JP3379066B2 - Adaptive equalizer and gain vector estimator used for it - Google Patents

Description

Detailed Description of the Invention

[0001]

The present invention is used in digital signal transmission. In particular, the present invention relates to an equalizer that automatically compensates for waveform distortion due to delay that occurs in a propagation path.

[0002]

2. Description of the Related Art In digital signal transmission, intersymbol interference may occur in a demodulated signal due to the influence of a delayed wave generated in a propagation path, and transmission characteristics may be significantly deteriorated. An adaptive equalizer is an effective technique for compensating for this intersymbol interference. In particular,
Even if the phase relationship between the preceding wave and the delayed wave dynamically changes as in digital mobile communication, the adaptive equalizer adaptively follows this phase fluctuation, and thus always achieves high transmission characteristics. There are several known methods for implementing an adaptive equalizer, but as an effective method in a mobile communication environment, maximum likelihood sequence estimation (MLSE: Maximum Likelihood Seq) is performed.
uence Estimation) equalizer and decision feedback equalizer (D
FE: Decision Feedback Equalizer).

FIG. 6 shows a configuration example of a maximum likelihood sequence estimation type equalizer. The equalizer includes an input terminal 61, a subtractor 62, a replica generator 63, a transmission path estimator 64, a squaring circuit 65, a maximum likelihood sequence estimator 66, and an output terminal 67. The received signal input from the input terminal 61 is input to the subtractor 62. The subtractor 62 obtains the error of the received signal with respect to the replica that is the estimated value of the received signal input from the replica generator 63. The squaring circuit 65 obtains the power of this error and inputs it to the maximum likelihood sequence estimator 66. Maximum likelihood sequence estimator 6
6 estimates the transmission sequence with the highest likelihood based on this error power, and uses the estimated sequence as a decoded signal at the output terminal 67.
Output more. On the other hand, the transmission path estimator 64 includes a subtractor 62
The impulse response of the transmission path is obtained based on the error signal obtained by the above and the transmission sequence obtained by the maximum likelihood sequence estimator 66. The replica generator 63 calculates the estimated value of the received signal based on this impulse response.

FIG. 7 shows a configuration example of a decision feedback equalizer.
This equalizer includes an input terminal 71, delay circuits 72 to 75, multipliers 76 to 80, an adder 81, a discriminator 82, and a subtractor 8.
3, a tap coefficient estimator 84 and an output terminal 85. The delay circuits 72 to 74, the multipliers 76 to 79, and the adder 81 perform a convolution operation on the received signal input from the input terminal 71. That is, the delay circuits 72, 73, 74
For sequentially delaying the received signal input from the input terminal 71. The multipliers 76 to 79 are delay circuits 72, 73, 7
Signals having different delay times obtained by 4 are multiplied by tap coefficients. The adder 81 adds the signals multiplied by the tap coefficient. The adder 81 also adds a signal that has passed through the delay circuit 75 and the multiplier 80 from the discriminator 82, that is, a weighted signal to the result of discriminating the output of the adder 81 one time before, and adds the result. Is output as a demodulation signal. The discriminator 82 discriminates the code of the demodulated signal and outputs it to the output terminal 85 as a decoded signal. On the other hand, the subtractor 83 obtains an error between the demodulated signal and the output of the discriminator 82, and the tap coefficient estimator 84 updates the tap coefficient multiplied by the multipliers 76 to 79 and 80 based on this error.

The maximum likelihood sequence estimation type equalizer is a parallel estimator that simultaneously performs channel estimation and maximum likelihood sequence estimation. Even in a multipath Rayleigh fading channel such as mobile communication, static Gaussian noise transmission is used. Even on the road, it achieves the best transmission characteristics. However, when dealing with large delayed waves, the decision feedback equalizer can equalize waveform distortion due to delayed waves by increasing the number of taps, whereas the maximum likelihood sequence estimation equalizer has a large memory length. There is a need to. In the decision feedback equalizer, the amount of calculation increases in proportion to the tap length, whereas in the maximum likelihood sequence estimation equalizer, the circuit scale increases exponentially with the memory length. .

On the other hand, in digital signal transmission, bursts are divided on the time axis, and a transmission channel is assigned to each burst. Time division multiple access (TDMA, Timme Division Mul)
The tiple access method is used. In this method, a preamble for synchronization or a training signal is added to the beginning or center of the burst. From the viewpoint of burst utilization efficiency, it is desirable that the number of synchronization signals is small. That is, the adaptive equalizer applied to burst transmission is required to have high-speed synchronization characteristics.

