JPH03132104A - Equalizer - Google Patents

Abstract

PURPOSE:To obtain an equalizer suitable for compensation of the characteristic of a transmission line whose delay distortion is timewise varied by selecting two adaptive equalization algorithms as error vector type and correlation matrix modification type. CONSTITUTION:Storage means 10, 11 storing respectively an input signal and a decision output are provided, and a means 7 switching two adaptive equaliza tion algorithms as error vector type and correlation matrix modification type and executing the algorithm and means 6-8 applying the training of correlation matrix modification type with a signal stored in the storage means 10, 11 when the algorithm is selected from the error vector type into correlation matrix modification type are provided. When the adaptive equalization algorithms is selected from the error vector type and correlation matrix modification type, a reproduction signal obtained by the equalization of the error vector type is used a reference signal and the reference signal is used to apply training of the correlation matrix modification type. Thus, since the algorithm of correla tion matrix modification type is applied when the characteristic change in the transmission line is fast, the equalization error is decreased and the deteriora tion in the transmission characteristic is suppressed small.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an equalizer that compensates for characteristics of a transmission line in which delay distortion of the transmission line changes over time. The present invention is particularly suitable for use in mobile radios such as portable radios.

The present invention uses two adaptive algorithms, a correlation matrix correction type and an error vector type, in an equalizer that compensates for the characteristics of the transmission path, thereby reducing power consumption when the characteristics of the transmission path vary slowly. Using an algorithm,
When the characteristic fluctuations are severe, an algorithm with small error is used.

[Conventional technology]

When a signal is subjected to linear distortion, an equalizer is conventionally used to extract the signal. Since the state of distortion changes over time, it is necessary to adaptively change the state of the equalizer.

FIG. 3 shows a circuit diagram of an example of a conventional equalizer.

A signal containing distortion is input to the input terminal 1. This signal is input to the equalization processing circuit 2.

The equalization processing circuit 2 is composed of a transversal filter, and includes delay circuits 21 connected in eight stages, N+1 multipliers 22 provided at each tap, and an adder 23 that adds the outputs of the multipliers 22. The delay circuit 21 delays each input signal by T. When the input signal at time n is x (n), each delayed output of the delay circuit 21 is x
(n-1)...x (n-N). input signal x
(n) and delayed output X (n-1) ・-x (n
-N) are each tapped coefficient co by the multiplier 22.
(n)...CM(n) is multiplied and added by the adder 23. As a result, an equalized signal Y (n) is obtained. The equalized signal Y(n) is expressed as Y(n)=Co(n) x(
t)+C+ x(n-1) +...+ C* (n
) x (n-N).

The equalized signal Y (n) is converted by the determination circuit 3 into a digital determination signal Z (n). If there is no judgment error, the judgment signal Z (n) is regarded as an ideal received waveform without deterioration due to distortion or noise.

The determination signal Z (n) is output from the output terminal 4 and is also supplied to the error calculation circuit 6 via the switch circuit 5 . The error calculation circuit 6 calculates the error e (n) of the equalized signal Y (n) using e (n)=Z (n) - Y(n). Using the error e (n) obtained in this way, the input signal x (n) and the delayed outputs X (n-1)...x (n-N), the adaptive processing circuit 7 uses the tap coefficient C3 ( n)...C
, (n).

In such a process, the tap coefficients are not set to correct values at the initial stage when the equalizer starts operating. Therefore, even if the output of the adder 23 is input to the determination circuit 3, the determination signal Z (n) will not be a correct signal. Therefore, in order to set the initial state, a training signal agreed upon between the transmitting side and the receiving side is transmitted from the transmitting side, and on the receiving side, the switch circuit 5 is switched, and the signal from the training signal generation circuit 8 is sent to the error calculation circuit. 6 to use human power.

Thereby, the error calculation circuit 6 compares the equalized signal Y (n) and the training signal to obtain an accurate error.

Through such a training process, the internal state of the equalizer can be set quickly.

As adaptive equalization algorithms for such equalizers, error vector type and correlation matrix correction type are well known.

In the error vector type adaptive equalization algorithm, the error e
(n) by the input signal of each multiplier 22, and the product is added to the tap coefficient value at that time to update the coefficient. In this algorithm, the error e (n) at that point, the input signal x (n) and the delayed output X
(n-1)...x (n-N) is used to calculate the tap coefficient update amount. No other signals are required for this calculation.

Typically, the input signal x (n) and the delayed output X (n-
1) . . . x (n-N) is represented as an input vector, the tap coefficient value is represented as a coefficient vector, and vector calculation is performed using these. At this time, in the error vector type algorithm, the evaluation function J obtained by squaring the error is partially differentiated by a coefficient vector, and the state of the equalizer is shifted in the direction of the maximum gradient of the evaluation function J obtained thereby.

