JPH03235511A - Automatic waveform equalizing system - Google Patents

Abstract

PURPOSE:To reduce a data error rate by reading an input waveform repeatedly from a memory storing the input waveform for a training period and using the input waveform for the updating of a tap coefficient. CONSTITUTION:A received signal waveform is tentatively stored in a memory 1 as it is. Then a signal waveform for a training period 41 is read repeatedly for plural number of times continuously one after another and inputted to this automatic waveform equalizer. Then the training period 41 is extended equivalently, the training operation is implemented for a long period to converge the tap coefficient. Thus, the convergence of the tap coefficient is enhanced without decreasing the efficiency of data transfer and the accuracy of an estimating signal is improved to reduce the data error rate.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an automatic waveform equalization method for an automatic waveform equalizer used to compensate for transmission line characteristics.

[Prior Art] Conventionally, there is a method of estimating the waveform of a transmitted signal using an automatic waveform equalizer to compensate for transmission path characteristics. Among these, in the transmission of digital signals, a decision feedback automatic waveform equalizer that uses an adaptive transversal filter is well known as one of the automatic waveform equalizers (I.E.I.
EEE TRANSACTIONS ON
COM-19, No. 3, 1971, pages 281-293).

An example of the decision feedback type automatic waveform equalizer will be shown below. Second
The figure is a block diagram of a decision feedback type automatic waveform equalizer.
Two transversal filters 31 and 32, two tap coefficient updaters 33 and 34 that respectively update the tap coefficients of these filters, and the transversal filter 31.
The subtracter 35 subtracts the signal of the transversal filter 32 from the output signal of the signal, an error signal calculator 37, a data determiner 36, and a reference signal memory 38. Note that the data demodulator 39 is not included in the equalizer, but is shown in the figure.

The input signal to the automatic waveform equalizer is first input to the transversal filter 31, where an output signal is created by a convolution operation. The output signal is the subtracter 35
In this step, the output signal created by the transversal filter 32 is subtracted to create a so-called estimated signal.

The estimated signal is sent to the data determiner 36 and the error signal calculator 37.
is input. The data determiner 36 determines the value of the estimated signal using a level comparator or the like according to the modulation method, and converts it into an ideal value. This value is input to the transversal filter 32, as well as to the data demodulator 39 and error signal calculator 37.

In the transversal filter 32, an output signal is created by a convolution operation similarly to the transversal filter 31, and is input to the subtracter 35.

The data demodulator 39 restores the signal to a digital signal according to the modulation method.

The error signal calculation method in the error signal calculator 37 is such that the difference between the output signal of the subtracter 35, that is, the estimated signal, and the reference signal read from the reference signal memory 38 is used as an error signal in the training period, and the estimated signal is used in the non-training period. The difference between the signal and the signal after the data determiner 36 is used as an error signal.

Note that here, a part of the transmission signal is previously set as a training period and a known signal is sent. A signal identical to the known transmission signal is stored in the reference signal memory 38 as a reference signal. The error signals thus calculated are input to the two tap coefficient updaters 33 and 34, respectively, and the update amount of the tap coefficients is calculated to update the tap coefficients.

Various methods have been proposed for updating tap coefficients, but a typical one is a basic method called the LMS method, which is performed using the following calculations.

If the update constant is G, the error signal is E, the input data stored in a transversal filter with n taps is Xk, the tap coefficient is Ck, and the updated tap coefficient is Ck', then Ck'-Ck-G-E -Xk (k=0.1.,,,,
n-1) (1) Here, select an appropriate value for the update constant G. If it is too small, convergence will be slow; if it is too large, convergence will be quick, but there may be divergence or unstable operation.

One way to quickly converge is to provide a training section in a part of the transmitted signal as described above, send out known data, and calculate the difference (error signal) between the estimated signal created by the equalizer and the known data. The convergence speed can be improved by accurately determining the correction value of the tap coefficient using the calculation formula shown in equation (1) based on this error signal and updating it.

