JPH0646008A - Decimation filter for oversampling type a/d converter - Google Patents

Abstract

PURPOSE:To provide a decimation filter of the oversampling type A/D converter where the circuit scale of the decimation filter and the waveform shaping filter is reduced. CONSTITUTION:A tap coefficient is set which expands zero points related to the characteristic of the pass band to the transmission function where at least one or two of these zero points are arranged in the vicinity of z=1 of the (z) plane and at least one of zero points is arranged on the outside of a unit circle of the (z) plane with respect to a transversal decimation filter 12 which takes a signal, which has n-fold speed ((n) is a positive integer) of the fundamental sampling frequency and consists of a small number of bits, as the input and removes the noise component of a high frequency and performs waveform shaping and thinning processing and outputs a signal of the fundamental sampling frequency, and this over-sampling type decimation filter is characterized by having the waving characteristic in which a polarity opposite to the main response exists before the main response with respect to the impulse response characteristic.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a transversal filter, and more specifically, it removes high frequency noise of a signal converted to digital data roughly at a frequency sufficiently higher than a basic sampling frequency and compared to a required accuracy. Then, the present invention relates to a decimation filter for converting into high-precision digital data at a speed of 1 to 2 times the basic sampling frequency.

[0002]

2. Description of the Related Art Over-sampling is used for signal reception of a digital subscriber line transmission interface device that transmits or receives high-speed digital data bidirectionally simultaneously or alternately by using an existing telephone subscriber line metallic pair cable. An A / D converter is used.

In the above-mentioned high-speed digital subscriber line transmission interface device, for example, four-valued amplitude codes corresponding to ± 1 and ± 3 are transmitted at a transfer rate of 80 KB / S. A signal that has been attenuated or waveform distorted due to the loss characteristics of the metallic pair cable is digitized by the signal receiving A / D converter of the partner device, and the attenuation distortion and the waveform distortion are corrected by digital processing. It

Further, due to the relationship between the size of hardware and the conversion accuracy of the A / D converter, an oversampling type A / D converter is used in a four-valued device having a transfer rate of about 80 KB / S.

The oversampling type A / D converter converts a signal into a digital signal roughly at a sampling frequency sufficiently higher than a basic sampling frequency determined by the baud rate frequency of the signal, and coarser than required accuracy. Then, it is subjected to a low-pass digital filter to remove high-frequency noise, and is converted into high-accuracy digital data at a speed of 1 or 2 times the basic sampling frequency or less. This filter is generally called a decimation filter.

The conventional decimation filter in the above-mentioned digital subscriber line transmission interface device is usually (1)
The transversal filter expressed by the formula is used. The characteristics of the stop band of this filter are the characteristics of a so-called comb-type filter in which the peaks of the loss characteristics due to m-fold zeros are arranged at equal intervals. Since the characteristics of the pass band of this filter are such that the loss increases as the frequency increases, a waveform shaping filter is required at the subsequent stage of this filter when flattening or controlling the pulse waveform.

[0007]

[Equation 1]

Where z is z = exp (j2πf / (nf _s )),
m is an integer of 2-3, n is an oversampling ratio, f _s
Is the basic sampling frequency, and f is the frequency of the signal.
The first precursor of the isolated pulse response characteristic of the received signal is used to extract the timing of the received signal.

Further, in order to use the first precursor for timing extraction, the sign of the isolated pulse response characteristic must change from negative to positive in the first precursor. Furthermore, it is desirable to make the post cursor small in order to reduce the load on the decision feedback equalizer in the subsequent stage.
The waveform shaping filter is used so that the pre-cursor and the post-cursor of the isolated pulse response characteristic have desired shapes.

Usually, a high-pass filter for cutting off a DC component is inserted after the waveform shaping filter.
Since this filter may be operated at the basic sampling frequency, it is processed by the digital signal processing unit together with the functions of the decision feedback equalizer and the like.

The input section of the conventional digital subscriber line transmission interface device is, as shown in FIG. 10, a ΔΣ converter 1, a decimation filter 2, a waveform shaping filter 3, an echo canceller, etc. in an oversampling A / D converter. Circuit 4.

