WO2003075482A1

WO2003075482A1 - Automatic equalizer and coefficient training method thereof

Info

Publication number: WO2003075482A1
Application number: PCT/JP2002/001085
Authority: WO
Inventors: Mitsuo Kakuishi; Nobukazu Koizumi; Hideyuki Araki
Original assignee: Fujitsu Limited
Priority date: 2002-02-08
Filing date: 2002-02-08
Publication date: 2003-09-12

Abstract

A coefficient update is performed by assuming a denominator function portion of a transfer function of an IIR (Infinite Impulse Response) type equalizer as a part of a numerator function portion of a transfer function of a FIR (Finite Impulse Response) type target equalizer (53, 54), so that a coefficient of the IIR type equalizer can be calculated stably and adaptively. In order to sufficiently compensate a steep gain valley, in addition to the FIR type equalizer (2), it is possible to use the IIR type equalizer which has not been able to be used conventionally.

Description

Description Automatic equalizer and its coefficient training method

The present invention relates to an automatic equalizer and a coefficient training method thereof, and is suitable for use, for example, in an apparatus for receiving a signal transmitted by a discrete multi-tone (Discrete Multi-Tone) modulation method using a subscriber line pair cable as a transmission path. The present invention relates to an automatic equalizer and its coefficient training method. Background art

In recent years, multimedia services such as the Internet have become widespread throughout society, including ordinary households, and the early provision of economical and highly reliable digital subscriber line transmission technology for using such services. Is strongly required. Here, enormous costs and time are required to lay a new communication line, and various methods for performing high-speed data communication using existing communication lines have been proposed.

For example, in recent years, XDSL (Digital Subscriber Line) has become widespread as a digital subscriber line transmission technology using an existing telephone line as a high-speed data communication line. This xDSL is a transmission method using the existing subscriber line (pair cable) and is one of the modulation and demodulation technologies. x DSL is broadly divided into the upstream transmission speed from the subscriber's home (hereafter, the subscriber side) to the accommodation station (hereafter, the office side) and the downstream transmission speed from the office side to the subscriber side. Are categorized as symmetric and asymmetric.

For example, a symmetric type has a high-bit-rate DSL (HD SL) with an uplink and downlink transmission speed of about 1.5 to 2.0 Mbps (megabit-noise), and a 160 k to 2 There are SDSL (Single-line DSL) at about 0 Mbps, and the asymmetric type is AD SL (Asymmetric DSL), which has recently been rapidly spreading in Japan.

Here, the AD SL further includes "G.dmt" having a downlink transmission speed of about 8 Mbps and "G.lite" (also called a simplified AD SL) having a downlink transmission rate of about 1.5 Mbps. , In both cases, a specific modulation scheme called DMT (Discrete Multi-Tone) modulation is adopted.

This DMT modulation method is, in a nutshell, a subcarrier whose transmission frequency band is about every 4 kHz (in the case of “G.lite”, depending on the conditions, up to 128 carriers in the downlink direction). And modulates each carrier. This DMT modulation method has a feature that it is resistant to noise of a specific frequency because communication with another subcarrier is possible even if a certain subcarrier becomes unusable due to the influence of noise having a specific frequency.

The details of the principle of the DMT modulation method are described in, for example, the document "JACBingham, Multicarrier Modulation For Data Transmission An Idea Whose Time has Come", IEEE Co. n. Mag., Pp5-14, May, 1990J. I have. The details of the digital subscriber line (ADSL) transmission system employing such a DMT modulation method are described below.

(1) Description of ADSL transmission system

As shown in Fig. 17, the ADSL transmission system focuses on its main components. For example, an AD SL device (AD SL modem) 110 installed on the subscriber side 100 and an AD SL device installed on the office 200 A device (AD SL modem) 210 is connected to each other via a transmission line (metallic line) 300 such as a subscriber line pair cable.

As shown in FIG. 17, the subscriber-side modem 110 includes a DMT modulator 101, a digital transmission filter 102, a digital Z-analog (D / A) converter 103, an analog band-pass filter as a transmission system. (BPF: Band Pass Filter) 104, etc., and the station-side modem 210 has a transformer 201, analog low-pass filter (LPF) 202, AZD converter 203, digital B PF 204, Time-domain Equalizer (TEQ) 205, Fast Fourier Transformer (FFT) 206, Frequency-domain Equalizer (FEQ) 207, Classifier 208 Etc. are provided.

Note that in Fig. 17, the configuration for the upstream direction from the subscriber side 100 to the office side 200 is shown. However, in practice, the ADSL modem 110 of the subscriber 100 (hereinafter referred to as the subscriber modem 110) receives the signal from the ADSL modem 210 of the office 200 (hereinafter referred to as the office modem 210). A receiving system having the same function as the system is provided, and the station-side modem 210 is provided with a transmitting system having the same function as the transmitting system in the subscriber-side modem 110. In principle, the communication of (2) is performed in the same manner as the uplink communication described below.

Here, in the subscriber side modem 110 described above, the DMT modulation section 101 is for performing DMT modulation on transmission data transmitted from a subscriber terminal such as a personal computer (PC). For example, as shown in Fig. 18, it is composed of a serial-to-parallel buffer (Serial to Parallel Buffer) 111, an encoder (Encoder) 112, and an inverse fast Fourier transformer (IFFT) 113. You.

The serial-to-parallel buffer 111 stores transmission data, which is serial data, for one symbol time (approximately l / 4 kHz), converts the stored data into parallel data, and outputs the parallel data. According to a predetermined transmission bitmap (not shown), the number of transmission bits is assigned to each carrier (subcarrier) (frequency band division).

For example, assuming that the number of carriers is Nc (= 1 to! 1; n is an integer of 2 or more), in FIG. 18, the M'fsbit / s transmission data (serial signal) is represented by a serial-parallel buffer. Is converted into parallel data of M pits of the symbol fs, and the mi bits of the bits are assigned as the i-th (i = l to n) bit groups to be transmitted on the same carrier. .