[Problems to be Solved by the Invention] However, even with this method, a long training period is required to obtain sufficient performance that satisfies the specifications, resulting in a reduction in data transmission efficiency.

An object of the present invention is to improve the convergence speed of an equalizer without lengthening the training period, that is, while maintaining data transmission efficiency, and as a result to reduce the data error rate.

[Means for Solving the Problems] In order to solve the above problems, the present invention provides a memory for storing received signal waveforms, and makes it possible to read and compare the signal waveforms of training sections multiple times. And so.

[Operation: The received signal waveform is stored as is in the memory. Among them, the signal waveforms of the above-mentioned training sections are repeatedly read one after another a plurality of times and inputted into the automatic waveform equalizer. The training interval is lengthened isometrically, the training operation is performed over a long period of time, and the tap coefficients are converged. Thereby, it is possible to increase the convergence of tap coefficients without lowering data transmission efficiency, improve the accuracy of the estimated signal, and reduce the data error rate.

[Example] The present invention will be described in detail below using an example shown in the drawings.

FIG. 1 is a block diagram showing an example of a decision feedback type automatic waveform equalizer using the present invention.

First, in this embodiment, since digital signal processing is used, all signals handled are sampled digital data. The signal used in this example is one of phase modulation.
This is a two-dimensional phase modulation method such as QPSK or π/4 shift QPSK. Therefore, the transversal filter is constructed in two dimensions. In addition, all calculations in the tap coefficient updater and error signal calculator for this purpose are
Perform operations on dimensions.

In FIG. 1, 1 is an input signal memory storing a received signal, 2 and 3 are two-dimensional transversal filters, 4 and 5 are tap coefficient updaters, 6.7 is a subtracter,
8 is a data determiner, 9 is an error signal calculator, 10 is a reference signal memory, and 11 is a data demodulator.

The sampled received signal signal and Q are once stored in the input signal memory 1 and then inputted to the transversal filter 2 as Xl and XQ, respectively. The transversal filter 2 performs processing using a two-dimensional convolution operation. The output signal is subtracted by the output signal from the transversal filter 3 in subtracters 6 and 7, respectively; 1. Calculate the estimated signal for each Q. These estimated signals are input to a data determiner 8 and an error signal calculator 9, respectively.

The output of the data determiner 8 is input to the error signal calculator 9, as well as to the transversal filter 3 and the data demodulator 11. The error signal calculator 9 calculates the errors EI and EQ of the estimated signal, multiplies them by an update gain -0, and sends the results to the tap coefficient updaters 4 and 5. Tap coefficient updaters 4 and 5 update the tap coefficients of transversal filters 2 and 3, respectively. Further, the error signal calculator 9 receives the reference signal read from the reference signal memory 1o and calculates an error signal. There are two methods for this calculation. This will be explained next.

FIG. 3(a) shows an example of the data format to be transmitted. Of these, 41 are training sections and 42 are data sections (non-training sections). In the training period, a known signal is sent in advance, and the same signal is stored in the memory on the automatic waveform equalizer side. This is used by the error estimator as a reference signal, which will be described next. In this embodiment, all of these signals are temporarily stored in the input signal memory 1. Then, as shown in the format shown in Figure 3(b), the training section is repeated multiple times (5 times in the figure).
times) and input it to the transversal filter 2. At this time, if the writing speed and reading speed in the input signal memory 1 are the same, the length of one frame becomes longer.

If the automatic waveform equalizer has enough calculation time, this is fine, but if not, set the read speed faster than the write speed, as shown in Figure 3(c), to reduce the time required for one frame length. There are methods to make them the same.

FIG. 4 shows a block diagram of the error signal calculator.