The received signal is input to the ΔΣ converter 1 through a simple low-pass filter (not shown). The received signal is ΔΣ
For example, the converter 1 has an oversampling frequency of 15.36M.
Converted to a 1-bit digital signal at Hz. This high-speed data is converted into low-speed multi-bit data by the decimation filter 2 having the transfer function of equation (1). However, it does not drop to the baud rate frequency. This is because the filter must be operated at a frequency twice or more the baud rate frequency in order to perform waveform shaping such as shaping of the cursor portion with the required accuracy. That is, the decimation filter 2 outputs data at 160 KHz intervals, that is, at 62.5 μs intervals, which is at least twice as high as the basic sampling frequency of 80 KHz, which is the baud rate frequency in the digital subscriber line transmission interface device. A specific example of equation (1) is shown in (2).

[0013]

[Equation 2]

Since the numerator of the formula (2) includes a term of 1-z ^-1 as a factor, the formula (2) becomes a function of only the numerator as a result. It becomes a tap transversal filter. 286 is obtained from (96-1) × 3 + 1.

[0015]

[Equation 3]

When the arrangement of zeros is written on the z plane, triple zeros are arranged at equal intervals on the unit circle except around z = 1 as shown in FIG. On the other hand, the waveform shaping filter is a transversal filter with about 6 to 8 taps that operates at a frequency twice the baud rate. Output is the baud rate frequency, namely 80K
Once a Hz is sufficient.

The transfer function of the waveform shaping filter is, for example, (4)
It becomes like a formula.

[0018]

[Equation 4]

Here, a _{0 to} a ₇ are tap coefficients of the transversal filter. FIG. 12 is a loss frequency characteristic diagram of a low frequency portion of a filter including a conventional decimation filter and a waveform shaping filter. The loss in the portion of 30 to 40 KHz is lower than the loss of direct current, and on the contrary, it has a little gain and becomes the stop band above 60 KHz. From a macro perspective, the higher the frequency, the greater the amount of attenuation. Figure 1
Although not shown in Fig. 2, the maximum is around 7.68 MHz.

The peak of the attenuation amount near the frequency of an integral multiple of 160 KHz in FIG. 12 is due to the zero point of the decimation filter, and the other peaks are due to the zero point of the waveform shaping filter. The reason why the peak of the frequency that is an integral multiple of 160 KHz is larger than the peaks of the bands other than the above is that the zero point of the decimation filter is the triple zero point.

A large amount of attenuation is locally obtained because of the triple zero, but the amount of attenuation at the frequency between the respective zeros is small, so that noise reduction, which is the purpose of the decimation filter, is suppressed. From this point of view, it is not efficient, and in order to reduce the noise below a certain value, the size of the filter, that is, the order must be increased.

FIG. 13 is a block diagram of a conventional decimation filter. Since the output of the ΔΣ converter 1 is only ± 1 binary value, the calculation of the equation (3) can be replaced by the addition / subtraction of the tap coefficient. In addition, since the input signal is added to the decimation filter at a cycle of 15.36 MHz, it is difficult to add and subtract multiplex signals. Therefore, in order to output once at 160 KHz, 285 /
From (15360/160) = 2.97, three adders are required.
For this reason, the processing of Eq. (2) is divided into three and calculated as in Eq. (5).

[0023]

[Equation 5]

Then, each term is executed by different hardware. FIG. 13 is a hardware configuration diagram of the decimation filter. The process of the first item of the equation (5) is processed in the first stage S1 of FIG. 13, and when the process is finished, the result is transferred to the register R2, and the process corresponding to the two items of the equation (5) is further illustrated. Second stage S2 in the middle part of 12
To process. Similarly, the result is transferred to the register R3, the remaining processing is processed in the third stage S3 of the circuit at the right end, and the result is used as the output of the decimation filter. FIG.
The tap coefficients are stored in the ROM M1 to ROM M3 of the circuit of FIG.
The process of inverting (subtracting) or adding as it is (adding) is performed. The circuit scale of this filter is about 1700.
It is a gate.

