CN101834652B - Downlink transmission method based on MIMO-OFDM (Multiple Input Multiple Output-Orthogonal Frequency Division Multiplexing) time domain feedback - Google Patents

Abstract

The invention discloses a downlink transmission method based on MIMO-OFDM (Multiple Input Multiple Output-Orthogonal Frequency Division Multiplexing) time domain feedback, comprising the following steps: feeding back a time domain tap delay line model parameters measured by a user to a base station (alternatively only feeding back path parameters with high power), obtaining frequency domain channel matrixes on all sub-carrier waves by the base station by using the parameters, and constructing the corresponding beam vectors for downlink transmission. As the number of the time domain parameters has no concern with the number of bandwidth and sub-carrier waves, the downlink transmission method based on MIMO-OFDM time domain feedback can effectively reduce the system feedback cost when the number of the bandwidth and the sub-carrier waves is very great; and meanwhile, the invention provides a feedback bite allocation method. The downlink transmission method based on MIMO-OFDM time domain feedback can effectively reduce the realization complexity by using scalar quantization compared with the frequency domain vector quantization.

Description

A kind of downlink transmission method based on MIMO-OFDM time domain feedback

Technical field

The present invention relates to wireless communication technology, be specifically related to the downlink transmission method of a kind of MIMO-OFDM (multi-I/O OFDM) system time domain feedback.

Background technology

In multi-user system, in order to obtain multi-user diversity gain, the base station need have user's channel condition information (CSI) separately.For FDD (FDD) system, just need the user that its CSI is fed back to the base station.In multi-I/O OFDM (MIMO-OFDM) system, each user need feed back the frequency domain CSI on each subcarrier, and feedback overhead is very big.Existing solution is to divide subband, in each subband, uses identical transmission parameter, and each user only feeds back the CSI of certain number of sub-carrier in each subband, can reduce feedback overhead so to a certain extent, reduce the system complex degree.To divide the performance loss that subband causes in order controlling, to require the coherence bandwidth of subband bandwidth less than channel usually, so the future communications system bandwidth is with number of subcarriers when very big, the effect of division subband is limited.

The frequency domain quantification technique requires base station and consumer premise justice code book, selects only code word from code book during quantification.When feedback bits more for a long time, the size of code book is very big, can take a lot of memory spaces, during quantification code book is carried out exhaustive search and also can cause great computing cost.

Optimum non-homogeneous scalar quantization criterion is provided by the Lloyd-Max condition, and optimum quantizer is to realize through alternative manner.In order to reduce complexity, can consider the best uniform quantizer on the average meaning, promptly quantization scheme is only relevant with the average power in each footpath, for example the best uniform quantizer of the Gaussian random variable of this paper employing.

In order to reduce feedback overhead, reduce system complexity, we can consider time domain feedback and quantification technique.Because time domain parameter number and bandwidth are irrelevant with the subcarrier number, therefore when bandwidth and subcarrier number were very big, the use time domain parameter fed back and can effectively reduce the system feedback expense; The storage and the computing cost that adopt scalar quantization to need simultaneously all quantize far below frequency domain vectors.Therefore research is to have very much practical significance based on the downlink transfer scheme of MIMO-OFDM system time domain feedback.

Summary of the invention

The objective of the invention is to, a kind of downlink transmission method based on MIMO-OFDM time domain feedback is provided, this method can effectively reduce the system feedback expense.

A kind of downlink transmission method provided by the invention based on MIMO-OFDM time domain feedback; The MIMO-OFDM system comprises a base station and a plurality of user, and the antenna amount of establishing the base station is M, and each user's antenna amount is N; It is characterized in that this method comprises the steps:

The 1st step base station channel estimating signal:

The 2nd step user utilizes the multiple tap coefficient of channel estimating calculated signals;

The 3rd step quantized respectively with imaginary part the real part of multiple tap coefficient and utilized greedy Bit distribution method to carry out the feedback bits distribution:

If the total feedback bits of each user is B, give 2L Gauss's tap coefficient with B Bit Allocation in Discrete, the step of greedy Bit distribution method is:

(3.1) quantizing bit number of all Gauss's tap coefficients of initialization is 0, and mean square error is the variance of each coefficient: b _k=0,

