CN101447969B - Channel estimation method of multi-band orthogonal frequency division multiplexing ultra wide band system - Google Patents

Abstract

The invention discloses a channel estimation method of multi-band orthogonal frequency division multiplexing ultra wide band system. The invention has the advantages that m sequence having better auto-correlation characteristics is taken as a time-domain training sequence and a cyclic prefix is appended; impulse response estimation value is obtained by doing mutual-correlation operation to receiving signals the cyclic prefix of which is eliminated and the training sequence and doing auto-correlation operation to each training sequence on a receiving terminal; that an auto-correlation matrix of the m sequence having diagonally dominant characteristics is utilized; one diagonal decomposition or three diagonal decomposition is respectively carried out to the auto-correlation matrix of the m sequence firstly; then an approximation method of first-order inverse matrix is adopted to effectively avoid complex inverse operation; then the operation amount is reduced by an order of magnitude; the performance approaches normal time-domain channel estimation method; and the invention is the fast and effective channel estimation method of the ultra wide band system, and is easy to be realized.

Description

A kind of channel estimation methods of multi-band orthogonal frequency division multiplexing ultra wide band system

Technical field

The present invention relates to a kind of channel estimation methods, especially relate to a kind of channel estimation methods of multi-band orthogonal frequency division multiplexing ultra wide band system.

Background technology

Ultra broadband (UWB, Ultra Wide Band) technology has the high speed of potentiality, an in-plant wireless personal communications technology as a kind of, all causes great concern in academia and industrial quarters in recent years, becomes the focus of present wireless communication field research and development.Super-broadband tech is in conjunction with multi-band orthogonal frequency division multiplexing (MB-OFDM, Multi-Band OrthogonalFrequency Division Multiplexing) technology constitutes multi-band orthogonal frequency division multiplexing ultra wide band (MB-OFDM UWB) technology, it can resist multipath fading and various narrow band interference effectively and to the characteristics such as flexible utilization of frequency spectrum resource, become one of super-broadband tech main flow implementation.Multi-band orthogonal frequency division multiplexing ultra wide band The Application of Technology prospect is very tempting, as all having a wide range of applications, especially in the application in digital home electronic product field in a high-speed radio territory net, wireless ether interface link, intelligent wireless local area network, outdoor peer-to-peer network and various fields such as sensing, location and recognition network.At present, the breach of the application of wireless family electronic product as the multi-band orthogonal frequency division multiplexing ultra wide band technology all selected by numerous companies.

Multi-band orthogonal frequency division multiplexing ultra wide band system will obtain perfect performance, the just essential technology such as coherent detection, demodulation, equilibrium that adopt, these technology all need to utilize the information of channel, therefore accurately channel estimating information for guaranteeing in the multi-band orthogonal frequency division multiplexing ultra wide band communication environment that transfer of data plays crucial effects reliably.Because the shared bandwidth of ultra-broadband signal is big, signal duration is short, transmission rate is high, this has just proposed estimated accuracy height, requirement that computation complexity is low to channel estimation technique.Therefore, how carrying out fast and effectively in multi-band orthogonal frequency division multiplexing ultra wide band system, channel estimating is a major challenge that present multi-band orthogonal frequency division multiplexing ultra wide band technology is faced.

Multi-band orthogonal frequency division multiplexing ultra wide band system has mostly adopted pilot tone frequency domain channel estimation approach, promptly inserts pilot tone at frequency domain, and carries out channel estimating at frequency domain.This class channel estimation methods may further comprise the steps: at first, insert pilot tone in the appropriate location of transmitting terminal frequency domain, utilize pilot data to obtain the channel information of pilot frequency locations by corresponding channel estimating criterion at receiving terminal

Pass through interpolater then, utilize the mode of interpolation right In whole frequency domain, carry out interpolation, so that obtain whole channel estimation value

At last channel estimation value and reception data are sent into equalizer, just can be to receiving the estimated value that data balancing obtain original transmission data.

At present, the channel information at above-mentioned pilot frequency locations normally obtains based on least square (LS, Least Squares) criterion or least mean-square error (MMSE, Minimum Mean Square Error) criterion.Wherein, based on the simple and realization easily of pilot tone frequency domain channel estimation method computational process of criterion of least squares, but this method is not considered The noise, thereby causes channel estimated accuracy not high.Based on the pilot tone frequency domain channel estimation method of minimum mean square error criterion owing to utilized the frequency domain autocorrelation performance of channel, so can obtain good performance, but relate to matrix inversion in the estimation procedure of this method, increase the computation complexity of this method, caused this method exploitativeness poor.In sum, there is the computation complexity height in more existing pilot tone frequency domain channel estimation methods, are difficult to reality, and the characteristic of channel of putting because of non-pilot symbol need use the mode of interpolation, cause problems such as computational accuracy is not high.

Present time-domain channel estimating method mainly contains based on discrete Fourier (DFT, Discrete Fourier Transform) filter method and maximum-likelihood criterion (ML, Maximum Likelihood) method of estimation, these two class methods can reduce the square mean error amount of channel estimating to a certain extent, but shortcoming is channel length (or limited time delay expansion of channel a) informational needs accurately to be obtained before channel estimating, thereby increased the duration and the computation complexity of channel estimation process, made these two class methods be restricted in actual applications.