As a method for realizing a high-speed synchronization characteristic, a transmission line estimator of an equalizer is used for successive least squares (RLS: Recursiv).
A method using the e Least Squares algorithm is known. For this algorithm, see S. Haykin, Adaptive filter theory, Prentice H.
See all, Englewook Cliffs.NJ, 1989.

In the recursive least squares algorithm, first, 1
A Kalman gain vector is obtained from the P matrix calculated before the symbol and the input signal vector. Next, the tap coefficient update amount calculation circuit obtains the tap coefficient update amount by the product of the Kalman gain vector and the error signal. Next, this output signal is added to the tap coefficient one symbol before to obtain an updated tap coefficient. At the same time, the P matrix is updated based on the Kalman gain vector, the signal vector, and the P matrix one symbol before. The recursive least squares algorithm is represented by the following mathematical formula.

[0009]

[Equation 1] In Equation 1, K _k is a Kalman gain vector, P _k is a P matrix, e _k is an error signal, H _k is a tap coefficient, λ is a forgetting coefficient, U _k is an input signal vector, and r _k is a received signal. k indicates the time. Also, H _k ^H is the vector H _k
Hermitian transpose of, * indicates complex conjugate, lowercase p
_{i, j} , k _i , h _i , and u _i are P _k , K _k , H _k , and U _{k, respectively.}
Indicates the i or j element of. In the recursive least squares algorithm, the optimum tap coefficient is obtained from all the signals received from the beginning of the burst, so high-speed synchronization characteristics can be obtained. However, in the successive least squares algorithm, the calculation amount increases in proportion to the square of the tap length, so that there is a problem that the calculation amount becomes too large when a large delay wave is dealt with.

There is a fast Kalman algorithm as a method for obtaining a gain vector in the recursive least squares algorithm, which has exactly the same characteristics as the recursive least squares algorithm and only increases the calculation amount as a linear function of the tap number. This is also detailed in the above mentioned literature. The fast Kalman algorithm is represented by the following mathematical formula.

[0011]

[Equation 2] FIG. 8 shows a configuration example of a gain vector estimator using the fast Kalman algorithm. This gain vector estimator includes input terminals 91-1 and 91-2, a vector expander 9
2, vector reducer 93, output terminals 94-1 and 94-
2, 95 and delay circuits 96, 97. The vector expander 92 includes a two-dimensional expander 101 and a forward estimator 1
02, and the front estimator 102 includes a front tap estimator 1
03 and a gain vector updater 104. The vector reducer 93 includes a two-dimensional reducer 105 and a backward estimator 106, and the backward estimator 106 includes a backward tap estimator 107 and a gain vector updater 108.

The vector expander 91 has an input terminal 91-
A reception signal r _k from 1, the identification signal d _{k-M + 1} at the time when the received signal r _k is input from the input terminal 91-2, obtained by the gain vector estimator before time The signal vector U _k-1 and the gain vector k _k-1 are input. Where M is the feedforward tap length,
This corresponds to the number of stages of the delays 72, 73 and 74 in the example of FIG. The two-dimensional expander 101 receives the received signal r _k and the identification signal d
The dimension of the signal vector is expanded by adding _{k-M + 1} and _{k-M + 1} as new elements to the signal vector U _k-1 . The forward estimator 102 uses the expanded signal vector U _k ⁽¹⁾ and the gain vector k _k−1 to obtain a gain vector k _k ⁽¹ required when the signal vector U _k ⁽¹⁾ is input. ⁾ Is calculated and output. At this time, the front tap estimator 103 in the front estimator 102 uses the signal vector U _k ⁽¹⁾ and the gain vector k _k−1 to perform the calculation of the first three mathematical expressions of the mathematical expression ^(2). To do. Gain vector updater 104
Using the obtained tap coefficient a _k and gain vector k _k−1 , the calculation of the fourth and fifth equations shown in Equation 2 is performed, and the gain vector is two-dimensionally longer than the gain vector k _k−1. Output k _k ⁽¹⁾ .

