CN101989864B - Blind multi-access interference suppression method for DS/CDMA system - Google Patents

Abstract

The invention relates to a signal interference processing method in the technical field of communication, in particular to a blind multi-access interference suppression method for a DS/CDMA (direct sequence/ code division multiple access) system. The method comprises the following steps of: establishing a convex set containing an optimal interference suppression filter coefficient vector in each iteration, approximating the projection of the current filter coefficient vector to the convex set by adopting the projection of a closed half plane containing the convex set, then updating the interference suppression filter coefficient vector, and updating an expansion coefficient in next iteration by adopting an adaptive adjusting strategy. Proved by experiments, the method can effectively inhibit multi-access interference in the DS/CDMA system in different noise, and has quick astringency, good convergence property in a stable state and low calculation and implementation complexity.

Description

Blind MAI suppression method in a kind of DS/CDMA system

Technical field

The present invention relates to the processing method that signal in the communication technical field disturbs, particularly relate to the blind MAI suppression method in a kind of DS/CDMA system.

Background technology

In direct sequence CDMA (DS/CDMA) communication system, each user has the spreading code of oneself, and is relevant by input signal and prior known user's spreading code are done at receiving terminal, thus the information of recovering.An its significant advantage be have a plurality of users of different spreading codes can be on same channel simultaneous transmission of signals, in other words, can share time and frequency resource and need not complicated frequency distribute and administrative mechanism.This system also has the characteristics such as the interference of anti-arrowband, anti-multipath fading and good confidentiality in addition.But also exist some problems simultaneously, except there was noise in receiving terminal, the factors such as propagation delay and received power caused the incomplete quadrature of the spreading code of different user easily, thereby made cdma receiver usually be subject to coming from the impact of other user's transmitted signal.Even if the cross correlation value impact is so unobvious, traditional matched filter also can be weaker than the interference user signal because of the signal of desired user and can't recover to obtain expectation information, this phenomenon is referred to as multiple access and disturbs (MAI), and the existence that multiple access disturbs has reduced power system capacity.Therefore, at the receiving terminal MAI that must take measures to suppress.

In the DS/CDMA of high-throughput communication system, usually adopt the blind MAI suppression method.Since need not training sequence, more and more extensive for research and the application of blind MAI suppression method.Document " Generalized projection algorithm for blind interferencesuppression in DS/CDMA communications " (IEEE Transactions on Circuits and Systems II for example, vol.44, no.6, pp.271-275) in, proposed a kind of simple set theory method and be referred to as space-alternating generalized projection (SAGP) method, suppressed multiple access with it and disturb.The performance of SAGP method under stable state is fine, and be slow but weak point is convergence rate.Document " Constrained normalizedadaptive filters for CDMA mobile communications " (EUSIPCO-European signal processing conference, band limit normalization minimum mean-square error (CNLMS) method has been proposed pp.1-5), restrictive condition in the Adaptive Multiple User Detection is joined in the renewal direction of interference suppression filter, but the CNLMS method is owing to only consider data in each iteration, thereby can not produce enough fast convergence rate.

In recent years, the subgradient projection method is extensively concerned in the research fields such as image recovery, Adaptive Signal Processing, Acoustic Echo Cancellation and digital deaf-aid.The method can be controlled the factors such as rate of convergence, operand and systematic function effectively, so that the overall performance of the system of institute's application promotes greatly.In cdma system, document " Efficient blind MAI suppression in DS/CDMA systems by embeddedconstraint parallel projection techniques " (IEICE Transactions on Fundamental, vol.E88-A, no.8, pp.2062-2071) a kind of parallel projection multiple access interference cancellation method (ECPP) has been proposed, method before comparing, the method improves in convergence rate and the convergence of system, but still exist two limitation: the one, need to be to the convex set of each parallel processing and the again weighting of common factor projection of limiting set during each iteration, because this common factor is compared more strict with the convex set of parallel processing, affect to a certain extent the constringency performance of the method, caused the computation complexity of method higher.The 2nd, the coefficient of expansion that affects system output signal under system's convergence rate and the stable state-interference plus noise ratio (SINR) immobilizes in the iteration renewal process, can't regulate according to concrete iteration situation self adaptation, thereby can't guarantee under any circumstance to obtain optimum systematic function.

