CN100521665C - Iterative decomposition method for fixed training sequence stuffing modulation system - Google Patents

Abstract

The invention belongs to the technical field of digital information transmission, and is characterized in that it is an iterative decomposition method based on fixed training sequence filling, and the current frame training sequence is skillfully regarded as a "virtual frame" composed of the current frame data and the next frame training sequence. The cyclic prefix of " makes the linear convolution of the "virtual frame" sent in the time domain and the channel impulse response into a circular convolution, which is convenient to use the frequency domain DFT method to obtain data and training sequence estimates, and facilitates the data and training sequences. separation and more accurate channel estimation. Through repeated iterations of channel estimation and data separation, more accurate channel estimation and more accurate data decomposition can be obtained. The computer simulation shows that the channel estimation and equalization of the present invention hardly loses the error performance of the system, and greatly simplifies the processing of channel estimation and equalization.

Description

A kind of iterative decomposition method that is used for fixing training sequence stuffing modulation system

Technical field

The invention belongs to digital information transmission technical field, be particularly related to OFDM (Orthogonal FrequencyDivision Multiplexing, OFDM) multicarrier system or single carrier (Single Carrier, SC) in the system based on fixed training sequence (Training Sequence, TS) iterative decomposition method of Tian Chonging.

Background technology

The problem that the current communication technology mainly will solve is how to improve transmission rate in limited bandwidth reliably.Because influence, the especially wireless channel of channel are time dispersive channels, signal can produce distortion and add noise, and therefore in effective transmission payload or data division, system also should be able to identify the variation of channel, and is offset or compensate.In addition, receiving terminal also will carry out clock recovery, carrier wave recovery and channel estimating etc. to received signal.In the transmission environment of complexity, for guaranteeing the reliability and the high efficiency of communication system, the method that communication system adopts usually is one section signal specific of transmission, and protection interval and data block that this signal also can be used as between the transmission data block are synchronous.

For multicarrier system, first and second kinds of formations of OFDM frame format are shown in Fig. 1 (a) and 1 (b) usually.In the form of Fig. 1 (a), DFT (or FFT) piece A2 is positioned at after the Cyclic Prefix section A1, and Cyclic Prefix is as the protection interval of DFT piece.OFDM modulation needs to use protection at interval or its equivalent, so that the multipath that may exist in the antagonism received signal disturbs, prevents the OFDM intersymbol interference, this structure be called Cyclic Prefix in OFDM System (Cyclic Padding OFDM, CP-OFDM).CP-OFDM is used widely at present, as digital audio broadcasting (Digital Audio Broadcasting, DAB), digital video broadcast-terrestrial (Terestrial Digital Video Broadcasting, DVB-T), IEEE 802.11a, HIPERLAN/2 WLAN standard etc.In the form of Fig. 1 (b), the zero padding section is followed in DFT piece back, and this zero padding section is as the protection interval of DFT piece.This structure be called zero padding OFDM (Zero Padding OFDM, ZP-OFDM).Can be referring to document [Muquet B, Wang Z, Giannakis G.B, Courville M.de, and Duhamel P, CyclicPrefixing or Zero Padding for Wireless Multicarrier Transmissions, IEEE Trans.onCommunications, 2002,50 (12): 2136-2148.].

For single-carrier system, frame structure form normally data block is protected the method for filling training sequence in the interval, and it constitutes shown in Fig. 1 (c) and 1 (d).In the form of Fig. 1 (c), data block C2 is between two training sequence C1, and previous training sequence can be regarded the Cyclic Prefix (CP) of a data block and a back training sequence as, eliminates the data inter-block-interference.In the form of Fig. 1 (d), data block D2 is positioned at after the training sequence D1.Can be referring to document [Witschnig H., Mayer, T., Petit M., Hutzelmann H., Springer A., Weigel R., The advantages of a unique wordfor synchronisation andchannel estimation in a SC/FDE system Personal Mobile Communications Conference, 2003.5th European (Conf.Publ.No.492)22-25 April 2003 Page (s): 436-440].