People such as Bowei Song have proposed a kind of time-domain channel estimating method based on the m sequence, use this method multi-band orthogonal frequency division multiplexing ultra wide band system workflow as shown in Figure 1.At transmitting terminal, the data-signal of input is through orthogonal phase shift modulation (QPSK, Quadrature Phase Shift Keying) obtains modulation signal, modulation signal is by string and conversion, inverse Fourier transform (IFFT, Inverse Fast Fourier Transform) and and go here and there and form a plurality of OFDM symbols after the conversion process, every the OFDM of fixed qty symbol, inserting a length is L _PThe training sequence estimated as time domain channel of m sequence s, and to add length according to the quality of the characteristic of channel be L _CCyclic Prefix (CP, Cyclic Prefix), and the supposition multi-band orthogonal frequency division multiplexing ultra wide band system be synchronous, the training sequence that obtains after the pended cyclic prefix and the data-signal of input transmit by ultra-wideband channel after carrier modulation treatment together; At receiving terminal, at first remove the channel decline that receives and the Cyclic Prefix in the training sequence after the white Gaussian noise influence, will remove the training sequence after the channel decline of Cyclic Prefix and white Gaussian noise influence then

With the m sequence s behind the m sequence s ring shift right i position ⁱMake related operation, $\begin{array}{c}C(i,j)=(1/L_P)\underset{k=0}{\overset{L_P-1}{\mathrm{\&Sigma;}}}\tilde{r}\left(k\right)s^i\left(k\right)\\ =\underset{j=0}{\overset{L_C-1}{\mathrm{\&Sigma;}}}{h}_j{C}_P(i,j)+(1/L_P)\underset{k=0}{\overset{L_p-1}{\mathrm{\&Sigma;}}}n\left(k\right)s^i\left(k\right)\end{array},$ Wherein, k=0,1 ..., L _p+ L _c-1, h represents the matrix-vector that the coefficient by each multipath of channel constitutes, $h={[h_0,{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="H2008101642240E00031.png"\; state="\{\{state\}\}">h</figure-callout>}}_1,\&CenterDot;\&CenterDot;\&CenterDot;h_{L_{C-1}}]}^T,$ h _jBe j multi-path coefficients of channel, h should satisfy condition: { h _j=0|L≤j≤L _C-1}, L are the exponent number of channel, C _P(i j) is m sequence s behind the m sequence s ring shift right j position ^jWith the m sequence s behind the ring shift right i position ⁱThe normalized autocorrelation coefficient, second is white Gaussian noise sequence n and m sequence s normalized crosscorrelation coefficient, n is a white Gaussian noise, n (k) is a k white Gaussian noise constantly.The amplitude of noise is compressed into original 1/L _PDoubly, promptly $(1/L_P)\underset{k=0}{\overset{L_p-1}{\mathrm{\&Sigma;}}}n\left(k\right)s^i\left(k\right).$ Like this can with $\begin{array}{c}C(i,j)=(1/L_P)\underset{k=0}{\overset{L_P-1}{\mathrm{\&Sigma;}}}\tilde{r}\left(k\right)s^i\left(k\right)\\ =\underset{j=0}{\overset{L_C-1}{\mathrm{\&Sigma;}}}{h}_j{C}_P(i,j)+(1/L_P)\underset{k=0}{\overset{L_p-1}{\mathrm{\&Sigma;}}}n\left(k\right)s^i\left(k\right)\end{array}$ Be approximated to C ≈ C _PH, wherein C _PBe the autocorrelation matrix of m sequence s, the length of inserting m sequence s is L _P, autocorrelation matrix C then _PNormalized autocorrelation functions in one-period satisfies: $\begin{array}{c}{C}_P(i,j)=(1/L_P)\underset{k=0}{\overset{L_P-1}{\mathrm{\&Sigma;}}}{s}^j\left(k\right)s^i\left(k\right)\\ =\left\{\begin{array}{cc}1,& i=j\\ -1/L_P,& i\&NotEqual;j\end{array}\right.\end{array}.$ Utilize the autocorrelation performance of m sequence to obtain the impulse response estimated value of channel thus $\tilde{h}=C_p^{-1}C.$ This method has utilized the autocorrelation performance of m sequence to obtain the estimated value of channel impulse response cleverly, its estimated accuracy is very high, and can adjust the expense of training sequence flexibly according to the needs of multi-band orthogonal frequency division multiplexing ultra wide band communication system transmits speed, to obtain the compromise of estimated accuracy and expense.But from $\tilde{h}=C_p^{-1}C$ As can be known, expect channel impulse response, Matrix C _PInversion operation be absolutely necessary, yet autocorrelation matrix C _P=[C _P(i, j)], i=0,1 ... L _p-1, j=0,1 ... L _p-1 is a L _PThe rank square formation is inverted to it if desired, and its computation complexity is very high, and (computation complexity is o (L _p ³)), high computation complexity has brought very big obstacle for the application of this method.

Summary of the invention

Technical problem to be solved by this invention is the deficiency that exists at prior art, and a kind of channel estimation methods that is applicable to multi-band orthogonal frequency division multiplexing ultra wide band system of low computation complexity is provided.