The gain vector k _k ⁽¹⁾ and the signal vector U _k ⁽¹⁾ are input to the vector reducer 93. The backward estimator 106 uses the gain vector k _k ⁽¹⁾ and the signal vector U _k ⁽¹⁾ to calculate and output a gain vector k _k required when the signal vector U _k is input. At the same time, in the two-dimensional reducer 105, the oldest received signal r _kM in the input signal vector U _k ⁽¹⁾
_And a signal vector U _k excluding the identification signal d _kLM is output. In this example, the received signal r _kM and the identification signal d
_kLM is output to output terminals 94-1 and _94-2 , respectively. Here, L is the feedback tap length, and
This corresponds to the number of stages of the delay circuit 75 in the above example. On the other hand, the backward estimator 106 receives the signal vector U _k ⁽¹⁾ and the gain vector k _k ⁽¹⁾ as input, and the backward tap estimator 10
The backward tap coefficient is updated by calculating the 6th to 8th equations of the equation 2 according to 7, and the gain vector updater 1
08 is the backward tap coefficient c _k and the gain vector k _k
^{Based on (1)} and, the gain vector k _k-1 is output by the ninth equation of the equation ⁽²⁾ . The gain vector k _k-1 and the signal vector U _k are input to the vector expander 92 at the next time.

FIG. 9 shows a configuration example of a tap coefficient estimator using gain vector estimation by the fast Kalman algorithm. This tap coefficient estimator includes an input terminal 111, a gain vector estimator 112, a multiplier 113, and an adder 11.
4, a delay circuit 115 and an output terminal 116. The error signal which is the output of the subtractor 83 shown in FIG. 7 is input to the input terminal 111. Gain vector estimator 112
Is the input signal from the input terminal 71 and the discriminator 82 in FIG.
And an output (not shown) are input, and the estimated gain vector is output. The multiplier 113 calculates the vector correlation between this gain vector and the error signal from the input terminal 111. The adder 114 and the delay circuit 115 constitute an integrator, and the arithmetic output of the multiplier 113 is integrated to output the average value of the vector correlation as a tap coefficient from the output terminal 116.

The fast Kalman algorithm utilizes the characteristic that data is shifted symbol by symbol in the delay line filter with taps and simplifies the operation of the sequential algorithm. Therefore, when several signals are input at one time like a decision feedback equalizer, the vector of Equation 2 becomes a matrix, and the scalar in the fourth equation becomes a matrix, so in the fifth equation Inverse matrix operation is required. Also, in the eighth formula, the reciprocal of the scalar is the inverse matrix operation. In general, a huge amount of calculation is required to realize the inverse matrix calculation by hardware.

[0016]

In the case of equalizing a large delay wave in a mobile transmission line, the maximum likelihood sequence estimation type equalizer having the best equalization ability needs to have a large memory length. There was a problem that a huge circuit scale was required. In addition, when a decision feedback equalizer is used, it is necessary to use a recursive least squares algorithm for tap coefficient estimation for high-speed pull-in. However, this algorithm requires a calculation amount proportional to the square of the tap length. Becomes large, and there is a problem that the amount of calculation also increases when equalizing a large delayed wave. Further, when the fast Kalman algorithm is used in the linear equalizer of the fractional sample or the decision feedback equalizer, there is a problem that the amount of calculation increases because the inverse matrix calculation is included.

The present invention solves such a problem and utilizes a high-speed Kalman algorithm which converges a transversal type equalizer at high speed and does not increase the amount of calculation relatively to the tap length. It is an object of the adaptive equalizer to eliminate the inverse matrix calculation in gain vector estimation and reduce the calculation amount.

[0018]

According to a first aspect of the present invention, a signal obtained by sequentially delaying an input signal and having different delay times is multiplied by a tap coefficient and added. A delay line filter with a tap, an error detection means for obtaining an error between the output of this tapped delay line filter and an identification signal obtained by sign-discriminating the output, and based on the output of this error detection means And a tap coefficient setting means for setting each tap coefficient of the delay line filter with taps, wherein the tap coefficient setting means includes the input signal up to that point of the delay line filter with taps and the new input signal and the error detection means up to that point. Gain vector estimating means for generating a gain vector representing a convergence direction for each tap coefficient of the tapped delay filter from the identification signal and the new identification signal, and And a tap coefficient estimating means for obtaining a new tap coefficient by multiplying the output of the error detecting means by the gain vector, wherein the gain vector estimating means is a signal having the input signal up to that point and the identification signal up to that point as elements. A forward estimation means for estimating the latest input signal from the expanded signal vector obtained by adding a new input signal and a new identification signal to the vector as elements and the gain vector in the previous stage, and updating the gain vector, and expansion The reduced signal vector obtained by removing a part of the past input signal and identification signal from the expanded signal vector is output as the signal vector in the next stage, and the expanded signal vector and the gain vector updated by the forward estimation means are output. A backward estimator that estimates the oldest input signal from and updates its gain vector In the adaptive equalizer including and, at least one of the forward estimation means and the backward estimation means includes means for repeating the update of the gain vector while enlarging or reducing the signal vector for each of the input signal and the identification signal. An adaptive equalizer is provided that is characterized by the above.