Summary of the invention

Purpose of the present invention just is to address the deficiencies of the prior art, and designs the blind MAI suppression method in a kind of DS/CDMA system.

Technical scheme of the present invention is:

Blind MAI suppression method in a kind of DS/CDMA system, it may further comprise the steps:

(1) initialization:

In initialization procedure, set the initial value of each parameter: iterations n=1; ${\tilde{A}}_{\mathrm{1,0}}=0,$ h ₁=s ₁, q, w _i, γ regulates relevant parameter ρ with the coefficient of expansion _Start, ρ _Stop, Δ, wherein ρ _StopMust be less than ρ _Start, total iterations N;

(2) set up convex set C _ρ ⁽ⁿ⁾[i] and corresponding convex function g _i(h _n):

(2-1) adopt following more new formula to estimate A ₁And b ₁[i]:

Here h _nIt is the coefficient vector of the n time interference suppression filter in the iteration

R[i] be the data sequence that receives,

With

Be respectively amplitude A in the n time iteration ₁Bit b with i transmission ₁The estimated value of [i], and γ ∈ (0,1] be forgetting factor; The sgn function definition is:

If i.e. a＞0, sgn a=1, otherwise

(2-2) introduce following convex set, this convex set comprises the optimal value h of interference suppression filter coefficient _Opt:

C in the following formula _ρ ⁽ⁿ⁾[i] is convex set, I _nBe the control sequence that contains q element, q is the number of the parallel processor of each iteration participation, and ρ is the coefficient of expansion.Corresponding with this convex set so convex function g _i(h _n) be:

$g_i\left(h_n\right)={(<h_n,r\left[i\right]>-{\tilde{A}}_{1,n}{\tilde{b}}_{1,n}\left[i\right])}^2-\mathrm{\&rho;}$

(3) calculate h _nTo convex set C _ρ ⁽ⁿ⁾The projection of [i]

Employing is to comprising convex set C _ρ ⁽ⁿ⁾The closed half-plane H of [i] _i(h _n) projection

Come close approximation

Namely $P_{C_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]}\left(h_n\right)\&cong;P_{H_i\left(h_n\right)}\left(h_n\right).$

H wherein _i(h _n) expression formula be:

▽ g in the following formula _i(h _n) be g _i(h _n) subgradient, $\&dtri;g_i\left(h_n\right)=2r\left[i\right](<h_n,r\left[i\right]>-{\tilde{A}}_{1,n+1}{\tilde{b}}_{1,n}\left[i\right])$

To close primitive formula as follows:

$P_{H_i\left(h_n\right)}\left(h_n\right)=\left\{\begin{array}{cc}{h}_n-\frac{g_i\left(h_n\right)}{\&dtri;g_i\left(h_n\right)},& h_n\&NotElement;H_i\left(h_n\right)\\ h_n,& h_n\&Element;H_i\left(h_n\right)\end{array}\right.$

(4) upgrade the interference suppression filter coefficient vector:

The iteration renewal process expression formula of interference suppression filter coefficient vector is as follows:

$h_{n+1}=P_{C_s}(h_n+{\mathrm{\&lambda;}}_n(\underset{i\&Element;I_n}{\mathrm{\&Sigma;}}{w}_i^{\left(n\right)}{P}_{C_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]}\left(h_n\right)-h_n))---\left(11\right)$

H in the following formula _nAnd h _N+1Represent respectively the n time and the filter coefficient vector during the n+1 time iteration.w _i ⁽ⁿ⁾Be h _nAt different convex set C _ρ ⁽ⁿ⁾Projection on [i]

The weight of giving, in the method, when each iteration is upgraded weight remain unchanged ( $w_i=w_i^{\left(n\right)}$ )。