The present invention solves the problems of the technologies described above the technical scheme that is adopted: a kind of channel estimation methods of multi-band orthogonal frequency division multiplexing ultra wide band system, may further comprise the steps:, at first the data-signal of importing is carried out the orthogonal phase shift modulation treatment and obtain modulation signal 1. at transmitting terminal; 2. then modulation signal is gone here and there successively and conversion, inverse Fourier transform and and go here and there conversion process, form a plurality of OFDM symbols; 3. again in a plurality of OFDM symbols that form, inserting a length every the OFDM symbol of setting quantity is L _PM sequence s, as a training sequence, and be L with m sequence s according to the characteristic of channel additional length before training sequence _CCyclic Prefix, obtain the training sequence after the pended cyclic prefix, represent with x, x=[x (0), x (1) ... x (L _P+ L _C-1)]; 4. last OFDM symbol with training sequence x after the pended cyclic prefix and formation transfers to receiving terminal by ultra-wideband channel together after carrier modulation treatment, training sequence x in transmission course after the pended cyclic prefix and OFDM symbol are subjected to the influence of channel fading and white Gaussian noise; 5. at receiving terminal, the channel decline that the definition receiving terminal receives and the training sequence x of the pended cyclic prefix after the white Gaussian noise influence are first received signal, channel decline that the definition receiving terminal receives and the OFDM symbol after the white Gaussian noise influence are second received signal, and first received signal is expressed as with tapped delay line model $r\left(k\right)=\underset{t=0}{\overset{L_C-1}{\mathrm{\&Sigma;}}}{h}_{t}x(k-t)+n\left(k\right),$ Wherein, k=0,1 ..., L _p+ L _c-1, r (k) is k first received signal constantly, and h represents the matrix-vector that the coefficient by each multipath of channel constitutes, $h={[h_0,{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="H2008101642240E00031.png"\; state="\{\{state\}\}">h</figure-callout>}}_1,\&CenterDot;\&CenterDot;\&CenterDot;h_{L_{C-1}}]}^T,$ h _tBe t multi-path coefficients of channel, h should satisfy condition: { h _t=0|L≤t≤L _C-1}, L are the exponent number of channel, and x is the training sequence after the pended cyclic prefix, and x (k-t) is the training sequence after the k-t pended cyclic prefix constantly, and n is a white Gaussian noise, and n (k) is a k white Gaussian noise constantly; 6. at first the first received signal r (k) is unloaded ripple modulation, and go circulation prefix processing to obtain first received signal of unloading after the ripple modulation treatment $\tilde{r}\left(k\right)=\underset{j=0}{\overset{L_C-1}{\mathrm{\&Sigma;}}}{h}_j{s}^j\left(k\right)+n\left(k\right),$ Wherein, k=0,1 ..., L _p+ L _c-1,

For going k first received signal constantly behind the Cyclic Prefix, h represents the matrix-vector that each multi-path coefficients by channel constitutes, $h={[h_0,{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="H2008101642240E00031.png"\; state="\{\{state\}\}">h</figure-callout>}}_1,\&CenterDot;\&CenterDot;\&CenterDot;h_{L_{C-1}}]}^T,$ h _jBe j multi-path coefficients of channel, h should satisfy condition: { h _j=0|L≤j≤L _C-1}, L are the exponent number of channel, and n is a white Gaussian noise, and n (k) is a k white Gaussian noise constantly, s ^jBe the m sequence behind the m sequence s ring shift right j position, s ^j(k) be the sequence in the k moment behind the m sequence s ring shift right j position; 7. calculate first received signal go behind the Cyclic Prefix then

With the m sequence s behind the m sequence s ring shift right i position ⁱCross-correlation matrix C and the autocorrelation matrix C of each training sequence s _P, C=[C (i, j)], (i is j) for going first received signal behind the Cyclic Prefix for C

With the m sequence s behind the m sequence s ring shift right i ⁱThe normalized crosscorrelation coefficient, $C(i,j)=(1/L_P)\underset{k=0}{\overset{L_P-1}{\mathrm{\&Sigma;}}}\tilde{r}\left(k\right)s^i\left(k\right),$ C _P=[C _P(i, j)], C _P(i j) is m sequence s behind the m sequence s ring shift right j position ^jWith the m sequence s behind the m sequence s ring shift right i position ⁱThe normalized autocorrelation coefficient, $\begin{array}{c}{C}_P(i,j)=(1/L_P)\underset{k=0}{\overset{L_p-1}{\mathrm{\&Sigma;}}}{s}^j\left(k\right)s^i\left(k\right)\\ =\left\{\begin{array}{cc}1,& i=j\\ -1/L_P,& i\&NotEqual;j\end{array}\right.\end{array}$ , wherein, i=0,1 ..., L _p, j=0,1 ..., L _p, k=0,1 ..., L _p+ L _c-1, For removing k first received signal constantly behind the Cyclic Prefix, s ^j(k) be the sequence in the k moment behind the m sequence s ring shift right j position, s ⁱ(k) be the sequence in the k moment behind the m sequence s ring shift right i position.8. again according to first received signal of going behind the Cyclic Prefix

With the m sequence s behind the m sequence s ring shift right i position ⁱCross-correlation matrix C and the autocorrelation matrix C of each training sequence s _P, the impulse response estimated value of calculating channel

$\tilde{h}=C_p^{-1}C,$ Wherein, C _p ^-1Be autocorrelation matrix C _PInverse matrix; According to described autocorrelation matrix C _PDiagonal dominance, with described autocorrelation matrix C _PBe decomposed into first matrix and the second matrix sum, described first matrix is designated as D, described second matrix is designated as E, C _P=D+E, satisfy in described first matrix D and the described second matrix E || D ^-1Described autocorrelation matrix C is calculated in E||＜1 o'clock _PInverse matrix C _P ^-1, $C_p^{-1}=(I-D^{-1}E+{\left(D^{-1}E\right)}^2+\&CenterDot;\&CenterDot;\&CenterDot;+{(-1)}^m{\left(D^{-1}E\right)}^m+\&CenterDot;\&CenterDot;\&CenterDot;)D^{-1},$ Wherein, symbol " || || " be the norm symbol, I is a unit matrix, D ^-1Be the inverse matrix of first matrix D, m=1,2 ..., ∞; Basis again $C_p^{-1}=(I-D^{-1}E+{\left(D^{-1}E\right)}^2+\&CenterDot;\&CenterDot;\&CenterDot;+{(-1)}^m{\left(D^{-1}E\right)}^m+\&CenterDot;\&CenterDot;\&CenterDot;)D^{-1}$ Calculate C _p ^-1The first approximation value, $C_p^{-1}\&ap;(I-D^{-1}E)D^{-1}\&ap;D^{-1}-D^{-1}{\mathrm{ED}}^{-1}.$

