CN103647735A - Method for determining equalizer tap length - Google Patents

Abstract

A method of determining a tap length of an equalizer is disclosed in accordance with

And

the length is adjusted according to the proportion of the length to the length, and the length is specifically divided into the following two conditions: if it is not

The equalizer tap length is increased (0 < alpha)_upLess than or equal to 1); if it is not

The equalizer tap length (alpha) is reduced_downNot less than 1); otherwise the equalizer tap length remains constant. The invention carries out self-adaptive adjustment on the tap length of the equalizer according to the parameters such as the error and the like output by the equalizer, and can determine an optimal equalizer length before the equalizer is used, thereby improving the performance of the equalizer.

Description

Determine the method for equalizer tap length

Technical field

The present invention relates to a kind of definite method, particularly relate to a kind of method of definite equalizer tap length.

Background technology

In the design of the High Speed Systems such as current computer and embedded device, use serial transceiver (SERDES), parallel digital signal is converted into serial transmission signal, can improves message transmission rate, influencing each other between modulus signal between isolation D/A and control module.In serial communication system, equalizer plays an important role, and by the correction to intersymbol interference, has guaranteed the integrality transmission of signal.

Traditional equalizer is divided into time-domain equalizer and frequency-domain equalizer.Frequency domain equalization often needs to proofread and correct respectively amplitude-frequency characteristic and group delay characteristic, and to the compensation ability of group delay distortion a little less than, especially conventionally helpless to non-minimum phase decline, so generally do not adopt in digital transmission system.Time domain equalization, considers from time domain impulse response, the signal waveform that the response that utilizes it to produce goes compensation to distort is all used widely in many fields of digital communication.As shown in Figure 1, equalizer length is the key factor that affects performance of filter and computation complexity to the model structure of adaptive equalization link.

Optimal filter theory is the solution that solves Wei Na-Hough (Wiener-Hopf) equation R ω 0=p.Solve to obtain ω 0=R-1p.Here R represents the autocorrelation matrix of input signal, also write r (i), p represents the cross-correlation matrix of input signal and filter output signal, ω (n) is that adaptive equalizer is at n tap coefficient or weight vector constantly, according to the direction of Gradient Descent, regulate tap coefficient, the Wei Na that ω 0 is called as Wei Na-Hough equation (Wiener-Hopf) separates.If use above formula to solve, require second-order statistics: (1) input signal autocorrelation matrix; (2) the cross-correlation matrix priori of input signal and filter desired output signal is known, and relates to matrix inversion, and operand is large, also infeasible in engineering.

A kind of feasible method is according to the feature of error performance curved surface, to carry out the search of mean square error smallest point.This method does not need to know input signal autocorrelation matrix, does not need to carry out matrix inversion yet, is usually used in dimension in adaptive equalization and receives coefficient and solve.Algorithm utilizes gradient information, analyze adaptive-filtering performance and follow the trail of optimum filtering state, make the tap coefficient of equalizer in the process of self adaptation adjustment, the variation of tap coefficient is moved towards error performance curved surface smallest point direction, finally realize best Wiener filtering.

In steepest gradient algorithm, there is more new formula (1) of equalizer tap coefficient:

$\mathrm{\&omega;}(n+1)=\mathrm{\&omega;}\left(n\right)+\frac{1}{2}\mathrm{\&mu;}[-\&dtri;\left(n\right)]\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(1\right)$

▽ (n) represents n N * 1 dimension gradient vector constantly, and N equals the number of equalizer tap coefficient here.μ is the renewal step-length of iteration.

