CN101056161A - A soft in and soft out detection method capable of reducing the complexity - Google Patents

Abstract

The present invention provides a soft-inlet-soft-outlet detecting method which can reduce the complexity through replacing the instantaneous covariance matrix by the average of the emitting symbol vector covariance matrix, namely one of said average channel values is used to replace the channel value of the whole region. Thus, Nt*T*F of matrix inverses are simplified to form one matrix inverse. In the present invention, only one time of matrix inverse is required in a high correlation channel region, and the complexity of the system is reduced greatly without any losing.

Description

A kind of soft in and soft out detection method that reduces complexity

Technical field

The invention belongs in the wireless communication system pilosity and penetrate the multiple receive antenna technical field, relate in particular to a kind of soft in and soft out detection method of the reduced complexity of iterative receiver when being applied to sky.

Background technology

Consider the MIMO-OFDM communication system of one a Nt transmitting antenna Nr reception antenna, contain following steps:

The information bit of transmitting terminal at first carries out chnnel coding, and for antiburst error, coded-bit is sent into interleaver and interweaved.Then the bit that obtains is carried out Space Time Coding and be mapped to a plurality of transmitting antennas, launch simultaneously by Nt transmitting antenna.

(1) model through the received signal after the OFDM demodulation is:

Y _t，f＝H _t，fx _t，f+n _t，f

H wherein _{T, f}(i, j) individual element is a tth OFDM symbol, j transmitting antenna on fth subcarrier is to the fading factor of the channel of i reception antenna, x _{T, f}=[x _{T, f}(1), x _{T, f}(2), K, x _{T, f}(Nt)] ^TBe the emission symbolic vector, n _{T, f}Be that average is zero, variance is σ ²Multiple Gaussian noise.

(3) at receiving terminal, detect with the Turbo iterative algorithm, at first minimum mean square error detector to send the likelihood information of coded-bit, passes through deinterleaving then, delivers to the soft soft channel decoder decoding that goes out of going into.

If P is (x=X _j) be symbol prior probability through the channel decoding feedback, for symbol constellation x={X ₀X ₁LX _M-1, M is the number of emission symbol constellation, we can obtain expectation emission symbol:

${\stackrel{\‾}{x}}_{t,f}\left(n\right)=E\left[x_{t,f}\left(n\right)\right]=\underset{j=0}{\overset{M-1}{\mathrm{\Σ}}}{X}_{j}P(x_{t,f}\left(n\right)=X_j)$

The emission symbolic vector of expectation is:

x _t，f(n)＝[ x _t，f(1) x _t，f(2)L x _t，f(Nt)] ^T

Through the emission symbolic vector after the counteracting of soft interference be:

Y _t，f(n)＝Y _t，f-H _t，f( x _t，f(n)- x _t，f(n)e _n)

E wherein _nBe complete zero column vector of Nt * 1, except being 1 at n element.Utilize Y _{T, f}(n) and emission symbol x _{T, f}(n) ask linear minimum mean-squared error to estimate.

$J\left(w_{t,f}\left(n\right)\right)=E{|x_{t,f}\left(n\right)-w_{t,f}^H\left(n\right){\stackrel{\‾}{Y}}_{t,f}\left(n\right)|}^2$

We can obtain weight vector thus:

$w_{t,f}\left(n\right)={(H_{t,f}{V}_{t,f}{H}_{t,f}^H+(1-v_{t,f}\left(n\right))h_{t,f}\left(n\right)h_{t,f}^H\left(n\right)+{\mathrm{\σ}}^{2}I)}^{-1}{h}_{t,f}\left(n\right)$

H wherein _{T, f}(n)=H _{T, f}e _n, V _{T, k}Be emission symbolic vector covariance matrix, be diagonal matrix.v _{T, f}(n) be n covariance of launching symbol to be detected.