Described first matrix D is by described autocorrelation matrix C _PThe diagonal matrix formed of diagonal entry, the described second matrix E is by described autocorrelation matrix C _PThe non-diagonal matrix formed of off diagonal element, described diagonal matrix is designated as D ₁, described non-diagonal matrix is designated as E ₁, obtain $C_p^{-1}\&ap;D_1^{-1}-D_1^{-1}{E}_1{D}_1^{-1};$ To described autocorrelation matrix C _PCoefficient carry out normalized, described diagonal matrix D after the normalized ₁Be a unit matrix I; According to $C_p^{-1}\&ap;D_1^{-1}-D_1^{-1}{E}_1{D}_1^{-1}$ Obtain $C_p^{-1}=I-{\mathrm{<figure-callout\; id="1"\; label="E"\; filenames="H2008101642240E00031.png"\; state="\{\{state\}\}">E</figure-callout>}}_1.$

Described first matrix D is by described autocorrelation matrix C _PThe triple diagonal matrix formed of three diagonal elements, the described second matrix E is by described autocorrelation matrix C _PThe non-triple diagonal matrix formed of the element except that three diagonal elements, described triple diagonal matrix is designated as D ₃, described non-triple diagonal matrix is designated as E ₃, obtain $C_p^{-1}\&ap;D_3^{-1}-D_3^{-1}{E}_3{D}_3^{-1};$ With described triple diagonal matrix D ₃Be decomposed into by described autocorrelation matrix C _PThe diagonal matrix formed of diagonal entry and by described autocorrelation matrix C _PDiagonal entry be the two diagonal matrix sums that 0 two diagonal elements are formed, described diagonal matrix is designated as D ₁, described two diagonal matrix are designated as D ₂, calculate described triple diagonal matrix D ₃Inverse matrix D ₃ ^-1, $D_3^{-1}=(I-D_1^{-1}{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="H2008101642240E00012.png,H2008101642240E00021.png"\; state="\{\{state\}\}">D</figure-callout>}}_2+{\left(D_1^{-1}{D}_2\right)}^2+\&CenterDot;\&CenterDot;\&CenterDot;+{(-1)}^m{\left(D_1^{-1}{D}_2\right)}^m+\&CenterDot;\&CenterDot;\&CenterDot;)D_1^{-1}$ , wherein, I is a unit matrix, D ₁ ^-1Be diagonal matrix D ₁Inverse matrix, m=1,2 ..., ∞; Basis then $D_3^{-1}=(I-D_1^{-1}{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="H2008101642240E00012.png,H2008101642240E00021.png"\; state="\{\{state\}\}">D</figure-callout>}}_2+{\left(D_1^{-1}{D}_2\right)}^2+\&CenterDot;\&CenterDot;\&CenterDot;+{(-1)}^m{\left(D_1^{-1}{D}_2\right)}^m+\&CenterDot;\&CenterDot;\&CenterDot;)D_1^{-1}$ Calculate D ₃ ^-1The first approximation value, $D_3^{-1}\&ap;(I-D_1^{-1}{D}_2)D_1^{-1};$ To described autocorrelation matrix C _PCoefficient carry out normalized, diagonal matrix D after the normalized ₁Be unit matrix I, according to $D_3^{-1}\&ap;(I-D_1^{-1}{D}_2)D_1^{-1}$ Obtain $D_3^{-1}=I-{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="H2008101642240E00012.png,H2008101642240E00021.png"\; state="\{\{state\}\}">D</figure-callout>}}_2;$ Basis again $C_p^{-1}\&ap;D_3^{-1}-D_3^{-1}{E}_3{D}_3^{-1}$ With $D_3^{-1}=I-{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="H2008101642240E00012.png,H2008101642240E00021.png"\; state="\{\{state\}\}">D</figure-callout>}}_2,$ Obtain $C_p^{-1}\&ap;(I-D_2)-(I-D_2)E_3(I-D_2).$

Compared with prior art, the invention has the advantages that adopt autocorrelation performance preferably the m sequence as time-domain training sequence, at receiving terminal by first received signal of removing Cyclic Prefix and training sequence being made computing cross-correlation and each training sequence being obtained the impulse response estimated value of channel as auto-correlation computation, and the autocorrelation matrix that utilizes the m sequence has the diagonal dominance characteristic, carry out diagonal angle decomposition or the decomposition of three diagonal angles by autocorrelation matrix at first respectively to the m sequence, adopt the approach method of single order inverse matrix then, effectively avoided complicated inversion operation, make operand reduce an order of magnitude, and performance is approached conventional time-domain channel estimating method, be a kind of channel estimation methods fast and effectively of radio ultra wide band system, be easy to realize.

Description of drawings

Fig. 1 is the workflow schematic diagram of multi-band orthogonal frequency division multiplexing ultra wide band system;

Fig. 2 is the curve chart that the bit error rate of the conventional time-domain channel estimating method of corresponding different length m sequence and LS algorithm changes with signal to noise ratio;

Fig. 3 is L _PThe performance comparison diagram of=31 o'clock LS algorithms, conventional time domain method of estimation, a diagonal angle decomposition method of the present invention and three diagonal angle decomposition methods;

Fig. 4 is L _PThe performance comparison diagram of=15 o'clock conventional time domain methods of estimation, a diagonal angle decomposition method of the present invention and three diagonal angle decomposition methods.

Embodiment

Embodiment describes in further detail the present invention below in conjunction with accompanying drawing.

A kind of channel estimation methods of multi-band orthogonal frequency division multiplexing ultra wide band system may further comprise the steps:

1. at transmitting terminal, at first adopt existing orthogonal phase shift modulation (QPSK) technology that the data-signal of importing is carried out the orthogonal phase shift modulation treatment and obtain modulation signal.

