CN101860504B - Channel equalization method for eliminating rear path interference by using movable tap - Google Patents

Abstract

The invention proposes a channel equalization method for eliminating rear path interferences by using a movable tap, which comprises the following steps of carrying out sliding correlation processing to each frame of received symbol data and local fixed frame header, computing correlation energy values and recording the maximum value of each frame of correlation energy, wherein the correlation length is the length of the frame header; according to the maximum value of the correlation energy obtained from the previous frame, setting a main path relative threshold and determining the position of a main path; according to the maximum value of the correlation energy obtained from the previous frame, setting a strong path relative threshold and detecting the quantity and the positions of rearpaths; and setting a window with a preset character length by each level movable tap by using the detected position of each rear path as a center to construct an equalizer for channel equalization processing. In the invention, the movable tap is used for eliminating the rear paths to greatly reduce the level number of a feedback filter, and 0 db rear paths with great delay can be eliminated by using a feedback filter with minimum level number.

Description

Adopt movable tap to eliminate the channel equalization method that disturb in the footpath, back

Technical field

The present invention relates to the communication channel balancing technology, especially relate to a kind of movable tap that adopts and eliminate the channel equalization method that disturb in the footpath, back.

Background technology

In digital communication system, can channel be described with following Mathematical Modeling:

$y\left(n\right)=\underset{k=-N}{\overset{M}{\mathrm{\Σ}}}h\left(k\right)\·s(n-k)+\mathrm{noise}\left(n\right)---\left(1\right)$

Wherein n represents the digital sample point, and h (k) represents channel impulse response, the data that s (n) expression sends, and the data that y (n) expression receives, noise (n) represents additive white Gaussian noise, the number of multipath is N+M+1.

Signal is in transmission course, because the existence of multipath channel, can bring intersymbol interference (Inter-Symbol Interference, ISI), and intersymbol interference is a major reason that produces signal errors, for solving the ISI problem, adopt equalizer to eliminate the influence of ISI usually, can come the ISI that multipath channel is brought is compensated from time domain, frequency domain respectively.

In general, single-carrier system can adopt the method for time domain equalization, and multicarrier system can adopt the method for frequency domain equalization.And in time domain equalization, can adopt the decision feedback equalization structure usually, as shown in Figure 1, the decision feedback filter device is made up of two filters, a feedforward filter, a feedback filter, the at interval equal is-symbol period T of the tap of two filters.The input of feedforward part is to receive burst { x _n, feedback filter be input as judgement sequence to detected symbol.

The output of DFF can be expressed as:

$y_n=\underset{k=-N}{\overset{0}{\mathrm{\Σ}}}{c}_k\·x_{n-k}+\underset{k=1}{\overset{M}{\mathrm{\Σ}}}{c}_k\·\stackrel{\‾}{x_{n-k}}---\left(2\right)$

Wherein, { c _kBe the tap coefficient of filter,

Be the decision value that previous information symbol is estimated, y _kBe the estimated value of k information symbol, this equalizer feedforward part has N+1 tap, and feedback fraction has M tap.

Being most commonly used to balanced linear filter is the FIR filter (FIR, Finite Impulse Response) of a transversary, and namely finite impulse response filter is also referred to as nonrecursive filter.Nonrecursive filter will be used input in the past when calculating each new output, and does not need output in the past.Supposing directly influences before the channel transmission has only, the footpath is the path that arrives prior to main footpath before the definition herein, if the symbol that n leads on the footpath constantly is x (n), preceding footpath is faster than the mark space in main footpath, the symbol that directly is superimposed upon so on the symbol of main footpath is x (n+1), is formulated signal y (n)=x (the n)+0.1x (n+1) that receives.We need use the influence of filter minimizing x (n+1), if our feedforward filter has three grades, as shown in Figure 2.

At a time, the signal of storing in three grades of registers is respectively: x (n+2), and x (n+1), x (n), this moment, the FIR filter was output as:

y(n)＝x(n)+0.1x(n+1)-0.1(x(n+1)+0.01x(n+2)+0.01(x(n+2)+0.1x(n+3))＝x(n)+0.001x(n+3) （3）

As can be seen, output valve has this moment reduced the shared proportion in preceding footpath, has eliminated most influence.