A plurality of input signals may be input at one time, and the repeating means may successively update the gain vector for each of the plurality of input signals while enlarging or reducing the signal vector one dimension at a time. For example, when two input signals are simultaneously input by diversity reception, the forward estimation means and the backward estimation means respectively repeat means corresponding to three signals of the two input signals and the corresponding common identification signal. Is preferably configured in three stages.

According to a second aspect of the present invention, there is provided a gain vector estimator used in such an adaptive equalizer. That is, in order to obtain the gain vector indicating the convergence direction of the filter coefficient by the high-speed Kalman algorithm, the signal that has been input so far is added to the signal vector that has been input to generate an enlarged signal vector. Vector expansion means, forward estimation means for estimating the latest input signal from this expanded signal vector and the gain vector estimated in the previous stage, and updating the gain vector, and one of the past ones from the expanded signal vector. The oldest input signal is estimated from the vector reduction means for outputting the reduced signal vector excluding the signal of the section as a signal vector in the next stage, the enlarged signal vector and the gain vector updated by the forward estimation means. A gain vector estimator having a backward estimation means for updating the gain vector. Then, a plurality of signals are input at one time, and the vector expansion means includes a plurality of one-dimensional expansion means for expanding a signal vector one dimension at a time for each of the plurality of signals, and the forward estimation means has the plurality of one-dimensional expansion means. The vector reduction means is provided in multiple stages so as to sequentially update the gain vector corresponding to each of the expansion means, and the vector reduction means includes a plurality of one-dimensional reduction means for reducing the signal vector one dimension at a time, and the backward estimation means is a plurality of one-dimensional reduction means. There is provided a gain vector estimator provided in multiple stages so as to sequentially update the gain vector corresponding to the signal vector reduced by the dimension reducing means.

In the case of gain vector estimation using the high-speed Kalman algorithm, the inverse matrix calculation is not included in the estimation calculation and may be the reciprocal calculation if one symbol is input at a time. Therefore, in the case of a fractional sample or a decision feedback equalizer in which a plurality of symbols are input at one time, the high-speed Kalman algorithm should be configured in multiple stages so that only one symbol is input in each stage at a time. Thus, the inverse matrix in the gain vector estimation calculation using the Kalman algorithm can be eliminated. When the gain vector estimator has a multi-stage configuration, the variables in each of the front / rear estimators increase in proportion to the number of input symbols. This also applies to the case where a plurality of symbols are input at one time in the conventional technique. Therefore, according to the present invention, the calculation amount does not increase in the parts other than the inverse matrix calculation.

According to the present invention, when a plurality of input signals are input at one time, the gain vector estimators utilizing the fast Kalman algorithm are connected in multiple stages to make the input signals of the gain vector estimators of each stage one. By doing,
It is possible to eliminate the inverse matrix calculation and reduce the calculation amount.

[0023]

1 is a block diagram showing an embodiment of the present invention, showing a configuration example of a gain vector estimator.

This gain vector estimator has an input terminal 1
A plurality of signals are input at one time from 1-1 to 11-N, and an enlarged signal vector is generated by adding a newly input signal to a signal vector having the signals input up to that time as elements 1 A vector expander 12-1 including a dimension expander 21 and a forward estimator 22 that estimates the latest input signal from the expanded signal vector and the gain vector estimated in the previous stage to update the gain vector. ~ 12-N
And a vector reducer 25 that outputs a reduced signal vector obtained by removing some past signals from the enlarged signal vector as a signal vector in the next stage, and an updated signal vector and forward estimator 22. And a vector estimator 13-1 to 13-N configured by a backward estimator 26 that estimates the oldest input signal from the obtained gain vector and updates the gain vector. The front estimator 22 includes a front tap estimator 23 and a gain vector updater 24, and the rear estimator 26 includes a rear tap estimator 27 and a gain vector updater 28.