To limiting set C _sOn projection.λ _nBe relaxation factor, its span is λ _n∈ [0,2M _n], in this scope, randomly draw acquisition in equiprobable mode when upgrading at every turn.M _nExpression formula as follows:

$M_n=\left\{\begin{array}{cc}\frac{\underset{i\&Element;I_n}{\mathrm{\&Sigma;}}{w}_i{\left|\right|P_{C_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]}\left(h_n\right)-h_n\left|\right|}^2}{{\left|\right|\underset{i\&Element;I_n}{\mathrm{\&Sigma;}}{w}_i{P}_{C_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]}\left(h_n\right)-h_n\left|\right|}^2}& h_n\&NotElement;{\&cap;}_{i\&Element;I_n{C}_{\mathrm{\&rho;}}^n\left[i\right]}\\ 1& h_n\&Element;{\&cap;}_{i\&Element;I_n{C}_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]}\end{array}\right.$

(5) adopt the self adaptation regulation strategy to upgrade ρ

Adopt the self adaptation regulation strategy to regulate and determine coefficient of expansion ρ in the next iteration, ρ _StartAnd ρ _StopBe respectively the value upper and lower bound of ρ; ρ=ρ when initial _Start, inspection condition in each iteration $h_n\&NotElement;{\&cap;}_{i\&Element;I_n}{H}_i\left(h_n\right)$ Whether satisfy, if satisfy then need not to change ρ; And work as $h_n\&Element;{\&cap;}_{i\&Element;I_n}{H}_i\left(h_n\right)$ The time, judge whether current ρ stops coefficient of expansion value ρ less than certain _StopIf: ρ＞ρ _Stop, then reduce ρ with step delta, namely ρ=ρ-Δ satisfies until find again $h_n\&NotElement;{\&cap;}_{i\&Element;I_n}{H}_i\left(h_n\right)$ ρ; And as ρ≤ρ _StopIn time, just no longer continue to reduce.

(6) iteration is upgraded and is finished judgement

After above-mentioned steps is finished, judge that whether the current iteration frequency n is less than the total iterations N that sets.If n＜=N then adds 1 with iterations n, enter the next iteration renewal process; Otherwise iterative process finishes.

According to technical scheme of the present invention, it has following many beneficial effects and remarkable advantage:

1. adopt a kind of iteration renewal process of new interference suppression filter coefficient vector, this renewal process has reduced in projection process the restriction of projection set and the computation complexity of system.

2. adopt the parallel projection technology, compare with non-parallel projection methods such as SAGP and CNLMS, this technology has improved the convergence rate of system and the constringency performance under the stable state.

3. will be converted into to the projection of convex set the projection to the half-plane that comprises this convex set, obtain simple projection formula, solve in the actual application to the definite computing formula complexity of convex set projection or the shortcoming that is difficult to obtain.

4. fully take into account the compromise of system's output SINR under the convergence rate of system and the stable state, adopt a kind of mechanism of as the case may be coefficient of expansion self adaptation being regulated, make system under different environment, can obtain simultaneously fast convergence and stable interference rejection.

Description of drawings

The overall flow figure of Fig. 1---the inventive method.

Fig. 2---parallel subgradient projection schematic diagram.

Fig. 3---the output SINR curve of system under the fixing coefficient of expansion condition.

Fig. 4---coefficient of expansion self adaptation regulation strategy flow chart.

Fig. 5---SAGP, CNLMS, the output SINR Performance Ratio of ECPP and APSP method is.

Fig. 6---SAGP, CNLMS, the BER Performance Ratio of ECPP and APSP method is.

Embodiment

Below in conjunction with drawings and Examples, technical solutions according to the invention are further elaborated.

