CN106789764A - The transform domain quadratic estimate method of the denoising of joint Weighted Threshold and balanced judgement - Google Patents

Abstract

The invention discloses joint Weighted Threshold denoising and the transform domain quadratic estimate method of balanced judgement, carry out as follows：1) joint overlying training sequence sends sequence with harmonic sequence generation；2) in OFDM baseband system, by the LS channel estimation methods of overlying training sequence, the rough estimate of channel frequency response is obtained；3) after channel frequency response rough estimate being counted as into amplitude and phase compensation, then widened window function, the M point dct transforms spent for M, and LPF, time domain denoising and sequence zero padding extension process are made to gained sequence, to the N point sequences after extensionMake N points idct transform, go window and secondary amplitude and phase compensation to obtain4) acquired results equilibrium is adjudicatedM points dct transform, LPF, time domain denoising, interpolation zero padding and N point idct transforms are repeated, the final estimation of frequency domain response is obtainedChannel estimating performance is influenceed this invention removes data sequence and training sequence, improves the performance of existing LS channel estimations, implementation complexity is relatively low.

Description

The transform domain quadratic estimate method of the denoising of joint Weighted Threshold and balanced judgement

Technical field

It is more particularly to a kind of to be applied to wireless and power line multi-carrier communication etc. the invention belongs to digital communication technology field The joint Weighted Threshold denoising in field and the transform domain quadratic estimate method of balanced judgement.

Background technology

In recent years, wireless communication technology is developed rapidly.Wherein, the time-varying characteristics of wireless channel are always the technology Study hotspot, is also the key for improving capability of wireless communication system and reliability.Meanwhile, propagation of the signal in wireless channel is One process of complexity：Signal dispersion, reflection and diffraction are included, surrounding enviroment and mobile station is accompanied by signal transmission Influence.Therefore, it is one of key of wireless communication technology that quick Real-time Channel is estimated.And in wire communication, low-voltage power line Carrier communication (LPLC) technology be it is a kind of by using already present widely distributed low-voltage power line as transmission signal matchmaker The high speed wide-band communication mode of Jie.It utilizes existing low-voltage power line network to carry out information transfer, without rewiring and Reduce cost.But in LPLC systems, its channel circumstance is complicated, and noise jamming is serious, to ensure the property of whole LPLC systems Can, also in the urgent need to efficient channel estimation technique and method.

OFDM (OFDM) technology is a kind of special multi-carrier modulation technology, and its principle is by going here and there and becomes Change, by parallel transmission in the data modulation of high speed serial transmission to multiple separate orthogonal sub-channels, with anti-interference energy The advantages of power is strong, the availability of frequency spectrum is high, transmission rate is fast.In Practical Communication System, ofdm system is divided into relevant ofdm system With two kinds of incoherent ofdm system, such as LPLC systems belong to relevant OFDM.Using relevant ofdm system, channel need to be known during demodulation Status information (CSI), and CSI can be obtained by channel estimation.So-called channel estimation is using reception by various methods The data estimation at end goes out the process of the parameter of channel.Existing channel estimation methods mainly have based on pilot tone/training sequence method Channel estimation methods, channel estimation and blind Channel Estimation based on decision-feedback method.For based on pilot tone or training sequence The two methods of channel estimation of row, wherein, it is low based on pilot channel estimation implementation complexity, and estimate that performance is more excellent, but because of letter Situations such as road time-varying, Selective intensity, be that quickly tracking channel status need to constantly send pilot tone, easily causes bandwidth and power is damaged Lose；, with data sequence be superimposed low-power training sequence in transmitting terminal by the channel estimation methods based on overlying training sequence, does not account for Remaining frequency spectrum resource is used, system bandwidth utilization rate is improve, channel state variations, but training sequence and data sequence can be quickly tracked Row can produce certain influence to channel estimation effect.

The content of the invention

For above two existing channel method of estimation exist shortcoming, present invention employs overlying training sequence with it is discrete Cosine transform (DCT) transform domain channel estimation methods, and it is secondary that balanced decision method completion is introduced in transform domain channel estimation The transform domain of denoising is estimated.Because overlying training sequence is not take up bandwidth, and joint training sequence and harmonic sequence to data sequence Transformation, can eliminate influence of the data sequence to channel estimation；Because DCT/ inverse discrete cosine transforms (IDCT) can effectively suppress high frequency Component is produced, and the channel response amplitude threshold for transmitting sequence is judged by setting weighted noise threshold value in its calculating process And balanced decision method is introduced, further eliminate noise jamming.Comprehensive overlying training sequence, DCT/IDCT transform domain channel estimations And balanced judgement advantage, the inventive method possesses moderate complexity, and performance is higher and the advantage such as good stability.Therefore, the invention Can be used for the application scenario such as real-time tracking channel state variations and fast signal detection under wireless and line communication transmission environment.

Present invention improves over the channel estimation methods of existing low performance complexity high, there is provided a kind of improved transform domain letter Channel estimation method, that is, combine the transform domain quadratic estimate method of Weighted Threshold denoising and balanced judgement, and it can be widely applied to nothing The occasions such as channel estimation and the fast signal detection of line and power line communication.

The present invention is achieved through the following technical solutions：

A kind of transform domain quadratic estimate method for combining Weighted Threshold denoising and balanced judgement, which employs joint superposition instruction Practice sequence and the method for sending sequence, the channel frequency estimated based on time statistic average least square (LS) are produced with harmonic sequence Response rough estimate method, and combine the transform domain quadratic estimate method of Weighted Threshold denoising and balanced judgement, specifically through following step It is rapid to realize：

1) joint overlying training sequence is produced with harmonic sequence and sends sequence, eliminates data sequence and training sequence to channel Estimate the influence for producing；

2) in general OFDM baseband system, by the LS channel estimation methods using existing overlying training sequence, Obtain the rough estimate of channel frequency response；

3) channel frequency response rough estimate is counted as amplitude and phase compensation, then widened degree for M window function (M is positive integer, And M OFDM symbol is a frame), after M point dct transforms, and gained sequence is made at LPF, time domain denoising and zero padding extension Reason, to N (N the is positive integer) point sequence after extensionMake N points idct transform, go window to process and secondary amplitude-phase benefit Repay

4) acquired results equilibrium is adjudicatedRepeat M points dct transform, LPF, time domain denoising, interpolation zero padding And N point idct transforms, obtain the final estimation of frequency domain response

Preferably, joint overlying training sequence produces the method for sending sequence with harmonic sequence, by binary system sequence { x (n) } is by after the front-end processing of OFDM (OFDM) baseband system, obtaining data sequence { d (n) }.And joined Overlying training sequence { t (n) } and harmonic sequence { a (n) } are closed, is produced and is sent sequence { s (n) }.Wherein, harmonic sequence { a (n) } is Linear complexity of random periodic sequences, and its cycle is identical with training sequence { t (n) }.Methods described may be used to lower detailed step and realize：

Step 2.1. set length as list entries x (n) of N (N is integer) it is encoded/interweave, modulation, 1:N serioparallel exchanges (A₁:A₂Represent A₁Road is converted to A₂Road, and A₁,A₂It is natural number, following presentation is identical) and N point quick Fourier inverse transformations Etc. (IFFT) after treatment, data sequence { d (n) } is obtained.Wherein, coding/be interweaved can be encoded using existing Turbo etc. respectively, And coordinate the interleaving modes such as pseudorandom；Modulation system is from M-ary Quadrature Amplitude modulation (M-QAM) etc.；And, Turbo is compiled Code, pseudo random interleaving, M-QAM and IFFT are prior art, the explanation in invention is related to technology.