For top described multicarrier/single-carrier system, we all can regard its frame structure as the structure that a training sequence adds data block, and no matter this data block is time domain (as single-carrier system) or (as the multicarrier system) of frequency domain.How to separate training sequence and data, eliminate data to the interference of training sequence with estimated channel impulse response better, and eliminate training sequence to the interference of data with restore data better, be the subject matter that this class frame structure runs in channel estimating and equilibrium.Document [L.Deneire; B.Gyselinckx; and M.Engels; Training sequence versuscyclic prefix-A new lookon single carrier communication; IEEE Commun.Lett.; vol.5; no.7; pp.292-294; Jul.2001.] propose a kind of new thinking, be used for analyzing the carrier wave communication system of filling, as shown in Figure 2 based on fixed training sequence (promptly the training sequence of protection interval filling is the same); regard current frame data and next frame training sequence as a virtual frames, and the training sequence of present frame is regarded as the Cyclic Prefix of this virtual frames.Repetitiousness like this, each training sequence is equivalent to all be used as the rear section frame of previous frame and the Cyclic Prefix of next frame, has improved the efficiency of transmission of system's payload.Because training sequence is regarded a Cyclic Prefix (CP) as, can remove the intersymbol interference (ISI) that multipath channel causes easily in addition, easier with training sequence and data separating, simplify the channel estimating and the equilibrium of multicarrier and single carrier transmitting system.

Summary of the invention

The objective of the invention is to propose a kind of iterative decomposition method that sequence is filled single carrier/multicarrier system that is used for fixing, eliminate the intersymbol interference that multipath channel causes, and training sequence and data are separated.

To the single carrier/multicarrier system of filling based on fixed sequence program, its baseband model as shown in Figure 3.S/P and P/S represent string and conversion and and string conversion respectively among the figure.

Suppose that at transmitting terminal the i frame data are

Symbol

Expression length is the sequence of N.

If through multicarrier system, then need to obtain behind the process inverse discrete Fourier transform (IDFT) earlier,

$s_{i,k}=\frac{1}{\sqrt{N}}\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{i,n}\exp \left\{\mathrm{<figure-callout\; id="2"\; label="j"\; filenames="C200710098532E00181.png"\; state="\{\{state\}\}">j</figure-callout>}\frac{2\mathrm{\&pi;nk}}{N}\right\},0\&le;k<N$

In the formula, i represents the OFDM frame number, and capitalization S represents time domain data, and small letter s represents frequency domain data.

Then, with the fixed training sequence (TS) that pre-defines

Be inserted into the output of each IDFT

In.If through single-carrier system, then directly with fixed training sequence (TS)

Be inserted into the i frame data

In, as Fig. 4. (a).

Because the frame structure of multicarrier system and single-carrier system all can be expressed as with the protection between fixed training sequence (TS) the alternate data piece at interval, and therefore data block can be used the unification of i frame data below in frequency domain or time domain

Expression.As Fig. 4. (b), transmission signal frame can be divided into not two parts of aliasing, i.e. fixed training sequence mutually

I 〉=0 and data ${\left\{d_{i,k}\right\}}_{k=0}^{N-1},i\&GreaterEqual;0$

Transmit and distortion can take place and introduce noise through transmission channel.For convenience of description, we suppose that the channel model that adopts is quasi-static L rank finite impulse responses (FIR) systems, and channel impulse response (CIR) is designated as

In the transmission system of reality, the fixed training sequence Design of length is the maximum delay that is not less than channel, and M 〉=L is promptly arranged.

Since the influence of multipath channel, receiving end signal

Aliasing can take place, and has the received signal of aliasing to comprise two parts: The linear convolution result of expression fixed training sequence and channel impulse response,

The linear convolution result of expression data block and channel impulse response is as Fig. 4. (c), 4. shown in (d).

With

Can be expressed as respectively:

$x_{i,k}=d_{i,k}*h_{i,k}=\underset{l=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{d}_{i,k-l}\&CenterDot;h_{i,l},0\&le;k<N+L-1$

$y_{i,k}=c_k*h_{i,k}=\underset{l=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{c}_{k-l}\&CenterDot;h_{i,l},0\&le;k<M+L-1$

Because The noise, received signal can be expressed as:

r _i，k＝u _i，k+n _i，k，0≤k<M+N

In the formula, n _{I, k}Expression additive white Gaussian noise (AWGN), wherein u _{I, k}Can be expressed as:

$u_{i,k}=\left\{\begin{array}{cc}{x}_{i-1,k+N}+y_{i,k},& 0\&le;k<L-1\\ y_{i,k},& L-1\&le;k<M\\ x_{i,k-M}+y_{i,k},& M\&le;k<M+L-1\\ x_{i,k-M},& M+L-1\&le;k<N+M\end{array}\right.$

When the receiving terminal demodulating data, need with

From Remove in the signal, stay remainder

Promptly eliminate the interference of training sequence to data.