And for feedback filter, the principle of footpath influence just had been more readily understood before it was eliminated, supposing directly influences after the channel transmission has only, footpath, definition back is the path that the back arrives in main footpath herein, if the symbol that n leads on the footpath constantly is x (n), preceding footpath is slower than the mark space in main footpath, and the symbol that preceding footpath is superimposed upon on the main directly symbol is x (n-1), the signal y (n) that receives=x (n)+0.1x (n-1).We need use feedback filter to eliminate the influence of x (n-1), and feedback filter only needs one-level, and this moment, feedback filter was output as:

y(n)＝x(n)+0.1x(n-1)-0.1x(n-1)＝x(n) （4）

Can be with finding out from top example, the coefficient of filter has determined balanced effect.Adjust equalizer coefficients so that average error probability minimum is called the filter coefficient optimization.Use in the communication system of equalizer at great majority, the characteristic of channel is priori the unknown, becomes when channel response is in many cases.In this case, be adjustable with the paired channel response of equalizer design; For time varying channel, need the tap coefficient of adaptive adjustment equalizer.

Have two kinds of algorithms can realize the tap coefficient optimization adaptively, a kind of is zero forcing algorithm, and another kind is lowest mean square (Least mean square, LMS) algorithm.Because the LMS algorithm has less mean square error (MSE, mean square error), and be easy to the hardware realization, used at large.

The LMS algorithm is a kind of application of steepest descent algorithm, and it optimizes criterion is least mean-square error.Generally being expressed as of steepest descent algorithm:

$\stackrel{\→}{w}(n+1)=w\left(n\right)-\frac{1}{2}\mathrm{\μ}\stackrel{\→}{g}\left(n\right)---\left(5\right)$

Being the n value of M tap coefficient of equalizer constantly, is the vector of a M * 1, the step-length (gain) of μ for arranging.

Also be the vector of a M * 1,

For the equalizer tap vector is

The time equalizer output y (n) and Expected Response value

The mean square error of difference.

$E\left[e^2\left(n\right)\right]=J\left(\stackrel{\→}{w}\right)$

$=E[d^2\left(n\right)-2d\left(n\right){\stackrel{\→}{w}}^T\left(n\right)\stackrel{\→}{u}\left(n\right)+{\stackrel{\→}{w}}^T\left(n\right)\stackrel{\→}{u}\left(n\right){\stackrel{\→}{u}}^T\left(n\right)\stackrel{\→}{w}\left(n\right)]$

$=E\left[d^2\left(n\right)\right]-2E\left[d\left(n\right){\stackrel{\→}{w}}^T\left(n\right)\stackrel{\→}{u}\left(n\right)\right]+E\left[{\stackrel{\→}{w}}^T\left(n\right)\stackrel{\→}{u}\left(n\right){\stackrel{\→}{u}}^T\left(n\right)\stackrel{\→}{w}\left(n\right)\right]$

$=E\left[d^2\left(n\right)\right]-2{\stackrel{\→}{w}}^T\left(n\right)E\left[d\left(n\right)\stackrel{\→}{u}\left(n\right)\right]+{\stackrel{\‾}{w}}^T\left(n\right)E\left[\stackrel{\→}{u}\left(n\right){\stackrel{\→}{u}}^T\left(n\right)\right]\stackrel{\→}{w}\left(n\right)$

$=E\left[d^2\left(n\right)\right]-2{\stackrel{\→}{w}}^T\left(n\right)\·P+{\stackrel{\→}{w}}^T\left(n\right)\·R\·\stackrel{\→}{w}\left(n\right)$

Right

About

Differentiate obtains

Can there be being derived as of any loss:

$\stackrel{\→}{g}\left(n\right)=-2\stackrel{\→}{p}\left(n\right)+2R\stackrel{\→}{w}\left(n\right)---\left(6\right)$