K=1 ..., 2L, b _kBe k the bit number that tap coefficient is assigned to, D _kBe the mean square error of k tap coefficient,

Expression D _kInitial value;

(3.2) suppose only to give k coefficient reallocation a quantization bit, with b ' _kExpression this moment k the bit number that tap coefficient is assigned to, then b ' _k=b _k+ 1, utilize formula I and II to obtain corresponding b ' _kOptimum quantization siding-to-siding block length Δ ' _kWith mean square error D ' _k

$\∂D=-\frac{1}{\sqrt{2\mathrm{\π}}}+2\mathrm{\Δ}[\frac{{(Q-1)}^2}{2}-0.25]-\frac{4}{\sqrt{2\mathrm{\π}}}\underset{i=1}{\overset{Q/2-1}{\mathrm{\Σ}}}\mathrm{Exp}(-\frac{i^2{\mathrm{\Δ}}^2}{2})-8\mathrm{\Δ}\underset{i=1}{\overset{Q/2-1}{\mathrm{\Σ}}}{\mathrm{IF}}_x\left(\mathrm{I\Δ}\right)$ Formula I

$D_k^{\′}=1-\frac{2{\mathrm{\Δ}}_k^{\′}}{\sqrt{2\mathrm{\π}}}+(\frac{{(Q-1)}^2}{2}-0.25){\left({\mathrm{\Δ}}_k^{\′}\right)}^2-4\underset{i=1}{\overset{Q/2-1}{\mathrm{\Σ}}}[{\mathrm{\Δ}}_k^{\′}{f}_x\left(i{\mathrm{\Δ}}_k^{\′}\right)+i{\left({\mathrm{\Δ}}_k^{\′}\right)}^2{F}_x\left(i{\mathrm{\Δ}}_k^{\′}\right)]$

Formula II

(3.3) other k-1 coefficients are repeated the operation in (3.2), find out mean square error and change the most significantly coefficient I:

This coefficient is to improve the most tangible coefficient to mean square error is comprehensive in this bit allocation procedures so, so 1 Bit Allocation in Discrete is given this coefficient and total bit number is subtracted 1:b _I=b _I+ 1, D _I=D ' _I, B=B-1.

(3.4) repeating step (3.2) and (3.3) finish up to all Bit Allocation in Discrete;

In the 4th step, the user is according to the Bit Allocation in Discrete in the 3rd step, and the bit that is assigned to each coefficient sends feedback information;

The frequency domain channel matrix vector is calculated in the 5th step base station;

The beam vector of the 6th step base station structuring user's:

The 7th step, base station selected corresponding close-to zero beam was launched data to relative users.

The present invention provides a kind of downlink transmission method based on MIMO-OFDM time domain feedback; The time domain tapped delay line model parameter that the user is measured feeds back to base station (can a feeding back path parameters with high power); The base station utilizes these parameters just can obtain the frequency domain channel matrix on all subcarriers, and then constructs corresponding beam vector and carry out downlink transfer.Because time domain parameter number and bandwidth have nothing to do with the subcarrier number, so when bandwidth and subcarrier number were very big, the present invention can effectively reduce the system feedback expense.The present invention adopts the feedback bits apportion design based on scalar quantization, promptly greedy Bit distribution method.Greedy algorithm is to be the excellent problem of asking of a series of parts with the optimum PROBLEM DECOMPOSITION of office of demanding perfection, and therefore can reduce computation complexity greatly, and performance loss is less.Bit distribution method of the present invention is all bits to be distributed to one by one the mean square error summation is improved the most significantly coefficient.

Description of drawings

Fig. 1 is the system framework figure of the inventive method;

Fig. 2 is the schematic flow sheet of the inventive method;

Fig. 3 is that emission of the present invention receives structure chart.

Embodiment

As shown in Figure 1, MIMO-OFDM comprises a base station and a plurality of user, and there is M root antenna the base station, and each user has N root antenna.

As shown in Figure 2, the step of the inventive method comprises:

S1, base station channel estimating signal:

Estimated signal can be a training symbol, also can be pilot signal.