The front tap estimator 23 obtains the inner product of the signal vector and the front tap vector estimated up to one time before, takes the difference (first subtraction value) between the inner product and the input signal, and obtains the difference. The correlation with the gain vector is calculated, the tap vector is added to the correlation output, the inner product of the obtained first addition vector and the signal vector is calculated, and the difference between the inner product and the input signal is calculated. The product of one subtraction value is taken, and the result is cumulatively added.

The gain vector updater 24 divides the first subtraction value by the cumulatively added value, calculates the correlation between the obtained signal and the first addition vector, and adds the gain vector to the correlation. , 1 in the obtained second addition vector
The dimension is expanded and the signal obtained by the above division is output as a new element.

The backward tap estimator 27 obtains the inner product of the signal vector and the backward tap vector estimated one time before, and takes the difference (second subtraction value) between the inner product and the oldest signal of the signal vector. , The difference is correlated with the second addition vector, the backward tap vector is dimensionally expanded, the vector correlation output is added to the vector to which 1 is added as a new element, and the backward tap vector of the vector correlation output is expanded. The division operation of each element of the vector is performed so that the calculated element is normalized to 1.

The gain vector updater 28 correlates the normalized output with the element corresponding to the expansion dimension of the tap vector among the gain vectors expanded one-dimensionally in the gain vector updater 24 and the one-dimensionally expanded vector. Subtract the vector correlation output from.

The signals from the input terminals 11-1 to 11-N are input to the vector expanders 12-1 to 12-N and supplied to the one-dimensional expander 21. These one-dimensional expanders 21 expand the signal vector by one dimension for each of the plurality of input signals. Forward estimator 2
2 sequentially updates the gain vector corresponding to the one-dimensional expander 21. The one-dimensional reducer 25 of each of the vector reducers 13-1 to 13-N reduces the signal vector one dimension at a time. The signals subjected to the reduction are output to the output terminals 14-1 to 14-N. Backward estimator 26
Updates the gain vector in sequence corresponding to the signal vector reduced by the one-dimensional reducer 25. The signal vector obtained by the one-dimensional reducer 25 of the vector reducer 13-N is input to the vector expander 12-1 as a signal vector at the next time point via the delay circuit 16. Further, the gain vector obtained by the backward estimator 26 of the vector reducer 13-N is output to the output terminal 15 and also passes through the delay circuit 17 as a gain vector at the next time point to be the vector expander 12-. Input to 1. The operation of this gain vector estimator will be described in detail below.

N signals u _k ^{(1) to} u _k ^(N) are input to the gain vector estimator at a time k at one time. The vector expander 12-1 inputs the input signal u _k ⁽¹⁾ input from the input terminal 11-1 and the gain vector k _k-1 and the signal vector U _k-1 estimated up to one time earlier. Then, the input signal u _k ⁽¹⁾ and the signal vector U _k-1 are input to the one-dimensional expander 21, and the output signal vector U _k-1
⁽¹⁾ and the gain vector k _k-1 are input to the forward estimator 22, and the signal vector U _k-1 ⁽¹⁾ and the gain vector k _k-1 ⁽¹⁾ output by the forward estimator 22 are input to the next stage. To the vector expander 12-2. 1D expander 21 enlarges the dimension of the signal vector U _k-1, and outputs a signal vector U _k-1 and the input signal u _k ⁽¹⁾ a new element ^(1). The front tap estimator 23 in the front estimator 22 uses the signal vector U
_The front tap coefficient a is calculated by performing the calculation of the first to third equations of Equation 2 from _k-1 ⁽¹⁾ and the gain vector k _k-1.
Update _k-1 . The gain vector updater 24 in the forward estimator 22 performs the operations of the fourth and fifth equations of Equation 2 from the forward tap coefficient a _k-1 and the gain vector k _k-1 , to obtain k _k Gain vector k which is one dimension longer than _-1
Calculates _k-1 ⁽¹⁾ .

The vector expander 12-2 outputs a signal vector U _k-1 which is an output signal from the vector expander 12-1.
⁽¹⁾ and gain vector k _k-1 ⁽¹⁾ and input signal u _k
^{From (2)} , by the first to fifth operations in Equation 2,
The signal vector U _k-1 ⁽²⁾ and the gain vector k _k-1 ⁽²⁾ are output.

By repeating the same processing, the signal expander 12-N finally outputs the signal vector U _k-1.
The gain vector k _k-1 from the N-dimensional significant signal vector U _k-1 ^(N) and the gain vector k _k-1 ^(N) are output.

