CN100544328C - A kind of improved serial interference deletion in optimal approach to zero detection method - Google Patents

Abstract

The invention provides a kind of improved serial interference deletion in optimal approach to zero detection method, this method is applicable to that (be abbreviated as: MIMO) system maybe can be modeled as the communication system of mimo system to multiple-input and multiple-output.Its basic principle is: the ZF weighting matrix that obtains when utilizing previous stage to detect accurately recursion goes out ZF weighting matrix and the ZF weighing vector that next stage detects.Compare with traditional serial interference deletion in optimal approach to zero detection method, detection method provided by the invention has reduced computational complexity effectively under the prerequisite of not losing performance.

Description

A kind of improved serial interference deletion in optimal approach to zero detection method

Technical field

The present invention relates to the signal detection technique of wireless communication system, relate in particular to multiple-input and multiple-output and (be abbreviated as: the MIMO) signal detection technique of system.

Background technology

Up-to-date studies show that: adopt a plurality of transmitting antennas and reception antenna can significantly improve the channel capacity of wireless communication system under the wireless fading channel environment.The system of a plurality of dual-mode antennas of this employing is commonly called multiple-input and multiple-output (MIMO) system.Because mimo system can be broken through the radio frequency resources restriction, effectively improves system spectral efficiency, therefore be considered to one of Main physical layer technology of following high-speed radiocommunication system.Chinese scholars has been done a large amount of further investigation work to the correlation technique of mimo system, and wherein the detection method of mimo system is an important research focus.

People such as the Golden of Bell Laboratory in 1998 and Foschini have proposed a kind of MIMO detection method (or being called the V-BLAST detection method) of serial interference deletion in optimal approach to zero.This method is carried out the continuous N level by putting in order of maximum signal to noise ratio to the M road parallel signal of launching and is detected, before every grade of detection, delete detection signal to the not interference of detection signal, need calculate the ZF weighing vector during every grade of detection, utilize the ZF weighing vector to recover the emission symbol of its correspondence then.Bell Laboratory utilizes its experiment porch of building to prove: under indoor rich reflection environment, can reach 20-40bit/s/Hz when average signal-to-noise ratio (SNR) adopts the spectrum efficiency of the mimo system of said method during for 24-34dB; It is to be beyond one's reach that high spectrum efficiency like this is utilized traditional technology.

Yet the serial interference deletion in optimal approach to zero detection method need carry out asking for M time pseudo-inverse operation, and under the many situations of antenna number, the complexity of this method is very high, is difficult to realize.Can the computational complexity that reduce this method be the key that can it practical application, also is a difficult problem of putting in face of the researcher at home and abroad.

Summary of the invention

Have the very high problem that computational complexity brought at the serial interference deletion in optimal approach to zero detection method, the invention provides a kind of detection method that reduces computational complexity significantly.

The invention provides a kind of improved serial interference deletion in optimal approach to zero detection method, the method for calculating the ZF weighting matrix in the conventional method is transformed into the method that increases by degrees, promptly utilize the ZF weighting matrix G that obtains in the detection of m-1 level _M-1The ZF weighting matrix G that derivation next stage (that is: m level) detects _mWith ZF weighing vector g _m

Make c _mBe that the m level detects pairing transmitting antenna at H _mMiddle corresponding row sequence number, H ₁=H, H represent mimo channel matrix, H _m(2≤m≤M) for deleting H _M-1C _M-1Be listed as resulting matrix; G ' _mThe ZF weighting matrix G that expression deletion m level detects _mC _mRow obtains matrix,

Expression H _mC _mRow, the mould of the Frobenius norm of ‖ ‖ representing matrix or vector.Then the basic step of described recurrence method comprises:

Calculate the ZF weighting matrix of the 1st grade of detection

, differentiate whether row full rank of H simultaneously, again with G ₁C ₁Row carries out obtaining behind the conjugate transpose ZF weighing vector g of the 1st grade of detection ₁Here, The pseudoinverse of expression H.