V _t，f＝diag(v _t，f(1)，v _t，f(2)，L，v _t，f(Nt))

Can obtain SIC-MMSE thus is output as:

$z_{t,f}\left(n\right)=w_{t,f}^H\left(n\right){\stackrel{\‾}{Y}}_{t,f}\left(n\right)=w_{t,f}^H\left(n\right)(Y_{t,f}-H_{t,f}{\stackrel{\‾}{x}}_{t,f}+{\stackrel{\‾}{x}}_{t,f}\left(n\right)h_{t,f}\left(n\right))$

By soft symbol z _{T, f}(n), the likelihood ratio that we just can the calculation code bit.These likelihood ratios are through after deinterleavings, send into softly to go into the soft channel decoder that goes out and decipher.Obtain the posteriority likelihood information of coded-bit, deduct by soft and go into the external information that the soft information of priori that the soft SIC-MMSE of going out detector sends into obtains coded-bit.Through after interweaving, external information is sent into the soft soft detector that goes out of going into, as prior information.Through iteration repeatedly, testing process is finished in the posterior probability convergence.

We can see by top process, and for each subcarrier and each transmitting antenna, we must calculate z _{T, f}(n).It need carry out matrix inversion operation, and complexity is O ((Nr) ³), complexity is very high.

Summary of the invention

Technical problem to be solved by this invention provides a kind of soft in and soft out detection method that reduces complexity, it is high relevant with the time, T*F*Nt matrix inversion of the high relevant MIMO-OFDM channel region of frequency is reduced to a matrix inversion, thereby reduces the complexity of algorithm greatly.If under the quasistatic mimo channel, NT*T matrix inversion is reduced to a matrix inversion.

On solving the problems of the technologies described above, the present invention has adopted following technical proposals: establish

$g=(H_{t,f}{V}_{t,f}{H}_{t,f}^H+{\mathrm{\σ}}^2{I}_{\mathrm{Nr}})$

Utilize the matrix inversion theorem:

$w_{t,f}\left(n\right)=\frac{g^{-1}{h}_{t,f}\left(n\right)}{(1+(1-v_{t,f}\left(n\right)){\left(g^{-1}{h}_{t,f}\left(n\right)\right)}^H{h}_{t,f}\left(n\right))}$

We can be reduced to a matrix inversion with Nt matrix inversion.

Consider a multipath channel, its complex representation is

$h(t,\mathrm{\τ})=\underset{k}{\mathrm{\Σ}}{\mathrm{\γ}}_k\left(t\right)\mathrm{\δ}(\mathrm{\τ}-{\mathrm{\τ}}_k)$

γ wherein _kBe the path fading coefficient, τ _kThe time delay in path.The time of channel and frequency dependence are

r(Δt，Δf)@E{H(t+Δt，f+Δf)H ^*(t，f)}

＝r _t(Δt)r _f(Δf)

Wherein:

$r_f\left(\mathrm{\&Delta;f}\right)=\underset{k}{\mathrm{\&Sigma;}}{\mathrm{\&sigma;}}_k^2{e}^{{-\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="A20061002557000121.png,A20061002557000122.png"\; state="\{\{state\}\}">j</figure-callout>}2\mathrm{\&pi;\&Delta;f\&tau;}}_k}$

r _t(Δt)＝J ₀(2πΔtT _ff _d)

J ₀(k) be first kind zeroth order Bei Saier function, T _fBe the sampling time, f _dIt is Doppler frequency.

If r _t(Δ t)＞ρ _t, r _f(Δ f)＞ρ _f, (ρ _tAnd ρ _fBe the dependent threshold of time and frequency).We suppose that this time-frequency region channel is a height correlation.We utilize following formula to calculate the length of high relevant range.

T and F are respectively the high correlation lengths of time and frequency, and K is the number of subcarrier.In this relevant district, the time-frequency channel value in the middle of we utilize replaces the channel value in whole zone.While according to existing document, can utilize the mean value of emission symbolic vector covariance matrix to replace instantaneous covariance matrix owing to the covariance matrix reflection is the statistical information of emission sign matrix.