Further, the decimation filter of FIG. 13 will be described in detail. The ROM M1, the selector S1, the adder A1, and the register R1 are blocks for calculating the first term of the equation (5). ROM M2, selector S2, adder A2,
The register R2 is a block for calculating the second term of the expression (5), and the ROM M3, the selector S3, and the register R3 are (5).
This is a block for calculating the third term of the equation.

The expression (5) is expressed by Z ^-1 , but in the actual processing, the signal data input at 15.36 MHz intervals is X _i , the filter coefficient is b _i , the product is calculated 288 times, and the product of the filter is calculated. This is the output. Since the actual input is only +1 or -1, the sum of products processing can be processed by the adder. 15.3
If you use an adder that can add once to 6MHz, 160KHz
Since you can add only 96 times during the time corresponding to
It is necessary to operate three adders in parallel in order to obtain the output every 160 KHz. For this, filter calculation is 3
It is divided into terms and processed. The first term of the equation (5) adds the filter coefficients of 0 to 95, and the second term of 96 to 19
1 is added and the third term has a filter coefficient of 19
The part of 2-287 is added.

In FIG. 13, filter coefficients 0 to 95 are stored in the ROM M1, filter coefficients 96 to 191 are stored in the ROM M2, and filter coefficients 192 to 287 are stored in the ROM M3. If the input signal X _i is positive, the coefficient from the ROM is input to the adder as it is, and if it is negative, the input from the ROM is inverted and added. The addition result is fed back to the input of the same adder and added 96 times.

Normally, the selector S4 and the selector S5 are set so that the output of the register R2 is input to the adder A2 and the output of the register R3 is input to the adder A3. However, only in the cycle corresponding to a multiple of 96, the selector S4 and the selector S5 respectively input the output of the register R1 to the adder A2 and the output of the register R2.
3 is set to be input. At the same time, the output of the register R3 is output as the output of the filter. Further, in this cycle, the route from the register 1 to the adder 1 is cut and the adder A1 starts addition from 0.

As a result, the content of the register R2 becomes the result of addition processing of the second term in addition to the addition result of the first term of the expression (5), and the content of the register R3 is the expression of the expression (5). The addition processing of the third term is performed in addition to the adders of the first term and the second term, and the addition result of the 96th addition of the register 3 becomes the calculation result of the expression (5). The time chart during this period is as shown in FIG.

FIG. 15 is a circuit diagram of a conventional waveform shaping filter. Although the same transversal filter is used, the input signal of the waveform shaping filter (output of the decimation filter) is different from that of FIG. 13 at 160 KHz.
This is because it is 18-bit data and not 1-bit input like the decimation filter. The tap coefficients a _{0 to} a ₇ of this filter require an accuracy of about 10 bits,
The processing of Eq. (4) is the sum of products operation. The multiplication process requires a multiplier, but the hardware scale of the multiplier becomes large. Therefore, in the circuit of FIG. 14, 256 kinds of output data when the input data is assumed to be 1 bit are stored in the ROM. If the input data is 18 bits, the solution corresponding to each digit of the input data is shifted by 1 bit and the output is obtained by adding 18 times. The input of this filter is 160 KHz
Thus, the output is once every 80 KHz, and one adder can be used multiple times. The circuit scale of this filter is 1500 gates excluding ROM.

The waveform shaping filter of FIG. 15 is used in the calculation of equation (4). Input the equation (4) into the data series X _i (i =
Rewriting using 0, 1, ..., 7) gives the following equation.

[0032]

[Equation 6]

B_ijIs 0 or 1, so 8 taps
In case of transversal filter C _ijIs 2⁸= 256 ways
Can take the value of. 256 values are stored in ROM
Save and correspond to each digit of 8 input signal data
C with bit data as address_ijFrom ROM MA4
By reading, C corresponding to each digit of the input signal string _ij
Can be output one after another to the adder.

In FIG. 14, eight consecutive signal data are stored in the registers RA1 to RA8. For example, the lower bit of each register is sequentially read from the ROM every 15.36 MHz.
It is input to the address of MA4 and the corresponding C _ij is output.

^Next , C _ij must be multiplied by 2 ^-j to obtain the sum. For example, when processing is performed from the lower bit of the data, the output of the register RA0 containing the addition result is set to 1
It is sufficient to shift down by one bit and add C _ij corresponding to the next bit. Since the result of adding the input data by the number of digits is the output of the filter, the direction of the selector is changed and the result is output.

