-user detection within the g-th group at time t. The coded bit d of the k user outputted by it_g，t ^(k)Has a posterior LOG likelihood of

${\mathrm{\&Lambda;}}_1\left(d_{g,t}^{\left(k\right)}\right)\stackrel{\mathrm{\&Delta;}}{=}\log \frac{P(d_{g,l}^{\left(k\right)}=+1|{\stackrel{\&RightArrow;}{x}}_g^n)}{P(d_{g,t}^{\left(k\right)}=-1|{\stackrel{\&RightArrow;}{x}}_g^n)}=\log \frac{p\left({\stackrel{\&RightArrow;}{x}}_g^n\right|d_{g,t}^{\left(k\right)}=+1)}{p\left({\stackrel{\&RightArrow;}{x}}_g^n\right|d_{g,t}^{\left(k\right)}=-1)}+\log \frac{p(d_{g,t}^{\left(k\right)}=+1)}{p(d_{g,t}^{\left(k\right)}=-1)},$

k＝1，…，K_g (23)

Wherein, the first term in equation (23) is defined as the extrinsic information (extrinsic information) given by the MAP multi-user detection, and is λ_1e ^k. The second term being a priori information, using λ_1o ^kTo indicate. They are obtained by channel decoding of the kth user of the last iteration. According to the equation (15),can be used for the conditional probability distribution of_gExpressed by a dimensional multivariate gaussian probability density function,

$p\left({\stackrel{\&RightArrow;}{x}}_g^n\right|{\stackrel{\&RightArrow;}{d}}_{g,t})=\frac{1}{\sqrt{{\left(2\mathrm{\&pi;}\right)}^{K_g}\det \left(R_{g,\stackrel{\&OverBar;}{Z}\stackrel{\&OverBar;}{Z}}\right)}}\exp [-\frac{1}{2}{({\stackrel{\&RightArrow;}{x}}_g^n-H_g{G}_{g,t}{\stackrel{\&RightArrow;}{d}}_{g,t})}^H{R}_{g,\stackrel{\&OverBar;}{Z}\stackrel{\&OverBar;}{Z}}^{-1}({\stackrel{\&RightArrow;}{x}}_g^n-H_g{G}_{g,t}{\stackrel{\&RightArrow;}{d}}_{g,t})]---\left(24\right)$

to calculate lambda_1e ^kNeeds x_g ⁿCoded bit d for the k-th user_g，t ^kJoint probability distribution of

$p({\stackrel{\&RightArrow;}{x}}_g^n,d_{g,t}^{\left(k\right)}=d)=\underset{{\stackrel{\&OverBar;}{d}}_{g,t};d_{g,t}^{\left(k\right)}=d}{\mathrm{\&Sigma;}}\mathrm{Pr}\left\{{\stackrel{\&RightArrow;}{x}}_g^n\right|{\stackrel{\&RightArrow;}{d}}_{g,t}\}\&CenterDot;\mathrm{Pr}\{{\stackrel{\&RightArrow;}{d}}_{g,t}\}---\left(25\right)$

Since the coded bits of different users are independent of each other, the conditional probability distribution of equation (25) can be written as

$p\left({\stackrel{\&RightArrow;}{x}}_g^n\right|d_{g,t}^{\left(k\right)}=d)=\frac{p({\stackrel{\&RightArrow;}{x}}_g^n,d_{g,t}^{\left(k\right)}=d)}{p(d_t^k=d)}=\underset{{\stackrel{\&RightArrow;}{d}}_t;d_t^k=d}{\mathrm{\&Sigma;}}\mathrm{Pr}\left\{{\stackrel{\&RightArrow;}{x}}_g^n\right|{\stackrel{\&RightArrow;}{d}}_{g,t}\}\&CenterDot;\underset{\underset{i\&NotEqual;k}{i=1}}{\overset{K_g}{\mathrm{\&Pi;}}}\mathrm{Pr}\{d_{g,t}^{\left(i\right)}\}.---\left(26\right)$