2. then modulation signal is gone here and there successively and conversion, inverse Fourier transform (IFFT) and and go here and there conversion process, form a plurality of OFDM symbols.Each OFDM symbol adopts 128 subcarriers in the present embodiment, the frequency interval 4.1254MHz between the adjacent sub-carrier, and the duration of each OFDM symbol is T ₀=242.4ns.

3. again in a plurality of OFDM symbols that form, inserting a length every the OFDM symbol of setting quantity is L _PM sequence s, as a training sequence, and be L with m sequence s according to the quality of the characteristic of channel additional length before training sequence _CCyclic Prefix (CP), obtain the training sequence after the pended cyclic prefix, the training sequence after this pended cyclic prefix is represented with x, x=[x (0), x (1) ... x (L _P+ L _C-1)], wherein, L _PFor m sequence s is the length of training sequence, L _CLength for Cyclic Prefix.In the present embodiment, the setting quantity of choosing is 4, and promptly inserting a length every 4 OFDM symbols is L _PM sequence s.

4. last OFDM symbol with training sequence x after the pended cyclic prefix and formation transfers to receiving terminal by ultra-wideband channel together after carrier modulation treatment, training sequence x in transmission course after the pended cyclic prefix and OFDM symbol will be subjected to the influence of channel fading and white Gaussian noise.

5. at receiving terminal, the channel decline that the definition receiving terminal receives and the training sequence x of the pended cyclic prefix after the white Gaussian noise influence are first received signal, channel decline that the definition receiving terminal receives and the OFDM symbol after the white Gaussian noise influence are second received signal, and first received signal is expressed as with tapped delay line model $r\left(k\right)=\underset{t=0}{\overset{L_C-1}{\mathrm{\&Sigma;}}}{h}_{t}x(k-t)+n\left(k\right),$ Wherein, k=0,1 ..., L _p+ L _c-1, r (k) is k first received signal constantly, and h represents the matrix-vector that each multi-path coefficients by channel constitutes, $h={[h_0,{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="H2008101642240E00031.png"\; state="\{\{state\}\}">h</figure-callout>}}_1,\&CenterDot;\&CenterDot;\&CenterDot;h_{L_{C-1}}]}^T,$ h _tBe t multi-path coefficients of channel, h should satisfy condition: { h _t=0|L≤t≤L _C-1}, L are the exponent number of channel, and x is the training sequence after the pended cyclic prefix, and x (k-t) is the training sequence after the k-t pended cyclic prefix constantly, and n is a white Gaussian noise, and n (k) is a k white Gaussian noise constantly.

6. at first the first received signal r (k) is unloaded ripple modulation, and go circulation prefix processing to obtain first received signal of unloading after the ripple modulation treatment $\tilde{r}\left(k\right)=\underset{j=0}{\overset{L_C-1}{\mathrm{\&Sigma;}}}{h}_j{s}^j\left(k\right)+n\left(k\right),$ Wherein, k=0,1 ..., L _p+ L _c-1,

For going k first received signal constantly behind the Cyclic Prefix, h represents the matrix-vector that each multi-path coefficients by channel constitutes, $h={[h_0,{\mathrm{<figure-callout\; id="1"\; label="h"\; filenames="H2008101642240E00031.png"\; state="\{\{state\}\}">h</figure-callout>}}_1,\&CenterDot;\&CenterDot;\&CenterDot;h_{L_{C-1}}]}^T,$ h _jBe j multi-path coefficients of channel, h should satisfy condition: { h _j=0|L≤j≤L _C-1}, L are the exponent number of channel, and n is a white Gaussian noise, and n (k) is a k white Gaussian noise constantly, s ^jBe the m sequence behind the m sequence s ring shift right j position, s ^j(k) be the sequence in the k moment behind the m sequence s ring shift right j position.

7. calculate first received signal go behind the Cyclic Prefix then

With the m sequence s behind the m sequence s ring shift right i position ⁱCross-correlation matrix C and the autocorrelation matrix C of each training sequence s _P, C=[C (i, j)], (i is j) for going first received signal behind the Cyclic Prefix for C

With the m sequence s behind the m sequence s ring shift right i ⁱThe normalized crosscorrelation coefficient, $C(i,j)=(1/L_P)\underset{k=0}{\overset{L_P-1}{\mathrm{\&Sigma;}}}\tilde{r}\left(k\right)s^i\left(k\right),$ C _P=[C _P(i, j)], C _P(i j) is m sequence s behind the m sequence s ring shift right j position ^jWith the m sequence s behind the m sequence s ring shift right i position ⁱThe normalized autocorrelation coefficient, $\begin{array}{c}{C}_P(i,j)=(1/L_P)\underset{k=0}{\overset{L_P-1}{\mathrm{\&Sigma;}}}{s}^j\left(k\right)s^i\left(k\right)\\ =\left\{\begin{array}{cc}1,& i=j\\ -1/L_P,& i\&NotEqual;j\end{array}\right.\end{array},$ Wherein, i=0,1 ..., L _p, j=0,1 ..., L _p, k=0,1 ..., L _p+ L _c-1,

For removing k first received signal constantly behind the Cyclic Prefix, s ^j(k) be the sequence in the k moment behind the m sequence s ring shift right j position, s ⁱ(k) be the sequence in the k moment behind the m sequence s ring shift right i position.