Step 2.2. sets the training sequence { t (n) } and harmonic sequence { a (n) } that length is N, the cycle is T, it is ensured that { d (n) + a (n) } range value of Fast Fourier Transform (FFT) (FFT) at cycle frequency P is 0.Wherein, P=lN/T, Q=N/T and N, P, T, Q are integer, l=0,1 ..., T-1, and FFT for prior art, the explanation in invention is related to technology.Data sequence { d (n) } meet following relation with harmonic sequence { a (n) }：

Then the expression formula of harmonic sequence { a (n) } is as follows：

Step 2.3. is to step 2.1 gained { d (n) }, joint training sequence { t (n) } and harmonic sequence { a (n) }, through formula (2.3) computing summation, produces time domain to send sequence { s (n) }.Take m-th n-th sampled value S of OFDM symbol_m(n), its expression Formula is：

Wherein, θ is real number, and value is：0<θ<1；D_mN () is that n-th sampled value of m-th OFDM symbol is corresponding discrete Quantized sequences value, m and n are positive integer.

Preferably, the channel frequency response rough estimate method estimated based on time statistic average least square (LS), folded Plus on the basis of training sequence, after seeking time statistic average to a frame in M (M is positive integer) individual OFDM symbol, believed using existing LS Road estimation, obtains channel frequency response rough estimate.Wherein, LS channel estimations are prior art, are said in invention is related to technology It is bright.The rough estimate is completed according to the following steps：

Step 2.3 gained time domain in claim 2 is sent sequence { s (n) } by step 3.1., and it is L that length is added successively_cp (L_cp>L, L are integer, represent channel length) Cyclic Prefix (CP), be placed in data block header, at utmost eliminate data block it Between inter-block-interference (IBI), N:1 parallel-serial conversion and after sending into channel, then through 1:N serioparallel exchanges and go Cyclic Prefix (reject number It is L according to the former addition length of block header_cpCP, be easy to receiving terminal demodulate receive) treatment after, obtain data sequence { y_m(n)}.Its expression Formula is as follows：

Wherein, data sequence vector is:y_m=[y_m(0),y_m(1),…,y_m(N-1)]^T；Channel impulse response is：h_m= [h_m(0),h_m(1),…,h_m(L-1)]^T；T and D_mThe equivalent cycle convolution matrix sum obtained by training sequence { t (n) } is corresponded to respectively According to sequence { D_m(n) } obtained by equivalent cycle convolution matrix；Noise column vector is:ω_m=[ω_m(0),ω_m(1),…,ω_m(N- 1)]^T, and subscript T representing matrix transposition, * represents product calculation.N-dimensional equivalent cycle convolution matrix t and D_mMatrix represent difference It is as follows：

Step 3.2. is to step 3.1 gained { y_m(n) }, must receive terminal sequence { Y through N points FFT_m(k)}；Wherein, Y_mK () is The frequency domain representation of m-th OFDM symbol, n-th sub-carrier signal, its expression formula is as follows：

And Y_m(k), H_m, W_mRespectively y_m(n)、h_m、ω_mN point FFTs.If diag [] represents diagonal matrix, and internal Element is diagonal data.Then T=diag [T (0), T (1) ..., T (N-1)], D=diag [D_m(0),D_m(1),…,D_m(N- 1)].In addition, element T (k) expression formula of matrix T is as follows：

In formula (3.5), FFT { } represents the operator of N point FFTs；Exp { } represents the index fortune of nature truth of a matter e Operator；It is imaginary unit.

Step 3.3. received data sequence { y after Cyclic Prefix and before N point FFTs for going_m(n) }, take a frame length The OFDM symbol (channel impulse response h is constant in the symbolic range) for M is spent, using time statistic average method, to being taken OFDM symbol seeks time domain average, the signal after obtaining averagely；Its time-domain expression is：

Step 3.4. uses LS channel estimations, obtains channel frequency response rough estimateWhen M takes infinity, { D_m (n) } time domain average is 0, and noise average is also 0.Therefore formula (3.6) can be deformed into：

To formula (3.7) equal sign both sides while premultiplicationTry to achieve the rough estimate of channel impulse responseWithWherein, subscript " ls " represents LS channel estimations,Expression formula it is as follows：

Wherein,ForN point FFTs, subscript " -1 " representing matrix inversion operation.

Preferably, the transform domain quadratic estimate method of Weighted Threshold denoising and balanced judgement is combined, using following steps reality It is existing：

Step 4.1. is by step 3.4 gained rough estimateThrough amplitude and phase compensation, windowed function, M point discrete cosines After conversion (DCT) treatment, obtainWherein, subscript " c " expression is expressed below identical in discrete cosine transform domain；Width Degree phase compensation is by being multiplied by a gain factor δ to rough estimate₁Complete；It is the window function of M that institute's windowed function can use width. And gain factor δ₁Can be respectively using sinusoidal window function SIN, expression formula with selected window function：

Step 4.2. is by obtained by step 4.1After being extended through threshold denoising, zero padding, N point sequences are obtained Wherein, time domain threshold denoising process is realized using following sub-step：

Step 4.2.1 concentrates on low-frequency range because of signal energy, by sequenceBy low pass filter (cut-off frequency P_c=L_cp- 1) after filtering high fdrequency component, sequence after being filtered

Step 4.2.2 is by obtained by step 3.4Flat-top is used to sample it, and the sampling period is T_s=T/ (N+ L_cp), obtain sample sequence { g_c(i) } (i.e. the corresponding channel impulse response of each sampled point), and seek each sampled point respective channels The amplitude mould of shock response.If sample sequence has a waits range value sampled point, range value correspondence weights are a.Now, Threshold value thresholding λ can be tried to achieve by formula (4.3)：

Wherein, g_cI () is the corresponding channel impulse response of ith sample point, i is integer, and i=1,2 ..., N；a₁+a₂ +···+a_q=N.