To simplify the complexity of the optimal MAP algorithm, we propose a simplified MAP multi-user detection algorithm based on sensitive bits. The basic idea of the MAP iterative multi-user detection algorithm based on sensitive bits is that we first distinguish the sensitive bits, which can obtain the "coarse" prior information. It gives information whether each coded bit is likely to estimate a pair or an error. With this a priori information, we can perform MAP detection in a small subset of corresponding resolved sensitive bits. Defining a likelihood metric as

$\mathrm{\&Psi;}\left({\stackrel{\&RightArrow;}{d}}_{g,t}\right)={({\stackrel{\&RightArrow;}{x}}_g^n-H_g{G}_{g,t}{\stackrel{\&RightArrow;}{d}}_{g,t})}^H{R}_{g,\stackrel{\&OverBar;}{Z}\stackrel{\&OverBar;}{Z}}^{-1}({\stackrel{\&RightArrow;}{x}}_g^n-H_g{G}_{g,t}{\stackrel{\&RightArrow;}{d}}_{g,t})---\left(27\right)$

And let

Wherein,

is an estimated multi-user coded bit vector. In addition, letIs defined as a new bit vector corresponding to the inversion

One and only one bit (and-1 → 1 or 1 → -1). We have demonstrated that when estimating a bit vectorWhen there are one or more bit errors, and we invertThe polarity of the erroneous bit in (1) is obtained

Then there is

Put another way, when

The larger the value of (d) the more likely the corresponding adjusted bit is to estimate an error, i.e., a sensitive bit. Generally, the number of erroneous bits in the estimated coded bit vector is small. For example, if the bit error rate of the coded bits is 10^-2This means that on average one bit out of every hundred bits is erroneous, and therefore the number of sensitive bits is not typically very large. We resolve the sensitive bits by: first we estimate the original individual user code bits using conventional single-user Matched Filtering (MF). The sensitive bits are then resolved according to inequality (29). We define the number of sensitive bits in the g-th group as f, at all K_gMetric of new adjusted bit vector

Middle search f (f < K)_g) The largest metric value. Sensitive bits are defined as bits that correspond to the adjustment in the f coded bit vectors. To implement the iterative MAP multi-user detection algorithm, we assume a priori probabilities of sensitive bits in the first iteration

Are equi-probability distributions. And the a priori probability of the non-sensitive bits is 1 because these bits are assumed to be correctly estimated. The MAP multi-user detection algorithm may consider only 2 corresponding to the f-sensitive bit^fAll possible coded bit vectors(where the phenanthrene sensitive bits in these vectors remain unchanged from the initial estimate). The calculation of the conditional probability of equation (26) can be simplified to the conventional optimal MAP criterion

$p\left({\stackrel{\&RightArrow;}{y}}_t\right|d_t^{\left(k\right)}=d)\&ap;\underset{{\tilde{d}}_t\&Element;{\left\{{\stackrel{\&RightArrow;}{d}}_t^s\right\}}_{s=\mathrm{1,}\&CenterDot;\&CenterDot;\&CenterDot;,2}f;d_t^{\left(k\right)}=d}{\mathrm{\&Sigma;}}\mathrm{Pr}\left\{{\stackrel{\&RightArrow;}{y}}_t\right|{\stackrel{\&RightArrow;}{d}}_t\}\&CenterDot;\underset{\underset{i\&NotEqual;k}{i=1}}{\overset{K}{\mathrm{\&Pi;}}}\mathrm{Pr}\{d_t^{\left(i\right)}\},k=1,\&CenterDot;\&CenterDot;\&CenterDot;K---\left(30\right)$

Of which only 2 is considered^f(2 when the kth coded bit is a sensitive bit^f-1) One significant coded bit vector, and the other 2^kg-f-1The vectors are insignificant vectors of coded bits, which can be ignored in calculating equation (30) because the equation depends primarily on 2^fAn importance vector.

