JPS631781B2 - - Google Patents

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to an automatic equalizer in a modem device used for data transmission, and in particular, to an automatic equalizer using a transversal filter. The present invention relates to a method for controlling automatic equalization of a transversal filter for convergence in data mode when it is lost, that is, when it becomes a divergent state.

[Conventional technology]

When transmitting digital data handled by computer terminals, faxes, etc. using existing analog voice lines, a modem with an automatic equalization function is essential, and a device that realizes this automatic equalization function is essential. Generally, a filter having a transversal type filter is used.

Furthermore, in high-speed data transmission, when a modem sends data to a terminal at the other end of the line, a special pattern called a training signal is sent out prior to the actual data transmission. This is an initial sequence necessary for gain control of the automatic gain adjustment circuit in the receiving section, timing regeneration, automatic equalizer presetting, carrier regeneration, etc.

On the other hand, when the automatic equalizer loses its equalization ability during the equalization process while receiving data, that is, when it enters a divergent state, it requests the transmitter to send out the training signal again. A common method is to try to converge again using the initial sequence.

However, in a system where several terminals are connected in parallel to one transmission path (called a multipoint system), each time the automatic equalizer in the modem receiver of one terminal diverges, the I guess I was repeating the training for levers.
The data processing efficiency of the entire system decreases.

Therefore, it is desired that the modem of the diverged terminal be able to converge again while receiving data, that is, when the system is in data mode.

[Problem that the invention is trying to solve]

The above will be explained in more detail as follows.

In devices that use multilevel orthogonal amplitude modulation, such as modems that comply with CCITT Recommendation V29, the signal coordinate arrangement is not symmetrical as shown in Figure 1, so a transversal filter is used as the basic configuration. When converging an automatic equalizer, the position of the center tap, which is the reference point, cannot be determined.

Therefore, to start training, all taps are grown from a constant value (e.g. 0) using local reference signals (usually
(Using absolute reference to train the receiver).

After this process, the position of the central tap and the magnitude of the tap coefficient are established.

These tap coefficients are generally variable values because they are adaptively modified in response to line fluctuations.

However, due to some disturbance (such as large impulsive noise or instantaneous interruption), the equalized waveform output from the transversal filter becomes significantly different from the reference signal waveform, and the error signal becomes large. If this error signal does not converge even after a predetermined period of time has elapsed in this state, the tap coefficient of the transversal filter becomes uncontrollable, and the equalization ability of the automatic equalizer is lost. This state is generally called a divergence state, but when this divergence state occurs, each tap coefficient that was secured in the converged state becomes a random value, and the tap coefficients must be controlled to return to the converged state again. becomes impossible.

Therefore, an object of the present invention is to provide a method that enables convergence in this data mode, which has not been possible in the past.

[Means for solving problems]

In the method of the present invention, when the transversal automatic equalizer is in a divergent state in a transversal automatic equalizer that normally controls the center tap and other taps by judgment error, the transversal automatic equalizer The center tap of the automatic equalizer is fixed at a constant value, and the equalizer output is determined for modifying other tap coefficients using data with little interference between data, and the transversal The average power (PF) of the reference signal with the average power (Po) of the automatic equalizer output signal at the receiver
automatic gain control (AGC) at the output stage of the transversal automatic equalizer.
It is characterized by applying .

〔Example〕

Hereinafter, the present invention will be explained in detail with reference to examples.

First, the convergence algorithm of the method of the present invention will be explained.

Figure 1 shows the range that each signal can take in terms of amplitude (distance from the center point) and phase (angle).

As shown in Figure 1, incoming data has a predetermined signal arrangement at a certain point in time (Figure 1
Therefore, if the line noise is Gaussian noise, the equalized data will be uniformly distributed within a certain circle for each signal point.

In addition, since the probabilities that these signal points are transmitted are all the same, if equalization is performed normally, the average power value (Po) of the equalized data will be equal to the predetermined signal arrangement. It perfectly matches a certain uniquely determined power value (PF).