Step 4.2.3 will filter after sequenceJudge by formula (4.4), sampled point is retained or zero setting：

Step 4.2.4 is to step 4.2.3 gained sequences { G_c(m) }, zero padding expands to N point sequencesAnd zero padding is expanded Exhibition process is in sequence { G_c(m) } end addition N-M zero；

Step 4.3. is by obtained by step 4.2.4Respectively through N points idct transform, go window process (divided by sinusoidal windows Function SIN) and secondary amplitude and phase compensation (be multiplied by gain factor δ₂), obtainWherein, gain factor δ₂Expression formula is：

Gained sequence { the Y of step 4.4. joint steps 2.4_m(k) } to step 4.3 gainedEquilibrium judgement, and by equilibrium As a resultRepeat M points dct transform, LPF, Weighted Threshold denoising, zero padding extension and N points idct transform and complete transform domain Quadratic estimate process, obtains the estimated result of channel frequency responseWherein, balanced judgement is completed by following sub-step：

Step 4.4.1. sets equilibrium resultIt is channel frequency response predicted value, then receives signal channel frequency response Sending signal frequency domain estimate is obtained after the zero forcing equalization of predicted valueWherein, zero forcing equalization is prior art, in hair It is bright to be related to explanation in technology.Expression formula is：

Step 4.4.2. is by the sending signal frequency domain estimate after equilibriumThrough data decision, qam constellation is mapped to Scheme on closest point, obtain sending signal decision valueFrequency channels response decision value is obtained simultaneouslyWherein,Table It is up to formula：

Judgement is according to as follows：

1) decision value is worked asWhen, then court verdict is correct, i.e. channel frequency response decision valueIt is letter Road actual frequency response H, X_mK () is modulation to k-th N point FFT data of subcarrier in m-th OFDM symbol；

2) decision value is worked asWhen, then decision valueThere is decision error with the response of channel actual frequency Δ, can make channel frequency response predicted value by decision error Δ feedback compensationStep wise approximation channel real response H.Its In, decision error Δ expression formula is：

And decision-feedback coefficient ξ is channel frequency response predicted valueModifying factor, and ξ for decision error Δ letter Number, i.e.,Wherein,

Step 4.4.3. joint first times DCT are estimatedDecision valueAnd decision-feedback coefficient ξ weighted sums, Obtain channel frequency response equilibrium resultWherein,Expression formula is：

Wherein, " * " represents product calculation.Q₁、Q₂、Q₃RespectivelyThe weights of ξ, value is interval The real number of [0,1], and Q₁+Q₂+Q₃=1.

It is as follows in prior art involved in the present invention：

Fast Fourier Transform (FFT)/inverse fast Fourier transform (FFT/IFFT) technology, Turbo code coding, pseudo random interleaving, M-ary Quadrature Amplitude modulation (M-QAM), least square (LS) channel estimation methods, based on discrete cosine transform/discrete cosine The transform domain channel estimation methods and zero forcing equalization technology of inverse transformation (DCT/IDCT).Each prior art principles illustrated is as follows：

FFT/IFFT technologies

FFT/IFFT technologies are the keys that OFDM technology realizes modulation /demodulation, and both inverse operations each other, are discrete fouriers The realization of the low complex degree of conversion/inverse discrete Fourier transform (DFT/IDFT).The modulation-demodulation technique of OFDM can be by FFT/ IFFT technologies are completed.Comprising multiple subcarriers through ovennodulation in one OFDM symbol, multiple subcarrier sums are represented by, I.e.：

Wherein, N is the number of subcarrier；T represents the OFDM symbol duration；d_iTo distribute to the data of every sub-channels Symbol；f_iIt is i-th carrier frequency of subcarrier；Rect (t) is rectangular function, and rect (t)=1 ,-T/2≤t≤T/2；And " * " is product calculation, and exp { } represents the exponent arithmetic symbol of nature truth of a matter e,It is imaginary unit, following presentation phase Together.When in formula (1), t_s=0 and during rect (t)=1, signal s (t) is sampled with the speed of T/N, have t=kT/N (k=0, 1 ..., N-1) obtain：

Find out from formula (3), s_kT () is equivalent to d_iIDFT computings, then in receiving terminal to s_kT () does DFT computings can recover Go out d_i, i.e.,：

Therefore, the modulation /demodulation of OFDM can realize that FFT/IFFT is calculating quickly soon for DFT/IDFT by FFT/IFFT technologies Method.

Turbo code is encoded

Turbo code encoder is made up of component coder, interleaver, residual matrix and multiplexer.The optimal choosing of component code It is selected as Recursive Systematic Convolutional (RSC) code.Usual two component codes use identical generator matrix.During coding, two component codes Input message sequence is identical, and length is the input message sequence { u of N_kEncoded in the 1st component coder of feeding Exported as system simultaneouslyMultiplexer is directly sent to, while { u_kInterleaved sequence { u after interleaved device π_nFeeding the 2nd Component coder.Wherein n=π (k), 0≤n, k≤N-1.π () is intertexture mapping function, and N is weaving length, that is, be input into information Sequence length.Two component coder list entries are only different symbol order, and the verification sequence of output is respectivelyWithBe improve code check and system spectral efficiency, by two verification sequences it is punctured after, obtainFinally, willWith System is exportedCodeword sequence { c is constituted together_k}。

Turbo coding principle block diagrams are as shown in Figure 8.

Pseudo random interleaving

Weaving length realizes step for the pseudo random interleaving of N：First, the random selection from set S={ 1,2 ..., N } One integer i₁, it is corresponding to choose to i₁Probability P (i₁)=1/N, the i that will be selected₁π (1) is designated as, while by i₁Deleted from set S Remove, obtain new set S₁；Secondly, in kth step, from set S_k-1={ i belongs to S, i ≠ i₁,i₂,…,i_N-k+1Middle random selection One i_k, it chooses probability P (i accordingly_k)=1/ (N-k+1), the i that will be selected_kπ (k) is designated as, while by i_kFrom set S_k-1In Delete, obtain new set, be designated as S_k；Finally, as k=N, π (N) is obtained, corresponding probability of choosing is P (i_N)=1, S_NFor Empty set, interleaving process terminates.

M-ary Quadrature Amplitude modulates (M-QAM)

Quadrature amplitude modulation (QAM) is a kind of Vector Modulation, is mutual to two just using the baseband signal of two-way independence The same frequency carrier wave handed over carries out carrier-suppressed double sideband amplitude modulation, using this modulated letter with spectrum orthogonal in same broadband Number, realize the transmission of the parallel digital information of two-way.Wherein, the principles of modulation and demodulation of M-QAM：Transmitting terminal, by serial to parallel conversion It is R by information rate_bInput binary signal be divided into two speed for R_b/ 2 binary signal, 2/L level translations are by two Speed is R_b/ 2 binary signal is changed into speed for R_bThe level signal of/[2lb (L)], then two quadrature carrier phases respectively Multiply, then be added M-QAM signals of suing for peace to obtain；Receiving terminal, using orthogonal coherent demodulation method, will receive signal point two-way and enters Enter two coherent demodulators of orthogonal carrier wave, form L binary signals and export binary signal decision device is respectively enterd, most By obtaining baseband signal after parallel-serial conversion.Wherein, lb () is represented with 2 as the logarithm operation at bottom is accorded with.M-QAM modulation /demodulation is former Reason figure is as shown in figure 9, and " LPF " represents low pass filter in diagram.