Based on the above analysis, we initialize a priori probabilities of the coded bits for MAP multiuser detection at the first iteration of

At the next iteration, the external information lambda fed back is decoded according to the MAP channel of each user_2e ^kThe MAP multi-user detection module can obtain more accurate prior probability and hard decision of transmission coding bit

Wherein, ${\mathrm{\&lambda;}}_{10}^k={\mathrm{\&lambda;}}_{2e}^k---\left(32\right)$

hard decision following transmission of coded bits

The sensitive bits will also be readjusted. Note that, unlike the first iteration, the code bit prior probability Pr { d ] for MAP multiuser detection is now the same_g，t ^(k)Is as

$\mathrm{Pr}\{d_{g,t}^{\left(k\right)}=d\}=\frac{\exp (d\&CenterDot;{\mathrm{\&lambda;}}_{1o}^k)}{1+\exp (d\&CenterDot;{\mathrm{\&lambda;}}_{1o}^k)}.---\left(33\right)$

Therefore, soft coding ratio estimation for the kth user in the kth group at time t

Can be obtained from the formulae (21) and (33).

In the g-th group, K follows the MAP multiuser detection module_gAnd (3) decoding the channel of each user, applying an MAP algorithm to give the posterior probability of the coded bit and the posterior probability of the information bit in the last iteration. Suppose we use a convolutional code with a code rate of R1/n, every n coded bits d_g，t ^(k)Corresponding to a pre-coding information bit b_g，j ^(k). The n channel bits are defined as $(d_{g,t}^{\left(k\right)},\&CenterDot;\&CenterDot;\&CenterDot;,d_{g,t+n-1}^{\left(k\right)})={\underset{\&OverBar;}{d}}_{g,j}^{\left(k\right)}.$ Therefore, we have

$\mathrm{Pr}\{d_{g,t^{\&prime;}}^{\left(k\right)}=d|{\underset{\&OverBar;}{x}}_g^{\left(K\right)}\}=\underset{m^{\&prime;}}{\mathrm{\&Sigma;}}\underset{{\underset{\&OverBar;}{d}}_{g,j}^k;d_{g,t^{\&prime;}}^k=d}{\mathrm{\&Sigma;}}\mathrm{Pr}\{-S_{j-1}=-m^{\&prime;};-{\underset{\&OverBar;}{d}}_{g,j}^{\left(k\right)}\left|{\underset{\&OverBar;}{x}}_g^{\left(k\right)}\right\}---\left(34\right)$

Wherein x is_g ^(k)Is the received signal sequence of the kth user in the g-th group. S_jIs the state at time j and m' covers all possible states. It is noted that equation (34) may be implemented using either an existing MAP channel decoding algorithm or a simplified log-MAP channel decoding algorithm. With the extrinsic information from the front-end MAP multi-user detection output, the branch metric between channel decoding MAP trellis decoding states is

${\mathrm{\&gamma;}}_j(m^{\&prime;},m)=\mathrm{Pr}\{S_j=m|S_{j-1}=m^{\&prime;}\left\}\underset{t^{\&prime;}=t}{\overset{t+n-1}{\mathrm{\&Pi;}}}\mathrm{Pr}\right\{{\stackrel{\&RightArrow;}{x}}_{g,t^{\&prime;}}\left|d_{g,t^{\&prime;}}^{\left(k\right)}\right\}---\left(35\right)$

Thus, the posterior LOG Likelihood (LLR) of the kth user is

${\mathrm{\&Lambda;}}_2\stackrel{\mathrm{\&Delta;}}{=}\log \frac{\mathrm{Pr}\{d_{g,t^{\&prime;}}^{\left(k\right)}=1|{\underset{\&OverBar;}{x}}_g^{\left(k\right)}\}}{\mathrm{Pr}\{d_{g,t^{\&prime;}}^{\left(k\right)}=-1|{\underset{\&OverBar;}{x}}_g^{\left(k\right)}\}}\&ap;{\mathrm{\&lambda;}}_{2e}^k+{\mathrm{\&lambda;}}_{1e}^k---\left(36\right)$

Wherein, the user decodes the output extrinsic information into ${\mathrm{\&lambda;}}_{2e}^k={\mathrm{\&Lambda;}}_2-{\mathrm{\&lambda;}}_{1e}^k.$ This information is fed back to the MAP multiuser detection module, and the a priori information for improved MAP multiuser detection can be obtained by equation (33).