As a result, the entire M-bit data is simultaneously transmitted on Nc carrier waves as a whole, and the receiving side (station side 200) transforms Ns time data (symbols) into a fast Fourier transform (FFT: Fast). Fourier mnsform) is used to convert to the frequency domain, and M bits of information are extracted from the amplitude and phase of the Nc carrier waves. Note that each carrier is placed at a distance of Afc (about 4 kHz) in the used frequency band. Considering the computation efficiency of IF FT and F FT, the number of symbols Ns is a power of 2 It is common to use "64" or "256".

Next, the encoder 112 performs quadrature amplitude modulation (for example, QAM (Quadrature Amplitude Modulation)) on the bit string output from the serial-parallel buffer 111 by changing the amplitude and phase for each carrier. The IFFT 113 performs inverse fast Fourier transform on the signal point data (data in the frequency domain) output from the encoder 112 to obtain the signal point. It converts the data into Ns time-domain signal sequences. As a result, the M-bit data is converted into a time-domain signal (symbol) consisting of Ns sample values and transmitted.

Next, in FIG. 17, the digital transmission filter 102 is for removing unnecessary components of the symbol sequence (digital data) obtained by the DMT modulator 101, and the 0/8 converter 103 The output of the digital transmission filter 102 is converted to analog data, and the analog BPF 104 is for removing unnecessary components of the output of the DZA converter 103.

On the other hand, in the station-side modem 210, the transformer 201 is for cutting off the direct current (dc) component of the signal received from the transmission line 300, and the analog LPF 202 is for cutting off the high frequency component signal and the noise component. The AZD converter 203 converts the received signal (analog signal) passed through the analog LPF 202 into a digital signal.

Further, the digital BPF 204 removes noise components and the like outside the desired band from the output of the AZD converter 203, and the TEQ 205 generates inter-symbol interference (ISI: Inter Symbol Interference) with the input signal on the transmission side ( This is to perform predetermined processing so that it falls within the cyclic prefix added on the subscriber side 100).

Here, one point of the DMT modulation method is that by providing a period called a guard time between symbol sequences, demodulation on the receiving side can be made less susceptible to transmission line distortion. At the guard time, a part (L sample) corresponding to the guard time at the end of the symbol sent after that is sent in advance. This is a cycle

(Cyclic Prefix). In general, an impulse with a time width of zero emitted from the transmission side becomes a response having a finite width in the time direction due to distortion of the transmission path and reaches the reception side. Therefore, even in the DMT modulation method, the reception symbol has a wider time width than the transmission symbol, and the amplitude direction is different from that of the transmitted symbol. However, if the spread in the time direction is less than the guard time and the selection of the FFT section is correct, the output of the FFT will be the same as the input of the transmitting IFFT multiplied by a complex fixed coefficient. However, transmission line distortion is not affected.

Therefore, the TEQ 205 adaptively updates its own coefficient (tap coefficient), and determines the length of the impulse response from the output of the transmitting IFFT 113 to the receiving FFT 206 via the transmission path 300 by the guard time. It works as follows. However, in practice, it is difficult to completely reduce the portion exceeding the guard time to zero, and the size is reduced as much as possible.

Thus, the purpose of the TEQ205 applied to the ADSL transmission system is quite different from that of the equalizer used in systems other than the DMT modulation system. (Generally, the frequency characteristics of the amplitude and group delay time are flattened. Is the main purpose). Note that the TEQ 205 itself is a FIR (Finite Impulse Response) type transversal equalizer, and has the same configuration as the equalizer used in the QAM system and the like. The coefficient is updated based on the difference between the equalized output of the TEQ 205 and the output of the channel target equalizer 205a, which is obtained by the adder 205b (the details will be described later). In addition, if the guard time is lengthened, the TEQ 205 is unnecessary or the required order can be reduced.However, if the guard time is lengthened, the transmission efficiency deteriorates, so the guard time is selected to be about 1Z16 of the symbol time. Is common.

Next, the FFT 206 is for converting the output after equalization by the TEQ 205 into frequency-domain data by the fast Fourier transform, and the FEQ 207 is a sub-channel (sub-carrier and By multiplying by a complex constant a + jb (j = -D), the FFT output is equalized in the frequency domain for each subchannel, and the amplitude of the received signal received by passing through the transmission path 300 And the effect on the phase is compensated for each carrier having a different frequency. The discriminator 208 includes a real part and an imaginary part of the output of the FEQ 207. Each of them is converted to digital data (digital I and Q values) (hereinafter referred to as identification processing).

The obtained digital data is output to a subsequent DMT demodulation section (not shown) as received data, and the DMT demodulation section modulates the data at a DMT modulation section 101 on the transmission side (subscriber side 100). After performing the reverse processing (decoding, parallel-to-serial conversion, etc.), the signal is orthogonally demodulated.

Further, in an actual device, the reception level greatly varies depending on the length of the transmission path 300, and therefore, an AGC converter for making the input level to the A / D converter 203 almost constant before the AZD converter 203 is used. (Automatic Gain Controlled) In many cases, a digital AGC is inserted after the AZD converter 203 in order to reduce the operation word length in digital signal processing. Illustration is omitted.

Hereinafter, the operation of the ADSL transmission system configured as described above will be described. First, in the subscriber-side modem 110, when transmission data is input to the DMT modulator 101, the transmission data is held in the serial-parallel buffer 111 for one symbol time (1/4 kHz). The stored data is divided for each predetermined number of transmission bits per carrier and output to the encoder 112. The encoder 112 converts each of the input bit strings into signal points for quadrature amplitude modulation, and outputs the signal points to the IFFT 113. In IFFT 113, quadrature amplitude modulation is performed on each signal point by performing an inverse fast Fourier transform on the output of encoder 112. A predetermined sample of the output of IFFT 113 is added to the head of the symbol as a cyclic prefix.

The IF FT output to which the cyclic prefix has been added as described above is input to a DZA converter 103 after unnecessary components have been removed by a digital transmission filter 102, where a predetermined sampling frequency (for example, 1.104 MHz ), Is converted into an analog signal, and is output to the transmission path 300. On the other hand, in the station-side modem 210, the analog signal received through the transmission path 300 is input to the analog LPF 202 via the transformer 201, and after the unnecessary components are removed by the analog LPF 202, the AZD converter 203 At digital Converted to a signal.