Fig. 1 is the overall flow figure (representing this method with APSP) of the inventive method.In a binary phase shift keying DS/CDMA system, suppose that for present receiving machine first user is desired user, the data sequence that receives so

(N is the length of spreading code) is:

$r\left[i\right]=A_1{b}_1\left[i\right]{\mathrm{<figure-callout\; id="1"\; label="s"\; filenames="H2009101830047E00011.png,H2009101830047E00021.png"\; state="\{\{state\}\}">s</figure-callout>}}_1+\underset{l=2}{\overset{L}{\mathrm{\&Sigma;}}}{A}_l{\stackrel{\&OverBar;}{b}}_l\left[i\right]{\stackrel{\&OverBar;}{s}}_l+n\left[i\right]---\left(1\right)$

A in the formula (1) ₁Signal amplitude (A for desired user ₁＞0); b ₁[i] is i bit of desired user transmission; ${\mathrm{<figure-callout\; id="1"\; label="s"\; filenames="H2009101830047E00011.png,H2009101830047E00021.png"\; state="\{\{state\}\}">s</figure-callout>}}_1\&Element;{\{-\frac{1}{N},\frac{1}{N}\}}^N$ Spread spectrum code sequence (the ‖ s that represents normalized desired user ₁‖=1);

It is the additive noise on i the bit.A _l(2≤l≤L) is the interference signal amplitude of l user except desired user,

With

Respectively i of l user spread spectrum code sequence that disturbs symbol and this user, so in the formula (1)

The expression multiple access disturbs (MAI).

For present receiving machine, r[i] and s ₁Being Given information, is the interference suppression filter of h as input by coefficient vector with it, thereby reduces the value of MAI item in the formula (1).

In addition, h must be at limiting set C _sIn, this set is defined as

So:

$<h,r\left[i\right]>=A_1{b}_1\left[i\right]+\underset{l=2}{\overset{L}{\mathrm{\&Sigma;}}}{A}_l{\stackrel{\&OverBar;}{b}}_l\left[i\right]<h,{\stackrel{\&OverBar;}{s}}_l>+<h,n\left[i\right]>---\left(2\right)$

In order to suppress the MAI of DS/CDMA system, must reduce as far as possible rear two value in the formula (2).So optimum interference suppression filter coefficient vector h _OptMust satisfy:

$h_{\mathrm{opt}}\&Element;\underset{h\&Element;C_s}{\arg \min }E\left\{{(<h,r[i]>-A_1{b}_1[i\left]\right)}^2\right\}---\left(3\right)$

E in the following formula represents mathematic expectaion.Purpose of the present invention just is, upgrades h by the iteration of limited number of time, makes it close to h _OptThereby, suppress to the full extent the MAI in the DS/CDMA system.The below will describe the concrete steps of the inventive method in detail.

(1) initialization

In initialization procedure, the initial value of each parameter of using in each step below needing to set: iterations n=1; ${\tilde{A}}_{\mathrm{1,0}}=0,{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="H2009101830047E00011.png,H2009101830047E00021.png"\; state="\{\{state\}\}">h</figure-callout>}}_1={\mathrm{<figure-callout\; id="1"\; label="s"\; filenames="H2009101830047E00011.png,H2009101830047E00021.png"\; state="\{\{state\}\}">s</figure-callout>}}_1,$ Q, w _i, γ regulates relevant parameter ρ with the coefficient of expansion _Start, ρ _Stop, Δ, total iteration update times N.The concrete meaning of each parameter is introduced hereinafter in detail.

(2) set up convex set C _ρ ⁽ⁿ⁾[i] and corresponding convex function g _i(h _n)

Because the A in the formula (3) ₁And b ₁[i] the unknown adopts following more new formula to estimate A ₁And b ₁[i]:

H in upper two formulas _nIt is the coefficient vector of the n time interference suppression filter in the iteration

With

Be respectively amplitude A in the n time iteration ₁Bit b with i transmission ₁The estimated value of [i], and γ ∈ (0,1] be forgetting factor; The sgn function definition is:

If i.e. a＞0, sgn a=1, otherwise

By formula (4) and formula (5), redefine mean square error:

$\underset{h_n\&Element;C_s}{\arg \min }E\left\{{(<h_n,r\left[i\right]>-{\tilde{A}}_{1,n}{\tilde{b}}_{1,n}\left[i\right])}^2\right\}---\left(6\right)$