The vector reducer 13-1 receives the signal vector U _k ^(N) and the gain vector k _k ^(N) , which are outputs from the vector expander 12-N, and inputs these two vectors to the backward estimator. 26, and at the same time signal vector U
_k ^(N) is input to the one-dimensional reducer 25, and the one-dimensional reducer 25
Of the vector vector U _k ^(N-1) and the gain vector k _k ^{(N-1) of the} backward estimator 26 are output from the vector reducer 13-1. The one-dimensional reducer 25 outputs a signal vector U _k ^(N-1) obtained by removing the oldest signal input in the past from the input signal vector. The backward tap estimator 27 in the backward estimator 26 detects the signal vector U _k.
^{By using (N)} and the gain vector k _k ^(N) , the fifth to eighth operations of the equation 2 are performed to output the backward tap coefficient c _k ^(N) . The gain vector updater 28 uses the gain vector k _k ^(N) and the backward tap coefficient c _k ^(N) to perform the ninth operation of Equation 2 to obtain a one-dimensionally smaller gain vector k _k ^(N-1). Is output.

The vector reducer 13-2 receives the signal vector U _k ^(N-1) and the gain vector k _k ^(N-1) and performs the same calculation as the vector reducer 13-1 to obtain the signal. A signal vector U _k ^(N-2) and a gain vector k _k ^(N-2) that are one-dimensionally smaller than the vector U _k ^(N-1) and the gain vector k _k ^(N-1) are output. By repeating this operation, the signal vector U _k-1 and the gain vector k
A signal vector U _k and a gain vector k _k having the same dimensions as _k−1 are obtained and are used as the input of the vector expander 12-1 at the next time. At the same time, the gain vector k _k
Is the output of this gain vector estimator.

This gain vector estimator is shown in FIG.
By using it as the gain vector estimator 92 of the tap estimator shown in FIG. 1, the gain vector obtained by this gain vector estimator is used to estimate the tap coefficient, and the tap coefficient is used to equalize the waveform distortion. You can

[0036]

FIG. 2 is a block diagram showing an embodiment in which the gain vector estimator of the present invention is used in a decision feedback equalizer for symbol period samples.

This gain vector estimator has an input terminal 1
1-1, 11-2, vector expanders 12-1, 12-
2, vector reducers 13-1, 13-2, output terminal 14
-1, 14-2, 15 and delay circuits 16, 17 are provided. The vector expanders 12-1 and 12-2 each include a one-dimensional expander 21 and a forward estimator 22, and the forward estimator 22 includes a front tap estimator 23 and a gain vector updater 24. In addition, the vector reducers 13-1 and 13
-2 is a one-dimensional reducer 25 and a backward estimator 26, respectively.
And the backward estimator 26 includes a backward tap estimator 27 and a gain vector updater 28.

This gain vector estimator has an input terminal 1
Received signal 1-1, the identification signal of the decision feedback equalization <br/> unit to the input terminal 11-2 is input to these signals, a larger signal vector by one-dimensional, the gain vector while reducing Repeat the update. Then, for the finally obtained gain vector, as shown in FIG. 9, the vector correlation with the error signal is calculated and integrated, thereby obtaining the tap coefficient of the tapped delay line filter shown in FIG. .

FIG. 3 shows a structural example of a combining type diversity equalizer, and FIG. 4 shows a structural example of a gain vector estimator used in this equalizer.

The synthetic diversity equalizer shown in FIG.
Input terminals 31 and 32 to which received signals from each diversity branch are input, delay circuits 33 to 35 that sequentially delay the signal from the input terminal 31, delay circuits 36 to 38 that sequentially delay the signal from the input terminal 32, and The delay circuit 39 for delaying the identification signal, the multipliers 40-43 for multiplying the signals from the input terminal 31 and the outputs of the delay circuits 33-35 by tap coefficients, and the signals from the input terminal 32 and the delay circuits 36-38, respectively. The output of the multiplier 47 and the output of the delay circuit 39 are multiplied by the tap coefficient, the multiplier 48 for multiplying the output of the delay circuit 39 by the tap coefficient, the adder 49 for adding the outputs of the multipliers 40 to 48, and the output of the adder 49 are identified. The discriminator 50 that determines the difference between the output of the adder 49 and the output of the discriminator 50, and the error output from the subtractor 51 Tap coefficient estimator 52 for estimating the
And an output terminal 53 for outputting the signal identified by the identifier 50.