The obtained digital signal is input to a TEQ 205 after further removing unnecessary components by a digital BPF 204, and the length of the impulse response of the reception path is determined by the TEQ 205 to the guard time (size). (The click prefix). The equalized signal is input to the FFT 206, where it is subjected to a fast Fourier transform to be converted into frequency-domain signal point data.

The obtained signal point data is subjected to the FEQ 207 to compensate for the influence on the amplitude and phase caused by passing through the transmission path 300 for each carrier having a different frequency. The data is converted to digital data (digital I and Q values) and output to the subsequent processing unit (DMT demodulation unit) as received data.

(2) How to calculate tap coefficient of TEQ205

Next, a method of obtaining the tap coefficient of TEQ 205 described above (coefficient update) will be described below. The method of obtaining such a coefficient is described in, for example, US Patent 5,285,474, 'Method for Equalizing A Multi-carrier Communication System'. In other words, the basic idea is to assume a circuit block (equalizer) called “channel target”. Such a “channel target” is a circuit with a transversal filter configuration having transfer characteristics equivalent to the transmission system from the IFFT output of the transmitting side (subscriber side 100) to the FFT input of the receiving side (station side 200). The block length is set to L or less, which is the number of taps equivalent to the guard time.

Then, the same signal (training signal) is sent to this “channel target” and the transmission system (hereinafter referred to as “real route”) from the transmitting IFFT output to the receiving FFT input, and the difference between their outputs Adaptively change the coefficients of the TEQ 205 included in the transversal filter of the channel target 205a (hereafter referred to as the target equalizer 205a) and the actual route so that is always zero. Has been implemented (see Fig. 19).

More specifically, a training time is provided, during which the same data sequence as the input data sequence of the transmission side (subscriber side 100) generated by the transmission side random bit sequencer 121 is sent to the pseudo random bit sequencer 205. Target equalization by c The output of the TEQ 205 and the output of the evening get equalizer 205a are compared (added) by the adder 205b, and the evening get equalizer is set so that the difference becomes small. The tap coefficients of 205a and TEQ205 are changed adaptively.

As a result, the impulse response between the transmitting-side IFFT output and the receiving-side FFT input is almost the same as that of the target equalizer 205a. The impulse response of the evening gate is the tap coefficient itself of the target equalizer 205a, and its time width is shorter than the guard time, so that the impulse response between the transmitter I FFT output and the receiver FFT input is The length is also equal to or less than the guard time, and the effect of transmission line distortion on DMT modulation can be suppressed.

Here, when calculating the coefficients of the target equalizer 205a and the TEQ 205, there are a method that is performed on the time axis and a method that is performed on the frequency axis. The method of obtaining the values on the time axis uses a conventionally well-known transversal filter coefficient updating method. That is, the tap coefficient is updated using the difference between the two outputs and the output value of the delay element in the transversal filter. It should be noted that when a solution is obtained in this time range, an extremely long transmission path 300 involves an extremely large time delay in an actual route. Therefore, such a delay is dealt with by correcting a delay of an integral multiple of the sampling period using a delay unit 205d shown in FIG.

On the other hand, a method performed on the frequency axis is described in the aforementioned US Pat. No. 5,285,474. Explaining with reference to Fig. 19, an FFT is connected downstream of the evening get equalizer 205a, and its output is compared with the output of the FFT 206 downstream of the TEQ 205 to calculate the difference for each sub-channel. , TEQ, and target equalizer 205a. Correction is made on the frequency axis by transforming the coefficient sequence of each coefficient into the frequency axis by FFT. After such correction, the operation of returning to the time domain axis again by IFFT and adjusting the number of tap coefficients in the time window is repeated, thereby correcting the coefficients of the target equalizer 205a and the TEQ 205.

As described above, in the conventional DMT modulation method, the characteristics of both the TEQ 205 and the target equalizer 205a are matched as much as possible regardless of the time axis and the frequency axis. However, in practice, it is good to start pulling from both initial values of zero No solution is obtained. As can be seen from Fig. 19, the target characteristic simulates the characteristic from the transmitter IFFT output to the receiver FFT input, so that the gain is sufficient in the frequency band where the transmission signal passes, compared to the other bands. High, that is, in a band through which the transmission signal passes, the difference in signal delay due to frequency in the pass band is small, and it may be incomplete, but it is desirable that delay equalization be performed.

For example, a range from about 26 kHz corresponding to sub-channel “6” to about 138 kHz corresponding to sub-channel “32” is assigned as a frequency band from the subscriber to the central office, and the maximum passband frequency is 138 kHz. The target equalizer 205a of the DMT modulation method processing at 552 kHz, which is four times the frequency of the bandpass filter, should have the initial value of the transfer function of the band-pass filter with a pass band of approximately 26 to 138 kHz. is there.

For a while after the start of the acquisition, the coefficient of the channel target 205 is fixed, the coefficient of the TEQ 205 is made variable, and when the TEQ coefficient stops moving, the coefficient of the target equalizer is made variable to further reduce the difference. By adopting it, the characteristics of TEQ205 can be obtained efficiently.

Here, what is the good characteristic of the DMT modulation method? The TEQ205 makes the amplitude of the part of the impulse response of the actual route that is longer than the guard time width extremely small, and each subchannel obtained as the output of the FFT206 The received data is not interfered with by other subchannel signals of the current symbol, nor by the signal of the previous symbol (that is, ISI is as small as possible).

In sub-channels with interference (ISI), the S / N deteriorates, and the amount of information (number of bits) transmitted through the sub-channels decreases. It goes without saying that the number of bits that can be arranged for each sub-channel is reduced by the noise of the transmission path 300 in addition to the above-mentioned intra-channel interference and inter-channel interference. The total number of bits that can be arranged for each subchannel is the total number of arranged bits per symbol, and the total number of arranged bits multiplied by the symbol repetition frequency is the throughput.