Output is 80KHz If input is 160KHz, 80KH
The above processing is performed once by shifting the input data twice for each z. FIG. 16 is a coefficient map of a conventional filter including a conventional decimation filter and a waveform shaping filter, which corresponds to an impulse response.

FIG. 17 and FIG. 18 are the output waveforms of the total filter of the conventional decimation filter and waveform generation filter in the map of FIG. 16 when the signal pulse that has passed through the longest cable is input. Pulse response characteristics, isolated pulse response characteristics when the cable length is zero.

The isolated pulse response characteristic when a signal that has passed through the longest cable is input is obtained by passing the output of the waveform shaping filter through a high-pass filter. Ie 1
It is the result of adding the process of -D.

The isolated pulse response characteristic when the cable length is zero is 1 / (1-0.8) in addition to 1-D processing.
75D). Where D is D = z ^-1
= Exp (-j2πf / 80) f: frequency (KHz). The filter coefficient of (1-D) or (1-D) / (1-0.875D) is a number whose filter coefficient is represented by 4 bits at most, so it is processed by a digital signal processing circuit without a multiplier in the subsequent stage. There is no problem in doing.

Further, since the conventional decimation filter is intended to suppress high frequency noise, its impulse response is the impulse response of a normal low pass filter, and as shown in FIG. There was only a response. Since there is no ringing before the main response in the characteristics of this figure, the precursor cannot be used for timing adjustment.

[0041]

As described above, the hardware of the decimation filter in the oversampling A / D converter of the conventional digital subscriber line transmission interface device and the hardware of the waveform shaping filter in the subsequent stage is about a total. A large circuit of 3200 gates was required.

The digital signal processing circuit in the latter stage for performing the processing of the above-mentioned high-pass filter, echo canceller, decision feedback equalizer, etc. must operate at a high speed of 15.36 MHz. Moreover, the processing of the decimation filter cannot be incorporated. Furthermore, since it does not have a multiplier, it cannot take in the processing of the waveform shaping filter.
If you have a multiplier, you can execute the processing of the waveform shaping filter, but the multiplier circuit scale must be 3000 gates or more,
On the contrary, the circuit scale is increased.

Therefore, the decimation filter and the waveform shaping filter must be realized as a dedicated circuit, which is one of the factors that hinder the reduction of the hardware scale of the entire apparatus and the reduction of the power consumption, and the reduction of these circuits. Was desired.

An object of the present invention is to provide a decimation filter for an oversampling A / D converter in which the circuit scale of the decimation filter and the waveform shaping filter is reduced.

[0045]

According to the present invention, a signal having a low bit number is input at a speed n times as high as a basic sampling frequency, and noise components having a high frequency are removed, waveform shaping and thinning and thinning processing are performed. , In a transversal type decimation filter that outputs a signal at the speed of the basic sampling frequency. Then, one or two points of the transversal-type taps relating to the characteristics of the pass band are arranged in the vicinity of Z = 1 on the Z plane, and at least one of the points is set in the unit circle of the Z plane. Set it so that the transfer function is placed outside.

By setting each of the transversal type taps as described above, the impulse response characteristic has a wave having a polarity opposite to that of the output response before the output response. Then, for example, the main points related to the stop band are arranged as one important point on the unit circle of the Z plane at substantially equal intervals. As a result, the same amount of attenuation is obtained while suppressing the increase of the order to twice or less.

[0047]

FIG. 1 is a basic configuration diagram of an embodiment of the present invention.
The central part is a decimation filter that also performs waveform shaping. This decimation filter has a transversal filter configuration like the conventional decimation filter.

In the decimation filter of the present invention, the impulse response, that is, the tap coefficient map is made negative once before the main response, and is designed so as to be smoothly connected so as to block higher frequency components. That is,
The decimation filter has the function of the waveform shaping filter. The decimation filter having the transversal filter has an order twice as large as that of the conventional one, but since the processing speed is 1/2, the hardware amount is almost the same.