For multicarrier system, remainder Just the situation with zero padding OFDM (ZP-OFDM) is identical, therefore the existing method that is applied to the ZP-OFDM system all is suitable for, document [Muquet B for example, Wang Z, Giannakis G.B, CourvilleM.de, and Duhamel P, Cyclic Prefixing or Zero Padding for Wireless Multicarrier Transmissions, IEEE Trans.on Communications, 2002,50 (12): 2136-2148.] ZF of mentioning in (ZF) and least mean-square estimate (MMSE) equalization algorithm.

For single-carrier system, remainder

Can utilize classical time domain equalization technology to eliminate ISI, for example adaptive algorithm of mentioning in the document [S.U.H.Qureshi, " Adaptive Equalization, " Proc.IEEE, vol.73, no.9, Sept.1985, pp.1349-87]; Perhaps use frequency-domain equalization technology, document [N.Benvenuto and S.Tomasin for example, " On the comparisonbetween OFDM and single carrier modulation with a DFE using afrequency-domain feed-forwardfilter; " IEEE Trans.Commun., vol.50, no.6, pp.947-955, June 2002] in the SC-DFE method mentioned.

In general analysis, we suppose that usually receiving terminal can access channel impulse response accurately, so fixed training sequence can disassemble fully with data.Yet in fact to expect channel impulse response accurately, the influence of data and training sequence need be separated fully.Therefore, accurately decomposition data and training sequence and accurate these 2 of estimated channel impulse responses condition each other.And because the time variation of channel, the error of system's introducing and the deviation of method of estimation own often can not obtain channel impulse response accurately.

The present invention proposes iterative decomposition method, makes channel estimating more accurate, and separating of training sequence and data is also more accurate.The invention is characterized in that this method may further comprise the steps successively in realization, as shown in Figure 5:

Step (1). supposing the system is finished synchronously, the training sequence of the data of present frame and next frame is considered as a virtual frames, and the training sequence of present frame is regarded as the Cyclic Prefix of this virtual frames, thereby described Cyclic Prefix and virtual frames are constituted a processed frame, wherein identical training sequence is made the protection usefulness of filling at interval, is defined as fixed training sequence, so repetitiousness, with when removing intersymbol interference, training sequence and data separating;

Step (2). differentiating described processed frame, represent with i, is also right and wrong the 0th frame of the 0th frame, estimates its channel impulse response respectively:

Step (2.1). when processed frame is the 0th frame:

The fixed training sequence in the processed frame

And the linear convolution result of this fixed training sequence and corresponding its channel impulse estimation response

Be N ₁The point discrete Fourier conversion, N ₁M+L, the not enough N if sequence is counted ₁, then zero padding is to N ₁The point.M is the length of fixed training sequence, and L is a channel impulse response length for the exponent number as the quasi-static L rank finite impulse response filter of channel model, thereby obtains

With

The channel estimating of the 0th frame

Obtain by following formula

${\hat{h}}_{i,k}^{\mathrm{iter}=0}=\mathrm{IDFT}\left\{\frac{{\mathrm{<figure-callout\; id="0"\; label="Y"\; filenames="C200710098532E00171.png"\; state="\{\{state\}\}">Y</figure-callout>}}_{0,k}}{C_k}\right\},$ 0≤k＜N ₁-1, k 〉=L item is 0;

Step (2.2). when processed frame is non-the 0th frame:

If for the first time iteration, the then channel impulse response of i frame

Channel impulse response with known i-1 frame

Equate, promptly ${\left\{{\hat{h}}_{i,l}^{\mathrm{iter}=0}\right\}}_{l=0}^{L-1}={\left\{{\hat{h}}_{i-1,l}\right\}}_{l=0}^{L-1}$ Iteration sequence number I is made as 0, the beginning iteration;

If not iteration for the first time is then the channel impulse response based on known i-1 frame

Last iteration result with the i frame

Linear interpolation obtains

Step (3). use the frequency domain equalization algorithm, estimate i signal frame data

Step (3.1). the data in the i signal frame With training sequence in the i+1 signal frame

Regard the i virtual frames as

The training sequence in the i signal frame be considered as the i virtual frames cyclic prefix CP, wherein N is the length of data, thereby obtains the i virtual frames