Be the cross correlation vector between desired signal and the input signal, i.e. M value of n equalizer register memory storage constantly (is designated as the vector of M * 1

) with the conjugation d of current time desired output ^*(n) cross correlation value of (scalar), that is:

R is

Autocorrelation matrix,

For

Conjugate transpose, the dimension of R matrix is M * M, that is:

$R=E\left[\stackrel{\→}{u}\left(n\right){\stackrel{\→}{u}}^H\left(n\right)\right]---\left(8\right)$

With R be desirable desired value, usually replace with statistical value in the practical application, and statistical average be similar to the average prerequisite of expectation be statistics enough fully, this is difficult to real-time satisfying in specific implementation, so adopt approximation usually, namely use the transient state of R and P to estimate to replace desired value, utilize the long-term training of equalizer and equilibrium to reach the sufficient requirement of statistics, formula is expressed as follows:

$\stackrel{\→}{p}\left(n\right)=\stackrel{\→}{u}\left(n\right)d^{*}\left(n\right)---\left(9\right)$

$R=\stackrel{\→}{u}\left(n\right){\stackrel{\→}{u}}^H\left(n\right)---\left(10\right)$

With above-mentioned approximate expression (9) and (10) formula substitution (6) formula and (5) formula, then obtain the tap renewal equation:

$\stackrel{\→}{w}(n+1)=w\left(n\right)-\frac{1}{2}\mathrm{\μ}\stackrel{\→}{g}\left(n\right)$ $=w\left(n\right)-\mathrm{\μ}[-\stackrel{\→}{u}\left(n\right)d^{*}\left(n\right)+\stackrel{\→}{u}\left(n\right){\stackrel{\→}{u}}^H\left(n\right)\stackrel{\→}{w}\left(n\right)]$

$=w\left(n\right)+\mathrm{\μ}\stackrel{\→}{u}\left(n\right)[d^{*}\left(n\right)-{\stackrel{\→}{u}}^H\left(n\right)\stackrel{\→}{w}\left(n\right)]$

$=w\left(n\right)+\mathrm{\μ}\stackrel{\→}{u}\left(n\right)e^{*}\left(n\right)$

Following formula has provided the computational methods of adaptive updates tap coefficient, i.e. input signal

What (every grade data in register) multiply by step size mu and error signal grips e altogether ^*(n), add previous tap coefficient numerical value.

If according to fixing tap pattern, in the position that does not have strong footpath, because the existence of tap can be introduced noise, can reduce balanced performance.

Summary of the invention

The present invention proposes a kind of movable tap that adopts and eliminates the channel equalization method that disturb in the footpath, back, adopt movable tap to eliminate footpath, strong back, greatly reduce the progression of feedback filter, and adopt the feedback filter of minimum progression can eliminate footpath behind the great 0db of time delay.

The present invention has adopted following technical scheme to realize: a kind of movable tap that adopts is eliminated the channel equalization method that disturb in the footpath, back, and it comprises step:

With every frame symbol data of receiving and the local fixedly frame head relevant treatment of sliding, correlation length is frame head length, and calculating correlation energy value records the maximum of every frame correlation energy;

According to the correlation energy maximum that previous frame obtains, a main footpath relative threshold value is set, determine the position in main footpath;

Correlation energy maximum according to previous frame obtains arranges strong footpath relative threshold value, detects bar number and the position in footpath, back;

Every grade of movable tap is respectively to detect centered by every strong back path position, offers the book character length window and makes up equalizer and carry out channel equalization.

Wherein, described main directly relative threshold value is the correlation energy peaked 1/4 that previous frame obtains.

Wherein, described strong footpath relative threshold value is the correlation energy peaked 1/128 that previous frame obtains.

Wherein, described book character length is 9 character lengths.

Compared with prior art, the present invention adopts movable tap, has reduced the progression of feedback filter, has improved balanced performance.