51 and 52 are subtracters whose negative input terminals are selectively connected to the reference signal and the output signal of the data determiner 8 by selection switches 56 and 57, respectively. Estimated signals are input to each positive input terminal. Also, the subtractor 51.
The output signals of 52 are input to multipliers 53 and 54, respectively. Here, the value -g selected by the selection switch 55
It is multiplied by l or -g2 and sent to the tap coefficient updater 4.5, respectively. Three selection switches 55.56.
The operations of 57 are linked and switched at the same time. At this time, gl is a larger value than g2. In other words, in the training section shown in Figure 3, the update coefficient is increased to achieve rapid convergence, but in the non-training section, the update coefficient is decreased to reduce the deterioration of the tap coefficients when the data is incorrect. It will be done.

Next, FIG. 5 shows an example of the configuration of a two-dimensional transversal filter. The sampled I and Q input signals are input to registers 60 and 61, respectively, and the values already in those registers are transferred to the next register. For this purpose, if these registers are composed of n registers, for example, they hold the sampled signals of 0 and n-1 samples before. An arithmetic unit for performing the operations described below is connected to each output of these registers.

Further, an arithmetic unit is also connected to the signal line leading to the first register 60, 61. This will be representatively explained below. These arithmetic units include multipliers 65, 66, 67, and 68 that multiply the Q signal and the tap coefficient, and an adder (subtractor) 6 that calculates the sum of the outputs of these multipliers.
Consists of 9.70. The tap coefficients consist of two numerical sequences, each of which is multiplied by the engineering and Q signals in four combinations. By subtracting and adding two of these four multiplication results in adders 69 and 70, respectively, two operation results are obtained. The results of these two calculations are sent to the next stage adder (subtractor) and added to the film formation calculation result. In this way, the calculation results of each stage are added one after another, and the output signal of this transversal filter is generated when the addition of the final stage is completed.

Next, the engineering and Q signals of the next sample are stored in register 60.6.
1, and the signals contained in these registers are successively shifted to the adjacent registers.

Then, all product-sum operations are performed in the same way as for the previous sample to create an output signal for the next sample. These signal sequences are estimated signals of the equalizer.

The above operation can be expressed as a two-dimensional convolution operation as shown below.

It is expressed as follows, where the estimated signals are YI and YQ, the input signals stored in the th register are XIk and XQk, and the tap coefficients are CIk and CQk.

Regarding the configuration method of this transversal filter,
Multiple multipliers, adders (subtracters), etc. can be configured by using a single multiplier or adder (subtracter) in time division, which is realistic considering the circuit scale of the multiplier. It is. Furthermore, even with other configuration methods, the estimated signal can be obtained by calculating equations (2) and (3).

Next, the configuration of the tap coefficient updater is shown in FIG.

If the number of taps of the transversal filter is n,
This tap coefficient updater is composed of an arithmetic block 80 that updates n+1 tap coefficients. Each calculation block is composed of four multipliers 81.82.83.84, two adders (subtractors) 85.86 and two integrators 88.87. The error signals -〇EI, -GHQ sent from the error signal calculator 9 are given as common input signals to each calculation block, and as shown in the figure, -GEI is sent to the multipliers 81 and 82,
are input to multipliers 83 and 84. On the other hand, the Q and Q signals sent from the transversal filter are input to multipliers 81, 84, 83, and 82, respectively. Multiplier 8
1 to 85 multiply these input signals to generate four output signals, which are input to adders 85 and 86. Adders 85 and 86 perform addition (subtraction), and integrators 87 and 86 respectively perform addition (subtraction).
．． 88. Integrators 88 and 87 add the input signal to the value of the signal generated at the time of the previous sample, and use the result as an output signal.

In this way, the tap coefficient update calculation is performed.

Expressing these operations as an expression is as follows.

If the tap coefficients of the next sample are CIk' and CQk', and the input signals stored in the registers are XIk and
) (4) CQk'= CQk-G(X
Ik-EQ-XQk-EI) (5) is performed. G is the update gain.

Note that it is also possible to configure various arithmetic units within this arithmetic block 8o to calculate coefficients by time-sharing processing, and in order to repeatedly perform the same processing for the configuration of n + 1 arithmetic blocks, It is possible to configure one calculation block to perform calculations by time-sharing processing.