On the other hand, if normal equalization is not performed and the equalizer diverges, the equalized data will be any value up to the hardware limit (for example, any value from the maximum value to the minimum value in terms of arithmetic accuracy). ), and its average power (Po) does not match PF.

Therefore, one condition is to make the average value Po of the power of this divergent data match PF.

However, simply applying this condition does not guarantee convergence of the equalizer. This is because correction using power alone is insufficient for vector data that includes phase.

That is, some kind of correction is also required for the phase. This uses a statistical correction method (maximum slope method) using the squared error of equalized data as an evaluation function.
When divergence occurs, it is not effective to perform error evaluation on all data because there is too much erroneous data.

Therefore, it is conceivable to use an error that is relatively correct as the second condition. In other words, in the signal arrangement shown in Figure 1, since the signal interval at the internal point A is shorter than that at the external point B, there is a high probability that it will be incorrectly determined, and there is little information to be used for correction. This means that we cannot expect it.

That is, the external points are the desired data for correction.

However, if only the external B signal point is used in a divergent state, the polarity of the error will be biased toward the external region, so that control will be unidirectional. That is, it begins to move in the direction A inside.

Therefore, the tap coefficient of the equalizer is modified in one direction (in this case, in the direction of lowering the gain of the equalizer), and the gain of the equalizer is not a normal gain. must be compensated in this way.

Therefore, at least the value of the center tap coefficient is fixed. Furthermore, an automatic gain control (AGC) circuit is placed at the equalizer output to satisfy the above condition (average power matching).

The above is the algorithm of the method of the present invention, and FIG. 2 is a block diagram of an embodiment implementing this method.

The embodiment shown in FIG. 2 will be described below.

As shown in the figure, the transversal automatic equalizer 10 equalizes the input waveform by convolving several tap coefficients {Ck} extracted from the delay circuit T with input data {Xi}.

Its general description is expressed as follows.

(Σ' indicates a cumulative value other than n=0) The tap coefficient {Ck} takes various values depending on the line characteristics, but when they diverge, each takes an arbitrary value independently. Moreover, if the value of the coefficient Co of the center tap is variable, it is impossible to converge again from this state.

First, to explain the tap control, the output signal A of the automatic equalizer 10 is transmitted through the multiplication circuit 1 (this circuit is related to the AGC circuit, which is one of the characteristics of the present invention, and will be described later). The equalized output B is obtained through this. This equalized output B is level-compared with the reference signal C in a judgment circuit 6, and the resulting judgment error (level difference) e is tapped as a judgment error e' via a multiplication circuit 7 that multiplies it by a predetermined weighting coefficient. Input to circuit 8. If it is determined that the variation amount (differential value) of the judgment error e or e' remains at a predetermined level or higher even after a predetermined period of time has elapsed, it is detected that the transversal filter has fallen into a divergent state. Then, the central tap coefficient Co is fixed to a certain value, for example 1. In this case, the reference signal C is 16 in Figure-1.
This is a signal that randomly generates point-like amplitude values and phases.

Here, in the present invention, whether the error signal determined by the determination circuit 6 is in the area A shown in FIG.
It is necessary to determine whether it is in area B. Therefore, at the same time as detecting the error signal, the circuit 6 compares the absolute value of the reference signal itself with a threshold value that separates the A and B areas. The error signal determined to be in the B area in this way is output to the multiplication circuit 7 together with a flag.

The internal A signal point and the external B signal point are distinguished by one of the following two methods.

(1) A method of dividing the coordinate space shown in FIG. 1 into N×N square regions and checking whether the reference signal falls within the external region B shown in FIG. 1.

(2) A method of dividing the coordinate space shown in Figure 1 radially into several line segments, further dividing each line segment into two, and checking whether it falls within the external area B shown in Figure 1.

The multiplier circuit 7 extracts only the error (e') regarding the outer B signal point from the input error signal and flag, and tap correction circuit 8 performs tap correction of the equalizer.
Enter.