If H row full rank, ZF weighting matrix and ZF weighing vector that the M-1 level of then utilizing first kind of recurrence method to calculate the back detects, that is:

Calculate G according to following recurrence formula earlier _m:

$G_m=G_{m-1}^{\′}-{\left|\right|g_{m-1}\left|\right|}^2{G}_{m-1}^{\′}{g}_{m-1}{g}_{m-1}^H;$

Again with G _mC _mRow carries out obtaining the ZF weighing vector g that the m level detects behind the conjugate transpose _m

If H Lieque order, ZF weighting matrix and ZF weighing vector that the M-1 level of then utilizing second kind of recurrence method to calculate the back detects, that is:

Calculate G according to following recurrence formula _m:

Calculate earlier ${\mathrm{\α}}_{m-1}=1-g_{m-1}^H{h}_{c_{m-1}};$

If α _M-1=0, then calculate G by following formula _m:

$G_m=G_{m-1}^{\′}-{\left|\right|g_{m-1}\left|\right|}^{-2}{G}_{m-1}^{\′}{g}_{m-1}{g}_{m-1}^H;$

If α _M-1≠ 0, then calculate G by following formula _m:

$G_m=G_{m-1}^{\′}+\frac{1}{{\mathrm{\α}}_{m-1}}{G}_{m-1}^{\′}{h}_{c_{m-1}}{g}_{m-1}^H;$

Again with G _mC _mRow carries out obtaining the ZF weighing vector g that the m level detects behind the conjugate transpose _m

Beneficial effect of the present invention is, the improved serial interference deletion in optimal approach to zero detection method that is proposed adopts the recurrence method of low complex degree, the operation result that utilizes previous stage to detect comes recursion to calculate ZF weighting matrix and ZF weighing vector that the back one-level detects, avoided complexity very high directly ask the pseudo-inverse operation of matrix.Compare with conventional method, this method is guaranteeing to have reduced the computational complexity of serial interference deletion in optimal approach to zero detection method significantly under the constant prerequisite of detection performance.

Description of drawings

Fig. 1 shows the theory diagram of mimo system;

Fig. 2 shows the flow chart of optimal approach to zero interference delete detection method;

Fig. 3 shows and adopts the Greville method to calculate the 1st grade of ZF weighting matrix and differentiate the whether flow chart of row full rank of mimo channel matrix;

Fig. 4 shows and detects p _mThe flow chart of road signal;

Fig. 5 shows the flow chart of the ZF weighting matrix that calculates the detection of m level;

Fig. 6 shows the performance comparison result of improved optimal approach to zero interference delete detection method provided by the invention and conventional method.

Embodiment

The present invention will be described in detail below by drawings and Examples.

Detection method of the present invention is applicable to mimo system, or can be modeled as other communication system of mimo system.For example, the present invention can directly be used on any one subcarrier of multi-input multi-output-orthogonal frequency-division multiplexing (MIMO-OFDM) system, also can be used for the Multiuser Detection of code division multiple access system.

Fig. 1 shows the theory diagram of mimo system.At transmitting terminal, data bit at first is mapped to and is the signal in the signal constellation (in digital modulation), through forming the baseband signal of multidiameter delay behind the serial to parallel conversion, launches simultaneously from many different antennas respectively after ovennodulation then; After the wireless channel decline, signal and noise stack back from different transmit antennas are received simultaneously by many antennas, through generating the multidiameter delay baseband signal after the demodulation, the channel condition information that the MIMO detector utilizes channel estimator to produce recovers initial data from baseband signal.In the real system, data bit can will pass through deinterleaving and decoding accordingly earlier through encoding and interweaving before the receiver dateout before mapping.The baseband signal input/output relation of this system can be represented as:

y＝Hx+ε

In the following formula, x=[x ₁x ₂X _M] ^TThe expression emission signal vector, M represents number of transmit antennas, [] ^TThe transposition of representing matrix or vector, x _mExpression is from the signal of m transmit antennas emission; ε=[ε ₁ε ₂ε _N] ^TThe expression noise vector, N represents reception antenna number, ε _nRepresent the noise that n root reception antenna receives; Y=[y ₁y ₂Y _N] ^TThe expression received signal vector, y _nRepresent the signal that n root reception antenna receives; H is the matrix of N * M dimension, expression mimo channel matrix, the element h of the capable m row of its n _{N, m}The baseband channel fading factor of expression from the m transmit antennas to n root reception antenna, before carrying out MIMO detection processing, at first to obtain the estimated value (for convenience of description, in the literary composition estimated value of mimo channel matrix still being designated as H) of channel matrix by channel estimator.The present invention relates to the MIMO detector portion of system shown in Figure 1.