LMS (Least Mean Square) algorithm is set up in nineteen sixty by Widrow the earliest, and it is the algorithm that obtains extensive use in current adaptive equalization field.The criterion that LMS adopts is desired output and the minimized criterion of the mean square error between real output value (MSE) that makes equalizer.First LMS algorithm on the basis of steepest gradient algorithm, obtains the instantaneous estimation of gradient vector, as formula (2):

$\widehat{\&dtri;}=\frac{\&PartialD;\left[e^2\left(n\right)\right]}{\&PartialD;\mathrm{\&omega;}\left(n\right)}=-2e\left(n\right)r\left(n\right)\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(2\right)$

E (n) is the error function of filter output with desired signal, and r (n) representative receives signal.This instantaneous estimation is without inclined to one side, therefore adopt the instantaneous estimation value of gradient vector

replace gradient vector ▽.Then, utilize the relation between adaptive equalizer tap coefficient and the estimated value of gradient vector, can obtain the more new formula (3) of tap coefficient:

$\widehat{\mathrm{\&omega;}}(n=1)=\widehat{\mathrm{\&omega;}}\left(n\right)+\frac{1}{2}\mathrm{\&mu;}[-\widehat{\&dtri;}]=\widehat{\mathrm{\&omega;}}\left(n\right)+\mathrm{\&mu;e}\left(n\right)r\left(n\right)\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(3\right)$

In digital communication, DFF (DFE) is used in conjunction with feed forward equalizer (FFE) conventionally.Feed forward equalizer is linear equalizer, and effect is the leading intersymbol interference in erasure signal; Structure of Decision-feedback Equalization and feed forward equalizer are similar, and just its input is the court verdict of feedback, and effect is the hysteresis intersymbol interference in erasure signal.Feedforward and feedback equalizer use same LMS algorithm to adjust tap coefficient according to e (n).DFF as shown in Figure 2 can at utmost be removed intersymbol interference, in the equilibrium of baseband transmission system, is used widely.

The size of MMSE (Minimum Mean Square Error least mean-square error) also depends on equalizer tap length, if used represent that Wei Na when equalizer length is separates, and uses J _{mmse, ∞}be illustrated in the MMSE obtaining, J _{mmse, ∞}expression formula is as shown in the formula (4):

Least mean-square error under stable state can be expressed as formula (5):

$J_N(\&infin;)=\frac{(J_{\mathrm{mmse},N}+{\mathrm{\&mu;NJ}}_{\mathrm{mmse},N}({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_y^2))}{2}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(5\right)$

Wherein, J _{mmse, N}that represent is the MMSE that equalizer length obtains while being N,

for the power of channel input signal d (n),

for the power of equalizer output signal y (n), what the infinity in bracket represented is equalizer coefficients iterations.From formula (5), can find out J _n(∞) size is subject to equalizer length N and J _{mmse, N}the restriction of two factors.J _{mmse, N}the dull nonincreasing function of N, when N increases, J _{mmse, N}determine and can not increase, may reduce.But when N is enough large, J _{mmse, N}reduce can not make up N and increase the J bring _n(∞) increase, if now continue to increase N, J _n(∞) will increase with the increase of N.If J _n(∞) N value hour is designated as N ₀, at interval [N ₀, ∞) upper, J _n(∞) be the dull nonincreasing function of N, and at interval N≤N ₀upper, J _n(∞) be the monotonic decreasing function of N.

So the selection of equalizer length is determining the performance of whole equalizer.If length is too short, can cause J _n(∞) excessive; If length is long, J _n(∞) increase on the contrary, and can cause the increase of computing complexity, thereby cause practical application very difficult.

When using equalizer, traditional method is rule of thumb, estimates a general tap length, then by test repeatedly, observes the situation of its error, finally determines a tap length.This test method wastes time and energy, and success depends on engineer's experience to a great extent.

Summary of the invention

Technical problem to be solved by this invention is to provide a kind of method of definite equalizer tap length, it is according to the parameters such as error of equalizer output, equalizer tap length is carried out to self adaptation adjusting, can be before equalizer uses, determine an optimal equaliser length, thereby improved the performance of equalizer.

