JPH0313763B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an automatic equalization circuit that automatically determines a tap coefficient sequence in an automatic equalizer using a transversal filter.

Automatic equalizers are essential for modems used in synchronous high-speed data transmission lines using public analog telephone lines, and CCITT has proposed recommendations for this in the V series. Such an automatic equalizer requires training to set its characteristics before starting actual data transmission.

Several methods have already been proposed for training automatic equalizers, but they can be roughly divided into methods using special signal sequences such as pseudorandom patterns, and methods using impulses. Can be divided. As an example of the former method, for example, CCITT Recommendation V29 and the like stipulates a method of transmitting a binary pseudo-Ram signal and using this to train an automatic equalizer.

The automatic equalizer is realized mainly using a transversal filter that has excellent phase characteristics from the standpoint of waveform transmission. The characteristics of an automatic equalizer must be the inverse characteristics of the line, but the process of obtaining the tap coefficient sequence of the transversal filter that determines these characteristics requires, for example, a minimum of 2 This can be explained using existing statistical methods such as the maximum slope method that uses the multiplicative error as an evaluation function.

In such conventional methods, in order to obtain the final tap sequence at high speed, it is necessary to generate a special training signal sequence that minimizes the variation in the eigenvalues of the cross-correlation matrix between incoming training signals, and to use an equalizer. Consideration has been given to performing cyclic equalization by high-speed calculation using conventional statistical methods with the same tap length and training signal period. However, when assuming a case where the line characteristics are extremely poor relative to the transmission signal band, the validity of the characteristics of the special signal train and the randomness of the cyclic signal period (frequency characteristics), etc.
It cannot be said that it is necessarily fully elucidated.

In addition, methods using impulses include directly solving the multi-dimensional simultaneous equations created by cross-correlation between the impulse train and the error signal, and other direct solutions include converting to the frequency domain using the Fourier transform method. There is a method to find the optimal tap coefficient sequence. This impulse can be obtained by modulating the original impulse with another signal sequence and then demodulating it on the receiving side, or it can be a pseudo-random sequence with a sufficiently long period or a special signal sequence equivalent to this. For example, the impulse can be regenerated by cross-correlating it with an incoming signal. Since the impulse has a flat frequency characteristic, it is convenient as a training signal. However, all of the conventional methods require processing a large amount of calculations, which not only increases the processing time but also requires an enormous amount of hardware.

The present invention attempts to propose a new automatic equalization circuit using impulses in an automatic equalizer, and its purpose is to perform automatic equalization using impulses but with relatively simple means. I can,
Another object of the present invention is to provide an automatic equalization method for a high-speed convergence automatic equalizer that can be realized on a normal hardware scale. To achieve this objective, the automatic equalization circuit of the present invention automatically sets tap coefficient values in a transversal filter to equalize transmission path distortion. a first register group that holds a complex impulse sequence, a second register group in which each conjugate complex number of the complex impulse sequence is set as an initial value of a tap coefficient, and an initial value of the complex impulse sequence and the tap coefficient. a correlation means for calculating an autocorrelation value up to a predetermined order of the complex impulse train from and a correction means for updating the contents of the first register group by correcting an initial value of the register group, and repeatedly updating the contents of the second register group by the correlation means and the correction means, The content of the finally obtained register group is used as a tap coefficient of a transversal filter,
In addition, in an automatic equalization circuit that automatically sets tap coefficient values in a transversal filter to equalize transmission path distortion, a first register group that holds a complex impulse sequence reproduced from a training signal, and a first register group that holds a complex impulse sequence reproduced from a training signal; A second register group in which each conjugate complex number of the impulse sequence is set as the initial value of the tap coefficient, and an autocorrelation between the complex impulse sequence and the initial value of the tap coefficient to a predetermined order of the complex impulse sequence. a correlation means for calculating the value; and a correlation means for correcting the initial value of the tap coefficient so that non-zero-order autocorrelation values become zero based on the autocorrelation value obtained by the correlation means, and correcting the initial value of the tap coefficient for the first register group. a correction means for updating the contents, the correlation means and the correction means repeatedly update the contents of the second register group, and the finally obtained contents of the register group are transversalized. It is characterized in that it is used as a tap coefficient of a shaped filter.

