JPH05160762A - Coefficient control system for echo canceller - Google Patents

Abstract

PURPOSE:To make capable of compressing a circuitry for an arithmetic operation by updating a tap coefficient with plural coefficient updation algorithm methods in the echo canceller of transversal filter of a transmitter. CONSTITUTION:A transmission signal a (n) is inputted to a D/A converter circuit 2 and a shift register 6. The sum of a reception signal and an echo is converted into digital data by an A/D converter circuit 5. An output of each stage of the shift register 6 is fed to multiplier circuits MO-MN, in which the signal is multiplied with a tap coefficient outputted from a coefficient updation control circuit 7 and the result is fed to an adder 9. The coefficient updation control circuit 7 updates a tap coefficient for each cycle by a 1st coefficient updation algorithm after the start of locking. Then each tap coefficient is updated only for once per plural cycles in a 2nd coefficient updation algorithm at least after a lapse of a prescribed time or after a prescribed locking state is reached. The adder 9 adds outputs of the multiplier circuits MO-MN to output a pseudo echo signal w (n).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo canceller of a transmission device, and more particularly to a coefficient control system of an echo canceller for controlling a tap coefficient of a transversal filter.

[0002]

2. Description of the Related Art A digital subscriber line transmission interface device equipped with an echo canceller is required to simultaneously transmit high-speed digital data in both transmitting and receiving directions by using a metallic pair cable of a telephone subscriber line.

A digital subscriber line transmission interface device using digital signal processing technology is a device for performing high-speed digital data communication using an existing telephone metallic cable, and a high speed (80 kbau) is provided in the device.
d), a hybrid device that simultaneously transmits bi-level (4-level) digital data in both directions is used. Since this hybrid apparatus uses bidirectional simultaneous transmission, a transversal filter type echo canceller is essential.

FIG. 5 is a processing flowchart of a conventional echo canceller. The process of FIG. 5 is executed every cycle of the clock applied to the shift register in the transversal filter.

When the execution is started, in step S1

[0006]

[Equation 1]

## EQU1 ## and the pseudo echo w (n) is obtained. Then, from the sum y (n) of the received signal and the echo, the pseudo echo w
The value obtained by subtracting (n) is defined as the error e (n). Here, n is the number of times of processing.

Then, in step S2, C _j (n) + αsgn [e (n) a (nj)] is obtained, and tap coefficient C _j (n +) in the next cycle is obtained.
1). Note that α is the step size. Step S2 is performed for each tap, and the calculation between j = 0 and N is performed.

The above steps S1 and S2 are sequentially executed every cycle. When the cable distance is long, the received signal from the far end is greatly attenuated. On the other hand, the echo in which the transmitted signal passes through the hybrid circuit to the receiving side has low dependence on the cable distance and is hardly attenuated. For this reason, the echo amplitude at the receiving end may be +46 dB, or 200 times the amplitude of the received signal. If the echo amplitude / far end reception signal is large as described above, it is necessary to attenuate the echo by at least 58 dB or more by the echo canceller. The residual echo when attenuated by 58 dB is -12 dB with respect to the received signal, making it possible to judge the received signal.

In order to obtain a wraparound attenuation amount of about 60 dB for a digital echo canceller, the coefficient word length of the tap coefficient of the echo canceller and the operation word length of the product-sum operation of the filtering process become problems.

The processing of the echo canceller is usually executed by digital signal processing including other processing, but in this case, in order to minimize the hardware, it is general to perform the operation by the fixed point method. However, in the case of fixed-point arithmetic, the precision depends on the minimum number to be handled, so it is necessary to carefully consider the coefficient word length.

In the operation of the transversal filter of the echo canceller, a long operation word length is required so that the operation word length does not matter, and the product-sum operation register 22 bits,
Other registers 16 bits or more are required. If the tap coefficient is a fixed value and the word length is 16 bits or more, the decrease in attenuation due to rounding error can be ignored.

However, in the case of a transversal filter type echo canceller, the coefficient of each tap is set to a step size based on the multiplication result of the signal error corresponding to each tap of each cycle and the output error of the echo canceller. A so-called coefficient updating process of changing only the corresponding value is performed and is not a fixed value. Also, the size of the step size cannot be smaller than the value corresponding to the least significant bit of the coefficient word length.

Since the coefficient update processing stochastically approximates the true value, the coefficient update processing approaches the true value in a zigzag manner, that is, with some blurring. Therefore, if the step size is large, the value may temporarily deviate considerably from the true target value, which causes an error, and the wraparound attenuation amount cannot be increased.