8. again according to first received signal of going behind the Cyclic Prefix

With the m sequence s behind the m sequence s ring shift right i position ⁱCross-correlation matrix C and the autocorrelation matrix C of each training sequence s _P, the impulse response estimated value of calculating channel

$\tilde{h}=C_p^{-1}C,$ C _p ^-1The autocorrelation matrix C that represents each training sequence s _PInverse matrix.In this step, in the impulse response estimated value of calculating channel

Before earlier according to autocorrelation matrix C _PDiagonal dominance, with autocorrelation matrix C _PBe decomposed into first matrix and the second matrix sum, first matrix is designated as D, second matrix is designated as E, C is then arranged _P=D+E,, satisfy in first matrix D and the second matrix E || D ^-1Autocorrelation matrix C is calculated in E||＜1 o'clock _PInverse matrix C _p ^-1, $C_p^{-1}=(I-D^{-1}E+{\left(D^{-1}E\right)}^2+\&CenterDot;\&CenterDot;\&CenterDot;+{(-1)}^m{\left(D^{-1}E\right)}^m+\&CenterDot;\&CenterDot;\&CenterDot;)D^{-1},$ Wherein, symbol " || || " be the norm symbol, I is a unit matrix, D ^-1Be the inverse matrix of first matrix D, m=1,2 ..., ∞; If only consider C _p ^-1The first approximation value, basis then $C_p^{-1}=(I-D^{-1}E+{\left(D^{-1}E\right)}^2+\&CenterDot;\&CenterDot;\&CenterDot;+{(-1)}^m{\left(D^{-1}E\right)}^m+\&CenterDot;\&CenterDot;\&CenterDot;)D^{-1}$ Calculate C _p ^-1The first approximation value, $C_p^{-1}\&ap;(I-D^{-1}E)D^{-1}\&ap;D^{-1}-D^{-1}E{D}^{-1},$ Utilize at last $C_p^{-1}\&ap;D^{-1}-D^{-1}E{D}^{-1}$ The impulse response estimated value of calculating channel

$\begin{array}{c}\tilde{h}=C_p^{-1}C\\ =(D^{-1}-D^{-1}E{D}^{-1})C\end{array}.$

In order to reduce computation complexity, the present invention proposes two kinds and find the solution C _PThe quick approach method of inverse matrix: a diagonal angle decomposition method and three diagonal angle decomposition methods.

One diagonal angle decomposition method: with autocorrelation matrix C _PBe decomposed into by autocorrelation matrix C _PThe diagonal matrix formed of diagonal entry and by autocorrelation matrix C _PThe non-diagonal matrix sum formed of off diagonal element, diagonal matrix is designated as D ₁, non-diagonal matrix is designated as E ₁, C then _P=D ₁+ E ₁, wherein,

L _PLength for the training sequence that inserts at transmitting terminal; Because autocorrelation matrix C _PDiagonal dominance, $\left|\right|D_1^{-1}{\mathrm{<figure-callout\; id="1"\; label="E"\; filenames="H2008101642240E00031.png"\; state="\{\{state\}\}">E</figure-callout>}}_1\left|\right|<1,$ So C _P ^-1Following expansion is arranged: $C_p^{-1}=(I-D_1^{-1}{\mathrm{<figure-callout\; id="1"\; label="E"\; filenames="H2008101642240E00031.png"\; state="\{\{state\}\}">E</figure-callout>}}_1+{\left(D_1^{-1}{E}_1\right)}^2+\&CenterDot;\&CenterDot;\&CenterDot;+{(-1)}^m{\left(D_1^{-1}{E}_1\right)}^m+\&CenterDot;\&CenterDot;\&CenterDot;)D_1^{-1},$ Wherein, || || be norm symbol, (D ₁ ^-1E ₁) ^mBe D ₁ ^-1E ₁The m power, I is a unit matrix, D ₁ ^-1Be diagonal matrix D ₁Inverse matrix, m=1,2 ..., ∞; According to C _P ^-1Expansion obtain C _P ^-1The first approximation value be $C_p^{-1}\&ap;D_1^{-1}-D_1^{-1}{E}_1{D}_1^{-1},$ From $C_p^{-1}\&ap;D_1^{-1}-D_1^{-1}{E}_1{D}_1^{-1}$ As can be seen, adopt a diagonal angle decomposition method, only relate to inverting of diagonal matrix, autocorrelation matrix C _PCoefficient carry out normalized, diagonal matrix D after the normalized ₁Be a unit matrix I, so C _P ^-1The first approximation value be $C_p^{-1}=I-{\mathrm{<figure-callout\; id="1"\; label="E"\; filenames="H2008101642240E00031.png"\; state="\{\{state\}\}">E</figure-callout>}}_1,$ From $C_p^{-1}=I-{\mathrm{<figure-callout\; id="1"\; label="E"\; filenames="H2008101642240E00031.png"\; state="\{\{state\}\}">E</figure-callout>}}_1$ Can draw and calculate C _P ^-1Do not need the process of inverting, computation complexity reduces greatly, the autocorrelation matrix C and the quality of this method performance places one's entire reliance upon _PDiagonal dominance.

Three diagonal angle decomposition methods: with autocorrelation matrix C _PBe decomposed into by autocorrelation matrix C _PThe triple diagonal matrix formed of three diagonal elements and by autocorrelation matrix C _PThe non-triple diagonal matrix sum formed of the element except that three diagonal elements, triple diagonal matrix is designated as D ₃, non-triple diagonal matrix is designated as E ₃, C then _P=D ₃+ E ₃, wherein,