With the channel impulse response of i virtual frames as the training sequence of Cyclic Prefix

Circular convolution

Step (3.2). respectively in the step (3.1)

Make N ₂The conversion of=M+N point discrete Fourier obtains

With

The not enough N if sequence is counted ₂, then zero padding is to N ₂Point, i virtual frames data estimation ${\left\{{\hat{v}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{N_1-1}$ Obtain by following formula

${\left\{{\hat{v}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{M+N-1}=\mathrm{IDFT}\left\{\frac{Z_{i,k}^{\mathrm{iter}=I}}{H_k^{\mathrm{iter}=I}}\right\},$ 0≤k<M+N-1

Take out the i virtual frames

The top n data be i signal frame data estimation:

${\left\{{\hat{d}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{N-1}={\left\{{\hat{v}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{N-1},$ 0≤k<N-1；

Step (4). obtain the linear convolution result's of i frame training sequence and channel impulse response estimation,

By calculating N ₂Leaf transformation obtains the data in the i signal frame in the=M+N point discrete Fourier

And channel impulse response

The linear convolution result be expressed as

The not enough N if sequence is counted ₂, then zero padding is to N ₂The point,

From i frame received signal The middle linear convolution result who removes i-1 frame data and channel impulse response

And

Thereby obtain the linear convolution result of i signal frame training sequence and channel impulse response ${\left\{{\hat{y}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{M+N-1}$ Estimation:

${\hat{y}}_{i,k}^{\mathrm{iter}=I}=\left\{\begin{array}{cc}{r}_{i,k}-{\hat{x}}_{i-1,k+N},& 0\&le;k<L-1\\ r_{i,k},& L-1\&le;k<M\\ r_{i,k}-{\hat{x}}_{i,k-M}^{\mathrm{iter}=I},& M\&le;k<M+L-1\end{array}\right.$

Step (5). adopt the frequency domain algorithm for estimating, estimate the channel impulse response of i signal frame,

Will

With Be N ₁Leaf transformation obtains in the point discrete Fourier

With

The not enough N if sequence is counted ₁, then zero padding is to N ₁Point, channel estimating Obtain by following formula

${\hat{h}}_{i,k}^{\mathrm{iter}=I}=\mathrm{IDFT}\left\{\frac{Y_{i,k}^{\mathrm{iter}=I}}{C_k}\right\},$ 0≤k<N ₁-1，

Then, will

In k 〉=L item be set to zero, thereby obtain

L is the length of channel impulse response;

Step (6) if. reached predefined iterations J, then stopped iteration,

With

Be respectively right

With

Final estimation, be designated as Then use the same method again and handle the i+1 frame signal; If do not reach predefined iterations J, I adds 1 with the iteration sequence number, returns step (1) and carries out iterative computation again.

Description of drawings

Fig. 1 is two kinds of frame structures that are applied to multicarrier system at present and two kinds of frame structures that are applied to single-carrier system: figure (a) is Cyclic Prefix OFDM frame structure, figure (b) is a zero padding OFDM frame structure, the frame structure one that figure (c) fills for fixed training sequence, the frame structure two that figure (d) fixed training sequence is filled;

Fig. 2 is the new processing thinking for above-mentioned four kinds of frame structures;

Fig. 3 is the baseband model based on the multicarrier/single-carrier system of training sequence filling;

Fig. 4 is the transmission of above-mentioned four kinds of frame structure systems and the Time Domain Decomposition of received signal frame: figure (a) is for sending signal frame, and figure (b) is for sending the Time Domain Decomposition of signal frame, and figure (c) is the Time Domain Decomposition of received signal, and figure (d) is the received signal frame; Wherein figure (b) illustrates that the training sequence and the data that send signal frame do not have aliasing, illustrate that because multipath disturbs, the training sequence of received signal frame and data are aliasings and scheme (c);

The flow chart of the iterative decomposition method of filling based on fixed training sequence that Fig. 5 proposes for the present invention;

The error sign ratio performance curve of iterative algorithm under channel model 1 that Fig. 6 proposes for the present invention:

Be QPSK, I=0,

Be QPSK, I=1,

Be QPSK, I=2,

Be 16QAM, I=0,

Be 16QAM, I=1,

Be 16QAM, I=2,

Be 64QAM, I=0,

Be 64QAM, I=1,

Be 64QAM, I=2,

Same under the curve representation method;

The error sign ratio performance curve of iterative algorithm under channel model 2 that Fig. 7 proposes for the present invention.