$\stackrel{\&OverBar;}{V}=\frac{1}{\mathrm{TF}}\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}\underset{f=1}{\overset{F}{\mathrm{\&Sigma;}}}{V}_{t,f}$

V wherein _{T, f}Be t _ThThe OFDM symbol, f _ThThe emission symbol covariance matrix of subcarrier, T and F are the high relevant ranges of channel.By this top approximate and matrix inversion theorem, in the time-frequency channel region of a height correlation, we only need a matrix inversion operation, have reduced the complexity of system.Thus:

$g_t=g@(H_{t_i,f_i}\stackrel{\&OverBar;}{V}{H}_{t_i,f_i}^H+{\mathrm{\&sigma;}}^2{I}_{\mathrm{Nr}})$

H wherein _{Ti, fi}Be the fading factor of channel, σ ²Be the power of white Gaussian noise, I _NrIt is the unit matrix of Nr * Nr.For the quasi-static flat fading channel of MIMO, we only need be just passable in the method above a quasi-static channel opens adopted in the duration.

Operation on the Matlab emulation platform proves: the present invention greatly reduces the complexity of system, and what loss is performance do not have simultaneously, reached its intended purposes.This invention is based on the SIC-MMSE criterion, and it has reduced the needed matrix inversion number of times of system.In high relevant a time-frequency channel region or quasi-static channel opens frame, only need matrix inversion one time, reduced system complexity, simultaneity factor performance not loss is little.

Description of drawings

Fig. 1 is the structured flowchart of existing MIMO BICM transmitter.

Fig. 2 is that existing MIMO BICM iterative receiver soft goes into the soft structured flowchart that goes out detector.

Fig. 3 is under the 4X4 antenna system, and QRD-M algorithm and low complex degree SIC-MMSE receiver complexity are relatively.

Fig. 4 represents to reduce the performance of iterative receiver from 1 time to 5 iteration of complexity.

Fig. 5 is that 1 iteration of traditional iterative receiver and the iterative receiver that reduces complexity and the performance of 5 iteration compare.

Fig. 6 adopts under the Turbo coded system, and the performance of the iterative receiver of traditional iterative receiver and reduction complexity relatively.

Fig. 7 is that 4 * 4 antenna QRD-M algorithms and low complex degree SIC-MMSE performance compare

Fig. 8 is that MIMO-OFDM Turbo-BLAST BER performance compares.

Embodiment

4 low complex degree Turbo iterative receivers that send 4 reception antenna mimo systems are provided below.

(1) at transmitting terminal, information sequence { b _m(m=1, L are { c through convolutional encoding K) _l(l=1, L, N).For the burst error of anti-channel, coded sequence { c _l(l=1, L are { d through random interleaving N) _l(l=1, L, N).Interweave the back bit sequence through going here and there and being converted to 4 subsequences, corresponding to each transmitting antenna.The transreceiver block diagram as shown in Figure 1.

(2) bit on each antenna is mapped as symbol, goes out by corresponding antenna transmission.The signal that the decline of process quasistatic flat channel obtains receiving terminal is:

y _t＝H _tx _t+n _t

H wherein _t(i, j) individual element is the t channel fading factor from j transmitting antenna to i reception antenna constantly, x _t=[x _{T, 1}, x _{T, 2}, L, x _{T, Nt}] be the emission symbolic vector, n _tBe that average is zero, variance is σ ²Multiple Gaussian noise.

(3) at receiving terminal, detect with the Turbo iterative algorithm, wherein soft goes into softly to go out detector and adopt new detection algorithm, reduces system complexity.Turbo iterative receiver soft goes into the soft block diagram that goes out detector as shown in Figure 2 when empty.

If P is (x=X _j) be symbol prior probability through the channel decoding feedback, for symbol constellation χ={ X ₀X ₁LX _M-1, M is the number of emission symbol constellation, we can obtain expectation emission symbol:

${\stackrel{\&OverBar;}{x}}_{t,k}=E\left[x_{t,k}\right]=\underset{j=0}{\overset{M-1}{\mathrm{\&Sigma;}}}{X}_{j}P(x_{t,k}=X_j)$

The emission symbolic vector of expectation is:

x _t＝[ x _t，1 x _t，2 L x _t，Nt]

Through the emission symbolic vector after the counteracting of soft interference be:

y _t，k＝y _t-H _t( x _t- x _t，ke _k)

E wherein _kBe complete zero column vector of Nt * 1, except being 1 at k element.Utilize y _{T, k}With emission symbol x _kAsk linear minimum mean-squared error to estimate.