L _PLength for the training sequence that inserts at transmitting terminal; A similar diagonal angle decomposition method is because autocorrelation matrix C _PDiagonal dominance, $\left|\right|D_3^{-1}{E}_3\left|\right|<1,$ So C _p ^-1Following expansion is arranged: $C_p^{-1}=(I-D_3^{-1}{E}_3+{\left(D_3^{-1}{E}_3\right)}^2+\&CenterDot;\&CenterDot;\&CenterDot;+{(-1)}^m{\left(D_3^{-1}{E}_3\right)}^m+\&CenterDot;\&CenterDot;\&CenterDot;)D_3^{-1},$ Wherein, || || be norm symbol, (D ₃ ^-1E ₃) ^mBe D ₃ ^-1E ₃The m power, I is a unit matrix, D ₃ ^-1Be triple diagonal matrix D ₃Inverse matrix, m=1,2 ..., ∞; According to C _P ^-1Expansion obtain C _p ^-1The first approximation value can represent $C_p^{-1}\&ap;D_3^{-1}-D_3^{-1}{E}_3{D}_3^{-1};$ With triple diagonal matrix D ₃Be decomposed into by autocorrelation matrix C _PThe diagonal matrix formed of diagonal entry and by autocorrelation matrix C _PDiagonal entry be the two diagonal matrix sums that 0 two diagonal elements are formed, diagonal matrix is designated as D ₁, two diagonal matrix are designated as D ₂, D is then arranged ₃=D ₁+ D ₂, wherein,

Calculate triple diagonal matrix D ₃Inverse matrix D ₃ ^-1, $D_3^{-1}=(I-D_1^{-1}{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="H2008101642240E00012.png,H2008101642240E00021.png"\; state="\{\{state\}\}">D</figure-callout>}}_2+{\left(D_1^{-1}{D}_2\right)}^2+\&CenterDot;\&CenterDot;\&CenterDot;+{(-1)}^m{\left(D_1^{-1}{D}_2\right)}^m+\&CenterDot;\&CenterDot;\&CenterDot;)D_1^{-1}$ , wherein, I is a unit matrix, D ₁ ^-1Be diagonal matrix D ₁Inverse matrix, m=1,2 ..., ∞; Basis then $D_3^{-1}=(I-D_1^{-1}{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="H2008101642240E00012.png,H2008101642240E00021.png"\; state="\{\{state\}\}">D</figure-callout>}}_2+{\left(D_1^{-1}{D}_2\right)}^2+\&CenterDot;\&CenterDot;\&CenterDot;+{(-1)}^m{\left(D_1^{-1}{D}_2\right)}^m+\&CenterDot;\&CenterDot;\&CenterDot;)D_1^{-1}$ Calculate D ₃ ^-1The first approximation value, $D_3^{-1}\&ap;(I-D_1^{-1}{D}_2)D_1^{-1};$ To autocorrelation matrix C _PCoefficient carry out normalized, diagonal matrix D after the normalized ₁Be unit matrix I, according to $D_3^{-1}\&ap;(I-D_1^{-1}{D}_2)D_1^{-1}$ Obtain $D_3^{-1}=I-{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="H2008101642240E00012.png,H2008101642240E00021.png"\; state="\{\{state\}\}">D</figure-callout>}}_2;$ Basis again $C_p^{-1}\&ap;D_3^{-1}-D_3^{-1}{E}_3{D}_3^{-1}$ With $D_3^{-1}=I-{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="H2008101642240E00012.png,H2008101642240E00021.png"\; state="\{\{state\}\}">D</figure-callout>}}_2,$ Obtain $C_p^{-1}\&ap;(I-D_2)-(I-D_2)E_3(I-D_2),$ Three diagonal angle decomposition methods calculate C as can be known _P ^-1Equally also can not need the process of inverting, computation complexity reduces greatly.

Table 1 has been listed the size of the computation complexity of existing time-domain channel estimating method based on the m sequence (being called the method for directly inverting in table 1), a diagonal angle decomposition method of the present invention and three diagonal angle decomposition methods.

Table 1 computation complexity comparison sheet

The method of directly inverting	One diagonal angle decomposition method	Three diagonal angle decomposition methods
			o(L p 3)	o(L p 2)	o(L p 2)

As shown in Table 1, the processing method that employing one diagonal angle decomposes and three diagonal angles decompose of the present invention's proposition can reduce the computation complexity of the existing method of directly inverting greatly.

It is the diagonal dominance characteristic of the autocorrelation matrix of m sequence that the present invention has utilized special training sequence, at first carry out diagonal angle decomposition and the decomposition of three diagonal angles by autocorrelation matrix to the m sequence, adopt the approach method of single order inverse matrix then, compare with traditional time-domain channel estimating method, the diagonal angle decomposition method that the present invention proposes and the operand of three diagonal angle decomposition methods have reduced an order of magnitude, and performance is approached traditional time domain channel estimated value.Computer artificial result has been verified validity of the present invention.

Insert a training sequence every 4 OFDM symbols, insert L respectively _P=15,31,63, the m sequence of 127 4 kind of different length.The curve that the conventional time-domain channel estimating method (method of promptly directly inverting) that Fig. 2 has compared corresponding different length m sequence and the bit error rate of frequency domain LS channel estimation methods change with signal to noise ratio.Know easily that from Fig. 2 under identical signal to noise ratio condition, the m sequence length is short more, the bit error rate of conventional time-domain channel estimating method is high more, and performance is poor more, and the m sequence length is long more, and its bit error rate is low more, and performance is good more.The purpose of Fig. 2 emulation is in order to choose the length of suitable training sequence.But consider operand and systematic function, pilot length is unsuitable long in the reality.Because the m sequence length is got L respectively _P=31, L _P=64 o'clock performance is close, therefore should consider to choose length L in practice _P=31 m sequence is proper.

Fig. 3 has compared m sequence length L _P=31 o'clock, conventional time domain method of estimation, a diagonal angle decomposition method, three diagonal angle decomposition methods, and the bit error rate of frequency domain LS channel estimation methods is with the change curve of signal to noise ratio.As can be seen from Figure 3, it is very close on performance with conventional time domain method of estimation with three diagonal angle decomposition methods to adopt a diagonal angle to decompose, but the computation complexity of preceding two kinds of methods has reduced an order of magnitude.