Introduce following random attribute collection with the mathematic expectaion in the replacement formula (6):

C in the following formula _ρ ⁽ⁿ⁾[i] is convex set, I _nBe the control sequence that contains q element, q is the number of the parallel processor of each iteration participation, and ρ is the coefficient of expansion, and it has determined to comprise h _OptThe size of set.Corresponding with this convex set so convex function is:

$g_i\left(h_n\right)={(<h_n,r\left[i\right]>-{\tilde{A}}_{1,n}{\tilde{b}}_{1,n}\left[i\right])}^2-\mathrm{\&rho;}---\left(8\right)$

(3) calculate h _nTo convex set C _ρ ⁽ⁿ⁾The projection of [i]

Next, need to obtain the coefficient vector h of current iteration _nTo each convex set C _ρ ⁽ⁿ⁾The projection of [i] Owing to obtain in the practical application

Accurate expression very difficult, so adopt here to comprising convex set C _ρ ⁽ⁿ⁾The closed half-plane H of [i] _i(h _n) projection

Come close approximation

Namely $P_{C_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]}\left(h_n\right)\&cong;P_{H_i\left(h_n\right)}\left(h_n\right).$

Convex function g in the formula (7) _i(h _n) everywhere can be little, its subgradient ▽ g _i(h _n) be $\&dtri;g_i\left(h_n\right)=2r\left[i\right](<h_n,r\left[i\right]>-{\tilde{A}}_{1,n+1}{\tilde{b}}_{1,n}\left[i\right]).$ Half-plane H so _i(h _n) expression formula be:

So,

Have the following primitive formula that closes:

$P_{H_i\left(h_n\right)}\left(h_n\right)=\left\{\begin{array}{cc}{h}_n-\frac{g_i\left(h_n\right)}{\&dtri;g_i\left(h_n\right)},& h_n\&NotElement;H_i\left(h_n\right)\\ h_n,& h_n\&Element;H_i\left(h_n\right)\end{array}\right.---\left(10\right)$

(4) upgrade the interference suppression filter coefficient vector

In the method, the iteration renewal process expression formula of interference suppression filter coefficient vector is as follows:

$h_{n+1}=P_{C_s}(h_n+{\mathrm{\&lambda;}}_n(\underset{i\&Element;I_n}{\mathrm{\&Sigma;}}{w}_i^{\left(n\right)}{P}_{C_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]}\left(h_n\right)-h_n))---\left(11\right)$

H in the following formula _nAnd h _N+1Represent respectively the n time and the filter coefficient vector during the n+1 time iteration.w _i ⁽ⁿ⁾For to projection

The weight of composing, it must satisfy two conditions: $w_i^{\left(n\right)}>0,\underset{i\&Element;I_n}{\mathrm{\&Sigma;}}{w}_i^{\left(n\right)}=1.$ When in the method, each iteration is upgraded weight remain unchanged ( $w_i=w_i^{\left(n\right)}$ )。λ _nBe relaxation factor, its span is λ _n∈ [0,2M _n], in this scope, randomly draw acquisition in equiprobable mode when upgrading at every turn.M _nExpression formula as follows:

$M_n=\left\{\begin{array}{cc}\frac{\underset{i\&Element;I_n}{\mathrm{\&Sigma;}}{w}_i{\left|\right|P_{C_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]}\left(h_n\right)-h_n\left|\right|}^2}{{\left|\right|\underset{i\&Element;I_n}{\mathrm{\&Sigma;}}{w}_i{P}_{C_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]}\left(h_n\right)-h_n\left|\right|}^2}& h_n\&NotElement;{\&cap;}_{i\&Element;I_n}{C}_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]\\ 1& h_n\&Element;{\&cap;}_{i\&Element;I_n}{C}_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]\end{array}\right.---\left(12\right)$

In addition, in the formula (11)

To limiting set C _sOn projection, it has guaranteed the h that each iteration has obtained when upgrading _N+1Satisfy＜h _N+1, s ₁〉=1, for any vector x,

Expression formula as follows:

So, in the iteration renewal process of formula (11), at first with h _nProject to the common factor of the convex set that participates in parallel processing

On, then use relaxation factor λ _nWeighting projects to C at last _sOn.