When the transmission line 300 is a monotonous line, good results are often obtained by the above method. For example, when the line length is 4 km, the in-band characteristics of the TEQ205 are gently high-pass filter characteristics that equalize the line characteristics of the line, and when the line length is zero, The in-band characteristics of the TEQ 205 are flat loss characteristics.

However, when a “bridge tap” exists in the transmission path 300, good results are often not obtained. A "bridge tap" is an open-ended branch line of an unspecified length for a branch in the middle of a track. When this “bridge tap” is present, the transfer characteristic often has a large loss in a specific frequency band of the passband, and cannot be handled by the conventional TEQ 205, and there is a large loss between the actual route and the target route. The difference remains, and as a result, the length of the impulse response of the actual route cannot be shortened, and the total number of arranged bits often decreases.

This is because TEQ 205 is an FIR-type adaptive equalizer, and therefore is not suitable for equalizing a peak of a loss centered on a specific frequency. That is, in general, in an FIR type adaptive equalizer such as a transpersal type equalizer, the transfer function is only a numerator function and no denominator function is present, so a steep loss peak can be realized, but a steep gain It is not suitable for realizing the characteristics of the peak (that is, compensating for the steep valley of the gain).

Therefore, in order to equalize (compensate) the loss peak, it is necessary to provide the same peak shape to the characteristic of TEQ205. Therefore, it is efficient to apply an IIR (Infinite Impulse Response) type adaptive equalizer, which is good at compensating for steep gain valleys, to TEQ205. Efficiency means that the required characteristics can be obtained with a small amount of computation. However, if the IIR type T EQ is used, there is no method for adaptively finding the coefficient of the denominator according to the transmission path 300 and the “bridge tap”.

For example, “Design of Digital Filters” (Miki Takebe, Tokai Shuppan, pp247-pp249) describes an IIR adaptive equalizer, but “convergence of successive approximation is not generally guaranteed. It does not always converge to the error ", and the IIR type adaptive equalizer cannot be used in practice.

As described above, conventionally, IIR type equalizers cannot be used as TEQ 205.Therefore, when a “bridge tap” is present, there is a problem that sufficient good throughput may not be obtained. Was.

For example, two “bridge taps” 1.5 km apart and 500 m apart When a transversal equalizer with 16 to 24 taps is used as the TEQ 205 as in the past for a certain line, the total number of arranged bits is as small as 83 to 88 bits.

On the other hand, if a second-order denominator function is added and its coefficient is optimized, the total number of arranged bits can theoretically be increased to 134-19.1 bits, or 1.6-2.3 times. For this reason, it has been thought that adding a second-order, or three-tap, recursive filter (IIR type equalizer) could significantly improve the characteristics.

However, as described above, since the method of adaptively finding the coefficient is not clear, for example, one having a simple F (frequency) characteristic such as an emphasis circuit that changes the coefficient stepwise in association with the AGC gain At present, it was not used for any other purpose. The present invention has been made in view of such a problem, and has found a method for stably and adaptively obtaining coefficients of an IIR type equalizer, thereby sufficiently compensating for a steep gain valley of a received signal. The purpose is to be able to use IIR type equalizers in addition to FIR type equalizers.

Disclosure of the invention

In order to achieve the above object, an automatic equalizer of the present invention adaptively performs equalization processing on a signal received from a predetermined transmission path, and has the following units. It is characterized by.

(1) An equalizer that shortens the length of the impulse response characteristics of the reception transmission line (from the transmission IFFT output to the FFT input) (hereinafter referred to as a real route equalizer)

(2) During the coefficient training period of this real route equalizer, the target equalizer that simulates the impulse response characteristics of the route up to the output of the real route equalizer including the transmission path

(3) During the coefficient training period, the tap coefficients of the target equalizer are updated based on the output of the real root equalizer and the output of the evening get equalizer, and a part of the updated tap coefficients ( ) Is used to set some tap coefficients of the real root equalizer.

In the automatic equalizer of the present invention configured as described above, the tap coefficient of the evening equalizer is calculated based on the output of the real root equalizer and the output of the target equalizer during the coefficient training period. Is updated and the transfer function of the target equalizer The tap coefficients of the IIR type equalizer can be set so that some reciprocals of the factorized factor are used as transfer functions.

Therefore, it is possible to use an IIR type equalizer having a denominator function part as a transfer function and suitable for compensating for a steep valley in the gain of a received signal, and thereby, regardless of the conditions of the transmission path, It is possible to always obtain a good transmission amount, that is, a throughput. BRIEF DESCRIPTION OF THE FIGURES

1 and 2 are block diagrams illustrating the principle of the present invention. FIG. 3 is a block diagram showing a configuration of the automatic equalizer according to the first embodiment of the present invention. 4 to 7 are block diagrams for explaining the function of the coefficient update block shown in FIG.

FIG. 8 is a block diagram showing a configuration example of the first-channel first-get equalizer shown in FIGS.

FIG. 9 is a block diagram showing a configuration example of the second channel sunset-equalizer shown in FIGS.

FIG. 10 is a block diagram showing a configuration example of the transversal equalizer (T EQ) shown in FIG.

FIG. 11 is a block diagram showing a configuration example of the recursive equalizer shown in FIG.

FIG. 12 is a diagram for explaining pole determination during coefficient training according to the present embodiment.

FIG. 13 is a block diagram showing a configuration in a case where the recursive equalizer shown in FIGS. 4 to 7 is arranged after TEQU.

FIG. 14 is a block diagram for explaining a function of a coefficient update block of the automatic equalizer according to the second embodiment of the present invention.

FIG. 15 is a diagram for explaining the operation of the coefficient update block according to the second embodiment.

FIG. 16 is a block diagram for explaining the function of the coefficient update block according to the second embodiment after the completion of coefficient training.

Figure 17 is a block diagram showing the configuration of a conventional digital subscriber line (AD SL) transmission system. It is a lock figure.

FIG. 18 is a block diagram showing a configuration of the DMT modulator shown in FIG.

FIG. 19 is a block diagram for explaining the method of updating the TEQ coefficient shown in FIG. BEST MODE FOR CARRYING OUT THE INVENTION

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

(A) Description of the first embodiment

The point of the present invention is to adaptively obtain the coefficients of the IIR type equalizer on the receiving system (real route) side input to the FFT using the target equalizer.