Received signals are input to the input terminals 31 and 32, respectively. The inputs from the input terminal 31 are delay circuits 33 to 35.
And a delay line with a tap composed of multipliers 40 to 43 and an adder 4
9 and the input from the input terminal 32 is delayed by the delay circuit 36.
.About.38 and the delay lines with taps composed of the multipliers 44 to 47, and the adder 49 respectively perform convolution calculations. 1 at the same time
A signal obtained by identifying the output signal of the adder 49 before the symbol is weighted by the multiplier 48 and input to the adder 49.

This configuration is a combination of an equalizer and diversity, and can realize maximum ratio combining in addition to the equalizing effect.

Since the combining-type diversity equalizer includes two branches, two reception signals are input at one time. In this case, it is necessary to input the two received signals and the discrimination signal obtained by the discriminator 50 to the vector estimator in the tap coefficient estimator 52. Therefore, the vector expanders 12-1 to 12-3 and the vector reducers 13-1 to 13-3 having a three-stage configuration are used. That is, the signal of the first branch is input to the vector expander 12-1, the signal of the second branch is input to the vector expander 12-2, and the identification signal is input to the vector expander 12-3. The vector reducer 13-1 outputs the oldest discrimination signal,
The vector reducer 13-2 outputs the oldest received signal of the second branch, and the vector reducer 12-3 outputs the oldest received signal of the first branch. This makes it possible to estimate the tap coefficient of the diversity decision feedback equalizer.

FIG. 5 illustrates the present invention with two feedback taps.
(L = 2), 4 feedforward taps (M = 4)
The results of the laboratory experiments when used in the decision feedback equalizer of are shown below. The modulation method is QPSK, the transmission path has two independent Rayleigh fadings, and the normalized maximum Doppler frequency is 0.
With 5 × 10 ⁻⁴ and 1.0 × 10 ⁻⁴ , the normalized amount of the delayed wave is 0.65. As the frame format, one having a unique word of 12 symbols at the beginning of one burst and data of 108 symbols after that was used. As a theoretical value, the characteristic of the adaptive equalizer in the presence of 1-symbol delay is shown. As you can see from this figure, 2d from the theoretical value
It can be confirmed that the deterioration is about B, and the high-speed synchronization characteristics similar to the successive least squares algorithm are obtained with a unique word of only 12 symbols.

[0045]

As described above, according to the present invention,
When multiple input signals are input at one time, the gain vector estimator using the fast Kalman algorithm is connected in multiple stages to make one input signal for each stage of the gain vector estimator and perform the inverse matrix operation. There is an effect that it can be eliminated and the calculation amount can be reduced. Further, in the successive least squares algorithm, the amount of calculation increases in proportion to the square of the tap length, whereas in the present invention, the amount of calculation increases only as a linear function of the tap length, so that there is a significant reduction in the amount of calculation. .

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a block diagram showing an embodiment in which the gain vector estimator of the present invention is used as a decision feedback equalizer for symbol period samples.

FIG. 3 is a block configuration diagram showing a configuration example of a combining-type diversity equalizer.

FIG. 4 is a block configuration diagram of a gain vector estimator used in this equalizer.

FIG. 5 is a diagram showing the results of an indoor experiment when the present invention is applied to a decision feedback equalizer having two feedback taps and four feedforward taps.

FIG. 6 is a block diagram showing a maximum likelihood sequence estimation type equalizer.

FIG. 7 is a block configuration diagram showing a configuration example of a decision feedback equalizer.

FIG. 8 is a diagram showing a configuration example of a conventional gain vector estimator using a fast Kalman algorithm.

FIG. 9 is a diagram showing a configuration example of a tap coefficient estimator using gain vector estimation by a fast Kalman algorithm.

[Explanation of symbols]

11-1 to 11-N, 31, 32, 61, 71, 91-
1, 91-2, 111 Input terminals 12-1 to 12-N, 92 Vector expanders 13-1 to 13-N, 93 Vector reducers 14-1 to 14-N, 15, 53, 67, 85, 94 −
1, 94-2, 95, 116 Output terminals 16, 17, 33-39, 72-75, 96, 97, 1
15 delay circuit 21 1-dimensional expander 22, 102 forward estimator 23, 103 forward tap estimator 24, 28, 104, 108 gain vector updater 25 1-dimensional reducer 26, 106 backward estimator 27, 107 backward tap estimator 40-48, 76-80, 113 Multiplier 49, 81, 114 Adder 50, 82 Discriminator 51, 62, 83 Subtractor 52, 84 Tap coefficient estimator 63 Replica generator 64 Transmission line estimator 65 Square circuit 66 Maximum likelihood sequence estimator 101 Two-dimensional expander 105 Two-dimensional reducer 112 Gain vector estimator