Description of drawings

Fig. 1 is the equaliser structure schematic diagram of decision-feedback;

Fig. 2 is the schematic flow sheet in footpath before the FIR filter is eliminated;

Fig. 3 is the generating structure schematic diagram of 10 rank m sequences;

Fig. 4 utilizes movable tap to eliminate the schematic flow sheet in footpath, back;

Fig. 5 is the structural representation that the decision feedback filter device of band movable tap is eliminated footpath, many backs;

Fig. 6 is the structural representation that the decision feedback filter device of band movable tap is eliminated footpath behind the 0db long delay.

Embodiment

The present invention includes two parts: the one, adopt and be correlated with, according to the dicision rules of correspondence, determine main footpath, bar number and the position in footpath after determining simultaneously; The 2nd, adopt movable tap to eliminate footpath, strong back, greatly reduce the progression of feedback filter, and adopt the feedback filter of minimum progression can eliminate footpath, the strong back of the great 0db of time delay.

In the present embodiment, the frame structure employing frame head of transmission data (such as, 595 symbol lengths)+frame (such as, 3780 symbol lengths) two parts time-domain signal composition, the baseband signalling rate identical (being 7.56Msps) of frame head and frame signal.The frame head part is by pseudorandom (PN, Pseudorandom Noise) sequence constitutes, generally be the pseudo-random binary sequence (brachymemma of m sequence) that adopts 10 rank maximum lengths, the length of header signal is 595 symbols, is that length is preceding 595 chips of 1023 m sequence.

The pseudo-random binary sequence of configuration frame head part is produced by the shift register group of 10 bits.The generator polynomial of the pseudo-random binary sequence of this maximum length is:

G ₁₀₂₃(x)＝1+x ³+x ¹⁰

The initial phase of the shift register group of this 10 bit is: 0000000001, and when beginning, each signal frame resets.Produce the structure of the pseudo-random binary sequence of this maximum length, as shown in Figure 3, preceding 595 chips of the pseudo random sequence of its generation, reaching " 1 " through " 0 " to+1 value is the binary character of non-return-to-zero to the mapping transformation of-1 value.

In addition, every transmission frame number adopts identical data according to the frame head of frame, therefore at receiving terminal the symbol data that receives and the fixedly frame head data of local generation is carried out relevantly, and the strong multipath of each bar all can produce a relevant peaks.

The present invention adopts movable tap that the footpath, back is handled, and can reduce the progression of equalizer, footpath behind the 0db of the long delay that disappears under the condition that does not increase equalizer progression.

Embodiment 1: supposing that now channel transmission has only the influence of footpath, back, be example with signal y (n)=x (n)-0.1x (n-8)+0.05x (the n-50)-0.08x (n-70) that receives, and utilizes movable tap to eliminate back specific implementation step directly as shown in Figure 4:

Step S11, receiving symbol data are relevant with the slip of local fixedly frame head, and correlation length is frame head length (present embodiment is 595 symbol lengths), calculate the correlation energy value, record the maximum of every frame correlation energy, with as directly the standard by force of frame judgement down;

Step S12, according to the correlation energy maximum that previous frame is preserved, a main footpath relative threshold value (such as correlation energy peaked 1/4) is set, determine that the position in main footpath is 595;

Step S13, according to the correlation energy maximum that previous frame is preserved, footpath, strong back relative threshold value (such as correlation energy peaked 1/128) is set, the bar number that detects footpath, strong back is 3, the position is respectively the position of the 603rd, 645,665 character;

Step S14, centered by detected 3 strong back path positions (position of the 603rd, 645,665 character), open length respectively and be 9 window (ideally, article one, footpath, strong back only need directly be disturbed after 1 grade of filter just can be finished and be eliminated, actual conditions are to have sampling deviation and decision error, therefore open a length in center, footpath, back and be the 9(center forward, each 4 grades backward) window eliminate these interference, length of window is determined according to simulation result), constitute equalizer as shown in Figure 5.

Grey is labeled as equalizer among the figure, dotted line is labeled as virtual equalizer (being equivalent to this place's tap coefficient is zero), every grade of movable tap is all centered by coherent detection goes out the position in footpath, back, the length of movable tap is 9 characters, and the position of sign Main_path, Back_path1, Back_path2, Back_path3 correspondence is respectively the position of the 595th, 603,645,665 character.