The principle and embodiments of the present invention will be explained below.

Now, the transmission signal is {a _K } (if it is a rising pulse, it is 0...010
...0) and the impulse response of the transmission path is {h _k }, then the signal {r _k } arriving at the equalizer can be expressed by the following equation.

r _o = _N 〓 ^i=-N a _i ·h _oi If the tap coefficient of the equalizer is {C _k }, then the equalized output {y _k } is expressed by the following equation.

y _o = _N 〓 ^i=-N c _i・r _oi = 〓 ⁱ C _i・( 〓 ^j a _j・h _oij )= 〓 ^j a _j・( 〓 ⁱ C _i・h _oji ) The signal {a _j } If it is a rising pulse (a _j =1 when j=0, a _j =0 otherwise), the output _yo is as follows.

y _o = _N 〓 ^i=-N C _i・h _oiHere , if C ₀ =1, C _i =0 (i≠0), then y _o =
h _o , and the impulse response of the transmission path itself appears in the equalizer output. Furthermore, n=-M,...
..., -2, -1, 0, 1, 2, ..., M (M is a sufficiently large value), if a pulse is sent when n = 0, then the output y _o (= h _o )
is the distortion itself in the transmission path. Therefore, this n≠
It is possible to correct the tap coefficient sequence {C _i } using the value at 0. That is, from the Nyquist condition, using y ₀ = 〓 ⁱ C _i・h _-i ＝1 y _o = 〓 ⁱ C _i・h _oi = 0, for example, using the least squares error as the evaluation function, tap using the maximum slope method, etc. It is sufficient to correct the coefficient sequence {C _i }.

By the way, when an impulse is transmitted through a distorted transmission line, if the system is linear, the autocorrelation value of the received impulse sequence will be as shown in FIG. In the same figure, A ₀ , A ₁ , A _-1 , A ₂ , A _-2 are respectively 0th order,
It shows first-order, -first-order, second-order, and -second-order correlations.
Note that the sign here indicates the temporal relationship, and (-)
indicates past values.

This indicates that some degree of equalization can be achieved by correlation operations. The automatic equalization method of the present invention attempts to equalize waveforms through correlation operations. Specifically, the automatic equalization method of the present invention attempts to equalize waveforms by correlation operation. Specifically, the automatic equalization method uses the correlation between the input signal sequence and the created equalized sequence. The aim is to make the correlation strongest when the two match perfectly, and to have no correlation in other cases.

The advantage of this method is that the optimal tap coefficient sequence {C _i } can be found without knowing where the median value of the input signal is (however, synchronization must be achieved at the end). That is, by doing so, phase distortion components included in the input signal are removed. However, although it is possible to remove phase distortion by this operation, it is not possible to completely equalize the amplitude, but after a certain number of operations,
The amplitude distortion can be corrected by the conventional maximum slope method using least squares error.

A method for obtaining an optimal equalization coefficient sequence by focusing on the autocorrelation of the incoming signal can be described by the following equations.

C _i △＝x ^* _i (1) A _j = 〓 ⁱ C _i・x _i+j (2) C ⁿ⁺¹ _i ＝C ⁿ _i −P・A _j・x ^* _i (3) Here, {C _i } is the equalizer tap coefficient sequence vector, {x _i } is the impulse sequence, A _j is the correlation between the tap coefficient and the input impulse, and P is a small positive number. Equation (1) means the initial setting of the impulse, the correlation A _j is obtained using Equation (2), this value is regarded as an error, and the tap coefficient is corrected using Equation (3). By repeating this series of operations for an appropriate degree of correlation, the final equalization coefficient sequence {C _i } can be obtained.

In this method, correction is started by determining the autocorrelation of the incoming signal, and processing is finally performed to eliminate higher-order correlations, and is therefore called the orthogonal correlation method.

FIG. 2 is a functional block diagram showing the configuration of an embodiment of the automatic equalization circuit of the present invention. In the same figure, 1
is an impulse reproduction circuit, 2 is a first tap delay register, 3 is a correlator, 4 is a register group, 5 is a tap coefficient corrector, 6 is a second tap delay register,
7 is a multiplier group, 8 is an accumulator, and 9 is a tap coefficient register for storing the tap coefficient sequence obtained by this method and used for subsequent equalization operations.