For example, if the training period of the coefficient is provided systematically as in the DMT modulation method and the signal transmitted from the subscriber side is known in advance, a pseudo transmission path (target route) is provided. Therefore, a system as shown in Fig. 1 can be considered because it can be compared with the signal from the actual transmission path (real route).

That is, the transfer function of the transmission path is HI (z)

(z—, the transfer function of TEQ is H

2 (Z (Z- ¹ ) / Hteq _d (z—, transfer function of target equalizer to H 3 (Z)

If (z " ¹ ) is set, the following equation (1) holds, assuming that the error between the target route and the actual route becomes zero after the coefficient is pulled.

HI (z) · H'2 (z) -H3 (z) = 0… (1) Here, if H2 (z) has no zero and both sides are divided by H2 (z),

HI (Z) -H3 (z) ZH2 (z) = 0 ... (2) holds. This is shown in a block diagram in FIG. From this, the transfer function 112 (z) of H _<< 3 is equal to the denominator function _Hteqd (z- ¹ ) of _Htequ (z " ¹ ) / _Hteqd (z" ¹ ). When it moves, it becomes a molecular function, and the transfer function of the target equalizer H 3 (Z)

(z " ¹ ).

This means that during training, the tap coefficients of the one-time equalizer are updated, and after the update, the transfer function corresponding to the reciprocal of the factor component when the transfer function of the target equalizer is factorized is the real root. To be inserted into the real root of the equalizer evening By setting the tap coefficients, it is possible to appropriately obtain the denominator function that the transfer function of the equalizer on the real root side should have.

Figure 3 shows an automatic equalizer configuration that achieves this. The configuration shown in FIG. 3 is also applied to, for example, the office-side modem 210 shown in FIG. 17, where 1 is a recursive equalizer, 2 is a TEQ equivalent to the TEQ 205 shown in FIG. 17, 3 is a window block, and 4 is a diagram. FFT equivalent to FFT 206 shown in 17 and 5 show a coefficient update block, respectively. Here, the recoverable equalizer 1 converts the input signal (the output of the digital BP F 204 in the case of FIG. 17 (the received signal from the transmission path 300)) into a predetermined denominator function (for example, H _teq _d (This is an IIR (Infinite Impulse Response) type equalizer using z as the denominator of the transfer function. Here, for example, it is configured as a 3-tap IIR type equalizer as shown in FIG. 11.

That is, for the data (output of the adder 11) held in the registers (delay elements) 13 and 14 one by one in time series, predetermined coefficients (tap coefficients) are respectively multiplied by the multipliers 15 and 16. The sum is obtained by the adder 12, and the result is added to the current input signal by the adder 11, so that an equalized output for the input signal can be obtained. ing.

On the other hand, the TEQ (real root equalizer) 2 equalizes the input signal (the output of the recursive equalizer 1) using a predetermined molecular function (for example, the above H _tequ (z- ¹ )) as a transfer function. It is configured as a well-known transversal type equalizer (FIR type equalizer) as shown in FIG. 10, for example, and a plurality of registers (delay elements) 21 corresponding to the number of taps are stored in a time-series manner. Multiplier 22 multiplies the sample data at a plurality of points on the time axis obtained by holding the input signal for each sample by a tap coefficient C (j) (j = —N to N). The sum by the adder 23 is obtained as an equalized output with respect to the input signal (the output of the reciprocating equalizer 1). The number of taps in TEQ2 is based on the number of samples L corresponding to the guard time length, but there is no problem if it is larger than that. Also, their initial values may all be "0.0".

Further, the window block 4 is for delaying the output of TEQ2 by a predetermined time and inputting it to the FFT 4, and the coefficient update block 5 is for recursive equalizer 1 And the coefficient of TEQ 2 are adaptively updated based on the output of TEQ 2. In the present embodiment, the recursive switching signal and the TEQ switching signal pass the recursive equalizer 1 through the input signal (through). ) The state can be changed, and the updated coefficient value can be reflected (set) in the reactive equalizer 1. Specifically, in the coefficient update block 5, during the coefficient training period, the functional units shown in FIGS. 4 and 6 are realized by software processing. That is, as shown in FIGS. 4 and 6, the coefficient update block 5 includes a pseudo-random bit sequencer 51, a delay block 52, a first channel target equalizer 53, and a second channel target equalizer. It has an adder 54, an adder 55 and an inverse second channel evening get function block 56.

Here, the pseudo-random bit sequencer 51 generates the same pseudo-random signal (training signal) as the pseudo-random bit sequencer 12 1 on the transmitting side described above with reference to FIG. 19, and the delay block 52 The delay time of the pseudo-random signal on the target route side input from the pseudo-random pit sequencer 51 is adjusted, and the pseudo-random bit sequencer 121 on the transmitting side is input to the TEQ 2 via the transmission path 300 by the pseudo-random bit sequencer 121. This is to match the delay time of the pseudo-random signal on the real route side with the delay time of the pseudo-random signal on the target route side. Note that the delay block 52 may be provided on the real route side (the output side of TEQ 2) (the delay time of the pseudo random signal on the real route side may be adjusted).

Also, the first channel evening get equalizer (hereinafter referred to as the first target equalizer) 53 acts on the output (pseudo-random signal) of the delay block 52, for example, as shown in FIG. In addition, it is configured as a transversal type (FIR type) equalizer similar to TEQ 2.

That is, in the first target equalizer 53, the number of registers (delay elements) 531 corresponding to the number of taps, the time axis obtained by holding the input signals one by one in time series, is obtained. The sample data at the above plural points are multiplied by a multiplier coefficient in a multiplier 532, respectively, and the sum of those multipliers is added to the input signal (output of the delay block 52). It can be obtained as output. The number of taps in the first evening get equalizer 53 is equal to the number of taps corresponding to the guard time length described above. A value close to the number L of pulls, and the initial value uses a value preset by the above-described method. Also, the second channel target equalizer (subordinate equalizer) 54 acts on the post-equalization output of the first evening equalizer 53. Here, the first target equalizer 54 is used. It is configured as an FIR type equalizer in the same way as 53.