LS channel estimation methods

The foundation criterion of LS channel estimation methods is under the influence of noise is not considered so that cost function J values are minimum, Cost function J is defined as：

J=(Y-XFh)^H(Y-XFh) (4)

Wherein, Y=[Y (0), Y (1) ..., Y (N-1)] from the output signal after the demodulation of OFDM symbol constitute to Amount；X=diag [X (0), X (1) ..., X (N-1)] by a frame signal institute group of output after binary system complex sequences x (n) mapping Into diagonal matrix, diag [] represent diagonal matrix；F is N-dimensional Fourier transform matrix, the corresponding n rows k column elements of matrix FThe span of n and k is all [0, N-1], and exp { } represents nature truth of a matter e Exponent arithmetic symbol,It is imaginary unit；H is channel impulse response to be estimated, the conjugation of subscript " H " representing matrix Transposition.

First, channel is write as matrix form：Y=XFh+v；

Secondly, to make cost function J values minimum, then need to meet conditionI.e.：

Finally, abbreviation formula (5) obtains time domain estimationAnd the frequency response of channel is obtained by H=Fh

Transform domain channel estimation methods based on DCT/IDCT

DCT is equivalent to the data sequence after mirror-extended compared to DFT, M point datas sequence through M point dct transforms 2M point DFT transforms, and DCT is a pair of DFT reality even functions.Different from DFT, DCT does not produce new high order component, and its sequence To expand in epoch edge be continuous cycle, while the characteristics of there is DCT energy to concentrate, performance and implementation complexity are all better than DFT.Dct transform domain channel estimation steps are as shown in Figure 10.

(1) the data sequence Y for pilot frequency locations being received_pK () obtains the channel at pilot sub-carrier through LS channel estimations The estimation of frequency response

(2) it is rightMake M point dct transforms, obtainIts expression formula is as follows：

(3) it is right in DCT domainZero padding is extended to N point sequences, obtainsIts expression formula is as follows：

(4) it is rightMake N point idct transforms, obtainIts expression formula is as follows：

Zero forcing equalization technology

Channel equalization technique can be divided into linear equalization and the class of nonlinear equalization two.Wherein, linear equalization is applied to channel frequently Rate response characteristic is flatter, intersymbol interference not severe case.Linear equalizer can be realized as shown in figure 11 by transversal filter.

Channel equalization is realized, the tap coefficient for calculating transversal filter is crucial.Zero forcing equalization is according to the characteristic of channel To adjust the tap coefficient of equalization filter, the total characteristic of balanced device and channel is set to be similar to desirable channel conditions, frequency domain upper table Now for output response only has value in central point, the influence of intersymbol interference is eliminated.Exist in the ofdm system of intersymbol interference, signal is passed The matrix of defeated process is represented：Y=HX+V, wherein X, Y, V are represented send sequence, receiving sequence and additive white Gaussian noise respectively Frequency domain form, H is the frequency domain form of channel impulse response.The basic thought of zero forcing equalization is the minimum of searching equation group Y=HX Norm Least Square, that is, work as | | Y-HX | |²During minimalization, X is solved.

OrderI.e.

Solve：Thus the equalizing coefficient matrix of zero forcing equalizer is：

Wherein, subscript " H " is Matrix Conjugate transposition, and subscript " -1 " is matrix inversion operation.

Combine overlying training sequence in the present invention and send sequence with harmonic sequence generation, and use existing overlying training sequence LS channel estimation methods obtain channel frequency response rough estimate, (thresholding is as sampled point is corresponding to set noise threshold thresholding The weighted average of channel response amplitude) and denoising, while being rung to channel frequency using the DCT/IDCT interpolation of adding window The transform domain quadratic estimate answered, and balanced decision method is introduced during quadratic estimate.Joint superposition training proposed by the present invention Sequence and the transform domain channel estimation methods of time domain denoising, eliminate data sequence and training sequence to channel estimating performance shadow Ring, improve the performance of existing LS channel estimations, implementation complexity is relatively low, can be answered in multi-carrier communication well With.

Brief description of the drawings

Fig. 1 is realization principle overall framework figure of the invention.

Fig. 2 is transmission sequence { s (n) } generation flow and transmission sequence { s (n) } frame structure signal in the embodiment of the present invention Figure.

Fig. 3 is that the embodiment of the present invention is thick using the frequency response that existing method obtains typical power line carrier multipath channel The structure chart of estimation.

Fig. 4 is the schematic block diagram of embodiment of the present invention weighting time domain threshold value setting.

Fig. 5 is that the embodiment of the present invention uses the DCT/IDCT interpolation of adding window to the transform domain quadratic estimate of channel frequency response Schematic flow sheet.

Fig. 6 is embodiment of the present invention joint receiving terminal data sequence { Y_m(k) } estimate with first time dct transform domain Obtain channel frequency response equilibrium valueSchematic diagram.

Fig. 7 is the schematic flow sheet of the embodiment of the present invention.

Fig. 8 is Turbo coding principle block diagrams.

Fig. 9 is M-QAM principles of modulation and demodulation figures.

Figure 10 is the figure of dct transform domain channel estimation.

Figure 11 is the linear equalizer schematic diagram realized by transversal filter.

Specific embodiment

The present invention is described in further detail below by way of preferred embodiment and with reference to accompanying drawing.

Joint Weighted Threshold denoising proposed by the invention can be applicable to the transform domain quadratic estimate method of balanced judgement In typical power line multicarrier wire communication or multi-carrier wireless communications system, when realizing that noise jamming is severe quick and precisely Ground completes channel estimation and signal testing function, is not only limited in the field that following examples are explained in detail.Hereinafter choose allusion quotation The transform domain quadratic estimate method of joint Weighted Threshold denoising with the balanced judgement of the power line multi-carrier communications systems of type, in detail Illustrate specific embodiment of the invention.

A kind of excellent embodiment of the present invention sequentially passes through following key steps and is achieved：

Joint overlying training sequence sends sequence with harmonic sequence generation；By typical power line multi-carrier communication multipath letter Road is incorporated into ofdm system, and (method is shown in " Liu Qiuge, Mu Xiaomin, Lu Yan by the LS channel estimation methods of existing amendment OFDM channel estimations [J] the computer engineering of brightness superposition Chirp training sequences and application, 2011,47 (31):97-100. "), Obtain the rough estimate of power line channel frequency responseAccording to gained rough estimateTo its amplitude and phase compensation (with Gain factor δ₁Be multiplied), adding window (it is the sinusoidal window function of M to use width), M point dct transforms and LPF (cut-off frequency P_c=L_cp- 1) process, sequence after being filteredAnd set Cyclic Prefix (CP, L_cpIt is the length of Cyclic Prefix) model The weighted arithmetic average for enclosing the channel response amplitude of the sampled point of self-energy concentration is threshold value thresholding, by the knot after filtering process Fruit performs threshold decision and completes first time denoising, if the corresponding channel response range value of filtered sample point is more than set Threshold value λ, then the sampled point is retained, conversely, by its zero setting；By the result { G after first time denoising_c(m) } extension benefit Zero one-tenth N point sequence(sequence { G_c(m) } 0) afterbody adds N-M, and by N points idct transform, (removal width is to remove window The sinusoidal windows of M), secondary amplitude and phase compensation is (with gain factor δ₂It is multiplied) obtain the estimation of first time transform domainJoint Y_m (k) (the corresponding frequency-region signal of m-th OFDM symbol, n-th sub-carrier signal) andEquilibrium judgement is obtainedIt is rightM points dct transform, LPF, Weighted Threshold denoising, interpolation zero padding and N point idct transforms are repeated, electric power is finally given The channel frequency response of line, completes transform domain quadratic estimate.