Since there is a coefficient updating process, in order to obtain the above-mentioned wraparound attenuation amount, a conventional echo canceller having 20 to 30 taps has a coefficient word length of 20 bits or more and a signal word length of 15 bits. ~ 16 bit is required.

When the tap coefficient is updated, the tap coefficient is distributed in a range having a considerable spread of about 10 times the true tap coefficient value ± step size even after the convergence, so that the substantial error is 3 to 4 bits. It gets worse by a value equivalent to.

[0017]

As described above, the coefficient word length of the echo canceller of the digital subscriber line transmission interface device is 16 bits without degrading the performance.
Since it is difficult to implement this, it is necessary to process all devices including the other processing with a 20-bit general-purpose digital signal processor (DSP), or to use a dedicated processor that performs 20-bit processing only for the echo canceller. The method of using it is done.

When a general-purpose digital signal processor of 20 bit processing is used, the processing other than the echo canceller such as the decision feedback equalizer is 16 bit processing, but only for the tap coefficient of the echo canceller, 20 b
There is a problem in that the device as a whole is inevitably inevitably processed by it.

When a device is constructed using dedicated hardware, the echo canceller and the decision feedback equalizer generally perform similar processing, so it is decided that the echo canceller can suppress the number of bits of the tap coefficient to 16 bits. One piece of hardware can be multiplexed and used for the feedback equalizer. This is effective in reducing the scale of the device. However, in the past, 20 bits were required, so the same as above.
I had to either use 20 bit hardware in common, or assign dedicated hardware only to the echo canceller. Therefore, there is a problem that it is not possible to reduce the hardware.

An object of the present invention is to provide a coefficient control system for an echo canceller, which processes all processing related to the echo canceller with 16 bits, for example.

[0021]

The present invention is a transversal filter type echo canceller of a transmission device.

Until the coefficient word length is affected by the training process of the echo canceller, the taps are updated every cycle by the first coefficient updating algorithm according to the first configuration,
When the error does not become small because the word length of the tap coefficient is short even if the convergence progresses, at least one of after the elapse of a certain time or after reaching a certain pull-in state, once in a plurality of cycles in the second step. Only update.

The update algorithm calculates the product of the transmission signal symbol value and the error output of the echo canceller for each cycle, accumulates only the code (sign), and is suitable for the accumulated value once in a plurality of cycles. A new tap coefficient is obtained by subtracting the value multiplied by the appropriate weight from the original tap coefficient. By this update, highly accurate coefficient update control can be performed by the same bit number calculation as the conventional one.

Further, among the tap coefficients of the echo canceller, there is a possibility that the tap coefficient may become large from 7 to 7.
There are 8 taps, those taps are controlled in the above-described first and second steps, and the other taps are updated by the first algorithm.

[0025]

The present invention will be described in detail below with reference to the drawings. FIG. 1 is a block diagram of an echo canceller used in a digital subscriber line transmission interface device or the like. The echo canceller is an Nth-order transversal filter,
The tap coefficient C _j adaptively changes to cancel the echo.

In FIG. 1, the transmission signal a (n) is ± after D / A conversion.
It is a signal having four values of 1, ± 3, and is normally pseudo-randomized by a scrambler. Here, n represents the time in baud rate units.

When the transmission signal a (n) is applied from the terminal 1,
The D / A conversion circuit 2 converts the digital transmission signal a (n) into an analog signal. The analog signal converted by the D / A converter 2 passes through the hybrid circuit 3 and the cable 4,
It is sent to the other party, but hybrid circuit 3 and cable 4
Echo comes back due to a mismatch between them. At the same time, the received signal from the far end is also added. This signal is converted into digital data by the A / D conversion circuit 5. Here, the sum (digital value) of the received signal and the echo is y (n)
And

On the other hand, the transmission signal a (n) is input to the shift register 6 in addition to the D / A conversion circuit 2. The shift register 6 is an N + 1 stage shift register that shifts input data at the same clock as the basic clock of the transmission signal. The output of each stage of the shift register 6 is a multiplication circuit M _0.
.About.M _N , multiplied by the tap coefficient output from the coefficient update control circuit 7, and added to the adder 9. The adder 9 adds the outputs of these multiplication circuits M _{0 to} M _N and outputs the result as a pseudo echo signal. The pseudo echo w (n) that is the output of the transversal filter of the echo canceller is expressed by the following equation.