Further, FIG. 3 is a diagram showing an example of the correction function calculation for one tap in the tap coefficient corrector 5 of FIG. 2. In the figure, 11 and 12 are multipliers, and 13 is a subtracter.

In FIG. 2, the impulse is modulated with another signal train and sent out on the transmitting side (not shown). The signal containing the incoming impulse is applied to the impulse regeneration path 1 and is regenerated into the original impulse form. The reproduced impulse is input to the first tap delay register 2. The complex conjugate vector of this impulse sequence is given to the register group 4 as the initial value of the tap coefficient according to equation (1), and is input to the correlator 3 together with the impulse sequence already latched in the tap delay register 2. 2) Correlation A _j is determined according to formula. Tap coefficient corrector 5
(3) for each tap by the correlation A _j.
A correction calculation is performed according to the formula.

In Figure 3, the correlation output from the correlator
A _j is multiplied by a constant value P in a multiplier 11 .
The output of multiplier 11 is further multiplied by impulse series vector x ^* _i in multiplier 12. The output of the multiplier 12 is subtracted from the tap coefficient vector C ⁿ _i in a subtracter 13 to obtain a new tap coefficient vector C ⁿ⁺¹ _i at the output of the subtracter 13.

Referring again to FIG. 2, the new tap coefficient sequence determined by the tap coefficient corrector 5 is latched into the register group 4 and used for the next calculation. After repeating this operation a certain number of times, the values of the tap coefficient strings finally latched in the register group 4 are transferred to the tap coefficient registers 9 of the second tap delay register 6, respectively.

After the tap coefficient string is set in the tap coefficient register 9, the signal on the transmission line is guided to the second tap delay register 6, and each tap output is cumulatively multiplied by its corresponding tap coefficient.
The outputs of the multiplier group 7 are accumulated in an accumulator 8;
This results in an equalized signal at the output of the accumulator 8.

In order to perform the arithmetic processing described above, arithmetic functions such as multipliers and adders/subtractors are required, but in order to efficiently realize these functions, it is necessary to time-share the arithmetic operation functions (AU). Possible use. In this case, data transfer with the AU can be controlled using a microprogram stored in a read-only memory (ROM) or the like.

FIG. 4 is a block diagram showing a specific configuration of an embodiment of the automatic equalization system of the present invention. In the figure, 20 is AU, and multiplicand register 21,
It is composed of a multiplier register 22, a multiplier 23, an adder 24, and an accumulator 25. 26
is a read/write memory (RAM), 27 is an arithmetic control decoder, 28 is a read-only memory (ROM), 2
9 is a counter (CTR), 30 is a RAM address latch, 31 is an address decoder, 32 is a counter (CTR), 33 is a delay circuit, 34 is a write bus,
35 is a read bus.

In FIG. 4, the CTR 29 is cleared by a clear signal given from the arithmetic control decoder 27 every time a series of processing is completed, and can repeat a certain number of counts. As the CTR 29 counts up, the binary count value is sequentially given to the ROM 28 as an address, so that the contents stored in the ROM 28 corresponding to the address are sequentially read out, and the arithmetic control decoder 27 and the RAM
address latch 30.

The arithmetic control decoder 27 controls the RAM 26 according to the content read from the ROM 28. It performs read/write (R/W) control, latch control for the multiplicand register 21 and multiplier register 22, and supplies a clear signal and a clock signal to the accumulator 25.

The RAM address latch 30 is composed of a shift register and the like, and depending on the contents read from the ROM 28, the address for the RAM 26 (old
RAM address) and address decoder 3
Enter 1.

The CTR 32 has a fixed number of digits, and by receiving its carry signal as a load input via the delay circuit 33, it repeatedly counts from the number determined by the preset information to the maximum number. The address decoder 31 is composed of a ROM, etc., and gives the old RAM address of the RAM address latch 30 an offset necessary for performing correlation calculations according to the count value of the CTR 32, and sets the old RAM address of the RAM address latch 30 to the new address.
Generate a RAM address and input it to RAM26.

Thereby, the data read from the RAM 26 is given to the multiplicand register 21 and the multiplier address 22 via the read bus 35, and the result of the required operation is written from the accumulator 25 to the RAM 26 via the write bus 34. The correlation calculation is performed by repeating the steps. RAM2
Writing external data to 6 is also done using write bus 3.
It is carried out through 4.