The present invention by the transformation of data sequence, the rough estimate of channel frequency response, joint Weighted Threshold denoising with it is balanced To noise second denoising, the wherein transformation of data sequence eliminates influence of the data sequence to channel estimation, it is contemplated that follow for judgement Noise in ring prefix ranges is not eliminated, and setting one is improved and realizes the preferable threshold value thresholding of effect and introduces equal Weighing apparatus decision method completes transform domain quadratic estimate, further eliminates noise jamming, reaches estimation performance and improves, and implementation complexity is fitted In joint Weighted Threshold denoising and balanced judgement transform domain quadratic estimate method.

The present invention proposes a kind of joint Weighted Threshold denoising and the transform domain quadratic estimate method of balanced judgement, is used In typical this embodiment of power line multi-carrier communication, specific embodiment can pass sequentially through following legend to illustrate.

As shown in figure 1, being realization principle overall framework figure of the present invention.Wherein, figure (a) is joint training sequence and mediation sequence Row are produced and send sequence, and using existing LS channel estimation methods, (method is shown in " Liu Qiuge, Mu on the basis of time statistic average Dawn is quick, and Lu Yan brightness is superimposed OFDM channel estimations [J] computer engineering and the application, 2011,47 (31) of Chirp training sequences: 97-100. "), obtain channel frequency response rough estimate；Figure (b) is channel frequency response rough estimateBecome through DCT/IDCT Change the process schematic that domain interpolation, time domain weighting threshold denoising, balanced decision process complete DCT/IDCT transform domain quadratic estimates； Figure (c) depicts link connection schematic diagram between figure (a) and figure (b), and represented linking relationship is：1) joint training sequence Produced with harmonic sequence and send sequence；2) sequence will be sent to send to channel, and using overlying training sequence at receiving terminal Time statistic average method, channel frequency response rough estimate is obtained according to LS estimation criterions3) by rough estimateThrough DCT/IDCT interpolation and Weighted Threshold denoising are crossed, first time transform domain is completed and is estimated, obtained4) receiving terminal number is combined According to sequence { Y_m(k) } and first time transform domain estimated resultIt is equalised to adjudicateAnd willPass through again DCT/IDCT interpolation and Weighted Threshold denoising, complete second transform domain and estimate.

As shown in Fig. 2 producing flow and transmission sequence { s (n) } frame to send sequence { s (n) } in the embodiment of the present invention Structural representation.It is in the present invention, special to eliminate influence of the data sequence { d (n) } to channel estimation shown in Fig. 2 (a) A harmonic sequence is not introduced, is acted on by the mediation of harmonic sequence { a (n) }, data sequence { d (n) } is transformed, so Joint training sequence { t (n) } is produced and sends sequence { s (n) } afterwards.Send sequence { s (n) } expression formula beWherein, " * " represents product calculation；θ is the work(of training sequence { t (n) } Rate, its value is：0<θ<1；Sequence { d (n) } be binary system sequence { x (n) } channel coding/interweave, M-QAM modulation, 1:N serial to parallel conversion and N points IFFT convert generated data sequence；Sequence { t (n) } is that length is the training sequence of N, its cycle It is T (T is positive integer).And Fig. 2 (b) depicts the frame assumption diagram for sending sequence { s (n) }, its frame structure is a length For data sequence { d (n)+a (n) } and length of N stack up transmission for the training sequence { t (n) } of N, and in their heads It is L that portion adds a length_cpCyclic Prefix (CP), and L_cpIt is the integral multiple of T.

In Fig. 2 (a), the circle for including "+" represents summation operation, and the circle for including "×" represents product calculation；Mediation sequence Row { a (n) } meet following condition：And It act as eliminating the influence of training sequence { t (n) } and data sequence { d (n) } to channel estimation, and Q=N/T is integer.

As shown in figure 3, for the present invention, using existing channel estimation methods, (method is shown in " Liu Qiuge, Mu Xiaomin, Lu Yan brightness It is superimposed OFDM channel estimations [J] computer engineering and the application, 2011,47 (31) of Chirp training sequences:97-100. "), it is complete Into the rough estimate of channel frequency response.Implementation step is：1) binary system sequence { x (n) }, through OFDM baseband system front end Data sequence { d (n) } is obtained after treatment, and combines harmonic sequence { a (n) }, training sequence { t (n) } and produce transmission sequence { s (n)}；2) it is L to adding length before each OFDM symbol_cpCyclic Prefix (CP) and through N:Sent after 1 parallel-serial conversion, led to Typical power line channel is crossed, through 1 at receiving terminal:N serioparallel exchanges and the added length of removal are L_cpCP after, to a certain frame OFDM Time statistic average is sought in symbol, channel impulse response time domain rough estimate is obtainedWillChannel is obtained through N point FFTs Frequency response rough estimateAnd result { the y after CP will be removed_m(n) } make N point FFTs and obtain { Y_m(k)}。

As shown in figure 4, being the schematic diagram of present invention setting noise threshold thresholding λ.Noise threshold thresholding λ setting up procedure is passed through Cross following steps completion：1) it is rightSampled using flat-top, and the sampling period is T_s=T/N, obtains sample sequence { g_c(i) }, g_c(i) be the corresponding channel impulse response of ith sample point and i=1,2 ..., N；2) the corresponding letter of each sampled point is sought successively Channel shock response amplitude | g_c(i) |, and all range values are sorted；3) record etc. range value number as weights (if a certain Individual range value identical quantity has a_qIt is individual, then by a_qAs weights), then by a_qCorresponding range value makees product, such as Fig. 4 It is shown；4) by all weights a_qSued for peace with the product of corresponding range value, and be multiplied by the inverse of weights sum, you can obtain noise threshold Value thresholding λ.

As shown in figure 5, the rough estimate for the present invention to channel frequency responseBy DCT/IDCT transform interpolations, The estimation of channel frequency response is obtained after time domain weighting threshold denoising, balanced decision process.Transform domain that Fig. 5 is described is secondary to be estimated Meter process, realizes according to the following steps successively：First, by the rough estimate of Fig. 2 gained channel frequency responsesBy to rough estimate Meter amplitude and phase compensation is (with gain factor δ₁It is multiplied), limited multiplied by sinusoidal window function SIN (window function length is M) Band, obtains resultSecondly, it is rightIt is P to make M point dct transforms and pass through a cut-off frequency_c(P_c=L_cp- 1, L_cpFor The length of Cyclic Prefix) low pass filter, after filtering high fdrequency component, by the amplitude corresponding to its result with weighting time domain threshold value λ makes comparisons, and when flat-top samples the corresponding channel response amplitude of gained sampled point more than λ, then retains the sampled point, otherwise, will adopt Sampling point zero setting；Again, the result zero padding after will determine that is extended to N point sequencesAnd make N points idct transform, go at window Reason and secondary amplitude and phase compensation are (with gain factor δ₂It is multiplied), obtain the frequency response of channelFinally, joint is received Hold receiving data sequence { Y_m(k) } and first time DCT/IDCT estimated resultEquilibrium judgement, secondary DCT/IDCT estimateWherein, the sampling period T of flat-top sampling_s=T/ (N+L_cp)；Sinusoidal window function SIN expression formulas are：K=0,1,2 ..., M；Secondary DCT/IDCT estimates to include to court verdictM points DCT Conversion, LPF, quadratic noise threshold decision, zero padding are extended to N point sequences, N point idct transforms.