The present invention solves above-mentioned technical problem by following technical proposals: a kind of method of definite equalizer tap length, is characterized in that its basis

with

between ratio adjust length, concrete minute following two kinds of situations:

If

just increase equalizer tap length (0 < α _up≤ 1);

If

just reduce equalizer tap length (α _down>=1);

Otherwise equalizer tap length remains constant, wherein

represent the least mean-square error that N-K rank Wei Na separates, represent the least mean-square error that N rank Wei Na separates, α represents threshold value, α _uprepresent whether to increase the decision threshold of filter order, α _downrepresent whether to reduce the decision threshold of filter order.

Preferably, with

between ratio meet with following formula:

$=1+\frac{\mathrm{\&mu;N}({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)}{2+\mathrm{\&mu;}(N-K)({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)}$

Wherein,

represent respectively the variance of input signal and the variance of the added noise of process passage.

Preferably, described α _downvalue meets with following formula:

${\mathrm{\&alpha;}}_{\mathrm{down}}=1+\frac{\mathrm{\&mu;<figure-callout\; id="2"\; label="K"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">K</figure-callout>}}{2+\mathrm{\&mu;}(N-K)}\&le;1+\frac{\mathrm{\&mu;K}({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)}{2+\mathrm{\&mu;}(N-K)({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)}$

Only work as

time just carry out α _downthe adjustment of value.

Preferably, described α _uplower limit can directly use the accumulation of estimation all just to estimate, meet following two formulas:

${\mathrm{\&alpha;}}_{\mathrm{up},\mathrm{lower}}\left(n\right)=\frac{{\mathrm{AMSE}}_H\left(n\right)}{{\mathrm{AMSE}}_{H-1}\left(n\right)};$

${\mathrm{AMSE}}_i={\mathrm{\&Sigma;}}_{j=1}^L\left[{\mathrm{\&beta;}}^{j-1}{|y_i(n-j+1)-d(n-j+1)|}^2\right]$

$={\mathrm{\&Sigma;}}_{j=1}^L\left[{\mathrm{\&beta;}}^{j-1}{e}_N^{\mathrm{ik}}{(n-j+1)}^2\right]$

Wherein, α _{up, lower}(n) α that representative is estimated constantly at n _up(n) lower limit, β is forgetting factor.

Preferably, described α _up(n) meet following two formulas:

α _up(n)=1-ρ _up,lower(n)

ρ be one much smaller than 1 normal number.

Preferably, the adjustment formula of described equalizer tap length is as follows:

$N(n+1)=N\left(n\right)-K\&times;\mathrm{\&Delta;}\{{\mathrm{\&alpha;}}_{\mathrm{up}}\left(n\right)\&le;\frac{\left[{\mathrm{\&Sigma;}}_{j-1}^L{\mathrm{\&beta;}}^{j-1}{\left(e_{N\left(n\right)}^{N\left(n\right)}(n-j+1)\right)}^2\right]}{\left[{\mathrm{\&Sigma;}}_{j=1}^L{\mathrm{\&beta;}}^{j-1}{\left(e_{N\left(n\right)}^{N\left(n\right)-N}(n-j+1)\right)}^2\right]}$

$\&le;{\mathrm{\&alpha;}}_{\mathrm{down}}\left(n\right)\}\&times;\mathrm{sgn}\{{\mathrm{\&Sigma;}}_{j=1}^L{\mathrm{\&beta;}}^{j-1}[{\left(e_{N\left(n\right)}^{N\left(n\right)}(n-j-1)\right)}^2-{\left(e_{N\left(n\right)}^{N\left(n\right)-K}(n-j-1)\right)}^2]\}$

Wherein, function Δ { } is a discriminant function, and when the conditional inquality in braces is set up, the return value of this function is 0, and when the conditional inquality in braces is false, the return value of this function is 1; Sgn{} is-symbol function, for the direction that determines that equalizer length is adjusted; L is called observation window length, is illustrated in a length adjustment process, and while calculating accumulation mean square error (AMSE), the number of past code element constantly; K is adjacent segment.