The Turbo spatio-temporal multi-user detection algorithm is briefly summarized below. Let f_maxThe maximum number of sensitive bits defined for each group and I is defined as the maximum number of iterations. The Turbo space-time multi-user detection algorithm based on the soft sensitive bit algorithm can be described as follows:initialization: obtaining spatial filtering weight vector of each user according to equation (7)All M users are then categorized by grouping criteria into G groups and corresponding "outer groups". At the receiving end, after beamforming and conventional matched filtering, the initial hard estimation and soft estimation of the coded bits of all M users can be obtained as

And ${\tilde{d}}_{\stackrel{\&OverBar;}{g},t}^{\left(k\right)}=\mathrm{real}\left(y_{m(k,\stackrel{\&OverBar;}{g})}\right)$

and assign it to the G group and corresponding "outer group" overlap process: forn is 1 to I

The first step is as follows: (Soft interference cancellation) to obtain the g-th "outer group" omega_g ^ISoft estimation of coded bits

By cancelling to Ω_g ^IMAI, soft interference cancellation expressed as

${\stackrel{\&OverBar;}{x}}_g^n=H_g{G}_{g,t}{\stackrel{\&RightArrow;}{d}}_{g,t}+H_{\stackrel{\&OverBar;}{g}}{G}_{\stackrel{\&OverBar;}{g},t}({\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g},t}-{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g},t}^n)+{\stackrel{\&RightArrow;}{N}}_g$

$=H_g{G}_{g,t}{\stackrel{\&RightArrow;}{d}}_{g,t}+{\stackrel{\&RightArrow;}{Z}}_g$

WhereinOf the co-correlation matrix

Is composed of

$E\left[{\stackrel{\&RightArrow;}{Z}}_g{\stackrel{\&RightArrow;}{Z}}_g^H\right]=H_{\stackrel{\&OverBar;}{g}}\&CenterDot;E[\mathrm{\&Delta;}{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g}}\&CenterDot;\mathrm{\&Delta;}{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g}}^H]\&CenterDot;H_{\stackrel{\&OverBar;}{g}}^H+\frac{1}{\sqrt{Q}}{H}_g{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="C0312082000212.png,C0312082000222.png"\; state="\{\{state\}\}">n</figure-callout>}}^2$

The second step is that: (simplified MAP iterative multiuser detection) i) finding f sensitive bits according to the sensitive bit algorithm. Based on these sensitive bits, the extrinsic information of the kth user in the g-th group output by the simplified MAP multi-user detection algorithm is

${\mathrm{\&lambda;}}_{1e}^k=\log \frac{p\left({\stackrel{\&RightArrow;}{x}}_g^n\right|d_{g,t}^{\left(k\right)}=+1)}{p\left({\stackrel{\&RightArrow;}{x}}_g^n\right|d_{g,t}^{\left(k\right)}=-1)}$