The operation of the automatic equalizer in the configuration of FIG. 4 is performed as follows in comparison with FIG. 2. In this case, all memories corresponding to the tap delay register and tap coefficient register are stored in RAM.
26 will do it. Further, various calculations are centrally processed in the AU 20.

The impulse data fetched from the write bus 34 is stored in a predetermined memory area of the RAM 26. The complex conjugate value of this impulse is stored in another memory area of the same RAM 26 as an initial tap coefficient.

Next, in order to find the correlation A _j (or A′ _j described later), the impulse data and tap coefficient data are
The read bus 35 is read out sequentially from the RAM 26.
The multiplicand register 21 and multiplier register 22 are latched through the multiplicand register 21 and the multiplier register 22. Multiplication and accumulation are performed using these two data, and a correlation value between the two is given to the accumulator 25 and stored in the RAM 26 via the write bus 34. When obtaining a value according to the order of correlation in this calculation, the address decoder 31 converts the addresses of the RAM 26 to give an offset according to the order to the addresses of the impulse data and tap coefficient data. .

The tap coefficients are corrected for each of the correlation values obtained in this way according to the above-mentioned descriptive formula. This operation is performed by appropriately reading tap coefficients, correlation values, etc. in accordance with address information from the microprogram, latching them into the multiplicand register 21 and the multiplier register 22, and performing arithmetic processing using the AU 20. After performing this correction a certain number of times, the corrected tap coefficient is
RAM 26 for storing tap coefficients as an equalizer
is stored in the area. Thereafter, the same circuit is operated as an equalizer by using the stored corrected tap coefficients. At this time, the processing content of the microprogram changes, and this is done by an external control signal EX-CONT1 that changes at a predetermined time. Also, in the address decoder 31, CTR3 is used to convert from the old RAM address to the new RAM address.
2 output data is external control signal EX-CONT2
Since each tap output and the corresponding tap coefficient are ignored by the data address, the arithmetic processing of each tap output and the corresponding tap coefficient is performed without any offset.

In the configuration shown in FIG. 4, the process of determining the correlation using data stored in each area of the RAM 26 is performed as follows. First, the impulse sequence is stored in a specific address area as an initial tap coefficient sequence {C _i }, and then the impulse sequence {x _i }
It is stored in another specific address area as .

FIG. 5 is a diagram showing the address space of the impulse sequence {x _i } and the tap coefficient sequence {C _i }. In the same figure, the impulse sequences x ₁ , x ₂ , ..., x _i , ..., x _N are respectively addresses AD ₀ , AD ₁ , ..., AD _i , ..., AD _N
are stored in the tap coefficient sequence C ₁ , C ₂ ,..., C _i ,...,
_CN are respectively addresses BD ₀ , BD ₁ ,..., BD _i ,
..., it is shown that it was stored in BD _N.

Therefore, when calculating correlation, impulse sequence x _i or tap coefficient sequence C _i is calculated by the number of correlation orders.
Just increment (or decrement) the address of and feed it to the multiplier. This is done by counting up the CTR 32, which is one input of the address decoder 31, each time tap correction is performed, and in the address decoder using the ROM.
This can be done by preparing an adder/subtractor to modify the RAM address or give an offset, and perform a logical operation with the address data.

The address decoder 31 has a function of giving an address offset by the output of the CTR 32 only when an address corresponding to either the impulse sequence x _i or the tap coefficient sequence C _i is input, and it is used for other addresses. outputs the address data as is.

The address decoder 31 can be implemented simply as a conversion table using a ROM. Another method is to shift the retrieved data so as to obtain correlation values of different orders, but the former method saves data memory.
This is advantageous in that less address information is required to be output from the microprogram.

In the above explanation, the correlation A _j between the tap coefficient and the input impulse is determined as one correlation between the tap coefficient and the input data, as shown in equation (2) above, and all tap coefficients are corrected using this correlation. However, the correlation A _j is generally calculated as follows.

A′ _j = _N 〓 ^j=-N ( 〓 ⁱ C _i・x _i+j ) (4) In the above equation, Σ′ indicates a sum other than j≠0.