Positive progressive effect of the present invention is: the equalizer tap length that the present invention calculates, no matter in any direction (being greater than optimum length starts, or being less than optimum length starts), can, along with the increase of iterations, finally converge on the length of a match channels response.Mean square deviation of the present invention can converge on a smaller value along with the increase of the iterations of this algorithm.The present invention makes the equalizer can be from the motion tracking characteristic of channel, adjust adaptively equalizer exponent number, make in error convergence to less scope of equalizer, filter order is moderate simultaneously, reaches the performance of the dynamic tracking characteristic of channel, reduction equalizer implementation complexity.

Accompanying drawing explanation

Fig. 1 is the structural representation of the model of existing adaptive equalization link.

Fig. 2 is the structural representation of existing adaptive equalizer.

Fig. 3 is the simulation result schematic diagram that equalizer exponent number of the present invention is adjusted curve (initial exponent number is 10).

Fig. 4 is the simulation result schematic diagram that equalizer exponent number of the present invention is adjusted curve (initial exponent number is 60).

Fig. 5 is the simulation result schematic diagram of equalizer error convergence curve of the present invention (initial exponent number is 10).

Fig. 6 is the simulation result schematic diagram of equalizer error convergence curve of the present invention (initial exponent number is 60).

Embodiment

Below in conjunction with accompanying drawing, provide preferred embodiment of the present invention, to describe technical scheme of the present invention in detail.

As shown in Figure 1, the present invention determines the method basis of equalizer tap length

(N-K rank Wei Na separate least mean-square error) with

ratio between (least mean-square error that N rank Wei Na separates) is adjusted length, concrete minute following two kinds of situations:

If

just increase equalizer tap length (0 < α _up≤ 1 o'clock); α _uprepresent whether to increase the decision threshold of filter order.K tap of each increase step-length.

If

just reduce equalizer tap length (α _down>=1 o'clock), α _downrepresent whether to reduce the decision threshold of filter order; Otherwise equalizer tap length remains constant, and α represents threshold value,

subscript represented in equalizer the tap coefficient using in order to estimate the symbol of transmission, yet subscript has represented the tap number that equalizer is total, the upper subscript in formula is below also the identical meaning.

represented stable state MSE (the Mean Square Error obtaining, mean square error), by using first N-K(N of a N tap number to represent equalizer total length, N-K represents front N-K tap of institute's investigation equalizer, and K represents to increase and decrease the step-length of tap number at every turn) individual tap.

In scale model above,

with

between ratio can be expressed as formula (6):

$\frac{J_N(\&infin;)}{J_{N-K}(\&infin;)}=\frac{J_{\mathrm{mmse},N}+\frac{\mathrm{\&mu;N}({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)J_{\mathrm{mmse},N}}{2}}{J_{\mathrm{mmse},N-K}+\frac{\mathrm{\&mu;}(N-K)({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)J_{\mathrm{mmse},\mathrm{<figure-callout\; id="2"\; label="N"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">N</figure-callout>}}}{2}}\&ap;\frac{2+\mathrm{\&mu;N}({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)}{2+\mathrm{\&mu;}(N-K)({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)}$

$=1+\frac{\mathrm{\&mu;N}({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)}{2+\mathrm{\&mu;}(N-K)({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(6\right)$

Wherein,

represent respectively the variance of input signal and the variance of the added noise of process passage.And then, definition (7):

$\mathrm{\&Delta;}{\mathrm{mes}}_k=J_N(\&infin;)-J_{N-K}(\&infin;)$

$=(J_{\mathrm{mmse},N}-J_{\mathrm{mmse},N-K})\&times;(1+\frac{\mathrm{\&mu;N}({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)}{2})$

$+\mathrm{\&mu;K}({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)\frac{J_{\mathrm{mmse},N-\mathrm{<figure-callout\; id="2"\; label="K"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">K</figure-callout>}}}{2}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(7\right)$

Thereby formula of obtaining (8):

$\frac{J_N(\&infin;)}{J_{N-K}(\&infin;)}=\frac{J_{N-K}(\&infin;)+{\mathrm{\&Delta;mse}}_K}{J_{N-K}(\&infin;)}=1+\frac{{\mathrm{\&Delta;mse}}_K}{J_{N-K}(\&infin;)}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(8\right)$