For a further relatively diagonal angle decomposes and the performance quality of three diagonal angle decomposition methods, and with the performance difference of conventional time domain method of estimation, Fig. 4 has provided these three kinds of methods (m sequence length L under the poor situation of m sequence autocorrelation performance _P=15 o'clock) performance simulation.As can be seen from Figure 4, the performance that the present invention proposes the diagonal angle decomposition method all slightly descends, but three diagonal angle decomposition method performances obviously are better than a diagonal angle decomposition method.

Claims (1)

1. the channel estimation methods of a multi-band orthogonal frequency division multiplexing ultra wide band system may further comprise the steps: 1. at transmitting terminal, at first the data-signal of importing is carried out the orthogonal phase shift modulation treatment and obtain modulation signal; 2. then modulation signal is gone here and there successively and conversion, inverse Fourier transform and and go here and there conversion process, form a plurality of OFDM symbols; 3. again in a plurality of OFDM symbols that form, inserting a length every the OFDM symbol of setting quantity is L _PM sequence s, as a training sequence, and be L with m sequence s according to the characteristic of channel additional length before training sequence _CCyclic Prefix, obtain the training sequence after the pended cyclic prefix, represent with x, x=[x (0), x (1) ... x (L _P+ L _C-1)]; 4. last OFDM symbol with training sequence x after the pended cyclic prefix and formation transfers to receiving terminal by ultra-wideband channel together after carrier modulation treatment, training sequence x in transmission course after the pended cyclic prefix and OFDM symbol are subjected to the influence of channel fading and white Gaussian noise; 5. at receiving terminal, the channel decline that the definition receiving terminal receives and the training sequence x of the pended cyclic prefix after the white Gaussian noise influence are first received signal, channel decline that the definition receiving terminal receives and the OFDM symbol after the white Gaussian noise influence are second received signal, and first received signal is expressed as with tapped delay line model

Wherein, k=0,1 ..., L _p+ L _c-1, r (k) is k first received signal constantly, and h represents the matrix-vector that the coefficient by each multipath of channel constitutes,

h _tBe t multi-path coefficients of channel, h should satisfy condition: { h _t=0|L≤t≤L _C-1}, L are the exponent number of channel, and x is the training sequence after the pended cyclic prefix, and x (k-t) is the training sequence after the k-t pended cyclic prefix constantly, and n is a white Gaussian noise, and n (k) is a k white Gaussian noise constantly; 6. at first the first received signal r (k) is unloaded ripple modulation, and go circulation prefix processing to obtain first received signal of unloading after the ripple modulation treatment

Wherein, k=0,1 ..., L _p+ L _c-1,

For going k first received signal constantly behind the Cyclic Prefix, h represents the matrix-vector that each multi-path coefficients by channel constitutes,

h _jBe j multi-path coefficients of channel, h should satisfy condition: { h _j=0|L≤j≤L _C-1}, L are the exponent number of channel, and n is a white Gaussian noise, and n (k) is a k white Gaussian noise constantly, s ^jBe the m sequence behind the m sequence s ring shift right j position, s ^j(k) be the sequence in the k moment behind the m sequence s ring shift right j position; 7. calculate first received signal go behind the Cyclic Prefix then

With the m sequence s behind the m sequence s ring shift right i position ⁱCross-correlation matrix C and the autocorrelation matrix C of each training sequence s _P, C=[C (i, j)], (i is j) for going first received signal behind the Cyclic Prefix for C

With the m sequence s behind the m sequence s ring shift right i ⁱThe normalized crosscorrelation coefficient, C _P=[C _P(i, j)], C _P(i j) is m sequence s behind the m sequence s ring shift right j position ^jWith the m sequence s behind the m sequence s ring shift right i position ⁱThe normalized autocorrelation coefficient,

Wherein, i=0,1 ..., L _p, j=0,1 ..., L _p, k=0,1 ..., L _p+ L _c-1,

For removing k first received signal constantly behind the Cyclic Prefix, s ^j(k) be the sequence in the k moment behind the m sequence s ring shift right j position, s ⁱ(k) be the sequence in the k moment behind the m sequence s ring shift right i position; 8. again according to first received signal of going behind the Cyclic Prefix

With the m sequence s behind the m sequence s ring shift right i position ⁱCross-correlation matrix C and the autocorrelation matrix C of each training sequence s _P, the impulse response estimated value of calculating channel

Wherein,

Be autocorrelation matrix C _PInverse matrix; It is characterized in that according to described autocorrelation matrix C _PDiagonal dominance, with described autocorrelation matrix C _PBe decomposed into first matrix and the second matrix sum, described first matrix is designated as D, described second matrix is designated as E, C _P=D+E, satisfy in described first matrix D and the described second matrix E || D ^-1Described autocorrelation matrix C is calculated in E||＜1 o'clock _PInverse matrix

Wherein, symbol " || || " be the norm symbol, I is a unit matrix, D ^-1Be the inverse matrix of first matrix D, m=1,2 ..., ∞; Basis again

Calculate

The first approximation value, $C_p^{-1}\&ap;(I-D^{-1}E)D^{-1}\&ap;D^{-1}-D^{-1}{\mathrm{ED}}^{-1};$

Described first matrix D is by described autocorrelation matrix C _PThe diagonal matrix formed of diagonal entry, the described second matrix E is by described autocorrelation matrix C _PThe non-diagonal matrix formed of off diagonal element, described diagonal matrix is designated as D ₁, described non-diagonal matrix is designated as E ₁, obtain

To described autocorrelation matrix C _PCoefficient carry out normalized, described diagonal matrix D after the normalized ₁Be a unit matrix I; According to

Obtain

Images