[0029]

[Equation 2]

This pseudo echo w (n) is added to the adder 10 and further added to the sum of the received signal converted by the A / D converter 5 and the echo. Since the output of the adder 9 is applied to the minus (-) terminal of the adder 10, the difference is obtained as a result. This output is the residual e (n), and is represented by e (n) = y (n) -w (n) (2). This residual e (n) is output from the terminal 11 as a received signal.

The residual contains the received signal from the far end. The residual e (n) and the transmission signal a (n) are coefficient update blocks, for example,

[0032]

[Equation 3]

However, the processing of j = 01, ..., N is performed, and the tap coefficient C _j changes by ± α. Here, α is a positive number and is called a step size.

In order to make a qualitative explanation below, the received signal y (n) is set as follows.

[0035]

[Equation 4]

Here, R (n) is a received signal from the far end. E _i is the amplitude of each cycle of the echo response characteristic when only one transmission signal is transmitted. An example is shown in FIG.
Shown in. This characteristic is also called the solitary wave response characteristic of the echo, but its amplitude does not become zero for a considerably long cycle.
Equation (4) is assumed to continue until NN + 1 cycles.

Substituting equations (1) and (4) into (2).

[0038]

[Equation 5]

Multiplying both sides of this equation by a (n-j)

[0040]

[Equation 6]

It becomes Considering how the right side of Eq. (6) changes over a long cycle, it is analyzed as follows. First, the received signal R (n) from the far end is included, but this signal is also pseudo-randomized by the scrambler, and this signal and the transmission signal a (n-j) randomized as described above. The average value of the product of and is 0 when viewed over a somewhat long period. Note that R (n) is no longer a discrete signal such as ± 1 or ± 3 due to the distortion of the transmission line,
Randomness is maintained. Also, regarding the second and third terms on the right side of the equation (6), the correlation between the transmission signals at different times n is 0, so the average value of the products is 0. Therefore,

[0042]

[Equation 7]

For the other terms, the probability of becoming positive and the probability of becoming negative are the same when viewed over a somewhat long period. As described above, many terms in Eq. (6) are equal in their probabilities of becoming positive and negative and adding them, i
The sum of all terms other than = j and k = j and the first term also has the same probability of becoming positive and negative.

Regarding the term for which (7) holds,

[0045]

[Equation 8]

It will be 0 if the following is true, but it will not be 0 other than that in the long term. When considered in light of the above, equation (3) by calculating the equation (6), the sign pulls the step size α positive if the current tap coefficient C _j, the current tap coefficients if the sign is negative C _j It means that the tap size α is added to. Therefore, in each cycle, the current tap coefficient C _j fluctuates positively and negatively depending on the values of the reception signal R (n) and the transmission signal from the far end, but in the long term, it depends on only the amplitude E _j , (8 ) It converges in the direction that the equation holds.

However, the current tap coefficient C _j derived from the equation (3) stochastically gradually converges to the correct value of the amplitude E _j by utilizing the randomness of the signal.
Assuming that the current tap coefficient C _j can be changed by α in the cycle, there is a case where it shifts to a value several times larger than α in both positive and negative directions with respect to the correct value in the short term even after convergence. In such cases,
The standard deviation σ, which is the average deviation, is

[0048]

[Equation 9]

It becomes as follows. Here, σ is a dispersion coefficient. As a result, 3σ, which is usually considered to be the worst case, is 7.35 · α, which means that there is a large deviation of about 3 bits from the accuracy of α. Therefore, the step size must be 3 to 4 bits smaller than the minimum bit required for fixed tap coefficients. This is because the maximum value of the coefficient deviation in the case of the fixed tap coefficient is only ± 1/2 of the minimum accuracy.

When the attenuation amount is 60 dB or more, since the word length of the tap coefficient determined by the rounding error as a transversal filter is about 16 bits, α has a value of 19 to 20 bits accuracy.

Therefore, in the present invention, a sufficient time has elapsed after the echo canceller started the pull-in, and the echo canceller is in a considerably good pull-in state, n ≧ E.
At N, change the algorithm.

In the case of n ≧ EN, the coefficient updating process is performed in place of the equation (3), for example,

[0053]

[Equation 10]

[0054]

[Equation 11]

[0055]

[Equation 12]

[0056]

[Equation 13]

The processing of is performed. Here, L is an integer greater than 2, and is usually a value of about 32 to 64, and γ is, for example, 0.25.
Is a positive number that is considerably smaller than 1. α is a step size, which is the minimum value allowed from the accuracy of the coefficient.