$J\left(w_{t,k}\right)=E{|x_{t,k}-w_{t,k}^H{\stackrel{\&OverBar;}{y}}_{t,k}|}^2$

We can obtain weight vector thus:

$w_{t,k}={({\mathrm{HV}}_t{H}^H+(1-v_{t,k})h_k{h}_k^H+{\mathrm{\&sigma;}}^{2}I)}^{-1}{h}_k$

H wherein _k=He _k, V _tBe t emission symbol covariance matrix constantly, be diagonal matrix.v _{T, k}Be k covariance of launching symbol to be detected.

V _t＝diag(v _t，1，v _t，2，L，v _t，Nt)

Can obtain SIC-MMSE is output as:

$z_{t,k}=w_{t,k}^H{\stackrel{\&OverBar;}{y}}_k=w_{t,k}^H(y-H{\stackrel{\&OverBar;}{x}}_t+{\stackrel{\&OverBar;}{x}}_{t,k}{h}_k)$

If

g _t＝(HV _tH ^H+σ ²I _Nr)

Utilize the matrix inversion theorem:

$w_{t,k}=({g_t}^{-1}-{g_t}^{-1}{h}_k\frac{1-v_{t,\mathrm{<figure-callout\; id="1"\; label="k"\; filenames="A20061002557000122.png,A20061002557000132.png"\; state="\{\{state\}\}">k</figure-callout>}}}{1+(1-v_{t,k})h_k^H{g_t}^{-1}{h}_k}{h}_k^H{g_t}^{-1})h_k$

$={(1+(1-v_{t,k}){\left({g_t}^{-1}{h}_k\right)}^H{h}_k)}^{-1}((1+(1-v_{t,k})h_k^H{g_t}^{-1}{h}_k){g_t}^{-1}-{g_t}^{-1}{h}_k(1-v_{t,k})h_k^H{g_t}^{-1})h_k$

$=\frac{1}{(1+(1-v_{t,k}){\left({g_t}^{-1}{h}_k\right)}^H{h}_k)}{g_t}^{-1}{h}_k$

We can be reduced to a matrix inversion with Nt matrix inversion.Because the covariance matrix reflection is the statistical information of emission sign matrix, according to existing document, we can utilize the time average of emission symbolic vector covariance matrix to replace instantaneous covariance matrix.

$\stackrel{\&OverBar;}{V}=\frac{1}{T}\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}{V}_t$

V _tBe f emission symbol covariance matrix constantly, V is the time average of emission symbolic vector variance matrix, and T is the time interval, and T is the duration length of quasi-static channel opens in this specific embodiment.By this top approximate and matrix inversion theorem, in a quasistatic fading channel duration, we only need a matrix inversion operation, have reduced the complexity of system.Thus:

g _t＝g@(H VH ^H+σ ²I _Nr)

Wherein H is the fading factor of the channel of reception antenna, σ ²Be the power of white Gaussian noise, I _NrIt is the unit matrix of Nr * Nr.By soft symbol z _{T, k}, the likelihood ratio that we just can the calculation code bit.These likelihood ratios are through after deinterleavings, send into softly to go into the soft channel decoder that goes out and decipher.Softly go into the soft channel decoder that goes out and adopt bcjr algorithm.Obtain the posteriority likelihood information of coded-bit, deduct by soft and go into the external information that the soft information of priori that the soft SIC-MMSE of going out detector sends into obtains coded-bit.Through after interweaving, external information is sent into the soft soft detector that goes out of going into, as prior information.Through iteration repeatedly, testing process is finished in the posterior probability convergence.

According to same theory, in the high relevant range of time of MIMO-OFDM system and frequency, utilize the mean value of emission symbolic vector covariance matrix to replace instantaneous covariance matrix, the channel value in the middle of promptly utilizing replaces the channel value in whole zone:

$\stackrel{\&OverBar;}{V}=\frac{1}{\mathrm{TF}}\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}\underset{f=1}{\overset{F}{\mathrm{\&Sigma;}}}{V}_{t,f}$

V wherein _{T, f}Be t _ThThe OFDM symbol, f _ThThe emission symbol covariance matrix of subcarrier, T and F are the high relevant ranges of channel.