As shown in fig. 6, being embodiment of the present invention joint receiving terminal data sequence { Y_m(k) } estimate with first time dct transform domain MeterObtain channel frequency response equilibrium valueSchematic diagram.The balanced decision steps that Fig. 6 describes：First, set balanced Court verdictIt is channel frequency response predicted value, using existing zero forcing equalization technology, joint receiving terminal data sequence { Y_m (k) } obtain sending signal frequency domain estimateAndExpression formula isIts It is secondary, by sending signal frequency domain estimateThrough data decision, it is mapped on the closest point of QAM constellation, obtains sending signal Decision valueAnd channel frequency response decision valueBoth sides relation formula is Finally, joint first time DCT estimationChannel frequency response decision valueAnd decision-feedback coefficient ξ (ξ is's Modifying factor, the function of decision error Δ) weighted sum, obtain channel frequency response equilibrium resultIts expression formula isWherein, " * " represents product calculation；Q₁、Q₂、Q₃Respectively The weights of ξ, are the real number in interval [0,1], and Q₁+Q₂+Q₃=1.

Data decision is according to being：When sending signal decision valueWith X_mK () (modulation is to k-th in m-th OFDM symbol The N points FFT data of subcarrier) it is equal when, court verdict is correct, i.e. channel frequency response decision valueIt is the actual frequency of channel Rate responds H；WhenWith X_mWhen () is unequal k, thenThere is decision error Δ with H, can be made by Δ feedback compensationStep wise approximation H.Wherein, decision errorDecision-feedback coefficient ξ is taken as sentencing Certainly error delta square onGradient, i.e.,

As shown in fig. 7, being the schematic flow sheet of embodiment of the present invention implementation.Fig. 7 depicts the master of the present embodiment realization Want step：

First stage, open channel estimates flow；

Second stage, initializes the parameter of Various types of data；

Phase III, joint overlying training sequence sends sequence with harmonic sequence generation；

Fourth stage, is sent sequence and is transmitted by typical power line multipath channel；

In 5th stage, obtain the rough estimate of channel response；

6th stage, front-end processing (including windowing process, amplitude and phase compensation, the M points of rough estimate that transform domain is estimated Dct transform and LPF)；

In 7th stage, set weighted noise threshold value thresholding；

In 8th stage, the channel response amplitude of sampled point is made comparisons with the noise threshold thresholding of setting, judge sampled point Whether retain；

In 9th stage, the result of reservation is extended to N point sequences, makees N points idct transform, go window to process and secondary amplitude phase Obtained after the compensation of positionJoint Y_m(k)、Frequency domain equalization judgement is carried out, weights Q is set₁、Q₂、Q₃, obtain and balanced sentence Certainly result

Tenth stage, to court verdictSecond denoising is realized, M points dct transform, LPF, Weighted Threshold is repeated After judgement, zero padding are extended to N point sequences, its result is made into N point idct transforms, obtain the result of channel frequency response

The dotted line with arrow is mainly court verdict in Fig. 7Realize that the zero padding after second denoising and threshold decision extends Process.

The present invention proposes a kind of transform domain quadratic estimate method for combining Weighted Threshold denoising and balanced judgement, that includes Joint overlying training sequence is produced the method for sequence that sends with harmonic sequence, is obtained using amendment least square (LS) channel estimation Channel frequency response rough estimate method, joint Weighted Threshold denoising complete the transform domain quadratic estimate side of rough estimate with balanced judgement Method.The present invention is completed through following steps successively：Joint overlying training sequence is produced with harmonic sequence and sends sequence；Using amendment LS Channel estimation methods obtain channel frequency response rough estimate；By acquired results successively amplitude and phase compensation, windowed function, discrete remaining String conversion (DCT), LPF, flat-top sampling processing, and using the weighted average of the channel response amplitude of sampled point as making an uproar Sound threshold value thresholding, threshold decision is made to sampled point, judges that whether sampled point retains；Result after threshold decision is expanded by zero padding Exhibition, existing inverse discrete cosine transform (IDCT) and improved frequency domain equalization decision method, repeat equilibrium result M points DCT and become Change, LPF, Weighted Threshold denoising, interpolation zero padding and N points idct transform complete DCT/IDCT transform domain quadratic estimates, obtain Final channel frequency response.Using the method for the invention, the multipath channel frequencies such as power line multicarrier, radio communication are capable of achieving Response estimates that joint Weighted Threshold is completed to noise denoising twice with balanced judgement, and estimates that performance is higher, moderate complexity. Therefore, the inventive method be applicable to wired or wireless channel noise jamming it is severe when channel estimation, noise jamming detection with The occasions such as suppression.

Although having described embodiments of the invention, to those skilled in the art, present invention side can not departed from These embodiments are carried out with various changes, modification, replacement and modification in the case of method principle and spirit, the scope of the present invention is by institute Attached claim and its equivalent restriction.I.e. by changing the length of the power θ of training sequence in the method for the invention, Cyclic Prefix Degree L_cp, weighted noise threshold value λ, DCT/IDCT conversion points, balanced judgement weights (Q₁、Q₂、Q₃) etc. parameter, still belong to institute of the present invention The category of method is stated, is still protected by this patent.

Claims (5)

1. the transform domain quadratic estimate method of Weighted Threshold denoising and balanced judgement is combined, it is characterised in that entered as follows OK：

1) joint overlying training sequence { t (n) } is produced with harmonic sequence { a (n) } and sends sequence { s (n) }；

2) in OFDM baseband system, by the LS channel estimation methods of overlying training sequence, the thick of channel frequency response is obtained Estimate；

3) after channel frequency response rough estimate being counted as into amplitude and phase compensation, then widened window function, the M point dct transforms spent for M, And make LPF, time domain denoising and sequence zero padding extension process to gained sequence, to the N point sequences after extensionMake N points idct transform, go window process and secondary amplitude and phase compensation obtain

4) acquired results equilibrium is adjudicatedRepeat M points dct transform, LPF, time domain denoising, interpolation zero padding and N points Idct transform, obtains the final estimation of frequency domain response

2. the transform domain quadratic estimate method of Weighted Threshold denoising and balanced judgement is combined described in claim 1, it is characterised in that： Step 1) combine overlying training sequence with harmonic sequence generation transmission sequence process：

By binary system sequence { x (n) } by after the front-end processing of OFDM baseband system, obtaining data sequence { d (n) }, and Joint training sequence { t (n) } and harmonic sequence { a (n) }, produce and send sequence { s (n) }, wherein, harmonic sequence { a (n) } be with Machine periodic sequence, and the cycle is identical with { t (n) }.