Fig. 2 shows the flow chart of serial interference deletion in optimal approach to zero detection method, is used for the MIMO detector portion of Fig. 1.Here the signal with i transmission antennas transmit abbreviates i road signal as, i=1, and 2 ..., M.Shown flow process enters step 203 and carries out the 1st grade of test initialization after detecting beginning 201, comprise the reception vector y of the 1st grade of detection of initialization ₁, equivalent channel matrix H ₁With the sequence number vector f ₁, its concrete steps are as follows:

The reception vector y of (1) the 1st grade of detection ₁Equal to receive vectorial y, i.e. y ₁=y;

The equivalent channel matrix H of (2) the 1st grades of detections ₁Just equal the mimo channel matrix H, i.e. H ₁=H;

The sequence number vector f of (3) the 1st grades of detections ₁Element be natural number 1 to M, and be that ascending order arranges, i.e. f ₁=[1 2 ... M] ^T

After this, in step 205, detect p ₁The road signal obtains

Its concrete steps are as follows:

(1) calculates ZF weighting matrix G ₁, promptly

Wherein

The pseudoinverse of representing matrix or vector; And the value of definite row full rank flag bit R is promptly worked as H ₁R=1 during for the row full rank, otherwise R=0;

(2) take out matrix G ₁The pairing capable sequence number c of row of middle mould value minimum ₁

(3) take out f ₁C ₁Individual element obtains the 1st grade and detects corresponding transmitting antenna sequence number p ₁

(4) to G ₁C ₁Row carries out conjugate transpose and obtains the 1st grade of ZF weighing vector g ₁, and utilize g ₁To receiving vectorial y ₁Carry out the ZF weighting, obtain p ₁The decision statistic amount of road signal , promptly ${\mathrm{<figure-callout\; id="1"\; label="z\; p"\; filenames="C200510080498E00122.png,C200510080498E00142.png"\; state="\{\{state\}\}">z</figure-callout>}}_{p_{}}=g_1^{\mathrm{<figure-callout\; id="1"\; label="H\; y"\; filenames="C200510080498E00122.png,C200510080498E00142.png"\; state="\{\{state\}\}">H</figure-callout>}}{y}_{}{}_1,$ Wherein, the conjugate transpose of [] H representing matrix or vector;

(5) to the decision statistic amount Carry out hard decision, obtain p ₁The decision value of road signal

In step 207, counting variable m is composed initial value, i.e. m=2.

In step 209, m judges to counting variable, if m≤M sets up, then enters step 211, otherwise then enters step 217.

In step 211, m level test initialization comprises the equivalent received vector y that initialization m level detects _m, equivalent channel matrix H _mWith the sequence number vector f _m, its concrete steps are as follows:

(1) from y _M-1Middle deletion

Other interference that transmits is obtained the equivalent received vector y that the m level detects _m, promptly $y_m=y_{m-1}-h_{c_{m-1}}{\hat{x}}_{p_{m-1}}.$ Here,

Expression H _M-1C _M-1Row.

(2) the equivalent channel matrix H of deletion m-1 level detection _M-1C _M-1Row obtain the equivalent channel matrix H that the m level detects _m

(3) the sequence number vector f of deletion m-1 level detection _M-1C _M-1Individual element obtains the sequence number vector f that the m level detects _m

In step 213, detect p _mThe road signal obtains decision value

, Fig. 4 has provided the idiographic flow of this step.

In step 215, counting variable m adds 1, i.e. m=m+1.

Then, this flow process withdraws from step 217.

Fig. 3 shows and adopts the Greville method to calculate the 1st grade of ZF weighting matrix and differentiate the whether flow chart of row full rank of mimo channel matrix H, is used for realizing (1) step of Fig. 2 step 205.Note A _kBe the submatrix that the preceding k row of H constitute, a _kBe the k row of H, k=1,2 ..., M.Flow process among Fig. 3 is to begin to enter step 301 in the compute matrix pseudoinverse.Then, step 303 is calculated A ₁Pseudoinverse, promptly

Here, the mould of the Frobenius norm of ‖ ‖ representing matrix or vector.In step 305, counting variable k is composed initial value, i.e. k=2.In step 307, k judges to counting variable, if k≤M sets up, then enters step 309, otherwise then enters step 317.In step 309, calculate d _kAnd q _k, promptly And q _k=a _k-A _K-1d _kThen, in step 311, according to q _k, d _kAnd A _K-1Calculate That is:

In the reality, because the restriction of computational accuracy, as ‖ q _kThe value of ‖ is less than certain minimum positive number (that is: ‖ q _k‖＜γ, γ is a minimum positive number here, its size is determined by the actual calculation precision) time just can think ‖ q _k‖=0, on the contrary then think ‖ q _k‖ ≠ 0.

In step 313, press the following formula recursion and calculate A _kPseudoinverse:

In step 315, counting variable k adds 1, i.e. k=k+1.