Tap coefficient when correlation is calculated using equation (4)
The correction of C _i is expressed by the following equation.

C ⁿ⁺¹ _i = C ⁿ _i −P・_N 〓 ^j=-N A′ _j・x ^* _i+j (5) Operations for calculating the equalization coefficient sequence using equations (4) and (5) The processing can also be realized using the same circuit as the embodiment shown in FIG. 4 by changing the microprogram.

Furthermore, by changing equations (3) and (5) as follows and adding them to the above-mentioned correction process, it is possible to correct the amplitude.

C ⁿ⁺¹ _i = C ⁿ _i −P(A ₀ −E ₀ ) x ^* _i (6) or C ⁿ⁺¹ _i = C ⁿ _i −P {( _N 〓 ^j=-N A′ ₀・x _i )−E′ ₀ }x ^* _i (7) However, A ₀ = A′ ₀ = 〓 ⁱ C _i x _i After repeating the above operation several times, the More reliable convergence is possible by applying the maximum slope method using a multiplicative error or the like as an evaluation function.

As explained above, according to the automatic equalization circuit of the present invention, automatic equalization can be performed by a relatively simple means using impulses, and a high-speed convergence automatic equalizer can be realized with ordinary hardware scale. Excellent effects can be obtained.

[Brief explanation of the drawing]

Fig. 1 is a diagram showing the autocorrelation value of an impulse sequence, Fig. 2 is a functional block diagram showing the configuration of an embodiment of the automatic equalization circuit of the present invention, and Fig. 3 is an example of correction function calculation for one tap. FIG. 4 is a block diagram showing the specific configuration of an embodiment of the automatic equalization circuit of the present invention, and FIG. 5 shows the address space of the impulse sequence {x _i } and the tap coefficient sequence {C _i }. FIG. DESCRIPTION OF SYMBOLS 1... Impulse reproduction circuit, 2... Tap delay register, 3... Correlator, 4... Register group, 5... Tap coefficient corrector, 6... Tap delay register, 7... Multiplier group, 8... Accumulator, 9... Tap Coefficient register, 1
1, 12... Multiplier, 13... Subtractor, 20... Arithmetic operation function (AU), 21... Multiplicand register, 22...
Multiplier register, 23... Multiplier, 24... Adder, 2
5... Accumulator, 26... Read/write memory (RAM), 27... Arithmetic control decoder, 28... Read only memory (ROM), 29... Counter (CTR),
30...RAM address latch, 31...address decoder, 32...counter (CTR), 33...delay circuit, 34...write bus, 35...read bus.

Claims (1)

[Claims] 1. In an automatic equalization circuit that automatically sets tap coefficient values in a transversal filter to equalize transmission path distortion, a first A register group 2, a second register group 4 in which each conjugate complex number of the complex impulse sequence is set as the initial value of the tap coefficient, and a register group 4 that stores the complex impulse sequence in advance from the complex impulse sequence and the initial value of the tap coefficient. Correlation means 3 for obtaining autocorrelation values up to a predetermined order, and correcting the initial value of the tap coefficient so that higher order autocorrelation values converge based on the autocorrelation values obtained by the correlation means. a correction means 5 for updating the contents of the second register group, and the correlation means and the correction means repeatedly update the contents of the second register group a plurality of times to finally obtain the results. An automatic equalization circuit characterized in that the contents of the register group are used as tap coefficients of a transversal filter. 2. In an automatic equalization circuit that automatically sets tap coefficient values in a transversal filter to equalize transmission path distortion, a first register group 2 that holds a complex impulse sequence reproduced from a training signal; A second register group 4 in which each conjugate complex number of the complex impulse sequence is set as the initial value of the tap coefficient, and registers from the complex impulse sequence and the initial value of the tap coefficient to the predetermined order of the complex impulse sequence. a correlation means 3 for calculating an autocorrelation value; and a correlation means 3 for calculating the second tap coefficient by correcting the initial value of the tap coefficient so that the autocorrelation values other than the 0th order become 0 based on the autocorrelation value calculated by the correlation means. a correction means 5 for updating the contents of the register group, and the register finally obtained by repeatedly updating the contents of the second register group by the correlation means and the correction means a plurality of times. An automatic equalization circuit characterized in that the contents of a group are used as tap coefficients of a transversal filter.