Same, in the frequency domain of MIMO-OFDM system, this moment, T=1 utilized the mean value of emission symbolic vector covariance matrix to replace instantaneous covariance matrix, and the channel value in the middle of promptly utilizing replaces the channel value in whole zone

$\stackrel{\&OverBar;}{V}=\frac{1}{F}\underset{f=1}{\overset{F}{\mathrm{\&Sigma;}}}{V}_f$

V wherein _fBe f _ThThe emission symbol covariance matrix of subcarrier, F are the high relevant ranges of the frequency of channel.

Below the present invention and existing algorithm complexity are analyzed:

The QRD-M algorithm

P: the status number M of reservation: the number of constellation symbol

Fig. 3 provides under 4 * 4 antenna systems, and QRD-M algorithm and low complex degree SIC-MMSE receiver complexity are relatively.We can see that the SIC-MMSE complexity of low complex degree is lower than the QRD-M algorithm.

Table 1.4 transmitting antenna, 4 reception antenna systems, QRD-M algorithm complex.

Complex multiplication	Complex addition	Real multiplications	The real number addition
				2*1*M	1*M	M*Log2(M)	M*(Log2(M)-1)+M
(Min(P，M)+M)+ Min(P，M)*M	Min(P，M)+Min(P，M)*M	M*Log2(M)	M*(Log2(M)-1)+ Min(P，M)*M
				(2*Min(P，M)+M)+ Min(P，M^2)*M	Min(P，M)+Min(P，M^2) +Min(P，M^2)*M	M*Log2(M)	M*(Log2(M)-1)+ Min(P，M*M)*M
(3*Min(P，M)+M)+ Min(P，M^3)*M	Min(P，M)+Min(P，M^2) +Min(P，M^3)+Min(P，M^3)*M	M*Log2(M)	M*(Log2(M)-1) +Min(P，M*M*M)*M

Table 2.4 transmitting antenna, 4 reception antenna systems, QPSK

The status number that P keeps	Complex multiplication	Complex addition	Real multiplications	The real number addition
					2	56	40	32	44
4	92	76	32	68
					6	108	98	32	84
8	124	120	32	100
					10	140	142	32	116
12	156	164	32	132
					14	172	186	32	148
16	188	208	32	164

Table 3. reduces complexity SIC-MMSE detector algorithm complexity

Complex multiplication	Complex addition	Real multiplications	The real number addition
				M/2+Nt*Nr+ (Nt+Nr*Nr+2+Nr+2*M)*Nt	M+(Nt1)*Nr+Nr+ [Nr*(Nr1)+(Nr1) +2+Nr+1]*Nt	Nt*[Log2(M)/2+M]	Nt*Log2(M)/2

Table 4.QPSK, 4 * 4 antennas reduce complexity SIC-MMSE algorithm complex

Complex multiplication	Complex addition	Real multiplications	The real number addition
				54	27	20	4

Table 5. reduces the per step complexity of complexity SIC-MMSE algorithm

	Complex multiplication	Complex addition	Real multiplications	The real number addition
					X	M/2	M	M

Y-H X	Nt*Nr	(Nt1)*Nr+Nr
					Wk’*(Y-H X)	Nr	Nr-1
LLR	(2*M)*Nt	1*Nt	(M+Log2(M)/2)*Nt	Nt*Log2(M)/2

(5) simulation result

We utilize simulation result that the performance of low complex degree iterative receiver is described.We mainly consider the performance of BER under the different signal to noise ratios.Suppose that channel is the quasistatic decline, constant when promptly channel in a frame is, and the fading coefficients between two frames is separate.At receiving terminal, we suppose that receiver has desirable channel information.Transmitting terminal adopts random interleaver, interleaver is not done any optimization.Chnnel coding adopts 1/2 code check, and the code word generator polynomial is [75] or [171 133], and perhaps Turbo encodes, the QPSK modulation.

In first experiment, we consider the system of 4 emissions and 4 reception antennas.Our more traditional iterative receiver and reduce the performance of the iterative receiver of complexity.

Fig. 4 represents to reduce the performance of iterative receiver from 1 time to 5 iteration of complexity.Find out that from figure in the time of 3 iteration, performance of BER begins convergence.