However, the second channel target equalizer (hereinafter, referred to as the second target equalizer) 54 has a lower order (2 to 4 order; 3 to 4) than the first target equalizer 53 described above. Considering that the tap coefficients of the recursive equalizer 1 that will be the denominator function part on the real route side are finally set using the tap coefficients, the tap coefficient of the first tap is considered. The numerical value is fixed to "1.0", and the other tap coefficients are initially set to "0.0". FIG. 9 shows the configuration of the second target equalizer 54 in the case of three taps. As shown in FIG. 9, also in the second target equalizer 54, the time axis obtained by holding the input signals one by one in time series in the two resistors (delay elements) 54 1 is obtained. Multipliers 542, 543, and 544 multiply the sample data of the above three points by tap coefficients, and the sum of those multipliers is added to the input signal (first target equalizer). (The output of the device 53).

Further, the adder 55 obtains a difference between the output of the TEQ 2 and the output of the second target equalizer 54 (an equalization error signal e; hereinafter, simply abbreviated as “error e”). The coefficients of the TEQ 2, the first target equalizer 53, and the second target equalizer 54 are updated so that the error e is minimized.

However, in updating the coefficients of the first evening equalizer 53, the error e 'taking into account the deformation in the second target equalizer must be used instead of the error e obtained by the adder 55. No. This is because the error used for updating the coefficients of the transversal equalizer must be the error of the (immediate) output terminal. For example, when an inverting amplifier (gain is “1.0”, but the phase is rotated by 180 degrees) is used instead of the second target equalizer 54, a positive Since the error becomes negative at the input of the inverting amplifier, the coefficient of the first target equalizer 53 is updated with a completely opposite sign error without considering the transfer function of the inverting amplifier (= -1. 0). .

Now, since the error e is measured at the output end of the second evening equalizer 54, The second target equalizer 54 can be used as it is for updating the coefficient of the second target equalizer 54, but the second target equalizer 54 is located between the output end of the first evening equalizer 53 and the measurement point of the error e. Therefore, if the error at the output point of the first target equalizer 53 is e 'and the transfer function of the second evening equalizer 54 is h2, e'h2 = e Holds. Therefore, the error e ′ = e / h 2 must be used as an error for updating the coefficient of the first target equalizer 53.

The error e ′ is obtained by the inverse second channel target function block (Yuichi get output point error calculation block) 5 6, and the inverse of the transfer function of the equalizer of the second target equalizer 54 is It has a transfer function, and the error obtained by the adder 55 is equalized in the time domain by the equalizer function to obtain an error e 'at the output point of the first target equalizer 53. It is used for updating the coefficient of the first target equalizer 53.

That is, the coefficient update block 5 of the present embodiment includes the adder 55, the inverse second channel target function block 56, the first evening get equalizer 53, and the second evening get equalizer 5. It is composed of a first difference calculation updating unit that updates each coefficient of 4 and TEQ 2.

Hereinafter, a coefficient updating process (coefficient training method) by the coefficient updating block 5 configured as described above will be described.

First, as shown in FIGS. 4 and 6, the pseudo-random bit sequencers 12 1 and 51 add the same pattern of pseudo-random signals to the real route (TEQ 2) and the target route (delay block 52). . Immediately after the start of the pull-in (the start of the coefficient update process), the recursive equalizer 1 is set to the through setting by the recursive switching signal, the coefficient of the first target equalizer 53 is fixed, and the TEQ 2 and the second evening are set. The coefficients of the one-get equalizer 54 are made variable, and the coefficients are updated adaptively (first acquisition process).

In this case, the coefficient updating method (automatic equalization algorithm) is that the target route output (the output of the second target equalizer 54) is subtracted from the actual route output (the output of TEQ 2) by the adder 55. Using the error e obtained by the above, a well-known method of adaptively approaching the coefficient value in the multiplier in the transversal type equalizers 2, 54 to the best value is used. is there.

For example, taking TEQ 2 having the configuration shown in FIG. 10 as an example, in a method called the LMS method, the second correction algorithm for the coefficient C (j) is expressed by the following equation (3).

C (;) ^{(v + 1)} = C () ^(v) -α 'e ^(v) … (3)

However, in this equation (3), Xj ^M represents the input signal sequence of TEQ2, and e ^(v) represents the error. The details of the automatic equalization algorithm are described in, for example, the document “Adaptive Signal Processing” (edited by Shokodo Shigeo Tsujii, first published on May 15, 1995, p. 24), and described in other documents other than the above. It is also possible to apply various algorithms.

Now, after the above-described first pull-in process, the coefficient update block 5 next performs a second pull-in process in which the coefficient of the first target equalizer 53 is also made variable and the error e is further reduced. . At this time, the recursive equalizer 1 and TEQ 2 are maintained in the through setting by the recursive switching signal.

Thereafter, as shown in FIGS. 5 and 7, the coefficient update block 5 performs the recursive equalizer 1 so that the transfer function of the recursive equalizer 1 becomes the reciprocal of the transfer function of the second target equalizer 54. Set the tap coefficient of. That is, the tap coefficients to be multiplied by multipliers 543 and 544 in FIG. 9 are set to the tap coefficients to be multiplied by multipliers 15 and 16 in FIG. 11, respectively. In addition, the second target equalizer 54 on the target route side performs a process of setting to through, and the recursive equalizer 1 is set to non-through by a recursive switching signal.

Further, thereafter, the coefficient update block 5 makes the coefficient of the TEQ 2 and the coefficient of the first target equalizer 53 variable, and updates the coefficient so that the error e is further reduced. The coefficient of the Rica equalizer 1 is fixed.