Next, a second embodiment will be explained using FIG. 7. 7th
The figure shows a training controller 9 in the first embodiment shown in FIG.
1. Tap coefficient memories 92 and 93 are provided. 1 to 11
Regarding, the operation is exactly the same as the first embodiment, but
The training controller 91 is configured to judge the output signal of the error signal calculator 9 and send a control signal to the tap coefficient memories 92 and 93 and the input signal memory 1. The last values of the tap coefficients of the transversal filter 2.3 are stored in the tap coefficient memories 92 and 93 each time the training period ends. The training controller 91 monitors the value of the error signal, reads out the tap coefficients of the training section where the error signal is the minimum from the tap coefficient memories 92 and 93, sets them in the transversal filter 2.3, and sets them in the training section. end. Another method is to stop repeating the training operation within that frame when the error signal becomes less than a certain value, that is, to end the I-learning section.

A two-dimensional decision feedback equalizer has been described above as an embodiment of the present invention. Of course, other automatic waveform equalizers can also be applied.

Depending on the modulation method, there are modulation methods that can be realized by one-dimensional processing, and the present invention can be applied to such cases as well. Further, although the sampling interval in the transversal filter has been described as the symbol interval (T), the present invention can also be applied to a so-called fractionally spaced equalizer that performs calculations at a sampling interval that is equal to or less than the symbol interval.

Note that all of the above configurations are in the form of functional blocks, and specific configuration methods include converting the functional blocks into hardware, or if the processing time is sufficient, M P
U (Micro Processing Unit)
Or D S P (Digital Signal
There is a method of configuring it with software using Processor signal processing L , S I ). Any case is included in the present invention.

As described above, according to this embodiment, the convergence speed can be improved and the data error rate can be reduced compared to the case where the input signal of the training section is used only once.

[Effects of the Invention] According to the present invention, it is possible to improve the convergence speed of tap coefficients and reduce the error rate without lengthening the training section of a transmission signal, that is, without reducing data transmission efficiency.

[Brief explanation of drawings]

Fig. 1 is a block diagram of a decision feedback type automatic waveform equalizer showing the first embodiment, Fig. 2 is a block diagram of a conventional decision feedback type automatic waveform equalizer, and Fig. 3 has a training section. An example of the data format of the transmitted signal, Fig. 4 is a block diagram of the error signal calculator, Fig. 5 is a block diagram of the two-dimensional transversal filter, Fig. 6 is a block diagram of the tap coefficient updater, and Fig. 7 is the block diagram of the error signal calculator. It is a block diagram showing a second example. Explanation of symbols 1...Input signal memory, 2.3...Transversal filter. 4.5...Tap coefficient updater, 9...Reference signal memory, 41...Training section, 42...Data section, 91...Training controller, 92.93. ...Tap coefficient memory Fig. 4

Claims (1)

[Claims] 1. Automatic compensation of transmission path characteristics adaptively by reducing the error signal while sequentially updating the tap coefficients of the transversal filter according to the error signal between the known signal and the estimated signal in the training section. An automatic waveform equalization method characterized in that the waveform equalizer has a memory for storing the input waveform of the training section, and the input waveform is repeatedly read from the memory and used for updating tap coefficients. 2. The automatic waveform equalization method according to claim 1, wherein the number of times the input signal of the training section is read from the memory is changed depending on the error signal. 3. The automatic waveform equalization method according to claim 1, wherein the training section is terminated when the error signal becomes less than a certain value. 4. The automatic waveform equalization method according to claim 1, further comprising a tap coefficient memory for storing the tap coefficients of the transversal filter each time the input signal waveform of the training section read out from the memory is read out from the memory, so that the error signal is minimized. An automatic waveform equalization method characterized in that the training interval is ended by reading out the tap coefficients of the training interval from the tap coefficient memory as the tap coefficients of the transversal filter.