In step 317, according to ‖ q _M‖ determines the value of full rank flag bit R, if ‖ q _M‖ ≠ 0 explanation matrix H is row full rank, that is: R=1; If ‖ q _M‖=0 explanation matrix H is Lieque order, that is: R=0.

Then, this flow process withdraws from step 319.

Fig. 4 shows and detects p _mThe road signal is to obtain

Idiographic flow, be used for realizing the step 213 of Fig. 2.Flow process among Fig. 4 is to enter step 401 after the step 211 in finishing Fig. 2.Then, in step 403, calculate ZF weighting matrix G _mAfter this, in step 405, take out matrix G _mThe pairing capable sequence number c of row of middle mould value minimum _mIn step 407, take out f _mC _mIndividual element obtains the m level and detects corresponding transmitting antenna sequence number p _mIn step 409, to G _mC _mRow is got conjugate transpose and is obtained m level ZF weighing vector g _m, and utilize ZF vector g _mTo receiving vectorial y _mCarry out linear weighted function, obtain p _mThe decision statistic amount of road signal

Promptly $z_{p_m}=g_m^H{y}_m.$ In step 411, to the decision statistic amount

Carry out hard decision, obtain p _mThe decision value of road signal

Then, this flow process withdraws from step 413.

Fig. 5 shows and calculates the ZF weighting matrix G that the m level detects _mIdiographic flow (2≤m≤M), be used for realizing the step 403 of Fig. 4.Enter step 503 after this flow process begins, m judges to counting variable, if m≤M-1 sets up, then enter step 505, otherwise (that is: when m=M) then enters step 517.

In step 505, delete the ZF weighting matrix G that the m-1 level detects _M-1C _M-1Row obtains

In step 507, to H ₁Full rank flag bit R judge, if R=1 sets up, then enter step 513, otherwise then enter step 509; In theory, as long as have certain independence (that is: not exclusively relevant) between the element of H, the probability of H row full rank just is 1, however in practice since channel undesirable, computational accuracy is limited etc., and factor may cause H ₁The situation of Lieque order.

In step 509, judge α _M-1Value whether be 0, if α _M-1=0 sets up, and then enters step 511, otherwise then enters step 513.Here, α _M-1Computing formula be: ${\mathrm{\&alpha;}}_{m-1}=1-g_{m-1}^H{h}_{c_{m-1}}.$ In the reality, because α is worked as in the restriction of computational accuracy _M-1Absolute value less than certain very little positive number (that is: | α _M-1|＜γ, wherein || the expression absolute value, γ is a very little positive number, its size is determined by the actual calculation precision) time just can think α _M-1=0; Otherwise, if | α _M-1| 〉=γ then thinks α _M-1≠ 0;

In step 511, utilize formula (1) recursion to calculate G _m, formula (1) is as follows:

$G_m=G_{m-1}^{\&prime;}-{\left|\right|g_{m-1}\left|\right|}^2{G}_{m-1}^{\&prime;}{g}_{m-1}{g}_{m-1}^H---\left(1\right)$

In step 513, utilize formula (2) recursion to calculate G _m, formula (2) is as follows:

$G_m=G_{m-1}^{\&prime;}+\frac{1}{{\mathrm{\&alpha;}}_{m-1}}{G}_{m-1}^{\&prime;}{h}_{c_{m-1}}{g}_{m-1}^H---\left(2\right)$

In step 515, directly calculate the ZF weighting matrix that the M level detects

Annotate: ZF weighting matrix G arbitrarily _m(2≤m≤M) all can be by G _M-1Recursion obtains; Yet work as m=M, $H_M=h_{c_M}$ Be a column vector, this moment, described recurrence method can be reduced to direct calculating M level ZF weighing vector, that is: $g_M=G_M^H={\left|\right|h_{c_M}\left|\right|}^{-2}{h}_{c_M}.$

Then, this flow process withdraws from step 517.

Fig. 6 shows the performance comparison result of method provided by the invention and conventional method.Abscissa is represented signal to noise ratio (snr), and that ordinate is represented is bit error rate (BER), and the transmitting antenna of this system and reception antenna number are 4, and channel is independent identically distributed MIMO flat Rayleigh fading channel, and the modulation system that is adopted is 16QAM.Simulation result shows that method provided by the invention and conventional method performance are identical.