S2, user utilize channel estimating calculated signals time domain parameter:

The wireless multipath channel of uniting of setting up departments has the analysable multipath of L bar (only considering non-zero tap), adopts tapped delay line model to simulate the equivalent low pass impulse response h (t of channel; τ):

$h(t;\mathrm{\τ})=\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{c}_l\left(t\right)\mathrm{\δ}(\mathrm{\τ}-\frac{m_l}{f_s})---\left(1\right)$

The t express time, τ representes time delay, δ () is a Dirac function, c _l(t) be the multiple tap coefficient of l bar non-zero tap when time t, can think the multiple gaussian variable of zero-mean, its time delay τ equals m _l/ f _s, m _lBe positive integer, f _sIt is the sample value frequency.

The duration of the OFDM symbol of supposing to choose is less than the coherence time of channel, to h (t; τ) make the frequency characteristic H (t that Fourier transform can obtain t multidiameter fading channel constantly; F):

$H(t;f)=\underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}{c}_l\left(t\right)\exp (-\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="HDA0000021782480000011.png"\; state="\{\{state\}\}">j</figure-callout>}2\mathrm{\&pi;}{\mathrm{fm}}_l/N_c)---\left(2\right)$

F representes frequency, and j representes imaginary unit, N _cThe quantity of expression subcarrier, the duration of establishing the OFDM symbol is Δ T, subcarrier spacing is Δ f, then the total N of ofdm system _c=f _s/ Δ f number of sub-carrier, the m subchannel frequency characteristic of u OFDM symbol is H _u(m):

$H_u\left(m\right)=H(\mathrm{u\&Delta;T};\mathrm{m\&Delta;f})=\underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}{c}_l\left(\mathrm{u\&Delta;T}\right)\exp (-\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="HDA0000021782480000011.png"\; state="\{\{state\}\}">j</figure-callout>}2\mathrm{\&pi;}{\mathrm{mm}}_l/N_c)---\left(3\right)$

C in the formula _l(u Δ T) expression multiple tap coefficient of l bar non-zero tap when time u Δ T, according to formula (3), each user can calculate N * M * L multiple tap coefficient c _l(u Δ T).The variation of multidiameter delay is slow relatively, therefore only pays close attention to tap coefficient.

S3, carry out scalar quantization and utilize greedy Bit distribution method to carry out the feedback bits distribution multiple tap coefficient:

The best uniform quantizer of Gaussian random variable is adopted in the quantification of multiple tap coefficient, and the real part and the imaginary part of multiple tap coefficient are carried out independent quantification.During uniform quantization, be divided into Q interval, Q=2 to whole real axis ^b, b is a quantizing bit number.Except first with last interval, other intervals all are isometric, represent with Δ, and most critical are to confirm Δ, make quantization error minimum.Mean square error distortion D can be defined as:

$D=1-\frac{2\mathrm{\&Delta;}}{\sqrt{2\mathrm{\&pi;}}}+(\frac{{(Q-1)}^2}{2}-0.25){\mathrm{\&Delta;}}^2-4\underset{i=1}{\overset{Q/2-1}{\mathrm{\&Sigma;}}}[\mathrm{\&Delta;}{f}_x\left(\mathrm{i\&Delta;}\right)+i{\mathrm{\&Delta;}}^2{F}_x\left(\mathrm{i\&Delta;}\right)]---\left(4\right)$

To the Δ differentiate, obtain the derivative of mean square error distortion D in the following formula about Δ

$\&PartialD;D=-\frac{1}{\sqrt{2\mathrm{\&pi;}}}+2\mathrm{\&Delta;}[\frac{{(Q-1)}^2}{2}-0.25]-\frac{4}{\sqrt{2\mathrm{\&pi;}}}\underset{i=1}{\overset{Q/2-1}{\mathrm{\&Sigma;}}}\exp (-\frac{i^2{\mathrm{\&Delta;}}^2}{2})-8\mathrm{\&Delta;}\underset{i=1}{\overset{Q/2-1}{\mathrm{\&Sigma;}}}i{F}_x\left(\mathrm{i\&Delta;}\right)---\left(5\right)$

Make

to obtain making the approximation of the minimum Δ of D.F wherein _x(i Δ) and F _x(i Δ) is respectively the probability density function values and the cumulative distribution functional value of an i Δ interval standardized normal distribution.