Arranging of threshold parameter is extremely important, very large performance that affects all in any distance to go algorithm.In order to reduce computation complexity, according to formula (6), α _downvalue can be expressed as formula (9):

${\mathrm{\&alpha;}}_{\mathrm{down}}=1+\frac{\mathrm{\&mu;<figure-callout\; id="2"\; label="K"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">K</figure-callout>}}{2+\mathrm{\&mu;}(N-K)}\&le;1+\frac{\mathrm{\&mu;K}({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)}{2+\mathrm{\&mu;}(N-K)({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="HDA0000420066850000012.png,HDA0000420066850000041.png"\; state="\{\{state\}\}">d</figure-callout>}}^2+{\mathrm{\&sigma;}}_v^2)}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(9\right)$

Should note only working as

time just carry out α _downthe adjustment of value.

α _uplower limit can directly use the accumulation mean square deviation (AMSE) of estimation to estimate, namely obtain formula (10) and formula (11):

${\mathrm{\&alpha;}}_{\mathrm{up},\mathrm{lower}}\left(n\right)=\frac{{\mathrm{AMSE}}_H\left(n\right)}{{\mathrm{AMSE}}_{H-1}\left(n\right)}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(10\right)$

${\mathrm{AMSE}}_i={\mathrm{\&Sigma;}}_{j-1}^L\left[{\mathrm{\&beta;}}^{j-1}{|y_i(n-j+1)-d(n-j+1)|}^2\right]$

$={\mathrm{\&Sigma;}}_{j=1}^L\left[{\mathrm{\&beta;}}^{j-1}{e}_N^{\mathrm{ik}}{(n-1+1)}^2\right]\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(11\right)$

α _{up, lown}(n) α that representative is estimated constantly at n _up(n) lower limit, β is forgetting factor.N represents constantly.I represents i the fragment of taking in H section altogether.J represents the index of every step.According to the change of lower limit, adjust α _up(n) ability of value is very important, because when equalizer tap length is less than optimization length, between stable state MSE and length, is non-linear relation.When equalizer tap length is much smaller than its optimal value time, α _upvalue larger, due to a large α _upaccelerating length gathers way in this case.When length approaches its optimal value, need a little α _up, can alleviate like this problem of fluctuation.

Based on formula (10), α _up(n) can be calculated as follows formula (12):

α _up(n)=1-ρα _up,lower(n) …………（12）

ρ be one much smaller than 1 normal number.When lower limit is little, α _up(n) value, close to 1, just becomes and is easy to so increase length for equalizer.When lower limit is large, mean that equalizer tap length approaches its optimization length, so α very much _up(n) value just reduces, and has prevented the problem of fluctuation.Yet, α _up(n) value may compare α _{up, lower}(n) value is little.But this will can not affect the performance of distance to go algorithm, because its main task increases equalizer tap length to its optimal value, completed, that is to say, equalizer tap length has enough approached its optimal value.It should be noted, only work as

time just carry out α _upadjustment.

The adjustment formula of final equalizer tap length can be written as formula (13):

$N(n+1)=N\left(n\right)-K\&times;\mathrm{\&Delta;}\{{\mathrm{\&alpha;}}_{\mathrm{up}}\left(n\right)\&le;\frac{\left[{\mathrm{\&Sigma;}}_{j=1}^L{\mathrm{\&beta;}}^{j-1}{\left(e_{N\left(n\right)}^{N\left(n\right)}(n-j+1)\right)}^2\right]}{\left[{\mathrm{\&Sigma;}}_{j=1}^L{\mathrm{\&beta;}}^{j-1}{\left(e_{N\left(n\right)}^{N\left(n\right)-K}(n-j+1)\right)}^2\right]}$