─────────────────────────────────────────────────── ─── Continuation of the front page (56) Reference JP-A-7-235896 (JP, A) JP-A-7-38470 (JP, A) JP-A-5-48503 (JP, A) JP-A-2- 252321 (JP, A) JP 9-8715 (JP, A) JP 7-226704 (JP, A) Satoshi Tano, Hiroshi Shirato, Yoichi Saito, “DFE applying high-speed Kalman algorithm for dimension expansion” Structure and Characteristics ”, 1996 IEICE Communications Society Conference Proceedings, August 30, 1996, p. 422, (B-421) (58) Fields investigated (Int.Cl. ⁷ , DB name) H04B 1/76 H04B 3/00 H04B 7/00 H03H 15/00-21/00 INSPEC (DIALOG) JISST file ( JOIS)

Claims (4)

(57) [Claims] 1. A tapped delay line filter for outputting a signal obtained by multiplying signals having different delay times obtained by sequentially delaying an input signal with respect to an input signal and adding the signals, and the delay with taps. Error detecting means for obtaining an error between the output of the line filter and an identification signal obtained by sign-identifying the output, and a tap for setting each tap coefficient of the delay line filter with taps based on the output of the error detecting means. Coefficient setting means, the tap coefficient setting means, from the previous input signal and a new input signal of the delay line filter with taps and the identification signal and a new identification signal of the error detection means, Gain vector estimating means for generating a gain vector representing a convergence direction for each tap coefficient of the tapped delay filter, and the gain vector And a tap coefficient estimating means for obtaining a new tap coefficient by multiplying the output of the error detecting means, wherein the gain vector estimating means is a signal having the input signal up to that point and the identification signal up to that point as elements. Forward estimation means for estimating the latest input signal from the enlarged signal vector obtained by adding the new input signal and the new identification signal to the vector as elements and the gain vector in the previous stage, and updating the gain vector. , Outputting a reduced signal vector obtained by removing some past input signals and identification signals from the enlarged signal vector as a signal vector in the next step, and by the enlarged signal vector and the forward estimation means. Estimate the oldest input signal from the updated gain vector and In the adaptive equalizer comprising a backward estimation means new to at least one of the forward estimation means and the backward estimation means for each of the input signals and identification signals,
An adaptive equalizer comprising means for repeating the update of a gain vector while enlarging or reducing a signal vector one dimension at a time. 2. A plurality of input signals are input at one time, and the repeating means includes means for sequentially updating the gain vector while enlarging or reducing the signal vector by one dimension for each of the plurality of input signals. The adaptive equalizer according to Item 1. 3. Diversity-received two input signals are simultaneously input, and the front estimation unit and the rear estimation unit respectively correspond to three signals of the two input signals and a corresponding common identification signal. 3. The adaptive equalizer according to claim 1, wherein the repeating means is configured in three stages. 4. An expanded signal vector obtained by adding a newly input signal to a signal vector having the signal input up to that time as a gain vector representing the convergence direction of the filter coefficient is obtained by a high speed Kalman algorithm. Vector expansion means for generating, and a forward estimation means for estimating the latest input signal from the expanded signal vector and the gain vector estimated in the previous stage to update the gain vector, and the expanded signal vector From the vector reducing means for outputting a reduced signal vector excluding a part of the past signals as a signal vector in the next stage, and the enlarged signal vector and the gain vector updated by the forward estimation means. Gain with backward estimation means for estimating the old input signal and updating its gain vector In the Coutl estimator, a plurality of signals are input at one time, and the vector expansion means includes a plurality of one-dimensional expansion means for expanding a signal vector one dimension at a time for each of the plurality of signals, and the forward estimation means is The vector reduction means is provided in multiple stages so as to sequentially update the gain vector corresponding to each of the plurality of one-dimensional enlargement means, and the vector reduction means includes a plurality of one-dimensional reduction means for reducing the signal vector one dimension at a time. A gain vector estimator, wherein the estimating means is provided in multiple stages so as to sequentially update the gain vectors corresponding to the signal vectors reduced by the plurality of one-dimensional reducing means.