Our more traditional iterative receiver of Fig. 5 and reduce by 1 iteration of iterative receiver of complexity and the performance of 5 iteration.Reducing the performance loss of the iterative receiver of complexity almost can ignore.

We relatively adopt Fig. 6 under the Turbo coded system, the performance of the iterative receiver of traditional iterative receiver and reduction complexity.No matter adopt convolutional encoding or Turbo coding, because in first time iteration, covariance matrix is I _{Nt * Nt}, the performance of two kinds of iterative receivers is identical.Iterative process below reduces the performance loss of the iterative receiver of complexity and almost can ignore.The performance of BER of two kinds of methods all improves along with the increase of iterations.

The performance that we see the QRD-M algorithm at Fig. 7 is along with the reserved state number is increased to 16 and improve from 2,4.From the analysis of (4), when the reserved state number was 16, the complexity of QRD-M algorithm was higher than the SIC-MMSE algorithm that reduces complexity.

We have provided under the MIMO-OFDM system of 4 emissions and 4 receptions at Fig. 8, [75] 1/2 code checks, convolutional encoding, QPSK modulation.The carrier frequency of system is 2.0GHz, and 512 subcarriers, circulating prefix-length are 64, user velocity 3Km/h.Under the PA channel, simplification soft goes into the soft iterative receiver performance that goes out.The relevant district of time-frequency is chosen as T=18 (OFDM symbolic number), F=52 (subcarrier number).The iterative receiver performance loss of low complex degree is little.

Claims (7)

1, a kind of soft in and soft out detection method that reduces complexity is characterized in that, utilizes the mean value of emission symbolic vector covariance matrix to replace instantaneous covariance matrix, thereby T*F*Nt matrix inversion is reduced to a matrix inversion.

2, the soft in and soft out detection method that reduces complexity according to claim 1 is characterized in that, under quasistatic mimo channel situation, utilizes the time average of emission symbolic vector covariance matrix to replace instantaneous covariance matrix, can be expressed as: $\stackrel{\&OverBar;}{v}=\frac{1}{T}\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}{v}_t,$ v _tBe t emission symbol covariance matrix constantly, v is the time average of emission symbolic vector variance matrix, and T is the time interval.

3, the soft in and soft out detection method that reduces complexity according to claim 2 is characterized in that, the duration length that described time interval T is a quasi-static channel opens.

4, the soft in and soft out detection method that reduces complexity according to claim 2 is characterized in that, described v _tBe diagonal matrix.

5, the soft in and soft out detection method that reduces complexity according to claim 1, it is characterized in that, in the high relevant range of time of MIMO-OFDM system and frequency, utilize the mean value of emission symbolic vector covariance matrix to replace instantaneous covariance matrix, the channel value in the middle of promptly utilizing replaces the channel value in whole zone:

$\stackrel{\&OverBar;}{v}=\frac{1}{\mathrm{TF}}\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}\underset{f=1}{\overset{F}{\mathrm{\&Sigma;}}}{v}_{t,f}$

V wherein _{T, f}Be t _ThThe OFDM symbol, f _ThThe emission symbol covariance matrix of subcarrier, T and F are the high relevant ranges of channel.

6. the soft in and soft out detection method that reduces complexity according to claim 1, it is characterized in that, in the frequency domain of MIMO-OFDM system, this moment T=1, utilize the mean value of emission symbolic vector covariance matrix to replace instantaneous covariance matrix, the channel value in the middle of promptly utilizing replaces the channel value in whole zone

$\stackrel{\&OverBar;}{v}=\frac{1}{F}\underset{f=1}{\overset{F}{\mathrm{\&Sigma;}}}{v}_f$

V wherein _fBe f _ThThe emission symbol covariance matrix of subcarrier, F are the high relevant ranges of the frequency of channel.

7, according to claim 2,5, the 6 described soft in and soft out detection method that reduce complexity, it is characterized in that intermediate variable g _t=g@ (H VH ^H+ σ ²I _Nr), wherein H is the fading factor of the channel of reception antenna, σ ²Be the power of white Gaussian noise, I _NrIt is the unit matrix of Nr * Nr.

Images