In the processing of equations (10) to (13), n ≠ n / L * L
In this case, β _j is the cumulative value of the polarities in Eq. (6) multiplied by α. Of course, β _j is cleared when n = n / L * L.
The tap coefficient is fixed while n ≠ n / L * L. When n = n / nL * L, the tap coefficient is updated by the value obtained by multiplying β _j by the proportional constant γ.

FIG. 2 is a flowchart of the first embodiment described above.
It is The shift register 6 shifts the transmission signal a (n).
Run every time. When execution starts, step S4
Then, the pseudo echo w (n) and the error e (n) are obtained. This
Step S4 is similar to conventional step S1. In addition, this
These are the adders 9 and 10 and the multiplier M in FIG.₀~ M _NTo
Therefore, it is required.

Then, it is determined in step S whether n is greater than or equal to EN.
Find in 5. When n is not equal to or greater than EN (NO), in other words, when n is less than En, the coefficient update control circuit 7 obtains the tap coefficient corresponding to each tap in step S6.
This step S6 is the same as the conventional step S2.
Then, after step S6, the processing corresponding to the clock of FIG. 1 is terminated. When n is less than EN, the same process as the conventional process is performed.

On the other hand, when it is determined in step S5 that n is equal to or greater than EN, it is determined in step S7 whether n = n / L * L. This formula is a formula for determining whether n is an integer multiple of L, and this formula holds when it is an integer multiple.

If it is not an integral multiple (NO), step S8
Is executed and the value of α · sgn [e (n) · a (nj)] is accumulated. That is, a process of storing B _j + α · sgn [e (n) · a (nj)] in B _j is performed. And C _j (n +
Let 1) be C _j (n). This means that the tap coefficient is not changed. Do this for 0 to N taps,
When step S8 ends, the process corresponding to one clock ends.

When n is an integral multiple of L (YES)
Has B accumulated so far_jΓ times (a positive constant)
Value is tap coefficient C_jThe value obtained by subtracting from (n)
Number C _j(N + 1) and further B_jTo clear. still,
This process is performed for 0 to N taps.

After step S9, the process corresponding to the clock of FIG. 1 is terminated. The above operation is sequentially executed corresponding to the clock. At this time, the tap coefficient is changed once every L times.

Taking L = 48 and γ = 0.25 as an example (1
Consider equation (0). From the above examination results, assuming a case close to the worst case of 3σ,

[0066]

[Equation 14]

It becomes It shows that when the coefficient is 16-bit precision, the tap coefficient may change by an amount close to the value corresponding to the minimum precision of the worst 15-bit precision.

As a result, when all the conventional processing is performed with 16 bits, the zigzag change up to 8 times the step size in the worst case is reduced to less than 2 times in the worst case, and the wraparound attenuation occurs. The quantity deterioration is also reduced.

As described above, the present invention can reduce the zigzag blurring caused by the stochastic processing by introducing the processing of averaging over L cycles while keeping the step size large. However, since the tap coefficient is updated only once in the L cycle, the update speed becomes slower than in the conventional case. However, it is already pulled in, and the tap coefficient is a value that is close to the target value, and after that, in the case of the step of making the step size smaller and making it more precise, and making the wraparound attenuation larger, , Its speed does not matter.

In the first embodiment, it is assumed that the processing of (10) to (13) is carried out for all tap coefficients, but it is also possible to carry out for only some tap coefficients. is there. It should be noted that the processing amount of these processes does not particularly increase as compared with the conventional tap coefficient updating process, but the required memory increases by introducing the parameter β _j .

In order to prevent this increase in memory, in the second embodiment, the processes of (10) to (13) are performed only for the taps that may have a large tap coefficient, and for the coefficients of other taps, For example, the step size is increased by updating the coefficient for a value that is eight times the true tap coefficient.

The shape of the solitary wave response characteristic of the echo, an example of which is shown in FIG. 4, varies depending on the length and type of the cable, the presence or absence of a branch cable, etc., but the amplitude is large at 7 to 8 taps in any case. Until now. If the maximum tap coefficient up to this point is ± 1.0, the tap coefficient after that is ± 1.0.
1/8 or less.

FIG. 3 is a flow chart of the second embodiment of the present invention. When the processing is started as described above, the pseudo echo w (n) is obtained in step S10, and the calculation at this time uses the equation (15). The value obtained by subtracting the pseudo echo w (n) from the echo y (n) is defined as the error e (n).