Of course, in updating the tap coefficient of the second target equalizer 54, the tap coefficient is later set as the tap coefficient of the recursive equalizer 1, and the pole of the transfer function of the recursive equalizer 1 is set on the complex plane. It is necessary to check the updated value so that it is not placed outside the unit circle (hereinafter simply referred to as “unit circle”). Here, the transfer function of the recursive equalizer 1 after tap coefficient setting, as for example, H (z) = 1 / ( l + a z- ^ b z '1), the 0-order coefficient "1.0" (Multiplier 542 If the tap coefficient to be multiplied by is = 1.0 (see Fig. 11), the only variables to be considered in the pole determination operation are a and b (the evening coefficient to be multiplied by the multipliers 543 and 544). Therefore, the amount of calculation for checking the tap coefficient of the second target equalizer 54 is reduced.

That is, when determining the pole position, the above function may be modified to confirm that a solution satisfying z ² + az + b = 0 exists in a unit circle on the complex plane. Considering this condition, b ^ O. 9, b≥a—0.9, b≥—a—0.9 is derived (actually, it is 1.0 instead of 0.9, In the case of actual operation, the margin is set to 0.9 in consideration of the effects of coefficient rounding, etc.). This indicates that it suffices that a and b exist within the triangular area shown in FIG.

The coefficient update block 5 controls the coefficient update by software so as to satisfy this condition. That is, the coefficient update block 5 sets the tap coefficients in the multipliers 542 and 543 to the multipliers 15 and 16 as they are when the pole exists outside the triangular area shown in FIG. If they exist outside the triangle area shown in Fig. 12, the vertices of the triangle shown in Fig. 12 are (-1.8,0.9), (1.8,0.9), and (0.0, -0.9), respectively. Depending on which side 6 to 8 of the triangle the pole (a, b) is near, the following three processes (1) to (3) are performed, and then a 'and b' are multiplied by multipliers 15, 16 Set to each.

(1) When it is close to the upper side 6

When b≥0.9, a≥1.8, a '= 1.8, b' = 0.9

When b≥0.9, 1.8> a≥-l, 8, a '= a, b' = 0.9

When b≥0.9, -1.8> a, a '= -1.8, b' = 0.9

(2) When close to the right side 7

When b <0.9, a> 0, a + b≥2.7, a '= 1.8, b' = 0.9

When b <0.9, a> 0, 2.7> a≥-0.9, a '= (a + b + 0.9) / 2, b' = (a + b-0.9) / 2 b <0.9, a> 0, -0.9> a + b, a '= 0.0, b' = -0.9

(3) Close to the left side 8

When b <0.9, a <0, -a + b≥2.7, a '= 1.8, b' = 0.9

When b <0.9, a <0, 2.7> -a + b≥-0.9, a '=-(-a + b + 0.9) / 2, b' = (-a + b-0.9) / 2 When b <0.9, -0.9> -a + b, a '= 0.0, b' = -0.9

As described above, according to the present embodiment, in addition to TEQ2, the introduction of the IIR type recursive equalizer 1 having the denominator function as the transfer function (equalizer function) reduces the number of orders, With a small amount of hardware, it is possible to effectively compensate for a steep gain valley that TEQ 2 is not good at. Therefore, a good transmission amount, that is, a throughput can always be obtained regardless of the condition of the transmission path 300. As for the coefficient of the recursive equalizer 1, the tap coefficient of the equalizer on the evening gate side is used at the time of training, so that the parameter can be obtained quickly and stably.

It should be noted that it is not indispensable to perform the attraction in three stages as described above. For example, since the improvement of the characteristics of the TEQ2 and the first target equalizer after the recursive equalizer 1 shown in FIG. The F (frequency) characteristic of the first target equalizer 53 mainly depends on the characteristics of an analog filter (for example, filters 104 and 202 shown in FIG. 17) applied to the transmission side and the reception side. When the frequency characteristics of the transmission path according to the length are mainly compensated by TEQ2, the second pull-in process may be omitted in some cases because an evening get equalizer with a fixed coefficient can be used substantially.

Further, the arrangement position of the above-described recursive equalizer 1 can be considered to be after the TEQ 2 as shown in FIG. 13, for example. In this case, the error e〃 for updating the tap coefficient of the TEQ 2 can be considered. In order to obtain the error e〃, the same theory as the configuration shown in Fig. 4 (Fig. 6) ("The error used for updating the coefficients of the trans-persal equalizer is the output immediately after it) The inverse recovery equalizer 1 'is required.

(B) Description of the second embodiment

In the second embodiment, the target equalizer is not separated from the first target equalizer 53 and the second target equalizer 54 as in the first embodiment, and the recursive equalization is performed as shown in FIG. One (L + L '— first-order) channel target equalizer 57 whose order is increased by the order L', which is the denominator function of unit 1, performs the pull-in process, and at the stage when the coefficient training progresses, this target Factor the transfer function of equalizer 57 (Step S 1 in Figure 15), convert it to the product of a first-order or second-order z- ¹ function, and select an appropriate one (L '— Extraction of linear function: As shown in step S 2) of FIG. 15 and FIG. 16, the coefficient updating block 5 performs processing to make the denominator function of the transfer function of the recursive equalizer 1 with the fixed coefficients on the real root side. For the remaining terms that are not set for relocation, function expansion is performed (step S4 in Fig. 15), and the tap coefficients are set so as to be the transfer function of the evening-equalizer 57 'with a predetermined order L or less. Try again.

That is, the coefficient update block 5 in the second embodiment updates the tap coefficients of the evening equalizer 57 so that the difference between the output of the TEQ 2 and the output of the target equalizer 57 is minimized. It has a function as a second difference calculation update unit, and after the training period ends, the transfer function of the target equalizer 57 is factorized and the zero point is desirably present in the unit circle. The tap coefficient of the recursive equalizer 1 is set so that the inverse function of the _z function becomes the transfer function of the recursive equalizer 1 with fixed coefficients.

Here, z = ej ' ^{27t i / fs} , fs represents the sampling frequency, f represents the frequency, and j = -1.

Here, the criteria for selecting an appropriate one should be as follows.

(1) In the case of the evening get equalizer, the transfer function has no denominator function and is only a numerator function, so its zero is allowed inside and outside the unit circle. However, in the case of the denominator function of the transfer function of recursive equalizer 1, Since is not allowed outside the unit circle, we must check the zeros of the factorized function and select from the terms whose zeros are inside the unit circle. The check of the zero point is as described above in the first embodiment.