Wherein,

$p\left({\stackrel{\&RightArrow;}{x}}_g^n\right|d_{g,t}^{\left(k\right)}=d)\&ap;\underset{{\stackrel{\&RightArrow;}{d}}_{g,t}\&Element;{\left\{{\stackrel{\&RightArrow;}{d}}_{g,t}^s\right\}}_{s=1,\&CenterDot;\&CenterDot;\&CenterDot;,2}f;d_{g,t}^{\left(k\right)}=d}{\mathrm{\&Sigma;}}\mathrm{Pr}\left\{{\stackrel{\&RightArrow;}{x}}_g^n\right|{\stackrel{\&RightArrow;}{d}}_{g,t}\}\&CenterDot;\underset{\underset{i\&NotEqual;k}{i=1}}{\overset{K_g}{\mathrm{\&Pi;}}}\mathrm{Pr}\{d_{g,t}^{\left(i\right)}\}.$

note that at the first iteration, the prior probability of the coded bits Pr { d }_g，t ^(k)The result is obtained by equation (31). And at the next iteration, the prior probability Pr { d_g，t ^(k)Is given by formula (33). ii) obtaining extrinsic information λ of MAP multi-user detection output_1e ^kThen, according to the formula (36) Outer information lambda of k-th user channel decoding_2e ^k. Then, the extrinsic information λ is processed_2e ^kAnd feeding back to the MAP multi-user detection module. When I is equal to I, the posterior LOG likelihood of the information bit is calculated, thereby decoding the received bit. The algorithm is ended.

iii) improved a priori information on the coded bits is obtained based on extrinsic information fed back from each user channel decoding (and ${\mathrm{\&lambda;}}_{1o}^k={\mathrm{\&lambda;}}_{2e}^k$ ) Thereby, more accurate hard and soft estimates of the user coded bits may be obtained.

And

the third step: obtain all M users improved hard and soft estimates, return to the first step.

This section presents simulation results and performance comparisons for the packet multiuser detection algorithm we propose under AWGN channel and frequency selective fading channel. In the simulation experiment, all users adopt the same convolutional code with code rate of 1/2. We use two convolutional codes: a convolutional code with constraint length of 5, octal generation factor of (23, 35) and constraint length of 3, octal generation factor of (5, 7). Each block of information bits is 128 long and a random interleaving method is used. All users equal the transmission power (and G ═ I). And sets a threshold beta for the grouping criterion₁And beta₂0.9 and 0.95, respectively. Assuming that the receiving end knows the noise variance σ_n ²And spreading sequences for each user. Finally, the signal-to-noise ratio is definedFor the ratio of information bit power to noise power, (AqBmIn) is defined in the simulation as q receive antennas, m sensitive bits and n iterations. It should be noted that n-1 means that there is no feedback information to improve system performance.

Fig. 4 shows the simulation performance of the Turbo multi-user detection algorithm based on the sensitive bit algorithm in the single-antenna coded multi-carrier CDMA system, where the number M of users in the system is 10 and the length L of PN sequence is 15. We use a convolutional code with a generation factor of (23, 35). It can be seen from the figure that the simplified MAP iterative multi-user detection algorithm works effectively even if the number of sensitive bits is much smaller than the total number of users, and when the number of sensitive bits f is 3 and the number of iterations n is 3, the performance of the simplified MAP iterative multi-user detection algorithm and the performance of the single-user coding system are 10 BER^-4Only 0.15dB difference. In addition, the complexity of the simplified MAP multi-user detection algorithm is reduced by 0 (K) of the optimal algorithm₁2^K1) Reduced to 0 ((K)₁-f/2)2^f) In which K is₁M, and only one. Specifically, when K₁When 10 and f is 3, the algorithm complexity is reduced from 0(10240) to 0 (78). Therefore, the simplified MAP iterative multi-user detection algorithm can be used as a sub multi-user detection algorithm in each group in Turbo space-time multi-user detection.

In the following simulation, we show the simulation performance of the Turbo space-time MUD algorithm in the coded multi-carrier CDMA system, where the system has 20 users and 3 receiving antennas. In addition, we use a low number of states (5, 7) convolutional code in order to reduce the simulation time. The size of the sector is 2 pi/3, the DOA incidence directions of each user are randomly distributed within (pi/6) < theta < (5 pi/6), and it is assumed that the base station can ideally estimate the DOA of the user. Finally, we set the maximum sensitive bit number f of the simplified Turbo multi-user detection algorithm in each group_max＝3。