Typical error vector algorithms include:
Least Mean Squares
s.

The rLSM method (hereinafter referred to as "rLSM method") is known. In practice, modified methods such as quantizing errors are used to simplify calculations.

On the other hand, in the correlation matrix modified algorithm, the input signal X (n) and the delayed output X (n-1>...x (n
-N) is used to calculate the autocorrelation matrix or its inverse matrix of the input signal, and update the tap coefficients while sequentially updating this matrix.

This method takes into account all signals from the initial setting up to the present time, that is, signals before the signal output from the delay circuit 21, and sequentially performs the least squares method so that the error e (n) is minimized. apply to Therefore, unlike the error vector type, the error e (n) at that point, the input signal x (n) and the delayed output x (n-1)...x
(n-N) alone cannot calculate the coefficient update amount. Furthermore, it requires an autocorrelation matrix or its inverse. A training process is required to obtain this matrix by averaging the signal sequence.

Typical correlation matrix modified algorithms include:
Recursive Least 5
qu-ares (hereinafter referred to as "rRLS method") is known.

In addition, the exponentially weighted RLS method that introduces an appropriate time constant, the high-speed RLS method that reduces the amount of calculation, the lattice method, and the like are known. For more information, see "Optimum Signal Processing" by Orfanidis, McGraw-Hill, 2nd edition (S, J, Orfani-dis: O
ptimum Signal Processing,
2nd editionMacGraw-Hill)
is explained in.

Below, in order to proceed with the explanation concretely, we will use the LMS method for the error vector type and the RLS method for the correlation matrix modified type.
This will be explained using the law as an example.

FIG. 4 shows the characteristics of the LMS method and the RLS method, and shows an outline of how the uniformity error changes from the initial reset state.

Comparing the LMS method and the RLS method, the LMS method is a simple method with a small amount of calculations, but the performance of equalization and adaptation is inferior to the RLS method. Referring to the example shown in FIG. 4, it takes t for the error to converge in the RLS method, while it takes t in the LMS method. However, the value of t is t.

This is several tens of times higher. In addition, the steady-state error ε of the LMS method is also RL
It is larger than the steady-state error ε8 of the S method.

[Problem to be solved by the invention]

In this way, the LMS method and the RLS method each have their advantages and disadvantages, so when the strain state is changing irregularly or rapidly changing, the RLS method is used, and when the fluctuation is gradual, the I
It is preferable to use the JIs method. In particular, when low power consumption of equipment is required, the LMS method with a simple algorithm is desirable, but when fluctuations are severe, the RLS method must be used. Therefore, it is desirable to perform equalization processing by switching between the LMS method and the RLS method. However, a method of switching the adaptive algorithm in the middle of equalization processing has not been known in the past.

An object of the present invention is to provide an equalizer that can switch between two adaptive equalization algorithms: an error vector type and a correlation matrix correction type.

[Means to solve the problem]

The equalizer of the present invention includes storage means for respectively storing input signals and determination outputs, and means for switching and executing two adaptive equalization algorithms, a correlation matrix correction type and an error vector type, as an adaptive processing means; The present invention is characterized by comprising means for training the correlation matrix modification type using a signal stored in the storage means when switching from the error vector type to the correlation matrix modification type.

[For production]

When switching the adaptive equalization algorithm from the error vector type to the correlation matrix correction type, the reproduced signal obtained when equalizing with the error vector type is used as a reference signal, and this reference signal is used to train the correlation matrix correction type algorithm. conduct.

This makes it possible to switch between different adaptive equalization algorithms instead of using a single algorithm as in the past.

〔Example〕

FIG. 1 is a block diagram of a fixture according to an embodiment of the present invention.

This equipment includes an equalization processing circuit 2 that equalizes the transmission characteristics of an input signal, a determination circuit 3 that determines the sign of the input signal from the equalized signal output from the equalization processing circuit 2, and this determination circuit 30. An error calculation circuit 6 and an adaptive processing circuit 7 are provided as adaptive processing means for adaptively setting the characteristics of the equalization processing circuit 2 so that the difference between the judgment output and the equalization signal becomes small.

Here, the feature of this embodiment is that it is equipped with a buffer memory 10 and a memory circuit 11 that store input signals and judgment outputs, respectively, and that the adaptive processing circuit 7 has two types of adaptive processing: a correlation matrix correction type and an error vector type. The present invention includes means for switching and executing the equalization algorithm, and means for training the correlation matrix correction type using signals stored in the buffer memory 10 and the memory circuit 11 when switching from the error vector type to the correlation matrix correction type. be.