It (is I that Fig. 3 has provided q=2 _n=n, n-1}), λ _nThe schematic diagram of=1 o'clock the inventive method when the n+1 time iteration.Known h _n(h _n∈ C _s), the shadow region represents convex set C _ρ ⁽ⁿ⁾[n] and C _ρ ⁽ⁿ⁾The common factor of [n-1], optimum filter coefficient value h _OptWithin this occurs simultaneously, while h _OptMust satisfy h _Opt∈ C _sUse respectively H _n(h _n) and H _N-1(h _n) representative comprises C _ρ ⁽ⁿ⁾[n] and C _ρ ⁽ⁿ⁾The closed half-plane of [n-1] is so to C _ρ ⁽ⁿ⁾[n] and C _ρ ⁽ⁿ⁾The projection of [n-1] is used to half-plane H _n(h _n) and H _N-1(h _n) projection approach, multiply by corresponding weight w _nAnd w _N-1And summation, obtain h _n', at last with h _n' project to C _sOn, obtain h _N+1

Adopt the benefit of this iteration renewal process to be: to adopt the parallel projection technology, compare with non-parallel projection methods such as SAGP and CNLMS, the constringent impact of q-1 iteration convexity set pair before in each iteration, considering, thereby improved the accuracy of projection, accelerated the convergence rate of system.In addition, in this method at first with h _nProject to the common factor of the convex set that participates in parallel processing

On, then use relaxation factor λ _nWeighting projects to C at last _sOn.And in the ECPP method, h _nAt first project to common factor $\underset{i\&Element;I_n}{\&cap;}(C_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]\&cap;C_s)$ Then weighting.And in the formula (11)

Compare, to common factor $\underset{i\&Element;I_n}{\&cap;}(C_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]\&cap;C_s)$ Projection require relatively strictly and difficult, and cause the required projection amount of calculation of ECPP method in each iteration to many 2qN multiplying than the APSP method.So the APSP method has reduced in projection process the restriction of projection set and the computation complexity of system in renewal process.

(5) adopt the self adaptation regulation strategy to upgrade ρ

By in formula (3) and the formula (7) as can be known, the size of coefficient of expansion ρ has determined respectively to participate in the iterative process convex set C of parallel computation _ρ ⁽ⁿ⁾[i], i ∈ I _nSize.When the numerical value of ρ increases, each convex set C _ρ ⁽ⁿ⁾The area of [i] also increases thereupon, causes its common factor

Increase, i.e. optimal filter h _OptProbability in the set that formula (3) is determined is larger, upgrades through the iteration of less number of times so, just can make $h_n\&Element;\underset{i\&Element;I_n}{\&cap;}{C}_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right],$ So the convergence rate of system is very fast; But simultaneously owing to work as $h_n\&Element;\underset{i\&Element;I_n}{\&cap;}{C}_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right]$ It is h that rear iteration is upgraded expression formula _N+1=h _n, namely fall into

H behind the shadow region shown in Figure 3 _nNo longer change, but because this region area is larger, so final h _nWith optimal coefficient vector h _OptDistant probability larger, thereby affect the constringency performance of system when stable state.Otherwise, when the numerical value of ρ hour, each convex set C _ρ ⁽ⁿ⁾The area of [i] is less, and the area of its common factor also diminishes, h _nNear h _OptProbability larger, thereby the constringency performance of system under stable state is better.But make h _nSatisfy $h_n\&Element;\underset{i\&Element;I_n}{\&cap;}{C}_{\mathrm{\&rho;}}^{\left(n\right)}\left[i\right],$ That is, make h _nFall into h _OptThe regional required iteration update times at place increases, thereby causes the convergence rate of system slow.