Then, it is determined in step S11 whether n is greater than or equal to EN. When n is not equal to or greater than EN (NO), the coefficient update control circuit 7 obtains the tap coefficient corresponding to each tap in step S12. Then, after step S12, the process corresponding to the clock in the figure ends.

On the other hand, when n is equal to or greater than EN, the step
In step S13, tap numbers with tap numbers from M to N
The value is the same as in equation (3). Then, n = n / L * L
If it is not satisfied (NO), it is determined in step S14.
, The tap coefficient from tap number 0 to M-1
Same as the value up to and B_jTo α ・ sgn (e (n) ・
a (n-j)] is added to B_jAnd And process
finish. Also, in step S14, n = n / L * L holds.
When step S16_j(N + 1) is C ₃-Γ
・ Β_jAnd further β_jIs set to 0. However, J
Is between 0 and M-1, and the tap coefficient C between them is_j(N
+1) and cumulative sum β of sine code_jTo update.

FIG. 3 shows a flowchart of the second embodiment. Here, for taps 0 to M-1, (1
Performing the processing of 0) to (13), assuming that the tap coefficient × B (B <1) is a true tap coefficient for M taps and thereafter.
In the product-sum processing for obtaining the output of the transversal filter of the echo canceller, the product-sum value after M taps is multiplied by B as shown in equation (15). In this way, the coefficient C _k after M taps becomes a large value, which is 1 / B times as large as that in the conventional case. Therefore, when the same α as that in the conventional case is used for that value, a true tap coefficient is obtained. For a certain B · C _k , α · B is equivalently the step size. For example, by setting B = 1/8, the substantial step size can be made 1/8, and therefore the deterioration of the wraparound attenuation amount caused by the stochastic processing can be reduced.

[0077]

[Equation 15]

In the above description of the present invention, the sign algorithm is used for all coefficient updates.
In fact, it is also possible to omit the sign process in equations (3) and (10). In some cases, the first step of the present invention is further divided into two sub steps, the sign process is omitted in the first step, and the sign algorithm is executed in the latter step. It is needless to say that the present invention is effective in such a case and the case where the sign code is not included is included in the scope of the claims of the present invention.

[0079]

By applying the present invention, the coefficient word length of the signal processing portion is, for example, 20 b because of the conventional echo canceller.
It can be shortened from it to a long one, for example, 16 bits, which has the effect of reducing the circuit scale of not only the memory but also the arithmetic circuit.

Since the processing speed can be increased when the low bit processing becomes possible, there is also an effect that the hardware design becomes easy. Furthermore, since 16-bit processing is possible, a digital subscriber line transmission interface device can be constructed by a general-purpose digital signal processor.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an echo canceller according to an embodiment of the present invention.

FIG. 2 is a processing flowchart of the first embodiment.

FIG. 3 is a processing flowchart of a second embodiment.

FIG. 4 is a diagram showing an example of a solitary wave response characteristic of an echo.

FIG. 5 is a processing flowchart of a conventional echo canceller.

[Explanation of symbols]

 1, 11 terminals 2 D / A conversion circuit 3 hybrid circuit 4 cable 5 A / D conversion circuit 6 shift register 7 coefficient update control circuit 9, 10 adder

Claims (4)

[Claims] 1. A transversal filter type echo canceller of a transmission apparatus, comprising a first step of updating a coefficient every cycle by a first coefficient updating algorithm after starting the pulling, and after a certain period of time or a certain pulling in. A coefficient control method for an echo canceller, comprising a second step of updating each tap coefficient only once in a plurality of cycles by a second coefficient updating algorithm after at least one of the states is reached. 2. The first coefficient updating algorithm changes by a specific value corresponding to an update sign which is a sign of a multiplication result of a signal corresponding to each tap and an output error of the echo canceller. The coefficient control system of the echo canceller according to claim 1. 3. The second coefficient update algorithm adds the update signs over a plurality of cycles for each tap, and subtracts the result multiplied by a weight from the original coefficient. The coefficient control system of the echo canceller according to claim 2. 4. Among the taps of the echo canceller of the transmission device, the first and second steps are performed only for the taps whose absolute value of the tap coefficient value may exceed a certain value, and for the other taps. The coefficient control system of the echo canceller according to claim 1, wherein the tap coefficient is updated every cycle by the first coefficient updating algorithm.