3. the transform domain quadratic estimate method of Weighted Threshold denoising and balanced judgement is combined described in claim 2, it is characterised in that： Joint overlying training sequence is produced with harmonic sequence and sends sequence by following steps realization：

Step 2.1. set length as the binary system sequence { x (n) } of N it is encoded/interweave, modulation, 1:N serioparallel exchanges, A₁:A₂ Represent A₁Road is converted to A₂Road, and A₁,A₂It is after natural number, and N point quick Fourier inversion process, to obtain data sequence {d(n)}；

Step 2.2. sets the training sequence { t (n) } and harmonic sequence { a (n) } that length is N, the cycle is T, it is ensured that { d (n)+a (n) } range value of Fast Fourier Transform (FFT) at cycle frequency P is 0；Wherein, P=lN/T, Q=N/T and N, P, T, Q are Integer, l=0,1 ..., T-1；Data sequence { d (n) } meets following relation with harmonic sequence { a (n) }：

${\mathrm{\&Sigma;}}_{k=0}^{Q-1}d(n+kT)={\mathrm{\&Sigma;}}_{k=0}^{Q-1}a(n+kT),n=0,1,...,T-1---\left(2.1\right)$

Then the expression formula of harmonic sequence { a (n) } is as follows：

$a\left(n\right)=a(n+kT)=\frac{1}{Q}{\mathrm{\&Sigma;}}_{k=0}^{Q-1}d(n+kT),n=0,1,...,T-1---\left(2.2\right)$

Step 2.3. is to step 2.1 gained { d (n) }, joint training sequence { t (n) } and harmonic sequence { a (n) }, through formula (2.3) Computing is sued for peace, and produces time domain to send sequence { s (n) }；Take m-th n-th sampled value S of OFDM symbol_mN (), expression formula is：

$\begin{array}{c}{S}_m\left(n\right)=\sqrt{\mathrm{\&theta;}}*t\left(n\right)+\sqrt{1-\mathrm{\&theta;}}*\&lsqb;d_m\left(n\right)+a\left(n\right)\&rsqb;\\ =\sqrt{\mathrm{\&theta;}}*t\left(n\right)+\sqrt{1-\mathrm{\&theta;}}*Dm\left(n\right),n=0,1,\mathrm{...},N-1\end{array}---\left(2.3\right)$

Wherein, θ is the power of training sequence { t (n) }, and is real number, and value is：0<θ<1；D_mN () is m-th OFDM symbol The corresponding discrete quantized sequential value of n-th sampled value, m and n are positive integer.

4. the transform domain quadratic estimate method of Weighted Threshold denoising and balanced judgement is combined described in claim 3, it is characterised in that： The rough estimate is completed according to the following steps：

Step 2.3 gained time domain is sent sequence { s (n) } by step 3.1., and it is L that length is added successively_cpCyclic Prefix, N:1 simultaneously String is changed and sends into channel, then through 1:N serioparallel exchanges and after going circulation prefix processing, obtain data sequence { y_m(n) }, expression formula is such as Under：

$y_m=\sqrt{\mathrm{\&theta;}}*{\mathrm{th}}_m+\sqrt{1-\mathrm{\&theta;}}*D_m{h}_m+N_m---\left(3.1\right)$

Wherein, data sequence vector is:y_m=[y_m(0),y_m(1),…,y_m(N-1)]^T；Channel impulse response is：h_m=[h_m(0), h_m(1),…,h_m(L-1)]^T；T and D_mThe equivalent cycle convolution matrix and data sequence obtained by training sequence { t (n) } are corresponded to respectively {D_m(n) } obtained by equivalent cycle convolution matrix；Noise column vector is:ω_m=[ω_m(0),ω_m(1),…,ω_m(N-1)]^T, and Subscript T representing matrix transposition, " * " represents product calculation；N-dimensional equivalent cycle convolution matrix t and D_mMatrix represent respectively it is as follows：

$t=\left[\begin{array}{ccccc}t\left(0\right)& t(N-1)& t(N-2)& \mathrm{...}& t(N-L+1)\\ t\left(1\right)& t\left(0\right)& t(N-1)& \mathrm{...}& t(N-L+2)\\ .& .& .& .& .\\ .& .& .& .& .\\ .& .& .& .& .\\ t(N-2)& t(N-3)& t(N-4)& \mathrm{...}& t(N-L-1)\\ t(N-1)& t(N-2)& t(N-3)& \mathrm{...}& t(N-L)\end{array}\right]---\left(3.2\right)$

$D_m=\left[\begin{array}{ccccc}{D}_m\left(0\right)& D_m(N-1)& D_m(N-2)& \mathrm{...}& D_m(N-L+1)\\ D_m\left(1\right)& D_m\left(0\right)& D_m(N-1)& \mathrm{...}& D_m(N-L+2)\\ .& .& .& .& .\\ .& .& .& .& .\\ .& .& .& .& .\\ D_m(N-2)& D_m(N-3)& D_m(N-4)& \mathrm{...}& D_m(N-L-1)\\ D_m(N-1)& D_m(N-2)& D_m(N-3)& \mathrm{...}& D_m(N-L)\end{array}\right]---\left(3.3\right)$

Step 3.2. is to step 3.1 gained { y_m(n) }, must receive terminal sequence { Y through N points FFT_m(k)}；Wherein, Y_mK () is m-th N-th sub-carrier signal y of OFDM symbol_mN the frequency domain representation of (), expression formula is as follows：

$Y_m=\sqrt{\mathrm{\&theta;}}*{\mathrm{TH}}_m+\sqrt{1-\mathrm{\&theta;}}*{\mathrm{DH}}_m+W_m---\left(3.4\right)$

Wherein, Y_m(k), H_m, W_mRespectively y_m(n)、h_m、ω_mN point FFTs；If diag [] represents diagonal matrix, and internal Element is diagonal data, then T=diag [T (0), T (1) ..., T (N-1)], D=diag [D_m(0),D_m(1),…,D_m(N- 1)]；In addition, element T (k) expression formula of matrix T is as follows：

$T\left(k\right)=FFT\left\{t\left(n\right)\right\}={\mathrm{\&Sigma;}}_{n=0}^{N-1}t\left(n\right)\exp \{-j2\mathrm{\&pi;}kn/N\},k=0,1,...,N-1---\left(3.5\right)$

In formula (3.5), FFT { } represents the operator of N point FFTs；Exp { } represents the exponent arithmetic symbol of nature truth of a matter e；It is imaginary unit；

Step 3.3. received data sequence { y after Cyclic Prefix and before N point FFTs for going_m(n) }, a frame length is taken for M OFDM symbol, using time statistic average method, taken OFDM symbol is averaging, the signal after obtaining averagely, its time domain expression Formula is：

$\frac{1}{M}{\mathrm{\&Sigma;}}_{m=1}^M{y}_m=\frac{\sqrt{\mathrm{\&theta;}}}{M}{\mathrm{\&Sigma;}}_{m=1}^{M}th+\left(\frac{\sqrt{1-\mathrm{\&theta;}}}{M}{\mathrm{\&Sigma;}}_{m=1}^{M}D\right)h+\frac{1}{M}{\mathrm{\&Sigma;}}_{m=1}^M{\mathrm{\&omega;}}_m---\left(3.6\right)$