This tap correction circuit 8 executes an algorithm for adaptively equalizing the equalizer 1 (such as the maximum slope method using the least square error as an evaluation function). The multiplication circuit 7 is a multiplication circuit that multiplies the equalized output by 1 if it is determined to be the above-mentioned outside signal, and multiplies it by 0 if it is determined to be another signal.

When a divergence state is detected by the tap control described above, the tap is first fixed and the equalization function is converged using the external B as the signal region.

However, if this control alone is used, the automatic equalizer 10 will only equalize the received signal from the outside B of the signal area, so it will move in one direction from the outside B to the inside A area. , regular equalization cannot be expected.

Therefore, the present invention focuses on performing automatic gain control between the automatic equalizer 10 and the equalized output B, and also compensates for the gain of the equalizer.

In Figure 2, the equalized output B is input to the squaring circuit 2 that calculates the magnitude of the vector, and the subtraction circuit 3 performs subtraction processing between the output D and the input E of the average power (PF) of the reference signal. , its output F is integrated circuit 4
(G), a predetermined constant ko is added in an adder circuit 5, and the equalized output A is multiplied in a multiplier circuit 1 to obtain an equalized output B subjected to gain control. has been done.

Therefore, after being level-adjusted by this AGC circuit, it is regarded as one of 16 points, and the determination circuit 6 calculates the error e with respect to the corresponding reference signal. Also, the equalized output signal B is input to the squaring circuit 2 to calculate the magnitude of its vector, and the squaring circuit 2 evaluates the power of the signal by (Real) ² + (Image) ² This is a circuit that performs the multiplication and accumulation calculations required for this purpose. The subtraction circuit 3 calculates the difference F between this and the unique average power E of the original reference signal arrangement. This difference value will be integrated by the integrating circuit 4.

This value is added to a certain value (for example, 1) by the adder 5, and becomes the value H by which the equalized output A is multiplied.

An example of the above content is as follows.

The power of the final equalized output B is the average power of the reference signal E
When G is larger than G, a negative value is accumulated in the integrator 4, and its output G also goes in the negative direction. Since this is added to the constant ko, the equalized output A
The value H by which the value H is multiplied is decreased, and the value of B becomes smaller.

This makes it possible to bring the average power of B closer to the reference power E.

In this way, tap control is performed by making a decision outside the signal region B, and gain compensation is performed at the output stage of the automatic equalizer to enable convergence in the data mode.

〔Effect of the invention〕

As described above in detail, according to the present invention,
When a divergence state is detected, the equalization output is controlled to match the average power of the reference signal, and the tap coefficient of the transversal filter is controlled by comparison with the reference signal at the outer signal position. This makes it possible to converge the automatic equalizer without a training signal even in a divergent state, and even in a multi-point system, it is possible to converge the automatic equalizer's divergence without reducing the processing efficiency of the entire system. , the effect is great industrially.

[Brief explanation of the drawing]

Figure 1 is a signal arrangement diagram for the present invention, Figure 2 is a block diagram of an embodiment applying the method of the present invention, 10 is an automatic equalizer, 1 is a multiplication circuit, and 2 is a vector magnitude calculation. 3 is a subtracter, 4 is an integrator, 5 is an adder, 6 is a subtracter, 7 is a multiplication circuit,
8 is the tap correction circuit.

Claims (1)

[Claims] 1. In a transversal automatic equalizer that normally controls the center tap and other taps based on judgment errors, when the transversal automatic equalizer enters a divergent state, the transversal The center tap of the automatic equalizer is fixed at a constant value, and the equalizer output is judged for correcting other tap coefficients using data with little interference between data, and the transformer Automatic gain control is performed at the output stage of the transversal automatic equalizer by controlling the average power (Po) of the output signal of the transversal automatic equalizer to match the average power (PF) of the reference signal in the receiver. A data mode convergence method characterized by multiplying .