Definition output signal-interference-to-noise ratio rate (SINR), this index can be weighed effectively through the systematic function after the interference suppression filter.Fig. 4 has compared under the prerequisite that does not adopt the self adaptation regulation strategy to upgrade ρ, the output SINR of system under four groups of fixing coefficient of expansion ρ=0.1,0.4,0.7,1.0.As can be seen from the figure ρ=1.0 o'clock system's convergence rate is the fastest, and the output SINR value that ρ=0.1 reaches after the stable state is maximum.Along with the increase of ρ, convergence rate is accelerated, but to disturbing the effect that suppresses also relatively relatively poor.The result that this figure reflects is consistent with above-mentioned analysis.So in actual applications, ρ's chooses the compromise that will consider convergence rate and stable state output SINR.

As the above analysis, after each interference suppression filter coefficient vector iteration is upgraded, adopt as shown in Figure 5 coefficient of expansion self adaptation regulation strategy to regulate and determine ρ (ρ ∈ [ρ in the next iteration _Stop, ρ _Start]).Select a larger coefficient of expansion value ρ at initial phase _Start(ρ _StartThe value upper limit for the coefficient of expansion).After each iteration is upgraded the interference suppression filter coefficient vector, the inspection condition $h_n\&NotElement;{\&cap;}_{i\&Element;I_n}{H}_i\left(h_n\right)$ Whether satisfy, if satisfy then need not to change ρ; And work as $h_n\&Element;{\&cap;}_{i\&Element;I_n}{H}_i\left(h_n\right)$ The time, judge whether current ρ stops coefficient of expansion value ρ less than certain _StopIf: ρ＞ρ _Stop, then reduce ρ (being ρ=ρ-Δ) with step delta, until find certain enough little ρ to make h _nAgain satisfy $h_n\&NotElement;{\&cap;}_{i\&Element;I_n}{H}_i\left(h_n\right);$ And as ρ≤ρ _StopThe time just no longer continue to reduce (ρ _StopValue lower limit for the coefficient of expansion).The purpose that adopts this regulation mechanism is to guarantee to utilize the larger coefficient of expansion when initial, thereby obtains comparatively faster convergence rate, when the later stage converges to stable state, and the coefficient of expansion that this strategy utilization is less, thus the system that guarantees has higher output SINR.

(6) iteration is upgraded and is finished judgement

After above-mentioned steps is finished, judge that whether the current iteration frequency n is less than the total iterations N that sets.If n＜=N then adds 1 with iterations n, enter the next iteration renewal process; Otherwise surperficial iterative process finishes, and finishes all flow processs of this method.

Estimate performance of the present invention below by experiment.In emulation experiment, the interference user number is 10, and its amplitude is expectation signal amplitude A ₁10 times.Selecting length is that 31 Gold sequence is as spreading code.Forgetting factor γ=0.01, q=16, w _i=1/q, $\&ForAll;i\&Element;I_n,$ ρ _start＝0.7，ρ _stop＝0.1，Δ＝0.05。SINR during the n time iteration is defined as follows:

${\mathrm{SINR}}_n=\frac{{\mathrm{\&Sigma;}}_{u=1}^U{<h_n^{\left(u\right)},{\mathrm{<figure-callout\; id="1"\; label="s"\; filenames="H2009101830047E00011.png,H2009101830047E00021.png"\; state="\{\{state\}\}">s</figure-callout>}}_1>}^2}{{\mathrm{\&Sigma;}}_{u=1}^U{\left[\frac{<h_n^{\left(u\right)},r^{\left(u\right)}\left[n\right]-A_1^{\left(u\right)}{b}_1^{\left(u\right)}\left[n\right]s_1>}{A_1^{\left(u\right)}}\right]}^2}---\left(14\right)$

Here h _n ^(u)And r ^(u)[n] is corresponding vector in the u time emulation, A ₁ ^(u)And b ₁ ^(u)[n] is respectively the amplitude of expectation subscriber signal in the u time emulation and n bit of transmission, simulation times U=1000.10 interference users are spreading rate T with respect to the propagation path time delay of desired user _cIntegral multiple, namely

Obtain at random equiprobably τ during each emulation _kSignal to noise ratio $\mathrm{SNR}=10{\log }_{10}(A_1^2/{\mathrm{\&sigma;}}_n^2)=15\mathrm{dB},$ σ _n ²Variance for additive noise.