Step 3.4. uses LS channel estimations, obtains channel frequency response rough estimateWhen M takes infinity, { D_m(n) } when Domain average is 0, and noise average is also 0, therefore formula (3.6) is deformed into：

$\frac{1}{M}{\mathrm{\&Sigma;}}_{m=1}^M{y}_m=\frac{\sqrt{\mathrm{\&theta;}}}{M}{\mathrm{\&Sigma;}}_{m=1}^{M}th---\left(3.7\right)$

To formula (3.7) equal sign both sides while premultiplicationTry to achieve the rough estimate of channel impulse responseWithIts In, subscript " ls " represents LS channel estimations,Expression formula it is as follows：

${\hat{h}}_{ls}\left(n\right)={\left(\sqrt{\mathrm{\&theta;}}t\right)}^{-1}\left(\frac{1}{M}{\mathrm{\&Sigma;}}_{m=1}^M{y}_m\right)---\left(3.8\right)$

Wherein,ForN point FFTs, subscript " -1 " representing matrix inversion operation.

5. the transform domain quadratic estimate method of Weighted Threshold denoising and balanced judgement is combined according to claim 4, it is characterised in that： Step 3), it is 4) specific as follows：

Step 4.1. is by step 3.4 gained rough estimateThrough amplitude and phase compensation, windowed function, M point discrete cosine transforms After treatment, obtainWherein, subscript " c " is represented in discrete cosine transform domain；Amplitude and phase compensation is by rough estimate It is multiplied by a gain factor δ₁Complete；It is the window function of M that institute's windowed function uses width；And gain factor δ₁With selected window function Using sinusoidal window function SIN, expression formula is respectively：

${\mathrm{\&delta;}}_1=\left\{\begin{array}{c}1/\sqrt{2},k=0\\ \exp \{-j\mathrm{\&pi;}k/\left(2M\right)\},1\&le;k\&le;M-1\end{array}\right.---\left(4.1\right)$

$SIN=sin\left(\mathrm{\&pi;}\right(k+\frac{1}{2})/M)---\left(4.2\right)$

Step 4.2. is by obtained by step 4.1After being extended through threshold denoising, zero padding, obtainWherein, time domain threshold value is gone Process of making an uproar is realized using following sub-step：

Step 4.2.1 concentrates on low-frequency range because of signal energy, by sequenceAfter low pass filter filters high fdrequency component, obtain Sequence after to filtering

Step 4.2.2 is by obtained by step 3.4Flat-top is used to sample it, and the sampling period is T_s=T/ (N+L_cp), obtain Sample sequence { g_c(i) }, and seek the amplitude mould of each sampled point respective channels shock response；If there is the amplitude such as a in sample sequence Value sampled point, then range value correspondence weights are a；Now, threshold value thresholding λ is tried to achieve by formula (4.3)：

$\mathrm{\&lambda;}=\frac{{\mathrm{\&Sigma;}}_{i=1}^q{a}_i*\left|g_c\left(i\right)\right|}{a_1+a_2+...+a_q},q=1,2,...,N---\left(4.3\right)$

Wherein, g_cI () is the corresponding channel impulse response of ith sample point, i is integer, and i=1,2 ..., N；a₁+a₂ +···+a_q=N.

Step 4.2.3 will filter after sequenceJudge by formula (4.4), sampled point is retained or zero setting：

Step 4.2.4 is to step 4.2.3 gained sequences { G_c(m) }, zero padding expands to N point sequencesAnd zero padding extends Process is in sequence { G_c(m) } end addition N-M zero；

Step 4.3. is by obtained by step 4.2.4Respectively through N points idct transform, go window process and secondary amplitude-phase Compensation, obtainsWherein, secondary amplitude and phase compensation is by going window result to be multiplied by a gain factor δ₂Complete, And gain factor δ₂Expression formula is：

${\mathrm{\&delta;}}_2=\left\{\begin{array}{c}\sqrt{2N/M},k=0\\ \sqrt{N/M}*\exp \{j\mathrm{\&pi;}k/\left(2N\right)\},1\&le;k\&le;N-1\end{array}\right.---\left(4.5\right)$

Gained sequence { the Y of step 4.4. joint steps 2.4_m(k) } to step 4.3 gainedEquilibrium judgement, and by equilibrium resultRepeat M points dct transform, LPF, Weighted Threshold denoising, interpolation zero padding and N points idct transform completion transform domain secondary Estimation procedure, obtains the estimated result of channel frequency responseWherein, balanced judgement is completed by following sub-step：

Step 4.4.1. sets equilibrium resultIt is channel frequency response predicted value, then receives signal channel frequency response prediction Sending signal frequency domain estimate is obtained after the zero forcing equalization of value Expression formula is：

$\hat{X}\left(k\right)=\frac{Y_m\left(k\right)}{\tilde{H}\left(k\right)},k=0,1,...,N-1---\left(4.6\right)$

Step 4.4.2. is by the sending signal frequency domain estimate after equilibriumThrough data decision, QAM constellation is mapped to most On neighbor point, sending signal decision value is obtainedFrequency channels response decision value is obtained simultaneouslyWherein,Expression formula For：

$\stackrel{\&OverBar;}{H}\left(k\right)=\frac{Y_m\left(k\right)}{\stackrel{\&OverBar;}{X}\left(k\right)},k=0,1,...,N-1---\left(4.7\right)$

The foundation of the judgement is as follows：

1) decision value is worked asWhen, then court verdict is correct, i.e. channel frequency response decision valueIt is channel reality Frequency response values H, X_mK () is modulation to k-th N point FFT data of subcarrier in m-th OFDM symbol；

2) decision value is worked asWhen, then decision valueThere is decision error Δ with the response of channel actual frequency, can lead to Decision error Δ feedback compensation is crossed, makes channel frequency response predicted valueStep wise approximation channel real response H；Wherein, adjudicate Error delta expression formula is：

$\mathrm{\&Delta;}=\stackrel{\&OverBar;}{X}\left(k\right)-\hat{X}\left(k\right)=\stackrel{\&OverBar;}{X}\left(k\right)-\frac{Y_m\left(k\right)}{\tilde{H}\left(k\right)}---\left(4.8\right)$

And decision-feedback coefficient ξ is channel frequency response predicted valueModifying factor, and ξ for decision error Δ function, I.e.

Step 4.4.3. joint first times DCT are estimatedDecision valueAnd decision-feedback coefficient ξ weighted sums, obtain letter Road frequency response equalization resultWherein,Expression formula is：

$\tilde{H}\left(k\right)=Q_1*{\hat{H}}_N\left(k\right)+Q_2*\stackrel{\&OverBar;}{H}\left(k\right)+Q_3*\mathrm{\&xi;}---\left(4.9\right)$

Wherein, Q₁、Q₂、Q₃RespectivelyThe weights of ξ, value is the real number of interval [0,1], and Q₁+Q₂+Q₃ =1.