Also, a decimation filter and waveform shaping
Decimation filter and wave using filter function
Forming filter is integrated. For equations (3) and (4)
Z ^-1If you take the product considering the difference of

[0050]

[Equation 7]

Which is a 957 tap transversal filter operating at 15.36 MHz. Since the output in this case is 80 KHz, the number of adders is 957 / (15360/80) =
It becomes 5 from 4.98. In other words, the circuit size is 5/3 times that of the circuit in FIG. 13, so the circuit size is about 2830 gates, which is smaller than 3200 gates. However, the scale of the ROM storing the tap coefficient becomes large.

The impulse response characteristic at this time is shown in FIG.
It is similar to 6. It is well known that the tap coefficient map in which the values of the tap coefficients of the transversal filter are graphed in order from the input side matches the impulse response. FIG.
The impulse response of the coefficient map shown in (2) has two ringings before and after the main response. The ringing on the inner side has a similar shape to the ringing characteristic on the front part of the isolated response characteristic shown in FIGS.

As described above, simply integrating the conventional decimation filter and the waveform shaping filter cannot sufficiently reduce the circuit scale. In the present invention, the amplitude characteristic and the phase characteristic of the pass band are controlled as follows so that the decimation filter operating at the oversampling frequency also has a waveform shaping function in addition to the cutting of high frequency components.

With respect to the transfer function of the filter, the zero points in the stop band are not overlapped. For example, single zeros are arranged at equal intervals from the vicinity of the basic sampling frequency to the oversampling frequency. As a result, the same amount of attenuation can be obtained while suppressing the increase of the order by 2 times or less as compared with the conventional case.

The characteristics of the pass band are designed in consideration of the impulse response. That is, the isolated pulse response characteristic is once set to a negative value before it reaches a peak value. Since the isolated pulse response characteristic and the impulse response correspond well to each other, the negative value should be set to a negative value before the impulse response of the filter reaches a peak value. Further, in order to make the impulse response have a negative value before reaching the peak value, one or two of the zero points of the transfer function of the filter, which are mainly related to the characteristics in the vicinity of the pass band, are placed on the real axis of the z plane. Only the two zeros are arranged outside the unit circle. As a result, the isolated pulse response waveform at the output of this filter can be shaped so as to rise once it has a negative value.

As a method different from the above, only two zeros of the transfer function of the filter, which are mainly related to the characteristics in the vicinity of the pass band, are arranged as conjugate complex roots near the point z = 1 on the z plane. Moreover, these two zeros are both placed outside the unit circle. As a result, the isolated pulse response waveform at the output point of the filter once becomes a negative value and then rises.
Furthermore, the amplitude at the time (second pre-cursor) before the time point corresponding to the baud rate from the first pre-cursor point, which is the time at which the value changes from negative to positive, can be surely set to zero.

A more specific description will be given. In general, the impulse response of the low pass filter of the minimum phase shift circuit is accompanied by ringing after the main response. However, there is no ringing before the main response, and the fact that the impulse response becomes negative before its amplitude reaches its maximum value is that there is ringing before the main response, which means that the filter has a minimum phase shift circuit. Is not equivalent to

Therefore, the shape of the first cursor portion of the isolated pulse response characteristic is made negative before the main response, so that the transfer function of the filter is not a minimum phase shift circuit. That is, at least one zero of the transfer function is provided outside the unit circle. In the transfer function of the conventional decimation filter, the zero point mainly relating to the characteristic of the pass band is not arranged as described above, but in the decimation filter of the present invention, one or two zero points mainly relating to the characteristic of the pass band are arranged, Further, at least one of them is arranged outside the unit circle, and the decimation filter is a filter which is not the minimum phase shift circuit.

At the same time, in order to secure the amount of attenuation in the stop band, all zero points on the unit circle are set as single zero points and are arranged at substantially equal intervals except for the pass band. When the zeros are arranged as described above, the transfer function of the filter is 2 with respect to the maximum z ^{−1 for} each term.
Since it is expressed as a product of terms that are the following, Σ [C _i z
^-1 ] to obtain the tap coefficient C _i .