The performance of the Turbo space-time multi-user detection algorithm in case of equal grouping under AWGN channel is given in fig. 5. All users are evenly divided into 4 groups with K _g5 and g 1, …, 4. And the minimum included angle of the DOAs of the users between different groups is 15 deg. in consideration of the constraint. This can be ensured by the management software of the base stationAnd switching interfering users not in the group to other time slots or frequency domain channels. Such spatial constraints ensure a partial spatial segmentation of users between groups. Therefore, the same PN sequence can be used repeatedly in different groups, with a PN sequence length of L-7 (L > K)_g). For comparison, we also show the performance of single-user coded MC-CDMA systems when using single and multiple antenna arrays. It can be seen from the figure that in the single-user case, a performance gain of 5dB can be obtained using the antenna array beamforming technique. Meanwhile, it can be seen that the Turbo space-time multi-user algorithm can approach the performance of multi-antenna single-user under a few iteration times (n is 3).

In fig. 6, we consider the case of equal cross-correlation between users and set the cross-correlation coefficient defined in equation (5) ${\mathrm{\&rho;}}_{m,m^{\&prime;}}^s=0.4,$ M is more than or equal to 1 and m' is more than or equal to 20. There is no spatial constraint for inter-group users at this point, and all users are randomly distributed within a sector. Simulation results show that the proposed Turbo space-time multi-user detection algorithm can obtain the performance of a multi-carrier CDMA system with almost single-user multi-antenna coding when the SNR is larger than-1 dB.

The performance of the Turbo space-time multi-user detection algorithm in frequency domain selective fading channel is given in figure 7. It can be seen that the present algorithm can achieve ideal results in a reduced number of iterations in a fading channel than in an AWGN channel. For example, with m-3 and n-2 (only one iteration), the proposed algorithm can approach the performance of a single user in a fading channel.

It can be seen that in a practical system, the algorithm we propose can still be implemented even when a large number of users (tens or even hundreds of users) access the base station simultaneously in a 120 ° sector. At this time, the algorithm has the algorithm complexity which is linear relation with the number of users

$O(\underset{g=1}{\overset{G}{\mathrm{\&Sigma;}}}(K_g-f/2)\&CenterDot;2^f),$ f≤f_max

The soft sensitive bit and space packet space-time iteration (Turbo) based multi-user detection algorithm can be applied to a multi-user access system coded in broadband wireless communication. The space-time iterative multi-user detection algorithm can approach the performance of a single-user multi-antenna coding system under a few iteration times under an AWGN channel and a frequency-selective fading channel when twenty users are accessed simultaneously. The algorithm complexity of the space-time iterative multi-user detection algorithm is in a linear relation with the number of users, and the MAP multi-user detection algorithm is proposed to be possible in practical application. The method can be applied to CDMA and SDM (Space Division Multiplexing) SDMA systems with codes.

Claims (1)

1. A space-time iterative multi-user detection method based on soft sensitive bits and space grouping comprises a maximum posterior probability (MAP iterative multi-user detection algorithm) combined with an intelligent antenna, and is characterized in that: classifying all users into a plurality of groups and corresponding groups except the groups, namely 'outer groups', according to spatial correlation, and then using a simplified MAP iterative multi-user detection algorithm based on soft sensitive bits for sub-multi-users in each group, namely firstly distinguishing the sensitive bits to obtain prior information, then carrying out MAP detection in a small subset corresponding to the distinguished sensitive bits, and then using outer information output by channel MAP decoding of users in each group after iterative processing to realize soft interference elimination outside the groups; it comprises the following steps in sequence:

1) receiving multiple users by using a multi-antenna matrix, setting M user access signals, and performing spatial filtering and frequency domain matching filtering according to the incident direction of user signals; according to the correlation of spatial filtering weight coefficients among users, all M users are classified and divided into G groups and corresponding 'outer groups'; namely: obtaining a spatial filtering weight vector of each user according to the following formula