Equate that with number of transmit antennas and reception antenna number (M=N) is example below, the computational complexity of brief analysis method provided by the invention.Here the amount of calculation with a complex multiplication is the unit of method complexity, ignores addition and subtraction, comparison, selection etc. and relatively simply handles, and only calculates the complexity of multiplication and division.Tradition serial interference deletion in optimal approach to zero detection method complexity reaches M ⁴Level; Because adopted simple recurrence method to calculate the ZF weighting matrix, method complexity provided by the invention only is M ³Level.Obviously, method provided by the invention can reduce the computational complexity of serial interference deletion in optimal approach to zero detection method significantly.

Claims (6)

1, a kind of improved serial interference deletion in optimal approach to zero detection method, this method is carried out the continuous N level by certain arrangement sequence to the M road parallel signal of launching and is detected, before every grade of detection, delete detection signal to the not interference of detection signal, need calculate the ZF weighing vector during every grade of detection, utilize the ZF weighing vector to recover the emission symbol of its correspondence then; The method is characterized in that it is that the ZF weighting matrix recursion of generation obtains when being detected by the m-1 level that the m level detects required ZF weighting matrix and ZF weighing vector, here, m=2,3 ..., M, the basic step of recursion is:

At first, calculate the ZF weighting matrix of the 1st grade of detection

, differentiate whether row full rank of H simultaneously, again with G ₁C ₁Row carries out obtaining behind the conjugate transpose ZF weighing vector g of the 1st grade of detection ₁, here, H represents the mimo channel matrix,

Represent its pseudoinverse, c ₁Be the 1st grade of row sequence number that detects pairing transmitting antenna correspondence in H;

Then, judge whether row full rank of H; If H row full rank, ZF weighting matrix and ZF weighing vector that the M-1 level of then utilizing first kind of recurrence method to calculate the back detects; If H Lieque order, ZF weighting matrix and ZF weighing vector that the M-1 level of then utilizing second kind of recurrence method to calculate the back detects.

2, detection method according to claim 1 is characterized in that, utilizes first kind of recurrence method to calculate the ZF weighting matrix G that the m level detects _mWith ZF weighing vector g _mStep as follows:

Delete the ZF weighting matrix G that the m-1 level detects _M-1C _M-1Row obtains matrix

Wherein, c _lBe that the l level detects pairing transmitting antenna at H _lMiddle corresponding row sequence number, l=1,2 ..., M, H when l=1 _l=H works as l=2, and 3 ..., H during M _lBe deletion H _L-1C _L-1Be listed as resultant matrix;

Calculate the ZF weighting matrix G that the m level detects according to following recurrence formula _m:

$G_m=G_{m-1}^{\&prime;}-{\left|\right|g_{m-1}\left|\right|}^{-2}{G}_{m-1}^{\&prime;}{g}_{m-1}{g}_{m-1}^H;$

With G _mC _mRow carries out obtaining the ZF weighing vector g that the m level detects behind the conjugate transpose _m

3, detection method according to claim 1 is characterized in that, utilizes second kind of recurrence method to calculate the ZF weighting matrix G that the m level detects _mWith ZF weighing vector g _mStep as follows:

Delete the ZF weighting matrix G that the m-1 level detects _M-1C _M-1Row obtains matrix

Calculate ${\mathrm{\&alpha;}}_{m-1}=1-g_{m-1}^H{h}_{c_{m-1}},$ Wherein, Expression H _lC _lRow, l=1,2 ..., M;

If α _M-1=0, then calculate G by following formula _m:

$G_m=G_{m-1}^{\&prime;}-{\left|\right|g_{m-1}\left|\right|}^{-2}{G}_{m-1}^{\&prime;}{g}_{m-1}{g}_{m-1}^H;$

If α _M-1≠ 0, then calculate G by following formula _m:

$G_m=G_{m-1}^{\&prime;}+\frac{1}{{\mathrm{\&alpha;}}_{m-1}}{G}_{m-1}^{\&prime;}{h}_{c_{m-1}}{g}_{m-1}^H;$

With G _mC _mRow carries out obtaining the ZF weighing vector g that the m level detects behind the conjugate transpose _m

According to claim 1,2 or 3 described detection methods, it is characterized in that 4, the ZF weighing vector method of calculating the detection of M level is: $g_M={\left|\right|h_{c_M}\left|\right|}^{-2}{h}_{c_M}.$

5, detection method according to claim 3 is characterized in that, when | α _M-1| during＜γ, then judge α _M-1=0; When | α _M-1| during 〉=γ, then judge α _M-1≠ 0; Here, the very little positive number of γ for setting according to the Practical Calculation precision.

6, detection method according to claim 1 is characterized in that, other communication system that this detection method is applicable to the multiple-input and multiple-output mimo system and can be modeled as mimo system.

Images