Regarding the position of the zero point mainly relating to the characteristics of the pass band, it is not concrete to arrange at least one outside the unit circle, but it is temporarily arranged near Z = 1 and the actual isolated pulse is generated. The waveform is input to the filter and the output waveform is viewed to determine the placement. That is, while paying attention to the isolated pulse response characteristic, the desired value is obtained by correcting the position of the zero point.

Generally, when there is only one zero point related to the pass band, it must be on the real axis of the z plane, so that the degree of freedom is only one, and the first pre-cursor point of the isolated pulse response characteristic and the main cursor. Find the position so that the (time) interval of the points is the desired size. Assuming that the first pre-cursor point is the point of zero amplitude at the forefront of the main response of the isolated pulse response characteristic, the time delayed by the time corresponding to the baud rate cycle with respect to that point is the main cursor. The position of the zero point is determined so that the amplitude is sufficiently large, that is, the main cursor point is near the center of the main response.

When there are two zero points related to the pass band,
If they are placed only on the real axis and one is placed outside the unit circle and the other zero is placed inside the unit circle, and the two zeros are so-called conjugates that are symmetric about the real axis outside the unit circle. There are cases where they are placed in different positions. In the former case, the position of the zero point is determined outside the unit circle so that the position of the main cursor is close to the position where the main response peaks and the amplitude is sufficiently large, as in the case of one piece. Since the position of the zero point in the unit circle affects the waveform behind the main response, it is controlled by observing its shape. If the zero points on the real axis are arranged so that they are mirror images of each other with respect to the unit circle, the impulse response becomes bilaterally symmetric and a tap coefficient distribution that is easy to handle is obtained. In the latter case, since the position is a complex number, there are two degrees of freedom, and the distance between the main cursor and the first pre-cursor and the distance between the first pre-cursor and the second pre-cursor can be controlled. The second precursor is an amplitude at a time advanced by a time corresponding to the baud rate cycle from the above-mentioned first precursor, and the amplitude of the second precursor point is preferably zero.

When the zero point mainly related to the pass band is arranged, the attenuation amount in the stop band is generally smaller than that when it is not arranged. However, in the present invention, the number of zero points related to the characteristics near the pass band is one or two. The number of taps can be reduced to less than twice that of the conventional one, even though it has a waveform shaping function by limiting the number to taps.

In the decimation filter of the present invention, the number of taps is twice as large as that of the conventional decimation filter, but a waveform shaping filter operating at a frequency twice or more of the basic sampling frequency, which is essential in the conventional system, is used. Since it is not necessary, the decimation filter of the present invention may output the output cycle twice as long as the conventional one, that is, the basic sampling frequency. However, although the ROM capacity is doubled in the hardware of the decimation filter shown in FIG. 12, the scale of other circuits is the same as that of the conventional decimation filter. Hereinafter, a detailed description will be given using two examples.

[0065]

[Equation 8]

[0066]

[Equation 9]

Where z ⁻¹ = exp (j2πf / 5360),
Π (1 + a _n z ^-1 + b _n z ^-2 ) is a function having a single zero point for each frequency equally spaced from the basic sampling frequency to the vicinity of the oversampling frequency.

In the equation (7), a ₀ = 1-α, 1 >>α>
If it is 0, the first and second terms on the right side of Π _A (Z ^-1 ) are terms that have a strong influence in the vicinity of the passband. Because the root of the first term is a
It is ₀ , and the root of the second term is 1 / a ₀ , and therefore both are arranged in the vicinity of z = 1. Further, the root of the first item is the zero point inside the unit circle, the root of the second term is the zero point outside the unit circle, and the two zero points are both on the real axis, and the product is
Since it is 1.0, it has a mirror object relationship with the unit circle.

On the other hand, the zero point of the first term on the right side of H _B (Z ⁻¹ ) is located at 1 + α ± jβ (where 1 >> α, β> 0), that is, in the vicinity of z = 1 outside the unit circle. It is assumed that two complex zeros are arranged at.

Loss characteristics of the decimation filter having the function shown in the equation (7) (the loss at f = 0 is standardized to be 0 dB), impulse response (equal to the coefficient map of the transversal filter) , Fig.2, Fig.3, Fig.4 and Fig.5 show the isolated pulse response characteristic which is the output waveform when the signal pulse that has passed through the longest cable of 7.5km is input and the isolated pulse response characteristic when the cable length is zero. Show.