A signal having waveform distortion is input to the input terminal 1, and this signal is stored in the buffer memory 10. Equalization processing circuit 2
reads the input signal x (n) from the buffer memory lO and outputs the equalized signal Y (n). The determination circuit 4 converts this equalized signal Y (n) into a digital determination signal Z (n) and outputs it to the output terminal 4.

The switch circuit 5 selects the determination signal Y (n) when performing a steady equalization operation, and connects it to the error calculation circuit 6. The error calculation circuit 6 regards the waveform of the signal supplied from the switch circuit 5 as an ideal waveform, and converts it into an equalized signal Y.
Calculate the error e (n) with (n).

The adaptive processing circuit 7 sets each tap coefficient of the equalization processing circuit 2 based on the value of this error e (n).

The above operation is common to the LMS method and the RLS method.

Next, the operation of this embodiment will be explained using an example in which a signal having a frame structure shown in FIG. 2 is equalized.

In the frame configuration illustrated in FIG. 2, the length is T.

training signal is placed at the beginning of the frame. For this training signal, the adaptive processing circuit 7 sets the tap coefficients of the equalization processing circuit 2 using the RLS method with good convergence performance. At this time, the switch circuit 5 selects the signal from the training signal generation circuit 8 and connects it to the error calculation circuit 6. After training, switch circuit 5
is switched, and the judgment output Z (n) of the judgment circuit 3 is connected to the error calculation circuit 6.

When the change in the transmission path is gradual, the adaptive processing circuit 7 changes its adaptive equalization algorithm from the RLS method to L at the time of Vl.
Switch to MS method. At this time, the tap coefficients determined by the RLS method are used as initial values, and the coefficients are simply updated by the LMS method.

Assume that the adaptive equalization algorithm is switched to the RLS method at time 720 because the transmission path situation has changed and, for example, the average value of the error e(n)' has gradually increased. In order to perform the RLS method, an autocorrelation matrix (IJ) is required. However, the residue cannot be obtained only at the time of V2. Therefore, in this case, training is performed again by going back by the necessary training time TL.

That is, ■ From the buffer memory 10 (V2 TL) to V2
The signal is traced back and sent to the equalization processing circuit 2.
Enter.

(2) The initial value of the tap coefficient of the equalization processing circuit 2 is the value at the time of V2 obtained by the LMS method.

(2) Switch the switch circuit 5 to connect the memory circuit 11 to the error calculation circuit 6, and perform training using the determined output value stored in the memory circuit 11 as a reference signal.

Through such an operation, an inverse matrix of the autocorrelation matrix can be obtained. Therefore, from the point of ■2, RLS
Equalization can be performed continuously.

〔Effect of the invention〕

As described above, the equalizer of the present invention can switch its adaptive equalization algorithm between an error vector type and a correlation matrix correction type. As a result, (i) when the simple algorithm of the error vector type is sufficient, the circuit that executes the adaptive processing of the correlation matrix modification type can be placed in a dormant state. Therefore, by configuring with CMO3 circuits, it is possible to reduce power consumption.

(11) Since the correlation matrix modification algorithm can be applied when the characteristics of the transmission path change rapidly, equalization errors can be reduced and deterioration of transmission characteristics can be suppressed to a small level.

There are effects such as Therefore, by applying the present invention to a portable wireless device for mobile communications in which the waveform distortion of the transmission path is large and the waveform changes quickly, a very large effect can be obtained.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an equalizer according to an embodiment of the present invention. FIG. 2 is a diagram explaining the operation of the embodiment. FIG. 3 is a block diagram of a conventional equalizer. FIG. 4 is a diagram showing the difference in characteristics between the LMS method and the RLS method. DESCRIPTION OF SYMBOLS 1... Input terminal, 2... Equalization processing circuit, 3... Judgment circuit, 4... Output terminal, 5... Switch circuit, 6
...Error calculation circuit, 7...Adaptive processing circuit, 8...
training signal generation circuit, 10... buffer memory, 11... memory circuit, 21... delay circuit, 22.
... Multiplier, 23... Adder.

Claims (1)

[Claims] 1. An equalization processing circuit that equalizes the transmission characteristics of an input signal; a determination circuit that determines the sign of the input signal from an equalized signal output from the equalization processing circuit; and this determination circuit. an equalizer comprising adaptive processing means for adaptively setting characteristics of the equalization processing circuit so that the difference between the judgment output of the input signal and the equalization signal is small; The adaptive processing means includes a means for switching and executing two adaptive equalization algorithms, a correlation matrix correction type and an error vector type, and a means for executing the adaptive equalization algorithm by switching between the two adaptive equalization algorithms, a correlation matrix correction type and an error vector type. an equalizer comprising means for training a modified correlation matrix using the signals stored in the storage means.