All M users are classified into G groups and corresponding 'outer groups' by grouping criteria:

${\stackrel{\&RightArrow;}{W}}_m=\frac{{\stackrel{\&RightArrow;}{a}}_m}{\left|\right|{\stackrel{\&RightArrow;}{a}}_m\left|\right|}=k\&CenterDot;{\stackrel{\&RightArrow;}{a}}_m,$ 1≤m≤M $k=\frac{1}{\sqrt{Q}}$

wherein: q is the number of antenna elements;

${\stackrel{\&RightArrow;}{a}}_m={[1,e^{-\mathrm{j\&pi;}\sin Q_m},.......,e^{-j(Q-1)\mathrm{\&pi;}\sin {\mathrm{\&theta;}}_m}]}^T$ for the mth user at an incident angle of Q_mA lower antenna array response; the grouping criterion is as follows:

1.1) when beta_μ，ν≥β₁And beta is_ν，μ≥β₂When the users are distributed to the same group omega_gI.e. { μ, ν, ν }. epsilon. OMEGA_g(ii) a Wherein, mu is 1, 1.. mu, v is 1, 2.. mu, v-1, beta₁，β₂Is a set threshold, and₁＞β₂，β_μ，νfor spatial filtering vectors W between users mu, v_μ， Such as:

${\mathrm{\&beta;}}_{\mathrm{\&mu;},\mathrm{\&nu;}}=\frac{\left|\right|{{\stackrel{\&RightArrow;}{W}}_{\mathrm{\&mu;}}}^H\&CenterDot;{\stackrel{\&RightArrow;}{W}}_{\mathrm{\&nu;}}\left|\right|}{\left|\right|W_{\mathrm{\&mu;}}\left|\right|\left|\right|W_{\mathrm{\&nu;}}\left|\right|}$

superscript (·)^HFor conjugate transpose, "·" denotes vector dot product;

1.2) when beta_μ，ν≥β₁And beta is_ν，ν≤β₂And then distributing the users mu and v to be the same group omega_gIs { mu, v }. epsilon.omega_gAnd is $\left\{\mathrm{\&nu;}\right\}\&NotElement;{\mathrm{\&Omega;}}_g,$

Thus, the total number of all users assigned to the G groups satisfies $\underset{g=0}{\overset{G=1}{\mathrm{\&Sigma;}}}{k}_g=M,$ Wherein k is_gThe total number of users in the g-th group, and the 'outer group' is the g-th group omega_gFor other users, using Ω_g ^IIndicates the number of users $k_g^I=M-k_g,k_g^I,$ Represents omega_g ^IThe total number of users;

2) calculating initial hard and soft estimates of the coded bits for all M users and assigning truth values to G groups and corresponding "outer groups";

after the receiving end is processed by beam forming and traditional matched filtering, the initial hard estimation and soft estimation of the coded bits of all M users are obtained, and the initial hard estimation and the soft estimation sequentially and respectively comprise:

${\hat{d}}_{\stackrel{\&OverBar;}{g},t}^{\left(R\right)}=\mathrm{sign}\left(\mathrm{real}\left(y_{m(k,\stackrel{\&OverBar;}{g})}\right)\right),$

${\tilde{d}}_{\stackrel{\&OverBar;}{g},t}^{\left(k\right)}=\mathrm{real}\left(y_{m(k,\stackrel{\&OverBar;}{g})}\right);$

y_{m(k， g)} is the received signal Matched Filtered (MF) output of the kth user classified as the g "outer group" by the mth user;

3) defining the maximum sensitive bit number and the maximum iteration number of each group;

3.1) according to the following inequality

To resolve estimated multi-user coded bit vectors

Likelihood metric of (d) and the original individual user code bits of a conventional single-user Matched Filter (MF) estimate

The upper limit of the likelihood metric difference value of (c),

the larger the value of (d), the more likely the corresponding adjusted bit is to estimate an error, i.e., a sensitive bit;