Loss characteristics of the decimation filter having the function shown in the above equation (8) (normalized so that the loss at f = 0 is 0 dB), impulse response (equal to the coefficient map of the transversal filter) , FIG. 6, FIG. 7, FIG. 8 and FIG. 9 show an isolated pulse response characteristic which is an output waveform when a signal pulse that has passed through the longest cable is input, and an isolated pulse response characteristic when the cable length is zero.

Comparing the loss characteristics of the decimation filter of the present invention shown in FIGS. 2 and 6 with the loss characteristics of the conventional filter (decimation filter + waveform shaping filter) shown in FIG. 12, the loss characteristics of the filter of the present invention are 70 to 4000 KHz. More attenuation is obtained, and higher attenuation is obtained with the conventional filter at higher frequencies. A / D
Since the S / N characteristic of the converter is determined by the sum of the leakage of noise from all frequencies, almost the same S / N characteristic can be obtained in this case.

The isolated pulse response characteristic shows what the signal waveform of the A / D converter looks like with respect to the transmission pulse from the far end. The waveform must be such that interference can be zero. That is, with the first pre-cursor point as the reference in the time direction, the amplitude at each post-cursor point following the main cursor must be 0.5 or less with respect to the amplitude of the main cursor point, which is the sampling time immediately after that. Further, the amplitude at the second pre-cursor, which is the sampling point before the first pre-cursor, cannot be corrected in the subsequent stage, so it must be as small as possible.

The above conditions are satisfied in the cases of FIGS. 4 and 5 and in the cases of FIGS. 8 and 9, and are similar to those of the conventional isolated pulse response characteristics of FIGS. It has excellent characteristics and there is no problem from the viewpoint of waveform transmission. However,
In the case of FIG. 5, the amplitude at the second pre-cursor point deviates by about 0.03 when the peak value is 1, but there is no influence on the error rate at this level.

Next, the coefficient map will be described. In the coefficient map of Fig. 3, there is one ringing before and after the main response, which is one of the two zero points related to the passband in Eq. (7).
This corresponds to that one is placed outside the unit circle and the other is placed outside the unit circle. Also, the coefficient map of FIG. 7 has two ringings only before the main response. this is
This corresponds to the fact that the two zero points related to the passband in Eq. (8) are both located outside the unit circle. Comparing FIG. 3 with FIG. 4 and FIG. 5 and comparing FIG. 7 with FIG. 8 and FIG. 9, the isolated pulse response characteristic can be obtained by setting a negative value in the coefficient map before the main response. At, the amplitude at the first precursor cursor point can be changed from negative to positive.

To summarize the present invention, the function of the decimation filter of the present invention is, for example, equations (7) and (8).

[0077]

[Equation 10]

As described above, the transversal filter process is performed, and when this is divided into three terms, the following equation is obtained.

[0079]

[Equation 11]

In the equation (10), the number of terms remains the same, but the number of terms in each term is double that in the case of the equation (5). The conventional decimation filter had to output every 160 KHz, but in the case of equations (7) and (8), it is only necessary to output every 80 KHz, so the processing time for each term is twice as long as before. Is.

Therefore, in the conventional decimation filter, the selector 4 and the selector 5 operate every 96 cycles, that is, every 160 KHz, but in the case of the present invention, they operate every 192 cycles. In addition, ROM is conventionally R
OM is 96 words, but the present invention requires 192 words. The hardware of the ROM can be configured to be sufficiently small so that it does not pose a problem.

[0082]

In the case of the embodiment, the decimation filter of the present invention is a 573tap transversal filter with an output of 80 KHz, so that the conventional decimation filter is 160
The scale of the hardware is almost the same as that of a 286 tap transversal filter with KHz output (because the speed is halved, multiplexing processing is possible). Therefore, it is possible to reduce the hardware equivalent to the conventional waveform shaping filter. The waveform shaping filter needs a logic circuit of about 1500 gates in the case of the above-mentioned conventional example of 160 KHz 8 tap, and it has an effect of surely reducing the number of circuits.