Utilizing greedy Bit distribution method to carry out feedback bits distributes:

If the total feedback bits of each user is B, give 2L Gauss's tap coefficient (multiple tap coefficient has L, and the real part imaginary part is separated independent quantitiesization, and then corresponding tap coefficient has 2L) with B Bit Allocation in Discrete.The step of greedy Bit distribution method is:

(1) quantizing bit number of all Gauss's tap coefficients of initialization is 0, and mean square error is the variance of each coefficient: b _k=0,

K=1 ..., 2L, b _kBe k the bit number that tap coefficient is assigned to, D _kBe the mean square error of k tap coefficient,

Expression D _kInitial value;

(2) suppose only to give k coefficient reallocation a quantization bit, with b ' _kExpression this moment k the bit number that tap coefficient is assigned to, then b ' _k=b _k+ 1, utilize formula (5) to obtain corresponding b ' _kOptimum quantization siding-to-siding block length Δ ' _kWith mean square error D ' _k

(3) other k-1 coefficients are repeated the operation in (2), find out mean square error and change the most significantly coefficient I:

This coefficient is to improve the most tangible coefficient to mean square error is comprehensive in this bit allocation procedures so, so should 1 Bit Allocation in Discrete given this coefficient and total bit number is subtracted 1:b _I=b _I+ 1, D _I=D ' _I, B=B-1;

(4) repeat (2) and (3), finish up to all Bit Allocation in Discrete.

S4, send feedback information through uplink feedback channel:

The user is according to the Bit Allocation in Discrete among the step S3, and the bit that is assigned to each coefficient sends feedback information;

S5, the frequency domain channel matrix vector is calculated in the base station:

User's frequency domain channel matrix vector is calculated according to user's feedback information in the base station.For user p and user q, can obtain their channel matrix H _pAnd H _q

S6, the beam vector of base station structuring user's:

Make up the close-to zero beam vector according to the respective channel matrix that obtains.If the close-to zero beam vector of user p and user q is w _pAnd w _q, owing to require H _pw _q=0, H _qw _p=0, so obtain

With

Expression is to channel matrix H _pAnd H _qConjugate transpose.

S7, downlink data is sent in the base station:

Base station selected corresponding close-to zero beam is launched data to relative users.If for user p and user q, the base station data are respectively s _pAnd s _q, the reception data y of user p then _pFor:

y _p＝H _pw _ps _p+H _pw _qs _q+n _p＝H _pw _ps _p+n _p，

n _pWhite Gaussian noise for user p.

As shown in Figure 3, transmitter and receiver of the present invention comprises: channel estimation module A1, time domain parameter computing module A2, quantification and greedy Bit Allocation in Discrete modules A 3, frequency domain channel vector calculate modules A 4, structure beam vector modules A 5 and user data transmission modules A 6.

In order to further describe the present invention, following mask body is considered following instance: there is M=4 root antenna the base station, and each user has the mimo system of N=2 root antenna, this transmission, and user k and user j are selected transmission user.(Fig. 2) tells about this instance below in conjunction with system flow chart.

In step S1, channel estimation module A1 transmitting channel estimated signal, estimated signal can be the training training, also can be pilot signals.

In step S2, time domain parameter computing module A2 calculates time domain parameter.In step S3,3 pairs of multiple tap coefficients of quantification and greedy Bit Allocation in Discrete modules A quantize.

Utilizing greedy Bit distribution method to carry out feedback bits distributes: establishing total feedback bits is B, and quantification and greedy Bit Allocation in Discrete modules A 3 are given 2L real Gauss's tap coefficient (the real part imaginary part is separated independent quantitiesization) with B Bit Allocation in Discrete.

In step S4, the user sends feedback information according to the Bit Allocation in Discrete among the step S3 with corresponding bits;

In step S5, frequency domain channel computing module A4 calculates user's frequency domain channel matrix vector according to feedback information.For user k and user j, can obtain their channel matrix H _kAnd H _j

In step S6, structure beam vector modules A 5 is according to the respective channel matrix structure close-to zero beam vector that obtains.If the beam vector of user k and user j is w _kAnd w _j, owing to require H _jw _k=0, H _kw _j=0, so obtain

In step S7, user data transmission modules A 6 selects corresponding wave beam to launch data to relative users.