The Performance Ratio that Fig. 6 has compared method of the present invention (representing with APSP) and other blind interference inhibition method.Can find no matter be in convergence rate, or the output SINR after reaching stable state, the performance of APSP all is better than other three kinds of methods.

Fig. 7 compared reach stable state after, be the present invention proposes under the 5dB-15dB condition APSP method and bit error rate (BER) performance of other three kinds of methods in signal to noise ratio (SNR).Carry out 500 experiments for every kind of method, 4000 bits of transmission are chosen last 2000 bits and are calculated BER in each experiment.Can find from the comparative result curve of Fig. 7: the performance of APSP method is better than SAGP and CNLMS method, compare with the ECPP method and also to have obvious performance advantage, this experimental result has illustrated that APSP can more effectively suppress MAI, all has higher robustness under different noise circumstances.

The scope that the present invention asks for protection is not limited only to the description of this embodiment.

Claims (1)

1. the blind MAI suppression method in the DS/CDMA system is characterized in that may further comprise the steps:

(1) initialization:

In initialization procedure, set the initial value of each parameter: iterations

, ,

,

,

Regulate relevant parameter with the coefficient of expansion ,

,

, wherein

Must less than

, total iterations

, s ₁The spread spectrum code sequence that represents normalized desired user,

,

,

When upgrading for each iteration to projection

The weight of composing;

(2) set up convex set

Convex function with correspondence

:

(2-1) adopt following more new formula to estimate

With :

Here Be

The coefficient vector of the interference suppression filter in the inferior iteration,

Be the data sequence that receives,

With

Be respectively

Amplitude in the inferior iteration

With

The bit of individual transmission

Estimated value,

Be forgetting factor;

Function definition is:

If, namely

, otherwise

(2-2) introduce following convex set, this convex set comprises the optimal value of interference suppression filter coefficient

:

In the following formula

Be convex set,

To contain

The control sequence of individual element,

The number of the parallel processor that participates in for each iteration,

Be the coefficient of expansion, corresponding with this convex set so convex function

For:

(3) calculate

To convex set

Projection

:

Employing is to comprising convex set

Closed half-plane

Projection

Come close approximation

, namely

Wherein

Expression formula be:

In the following formula

For

Subgradient,

To close primitive formula as follows:

(4) upgrade the interference suppression filter coefficient vector:

The iteration renewal process expression formula of interference suppression filter coefficient vector is as follows:

（11）

In the following formula

With

Represent respectively

Inferior and

Filter coefficient vector during+1 iteration;

For

In different convex sets

On projection The weight of giving, in the method, weight remained unchanged when each iteration was upgraded, namely

To limiting set

On projection, limiting set

Refer to satisfy

Set;

Be relaxation factor, its span is

, in this scope, randomly draw acquisition in equiprobable mode when upgrading at every turn;

Expression formula as follows:

(5) adopt the self adaptation regulation strategy to upgrade

Adopt the self adaptation regulation strategy to regulate and determine the coefficient of expansion in the next iteration

, With

Be respectively

The value upper and lower bound; When initial

, inspection condition in each iteration

Whether satisfy, change if satisfy then need not

And work as

The time, judge current

Whether stop coefficient of expansion value less than certain

If:

, then with step-length Reduce

, namely

, satisfy until find again

And work as

In time, just no longer continue to reduce;

(6) iteration is upgraded and is finished judgement

After above-mentioned steps is finished, judge the current iteration number of times

Whether less than total iterations of setting

If

, then with iterations

Add 1, enter the next iteration renewal process; Otherwise iterative process finishes.

Images