$\&le;{\mathrm{\&alpha;}}_{\mathrm{down}}\left(n\right)\}\&times;\mathrm{sgn}\{{\mathrm{\&Sigma;}}_{j=1}^L{\mathrm{\&beta;}}^{j-1}[{\left(e_{N\left(n\right)}^{N\left(n\right)}(n-j-1)\right)}^2-{\left(e_{N\left(n\right)}^{N\left(n\right)-K}(n-j-1)\right)}^2]\}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(13\right)$

Wherein, e Representative errors, be mean square error square.Function Δ { } is a discriminant function, and when the conditional inquality in braces is set up, the return value of this function is 0, and when the conditional inquality in braces is false, the return value of this function is 1; Sgn{} is-symbol function, for the direction that determines that equalizer length is adjusted; L is called observation window length, is illustrated in a length adjustment process, and while calculating accumulation mean square error (AMSE), the number of past code element constantly; K is adjacent segment, i.e. the difference of H-1 fragment of equalizer and these two sections of tap number of H section.Equalizer can be self-adaptive variable-order equalizer.

The equalizer tap length that the present invention calculates, no matter in any direction (be greater than optimum length and start, or be less than optimum length start), can be along with the increase of iterations, the length that finally converges on a match channels response, simulation result as shown in Figure 3 and Figure 4.Mean square deviation of the present invention can converge on a smaller value along with the increase of the iterations of this algorithm, and simulation result as shown in Figure 5 and Figure 6.The present invention makes the equalizer can be from the motion tracking characteristic of channel, adjust adaptively equalizer exponent number, make in error convergence to less scope of equalizer, filter order is moderate simultaneously, reaches the performance of the dynamic tracking characteristic of channel, reduction equalizer implementation complexity.

Above-described specific embodiment; the technical problem of solution of the present invention, technical scheme and beneficial effect are further described; institute is understood that; the foregoing is only specific embodiments of the invention; be not limited to the present invention; within the spirit and principles in the present invention all, any modification of making, be equal to replacement, improvement etc., within all should being included in protection scope of the present invention.

Claims (6)

1. a method for definite equalizer tap length, is characterized in that, its basis

with

between ratio adjust length, concrete minute following two kinds of situations:

If

just increase equalizer tap length (0 < α _up≤ 1);

If

just reduce equalizer tap length (α _down>=1);

Otherwise equalizer tap length remains constant, wherein

represent the least mean-square error that N-K rank Wei Na separates,

represent the least mean-square error that N rank Wei Na separates, α represents threshold value, α _uprepresent whether to increase the decision threshold of filter order, α _downrepresent whether to reduce the decision threshold of filter order.

2. the method for definite equalizer tap length as claimed in claim 1, is characterized in that, described in,

with

between ratio meet with following formula:

Wherein,

with

represent respectively the variance of input signal and the variance of the added noise of process passage.

3. the method for definite equalizer tap length as claimed in claim 2, is characterized in that, described α _downvalue meets with following formula:

Only work as

time just carry out α _downthe adjustment of value.

4. the method for definite equalizer tap length as claimed in claim 1, is characterized in that, described α _uplower limit can directly use the accumulation of estimation all just to estimate, meet following two formulas:

Wherein, α _{up, lower}(II) α that representative is estimated constantly at n _up(n) lower limit, β is forgetting factor.

5. the method for definite equalizer tap length as claimed in claim 4, is characterized in that, described α _up(n) meet following two formulas:

α _up(n)=1-ρα _up,lower(n)

ρ be one much smaller than 1 normal number.

6. the method for definite equalizer tap length as claimed in claim 5, is characterized in that, the adjustment formula of described equalizer tap length is as follows:

Wherein, function Δ { } is a discriminant function, and when the conditional inquality in braces is set up, the return value of this function is 0, and when the conditional inquality in braces is false, the return value of this function is 1; Sgn{} is-symbol function, for the direction that determines that equalizer length is adjusted; L is called observation window length, is illustrated in a length adjustment process, and while calculating accumulation mean square error (AMSE), the number of past code element constantly; K is adjacent segment.

Images