The present invention not only is confined to above-mentioned embodiment; Persons skilled in the art are according to embodiment and the disclosed content of accompanying drawing; Can adopt other multiple embodiment embodiment of the present invention, therefore, every employing project organization of the present invention and thinking; Do some simple designs that change or change, all fall into the scope of the present invention's protection.

Claims (1)

1. downlink transmission method based on MIMO-OFDM time domain feedback, the MIMO-OFDM system comprises a base station and a plurality of user, and the antenna amount of establishing the base station is M, and each user's antenna amount is N, it is characterized in that, and this method comprises the steps:

The 1st step base station channel estimating signal;

The 2nd step user utilizes the multiple tap coefficient of channel estimating calculated signals;

The 3rd step quantized respectively with imaginary part the real part of multiple tap coefficient and utilized greedy Bit distribution method to carry out the feedback bits distribution; If the total feedback bits of each user is B, give 2L Gauss's tap coefficient with B Bit Allocation in Discrete, the step of greedy Bit distribution method is:

(3.1) quantizing bit number of all Gauss's tap coefficients of initialization is 0, and mean square error is the variance of each coefficient: b _k=0, K=1 ..., 2L, b _kBe k the bit number that tap coefficient is assigned to, D _kBe the mean square error of k tap coefficient,

Expression D _kInitial value;

(3.2) suppose only to give k coefficient reallocation a quantization bit; Expression k bit number that tap coefficient is assigned to this moment with

, then

utilizes formula I and II to obtain the optimum quantization siding-to-siding block length of correspondence

and mean square error

During uniform quantization, be divided into Q interval, Q=2 to whole real axis ^b, b is a quantizing bit number; Except first with last interval, other intervals all are isometric, represent with △; To the △ differentiate, obtain the derivative of mean square error distortion D about △

$\&PartialD;D=-\frac{1}{\sqrt{2\mathrm{\&pi;}}}+2\mathrm{\&Delta;}[\frac{{(Q-1)}^2}{2}-0.25]-\frac{4}{\sqrt{2\mathrm{\&pi;}}}\underset{i=1}{\overset{Q/2-1}{\mathrm{\&Sigma;}}}\mathrm{Exp}(-\frac{i^2{\mathrm{\&Delta;}}^2}{2})-8\mathrm{\&Delta;}\underset{i=1}{\overset{Q/2-1}{\mathrm{\&Sigma;}}}{\mathrm{IF}}_x\left(\mathrm{I\&Delta;}\right)$ The formula I

$D_k=1-\frac{{2\mathrm{\&Delta;}}_k^{\&prime;}}{\sqrt{2\mathrm{\&pi;}}}+(\frac{{(Q-1)}^2}{2}-0.25){\left({\mathrm{\&Delta;}}_k^{\&prime;}\right)}^2-4\underset{i=1}{\overset{Q/2-1}{\mathrm{\&Sigma;}}}[{\mathrm{\&Delta;}}_k^{\&prime;}{f}_x\left({\mathrm{I\&Delta;}}_k^{\&prime;}\right)+i{\left({\mathrm{\&Delta;}}_k^{\&prime;}\right)}^2{F}_x\left({\mathrm{I\&Delta;}}_k^{\&prime;}\right)]$ The formula II

F wherein _x(i Δ) and F _x(i Δ) is respectively the probability density function values and the cumulative distribution functional value of an i Δ interval standardized normal distribution;

(3.3) other k-1 coefficients are repeated the operation in (3.2), find out mean square error and change the most significantly coefficient I:

This coefficient is to improve the most tangible coefficient to mean square error is comprehensive in this bit allocation procedures so, so 1 Bit Allocation in Discrete is given this coefficient and total bit number is subtracted 1:b _I=b _I+ 1, B=B-1.

(3.4) repeating step (3.2) and (3.3) finish up to all Bit Allocation in Discrete;

In the 4th step, the user is according to the Bit Allocation in Discrete in the 3rd step, and the bit that is assigned to each coefficient sends feedback information;

The frequency domain channel matrix vector is calculated in the 5th step base station;

The beam vector of the 6th step base station structuring user's:

The 7th step, base station selected corresponding close-to zero beam was launched data to relative users.

Images