[Brief description of drawings]

FIG. 1 is a basic configuration diagram of an embodiment of the present invention.

FIG. 2 is a diagram showing a loss characteristic of the decimation filter of the present invention having one zero outside the unit circle.

FIG. 3 is a coefficient map (impulse response) of the decimation filter of the present invention having one zero outside the unit circle.
FIG.

[Fig. 4] Isolated pulse response characteristics of a decimation filter having one zero outside the unit circle (cable length 7.5Km amplitude is
It is a figure which shows the standardization with a peak value.

FIG. 5: Isolated pulse response characteristics of a decimation filter with one zero outside the unit circle (cable length 0 km amplitude is
It is a figure which shows standardization with a peak value.

FIG. 6 is a diagram showing a loss characteristic of a decimation filter having two zeros outside the unit circle.

FIG. 7 is a diagram showing a coefficient map (impulse response) of the decimation filter of the present invention having two zeros outside the unit circle.

FIG. 8: Isolated pulse response characteristics of a decimation filter having two zeros outside the unit circle (cable length 7.5Km amplitude is
It is a figure which shows the standardization with a peak value.

FIG. 9: Isolated pulse response characteristics of a decimation filter having two zeros outside the unit circle (cable length 0 km amplitude is
It is a figure which shows standardization with a peak value.

FIG. 10 is a diagram showing an input unit of a received signal of a conventional digital subscriber line transmission interface device.

FIG. 11 is a diagram showing a zero point arrangement of a transfer function of a conventional decimation filter.

FIG. 12 is a diagram showing loss frequency characteristics of a conventional filter including a decimation filter and a waveform filter.

FIG. 13 is a hardware configuration diagram of a decimation filter.

FIG. 14 is a timing chart of a decimation filter.

FIG. 15 is a configuration diagram of a waveform shaping filter.

FIG. 16 is a diagram showing a coefficient map (impulse response) of a conventional filter including a decimation filter and a waveform shaping filter.

FIG. 17 is a diagram showing an isolated pulse response characteristic (cable length 7.5 Km amplitude is standardized by a peak value) of a conventional filter including a decimation filter and a waveform shaping filter.

FIG. 18 is a diagram showing an isolated pulse response characteristic (cable length 0 Km amplitude is standardized by a peak value) of a conventional filter including a decimation filter and a waveform shaping filter.

FIG. 19 is a diagram showing an example of a coefficient map of a conventional decimation filter.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Nobukazu Koizumi 1015 Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture, Fujitsu Limited (72) Inventor, Yutaka Awata 1015, Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture, Fujitsu Limited

Claims (6)

[Claims] 1. A signal having a low bit number is input at a speed n times (a positive integer) the basic sampling frequency, noise components of high frequency are removed, waveform shaping and decimation are performed,
In a transversal-type decimation filter that outputs a signal at the speed of the basic sampling frequency, one or two zero points relating to the characteristics of the pass band are arranged near z = 1 on the z plane, and the zero point A tap coefficient that expands at least one of them into a transfer function arranged outside the unit circle on the z-plane is set, and the impulse response characteristic has a waviness characteristic with a polarity opposite to that of the main response before the main response. Decimation filter for oversampling A / D converter. 2. The amplitude of the second precursor of the isolated pulse response characteristic is set to zero before the amplitude of the impulse response becomes maximum, and the amplitude of the second precursor of the isolated pulse response characteristic is set to zero. The decimation filter according to claim 1, wherein the tap coefficients are set so that the values have the same polarity. 3. The decimation filter according to claim 1, wherein the zero points related to the stop band in the transfer function are arranged as single zero points at substantially equal intervals on the unit circle of the z plane. 4. A zero point relating to a characteristic of a pass band arranged outside the unit circle is arranged inside and outside the unit circle so as to have a mirror image relationship with respect to the unit circle. The decimation filter according to claim 1. 5. The decimation filter according to claim 1, wherein one zero point mainly arranged on the circumference of the unit circle and relating to the characteristics of the pass band is provided outside the unit circle. 6. The decimation switch according to claim 2, wherein two zero points mainly arranged on the circumference of the unit circle and relating mainly to the characteristics of the pass band are arranged outside the unit circle.