Embodiment

The iterative decomposition method of the present invention proposes a kind of multicarrier/single-carrier system of filling based on fixed training sequence, it may further comprise the steps successively:

Step (1). supposing the system is finished synchronously, the training sequence of the data of present frame and next frame is considered as a virtual frames, and the training sequence of present frame is regarded as the Cyclic Prefix of this virtual frames, thereby described Cyclic Prefix and virtual frames are constituted a processed frame, wherein identical training sequence is made the protection usefulness of filling at interval, is defined as fixed training sequence, so repetitiousness, with when removing intersymbol interference, training sequence and data separating;

Step (2). differentiating described processed frame, represent with i, is also right and wrong the 0th frame of the 0th frame, estimates its channel impulse response respectively:

Step (2.1). when processed frame is the 0th frame:

The fixed training sequence in the processed frame

And the linear convolution result of this fixed training sequence and corresponding its channel impulse estimation response Be N ₁The point discrete Fourier conversion, N ₁M+L, the not enough N if sequence is counted ₁, then zero padding is to N ₁The point.M is the length of fixed training sequence, and L is a channel impulse response length for the exponent number as the quasi-static L rank finite impulse response filter of channel model, thereby obtains

With

The channel estimating of the 0th frame

Obtain by following formula

${\hat{h}}_{i,k}^{\mathrm{iter}=0}=\mathrm{IDFT}\left\{\frac{{\mathrm{<figure-callout\; id="0"\; label="Y"\; filenames="C200710098532E00171.png"\; state="\{\{state\}\}">Y</figure-callout>}}_{0,k}}{C_k}\right\},$ 0≤k＜N ₁-1, k 〉=L item is 0;

Step (2.2). when processed frame is non-the 0th frame:

If for the first time iteration, the then channel impulse response of i frame

Channel impulse response with known i-1 frame

Equate, promptly ${\left\{{\hat{h}}_{i,l}^{\mathrm{iter}=0}\right\}}_{l=0}^{L-1}={\left\{{\hat{h}}_{i-1,l}\right\}}_{l=0}^{L-1}$ Iteration sequence number I is made as 0, the beginning iteration;

If not iteration for the first time is then the channel impulse response based on known i-1 frame

Last iteration result with the i frame

Linear interpolation obtains

Step (3). use the frequency domain equalization algorithm, estimate i signal frame data

Step (3.1). the data in the i signal frame

With training sequence in the i+1 signal frame

Regard the i virtual frames as

The training sequence in the i signal frame be considered as the i virtual frames cyclic prefix CP, wherein N is the length of data, thereby obtains the i virtual frames

With the channel impulse response of i virtual frames as the training sequence of Cyclic Prefix

Circular convolution

Step (3.2). respectively in the step (3.1)

Make N ₂The conversion of=M+N point discrete Fourier obtains

With

The not enough N if sequence is counted ₂, then zero padding is to N ₂Point, i virtual frames data estimation ${\left\{{\hat{v}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{N_1-1}$ Obtain by following formula

${\left\{{\hat{v}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{M+N-1}=\mathrm{IDFT}\left\{\frac{Z_{i,k}^{\mathrm{iter}=I}}{H_k^{\mathrm{iter}=I}}\right\},0\&le;k<M+N-1$

Take out the i virtual frames

The top n data be i signal frame data estimation:

${\left\{{\hat{d}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{N-1}={\left\{{\hat{v}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{N-1},0\&le;k<N-1;$

Step (4). obtain the linear convolution result's of i frame training sequence and channel impulse response estimation,

By calculating N ₂Leaf transformation obtains the data in the i signal frame in the=M+N point discrete Fourier

And channel impulse response

The linear convolution result be expressed as

The not enough N if sequence is counted ₂, then zero padding is to N ₂The point,

From i frame received signal

The middle linear convolution result who removes i-1 frame data and channel impulse response

And

Thereby obtain the linear convolution result of i signal frame training sequence and channel impulse response ${\left\{{\hat{y}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{M+N-1}$ Estimation:

${\hat{y}}_{i,k}^{\mathrm{iter}=I}=\left\{\begin{array}{cc}{r}_{i,k}-{\hat{x}}_{i-1,k+N},& 0\&le;k<L-1\\ r_{i,k},& L-1\&le;k<M\\ r_{i,k}-{\hat{x}}_{i,k-M}^{\mathrm{iter}=I},& M\&le;k<M+L-1\end{array}\right.$

Step (5). adopt the frequency domain algorithm for estimating, estimate the channel impulse response of i signal frame,

Will With

Be N ₁Leaf transformation obtains in the point discrete Fourier

With

The not enough N if sequence is counted ₁, then zero padding is to N ₁Point, channel estimating Obtain by following formula

${\hat{h}}_{i,k}^{\mathrm{iter}=I}=\mathrm{IDFT}\left\{\frac{Y_{i,k}^{\mathrm{iter}=I}}{C_k}\right\},$ 0≤k<N ₁-1，

Then, will

In k 〉=L item be set to zero, thereby obtain

L is the length of channel impulse response;

Step (6) if. reached predefined iterations J, then stopped iteration,

With Be respectively right

With

Final estimation, be designated as

Then use the same method again and handle the i+1 frame signal; If do not reach predefined iterations J, I adds 1 with the iteration sequence number, returns step (1) and carries out iterative computation again.

Based on foregoing description, the iterative decomposition method based in the multicarrier/single-carrier system of fixed training sequence filling proposed by the invention has been carried out Computer Simulation, main simulation parameter is as shown in table 1.Adopt two kinds of channel models 1 shown in the table 2 and 3 and 2 in the emulation.Wherein, multipath channel shown in the table 2 is the fixedly receive channel model of European DVB-T standard, multipath channel shown in the table 3 is Chinese pbs film TV management (the State Administration of Radio Film and Television of general bureau, SARFT) single frequency network that proposes in the Digital Television test report (SFN) model, this multipath channel comprise the 0dB echo that time-delay reaches 30us.In simulation process, adopt parameter N=3780, M=L=420, N ₁=840, N ₂=4200, fixed training sequence adopts one section pseudo random sequence, and (Pseudorandom Noise Sequence PN) fills.It is that (Symbol Error Rate, SER) performance relatively for the error sign ratio of 0,1,2 o'clock QPSK, 16QAM, three kinds of planisphere modulated of 64QAM system that Fig. 6 and Fig. 7 have listed under two kinds of simulated channel no iteration and iterations respectively.As seen, the method for the present invention's proposition just makes systematic function be greatly improved when an iteration.

The main simulation parameter of table 1

Symbol rate	7.56M symbol/second
		The subcarrier planisphere	QPSK，16QAM，64QAM
OFDM sub-carrier number N	3780
		Subcarrier spacing	2KHz
The training sequence length M	420

The channel impulse response of table 2 channel model 1

Tap	Postpone (T s)	Normalized power
			1	0	1
2	2	0.22589
			3	4	0.15034

4	5	0.051534
			5	6	0.14972
6	7	0.17099
			7	13	0.29572
8	16	0.40716
			9	18	0.25878
10	19	0.22116
			11	26	0.26291
12	28	0.24014
			13	30	0.057662
14	31	0.061831
			15	41	0.25973
16	42	0.11659
			17	59	0.40097
18	83	0.30359
			19	98	0.35083
20	101	0.18507
			21	165	0.17681

The channel impulse response of table 3 channel model 2

Tap	Postpone (T s)	Normalized power
			1	14	1
2	0	0.126
			3	15	0.1
4	27	0.1
			5	57	0.316
6	241	1

Claims (1)

1, the iterative decomposition method in a kind of modulating system of filling based on fixed training sequence is characterized in that this method is to realize according to the following steps successively in the digital integrated circuit of receiving terminal:

Step (1). suppose that modulating system finishes synchronously, the training sequence of the data of current demand signal frame and next frame is considered as a virtual frames, and the training sequence of current demand signal frame is regarded as the Cyclic Prefix of this virtual frames, thereby described Cyclic Prefix and virtual frames are constituted a processed frame, wherein identical training sequence is made the protection usefulness of filling at interval, is defined as fixed training sequence, so repetitiousness, with when removing intersymbol interference, training sequence and data separating;

Step (2). differentiating described processed frame, represent with i, is also right and wrong the 0th frame of the 0th frame, estimates its channel impulse response respectively:

Step (2.1). when processed frame is the 0th frame:

The fixed training sequence in the processed frame

And the linear convolution result of this fixed training sequence and corresponding its channel impulse estimation response

Be N ₁The point discrete Fourier conversion, N ₁M+L, the not enough N if sequence is counted ₁, then zero padding is to N ₁Point, M are the length of fixed training sequence, and L is a channel impulse response length for the exponent number as the quasi-static L rank finite impulse response filter of channel model, thereby obtains

With

The channel impulse response of the 0th frame

Obtain by following formula

${\hat{h}}_{i,k}^{\mathrm{iter}=0}=\mathrm{IDFT}\left\{\frac{Y_{0,k}}{C_k}\right\},$ 0≤k＜N ₁-1, k 〉=L item is 0;

Step (2.2). when processed frame is non-the 0th frame:

If for the first time iteration, the then channel impulse response of i frame

Channel impulse response with known i-1 frame

Equate, promptly ${\left\{{\hat{h}}_{i,l}^{\mathrm{iter}=0}\right\}}_{l=0}^{L-1}={\left\{{\hat{h}}_{i-1,l}\right\}}_{l=0}^{L-1}$ Iteration sequence number I is made as 0, the beginning iteration;

If not iteration for the first time is then the channel impulse response based on known i-1 frame Last iteration result with the i frame

Linear interpolation obtains

Step (3). use the frequency domain equalization algorithm, estimate i signal frame data

Step (3.1). the data in the i signal frame

With training sequence in the i+1 signal frame

Regard the i virtual frames as

The training sequence in the i signal frame be considered as the i virtual frames cyclic prefix CP, wherein N is the length of data, thereby obtains the i virtual frames

With channel impulse response as the training sequence of the cyclic prefix CP of i virtual frames

Circular convolution

Step (3.2). respectively in the step (3.1) Make N ₂The conversion of=M+N point discrete Fourier obtains With The not enough N if sequence is counted ₂, then zero padding is to N ₂Point, i virtual frames data estimation

Obtain by following formula

${\left\{{\hat{v}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{M+N-1}=\mathrm{IDFT}\left\{\frac{Z_{i,k}^{\mathrm{iter}=I}}{H_k^{\mathrm{iter}=I}}\right\},$ 0≤k<M+N-1

Take out the i virtual frames

The top n data be i signal frame data estimation:

${\left\{{\hat{d}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{N-1}={\left\{{\hat{v}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{N-1},$ 0≤k<N-1；

Step (4). obtain the linear convolution result's of i frame training sequence and channel impulse response estimation,

By calculating N ₂Leaf transformation obtains the data in the i signal frame in the=M+N point discrete Fourier

And channel impulse response

The linear convolution result be expressed as

The not enough N if sequence is counted ₂, then zero padding is to N ₂The point,

From i frame received signal The middle linear convolution result who removes i-1 frame data and channel impulse response

And

Thereby obtain the linear convolution result of i signal frame training sequence and channel impulse response ${\left\{{\hat{y}}_{i,k}^{\mathrm{iter}=I}\right\}}_{k=0}^{M+N-1}$ Estimation:

${\hat{y}}_{i,k}^{\mathrm{iter}=I}=\left\{\begin{array}{cc}{r}_{i,j}-{\hat{x}}_{i-1,k+N}& 0\&le;k<L-1\\ r_{i,k}& L-1\&le;k<M\\ r_{i,k}-{\hat{x}}_{i,k-M}^{\mathrm{iter}=I}& M\&le;k<M+L-1\end{array}\right.$

Step (5). adopt the frequency domain algorithm for estimating, estimate the channel impulse response of i signal frame,

Will

With

Be N ₁Leaf transformation obtains in the point discrete Fourier

With

The not enough N if sequence is counted ₁, then zero padding is to N ₁Point, channel impulse response

Get by following formula

${\hat{h}}_{i,k}^{\mathrm{iter}=I}=\mathrm{IDFT}\left\{\frac{Y_{i,k}^{\mathrm{iter}=I}}{C_k}\right\},$ 0≤k<N ₁-1，

Then, will

In k 〉=L item be set to zero, thereby obtain

L is the length of channel impulse response;

Step (6) if. reached predefined iterations J, then stopped iteration, With

Be respectively right

With

Final estimation, be designated as

Then use the same method again and handle the i+1 frame signal; If do not reach predefined iterations J, I adds 1 with the iteration sequence number, returns step (1) and carries out iterative computation again.

Images