Embodiment 2: be example with 0db long delay multipath signal y (n)=x (n)-x (n-2000), it is as follows to utilize movable tap to eliminate behind the long delay of 0db the specific implementation step in footpath:

The receiving symbol data are relevant with the slip of local fixedly frame head, and correlation length is 595, calculate the correlation energy value, record the maximum of every frame correlation energy, with as directly the standard by force of frame judgement down;

According to the correlation energy maximum that previous frame is preserved, a main footpath relative threshold value (such as correlation energy peaked 1/4) is set, determine that the position in main footpath is 595;

Correlation energy maximum according to previous frame is preserved arranges strong footpath relative threshold value (such as correlation energy peaked 1/128), and the bar number that detects footpath, strong back is 1, position 2595;

Centered by the path position 2595 of detected back, open length respectively and be 9 window, constitute equalizer as shown in Figure 6.Grey is labeled as equalizer among the figure, and dotted line is labeled as virtual equalizer (be equivalent to this place's tap coefficient be zero) because this channel is 0db long delay multipath channel, have only one after the footpath, the position of Back_path correspondence is the position of the 2595th character.

To sum up, the present invention is directed to the footpath, back handles, adopt movable tap, can reduce the progression of equalizer, footpath behind the 0db of the long delay that disappears under the condition that does not increase equalizer progression: such as, hypothesis channel transmission now has only the influence of footpath, back, the signal y (n) that receives=x (n)-0.1x (n-8)+0.05x (n-50) uses feedback filter to eliminate the influence in footpath, back, and feedback filter only needs 2 grades in theory, the time delay that disappears respectively is 8,50 back footpath, and this moment, feedback filter was output as:

y(n)＝x(n)+0.1x(n-8)-0.05x(n-50)-0.1x(n-8)+0.05x(n-50)＝x(n)

In the conventional mode, if disturb in the footpath after eliminating two that introduce in this channel transmission, the length of feedback filter need be accomplished 50 grades, and the tap of employing activity, the length of feedback filter only needs 2 grades to get final product in theory, consider near the existence of the secondary lobe in strong footpath, before and after footpath, strong back, respectively getting 4 grades of interference of eliminating secondary lobe.Take all factors into consideration in this case that the progression of feedback filter is 18 grades, well below traditional 50 grades; And according to fixing tap pattern, in the position that does not have strong footpath, because the existence of tap can be introduced noise, can reduce balanced performance.

Claims (4)

1. one kind is adopted movable tap to eliminate the channel equalization method that disturb in the footpath, back, it is characterized in that, comprises step:

With every frame symbol data of receiving and the local fixedly frame head relevant treatment of sliding, correlation length is frame head length, and calculating correlation energy value records the maximum of every frame correlation energy;

According to the correlation energy maximum that previous frame obtains, a main footpath relative threshold value is set, will be greater than the most preceding footpath of this main footpath relative threshold value as main footpath, and the position in the main footpath of record;

According to the correlation energy maximum that previous frame obtains, footpath, strong back relative threshold value is set, will be greater than all footpaths, back of this footpath, strong back relative threshold value all as footpath, strong back, and record all strong after directly positions;

Every grade of movable tap is respectively to detect centered by every strong back path position, offers the book character length window and makes up equalizer and carry out channel equalization;

Wherein: described Preceding footpath is the path that arrives prior to main footpath; Footpath, described back is the path that the back arrives in main footpath.

2. eliminate the channel equalization method that disturb in the footpath, back according to the described employing movable tap of claim 1, it is characterized in that, described main footpath relative threshold value is the correlation energy peaked 1/4 that previous frame obtains.

3. eliminate the channel equalization method that disturb in the footpath, back according to the described employing movable tap of claim 1, it is characterized in that, footpath, described strong back relative threshold value is the correlation energy peaked 1/128 that previous frame obtains.

4. eliminate the channel equalization method that disturb in the footpath, back according to the described employing movable tap of claim 1, it is characterized in that described book character length is 9 character lengths.

Images