3.2) then k_gAn

Middle search f (f < k)_g) F is the number of sensitive bits in the g group;

4) a simplified iterative MAP multi-user detection algorithm based on soft-sensitive bits is performed in the g-th group, which in turn has the following steps:

4.1) realizing soft interference elimination according to the 'outer group' soft estimation;

according to "Outer group omega_g ^ICoded bits for inter-user transmission

To achieve soft interference cancellation: the above soft estimation

${\tilde{d}}_{\stackrel{\&OverBar;}{g},k}^{\left(k\right)}=\mathrm{real}\left(y_{m(k,\stackrel{\&OverBar;}{g})}\right);$

After the interference is eliminated, there is ${{\stackrel{\&RightArrow;}{X}}_g}^n=H_g{G}_{g,t}{\stackrel{\&RightArrow;}{d}}_{g,t}+{\stackrel{\&RightArrow;}{Z}}_g,$ While

Is self-correlation matrix ofIs composed of

$E\left[{\stackrel{\&RightArrow;}{Z}}_g{{\stackrel{\&RightArrow;}{Z}}_g}^H\right]=H_{\stackrel{\&OverBar;}{g}}\&CenterDot;E[\mathrm{\&Delta;}{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g}}\&CenterDot;\mathrm{\&Delta;}{{\stackrel{\&RightArrow;}{d}}_{\stackrel{\&OverBar;}{g}}}^H]\&CenterDot;{H_{\stackrel{\&OverBar;}{g}}}^H+\frac{1}{\sqrt{Q}}{H}_g{{\mathrm{\&sigma;}}_n}^2;$

Wherein H_gIs the g-th group omega_gInner users and correlation matrix, H _g Is group omega_gInner users and corresponding "outer groups" omega_g ^ICorrelation matrix between inner users, G_g，tIs an internal omega of the group_gThe average power of the users is then calculated,is a group omega_gThe vector of coded bits transmitted by the users,

is an outer group omega_g ^ICoded bits for inter-user transmission

The soft-state estimation of (a) is performed,

is an "outer group" omega_g ^IEstimation error, σ, of MAP channel decoded output of a user_n ²Is the variance of additive white gaussian noise, AWGN;

4.2) finding out f sensitive bits by using a sensitive bit algorithm, and calculating the output extrinsic information of each user in the g group by using a simplified MAP multi-user detection algorithm according to the sensitive bits;

when we want to beWhen the prior probability of the sensitive bits is equal probability distribution and the prior probability of the non-sensitive bits is 1 in one iteration, the MAP multi-user detection algorithm can only consider 2 corresponding to f sensitive bits^fA vector of coded bits

4.2.1) computing extrinsic information λ of the kth user in the g-th group_1e ^k；

4.2.2) calculating the posterior LOG Likelihood Ratio (LLR) of the kth user by using MAP algorithm and using Lambda₂Indicating that the user decodes the output extrinsic information into ${\mathrm{\&lambda;}}_{2e}^k={\mathrm{\&Lambda;}}_2-{\mathrm{\&lambda;}}_{1e}^k;$

4.2.3) treatment of Lambda_2e ^kFeeding back to MAP multi-user detection module, calculating posterior LOG likelihood of information bit when iteration is finished, so as to decode received bit;

4.2.4) obtaining improved prior information of the coding bits according to the external information fed back by each user channel decoding, namely: ${\mathrm{\&lambda;}}_{1\mathrm{\&sigma;}}^k={\mathrm{\&lambda;}}_{2e}^k,$ thereby obtaining more accurate hard and soft estimates of the user coded bits;

4.3) judging whether the iteration is finished:

if not, each user MAP channel decoder calculates the extrinsic information of the coded bit and returns to the step (2);

if the calculation is finished, namely G groups are calculated, the MAP channel decoder of each user calculates the extrinsic information of the information bit to be output as a multi-user detection signal;

5) and (4) obtaining improved hard estimation and soft estimation of all M users, and returning to the step (3).

Images