② The frequency characteristics of the gain of the final target route should be as flat as possible in the transmission frequency band or as close to single-line characteristics as possible, and the gain should decrease monotonically with the frequency gap. Choose so that no large gain dips or peaks remain, or small gain dips or peaks.

If the initial target equalizer 57 also includes a term corresponding to the recursive equalizer 1 as in this example, it is difficult to specify the initial value of the target equalizer 57. Therefore, in this case, it is desirable to provide an initial value to TEQ 2 and to initially fix the coefficient of TEQ 2 and make the coefficient of the target equalizer 57 variable. In FIG. 16, the coefficient update block 5 updates the coefficient of the TEQ 2 and the target equalizer 5 7 ′ while fixing the coefficient of the recursive equalizer 1, and further compresses the error e. .

In the second embodiment, as in the first embodiment, a fixed-coefficient recursive equalizer 1 updated at the evening-get-point side during training is introduced, thereby reducing the amount of hardware. Therefore, it is possible to sufficiently compensate for the steep gain valley of the received signal that TEQ 2 is not good at. Therefore, irrespective of the conditions of the transmission path 300, a good transmission amount, that is, a throughput can always be obtained, and all parameters of the equalizer system can be obtained at high speed and in a stable manner.

The present invention is not limited to the above-described embodiment, and can be implemented with various modifications without departing from the spirit of the present invention.

For example, either the time domain or the frequency domain calculation method can be applied to update the coefficients of the equalizer system. Industrial applicability

As described above, according to the present invention, the FIR type is realized by introducing a fixed coefficient IIR type equalizer that uses the tap coefficients of the Eich device that is updated stably on the target route side during training. Can eliminate the problem that the steep gain valley of the received signal cannot be sufficiently compensated for using the equalizer alone, and it is good regardless of conditions such as the presence of a bridge tap in the transmission path such as a paired subscriber cable. High throughput can be secured. Therefore, high-speed, high-quality services can be provided to subscribers in ADSL services and the like, and their usefulness is considered to be extremely high.

Claims

The scope of the claims

1. An automatic equalizer that adaptively performs equalization processing on a received signal from a predetermined transmission path,

An equalizing means that operates on a reception signal from the transmission path;

An evening equalizer that simulates an impulse response characteristic up to an output of the equalizing means including the transmission path;

The tap coefficients of the equalizer and the target equalizer are updated based on the output of the equalizer based on the training signal from the transmission path and the output of the target equalizer. An automatic equalizer, comprising: a coefficient update control unit that sets a tap coefficient of a part of the equalizing means using a tap coefficient of the target equalizer.

2 · The equalization means

Including two or more different equalizers for equalizing the received signal in a time domain,

2. The automatic equalizer according to claim 1, wherein the coefficient update control unit performs the setting by a pre-stage equalizer among the different equalizers.

3. When updating the tap coefficients of the equalizers that are not subjected to the setting after the setting, the coefficient update control unit sets the tap coefficient of the equalizer to be set to be fixed. 3. The automatic equalizer according to claim 2, wherein:

4. The equalizing means is

An IIR (Infinite Impulse Response) type equalizer for equalizing the received signal in the time domain, and a FIR (Finite Impulse Response) type equalizer for equalizing the output of the IIR type equalizer in the time domain. As well as

The setting performed by the coefficient update control unit includes:

This is performed by setting tap coefficients of the IIR type equalizer so as to have a transfer function corresponding to the reciprocal of a factor obtained by factorizing the transfer function of the target equalizer after the update. The automatic equalizer according to claim 1.

5. The coefficient update control unit

The automatic equalizer according to claim 4, wherein after the update, the evening equalizer is controlled such that a transfer function of the target equalizer is obtained by dividing the factor. Chemist.

6. The target equalizer has a plurality of equalizers connected in series,

The factor corresponds to a transfer function of any of the plurality of equalizers;

The automatic equalizer according to claim 4, wherein after the update, the target equalizer is controlled so as to perform setting such that an equalizer having the corresponding transfer function is set to be through. Chemist.

7. The coefficient update control unit:

An adder for obtaining, as an equalization error signal, a difference between the output of the equalization means and the output of the equalizer having the above-described transfer function;

A target output point error calculator for obtaining an equalization error signal for an output point of the equalizer having the above equivalent transfer function from the equalization error signal obtained by the adder;

Based on the equalization error signal obtained by the adder, each tap coefficient of the FIR equalizer and the equalizer having the above-mentioned corresponding transfer function is updated, and obtained by the target output point error calculator. 7. The method according to claim 6, wherein a tap coefficient of an equalizer other than the equalizer having the corresponding transfer function is updated based on the obtained equalization error signal. Automatic equalizer.

8. The coefficient update control unit:

During a certain period, the equalizers other than the equalizer having the corresponding transfer function of the target equalizer are set to the through setting, and the tap coefficients of the equalizer having the corresponding transfer function and the equalizer are updated. After the lapse of the certain period, the equalizers other than the equalizer having the above-described corresponding transfer function are set to the non-through setting, and the equalizers constituting the target equalizer and 7. The automatic equalizer according to claim 6, wherein each coefficient of said FIR equalizer is configured to be updated.

9. The coefficient update control unit:

The IIR equalizer is set so that the poles of the factor obtained by factorizing the transfer function of the target equalizer after the update have the inverse function of the desired z function existing in the unit circle on the complex plane. 5. The automatic equalizer according to claim 4, wherein the automatic equalizer is configured to set a tap coefficient.

10. A coefficient training method for an automatic equalizer that adaptively performs equalization processing on a received signal from a predetermined transmission path,

Based on the output of the equalizing means acting on the received signal from the transmission path and the output of the target equalizer simulating the impulse response characteristics up to the output of the equalization means including the transmission path, Updating the tap coefficients of the equalizing means and the target equalizer,

A coefficient training method for an automatic equalizer, characterized in that a tap coefficient of a part of the equalizing means is set using a tap coefficient of the evening get equalizer after update.