JPH046132B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention sets the sampling interval to T seconds or less in contrast to the conventionally known frequency sampling type automatic equalizer in which the sampling interval is equal to the data symbol repetition period (T seconds). It is.
The present invention is characterized by being able to absorb timing phase shifts in the equalizer and suppressing noise outside the signal band, as compared to normal frequency sampling type automatic equalizers. Furthermore, the convergence speed in initial training is significantly improved compared to the conventional double sampling automatic equalizer using a transversal filter. below,
To distinguish terminology, a system in which the sampling interval of the received signal is equal to the data symbol repetition period T seconds will be referred to as a Nyquist sampling automatic equalizer.

In general, in data transmission using a waveform with roll-off characteristics, the frequency characteristic of the input signal to the Nyquist sampling automatic equalizer is |W|>π/T
aliasing distortion occurs due to the roll-off portion corresponding to . Due to this aliasing distortion, a zero point of the spectrum occurs at a specific timing phase shift, making it impossible to realize the inverse characteristics of the communication path in the automatic equalizer. This phenomenon commonly appears in automatic equalizers using transversal filters or frequency sampling filters. To solve this problem, there is a method of making the sampling interval shorter than T seconds in an automatic equalizer using a transversal filter (Fractionary Tap
Spacing equalizer) is known. In particular, since it is easy to implement in a digital circuit, a double sampling automatic equalizer with a sampling interval of T/2 seconds is used. However, it is known that a double sampling automatic equalizer using a transversal filter requires a very large number of repeated corrections of the tap coefficients in order to converge to the ideal tap coefficients. Theoretically, this is because if the eigenvalues are determined from the correlation matrix of the received signal, a group of eigenvalues that are very close to zero will be generated. The present invention is a method of increasing the convergence speed without causing such a decrease in the convergence speed, while at the same time absorbing the timing phase shift and suppressing out-of-band noise.

The frequency sampling type Nyquist sampling automatic equalizer uses a frequency sampling circuit as shown in FIG. 1A or FIG.
After expanding into 7 to 109, the coefficients of the filter are determined by the gradient method while being normalized according to the energy of each DFT coefficient.
Here, the frequency sampling circuit is a circuit that sequentially performs discrete Fourier transform DFT on the received signal.
This is a circuit that satisfies the definition formula of discrete Fourier transform DFT. 101 in FIG. 1A is the input signal xl at time l. 102 is the delay at time N, 103 is 1
It represents a time delay. Also, when W is set as W=exp(-j2π/N) (1), 104 is W ^-0 , 105 is W ^-1 , 1
06 is a coefficient value of W ^{- (N-1)} . 110 is 1/N. At this time, the output of this circuit is l
DFT coefficients 107-109 at time points are obtained.

When applying a method in which the sampling interval is T seconds or less to a frequency sampling type automatic equalizer, the transversal filter in the commonly used transversal type double sampling automatic equalizer is shown in Figure 1. Basically, it is replaced with a frequency sampling circuit like this. The filter coefficients are modified using the gradient method similar to the Nicequist sampling automatic equalizer. Here, the output 10 of the frequency sampling circuit
Looking at 7 to 109, these are the DFT coefficients themselves, and correspond to the output of components for each frequency band. Therefore, since sampling is performed at an interval shorter than the repetition interval of data symbols, DFT coefficients 206 to 207 corresponding to components outside the signal band
exists, and it can be considered that this component does not contain information regarding the transmitted data, and the filter coefficient by which this DFT coefficient is multiplied is zero. In other words, the circuit corresponding to this frequency component may be omitted. However, if a characteristic of the frequency sampling filter is that it is necessary to prepare a transition sample point in the cut-off frequency region, the filter coefficient corresponding to this transition sample point must not be set to zero.

As mentioned above, the signal corresponds to the out-of-band component.
By setting the filter coefficient corresponding to the DFT coefficient to zero, it is possible to completely remove noise outside the signal band. At the same time, this operation can completely prevent a reduction in convergence speed caused by a group of eigenvalues very close to zero in the transversal automatic equalizer.

An example of the circuit configuration of this frequency sampling type automatic equalizer is shown in FIG. 2 in the case where the input signal is a complex number in the QAM transmission system. Let N be the number of discrete Fourier transform points (the number of output stages of the frequency sampling circuit) in this circuit. 201 in the figure is an input signal, and the input signal at time l is x(l)
shall be. Reference numeral 202 denotes an input signal sampler, and the sampling period of this sampler is set to less than T seconds. In particular, when it is set to T/2 seconds, it is a double sampling equalizer. Furthermore, the sampled signal is converted into DFT coefficients by the frequency sampling circuit 203 (Fig. 1B), and is converted into DFT coefficients 204 to 209.
is output as kth at time l
If the DFT coefficient is Xk (l), then 204 is Xo
(l), 205 is X _M (l), 206 is X _M+1 (l), 2
07 is X _NM-1 (l), 208 is X _NM (l), 209
is expressed as X _N-1 (l). Of these, 206-2
07, that is, the components from X _M+1 (l) to X _NM-1 (l) correspond to components outside the signal band, and the filter coefficient by which this DFT coefficient is multiplied is zero. Therefore, the circuit corresponding to this component is omitted. 210 to 213 are samplers, and the data symbol repetition period T
Sampled every second. 214 to 217 are filter coefficients of the kth value at time l.
Let it be Hk(l). 218 is the output of this automatic equalizer and is designated as y(l). 219 is a reference signal;
Let it be d(l). 220 is the difference between the output signal 218 and the reference signal 219, and this is expressed as e(l).
At this time, the filter coefficient correction algorithm is given as follows.

Hk(l+1)=Hk(l)−()e(l) /gk(l) (2) y(l)= _P 〓 ^K=0 Hk(l)Xk(l) + _P 〓 ^K=0 Hk( l)×k(l) (3) e(l)=y(l)−d(l) (4) gk(l)=(1−α)gk(l−1)+|Xk(l)| ² (5) gko=<x ² > (6) where <x ² > represents the root mean square value of the input signal x(l).

In the above equation, gk(l) is a coefficient for estimating the energy of each frequency component, and other values such as a constant value, an average of absolute values, etc. may be used as gk(l). Furthermore, instead of normalizing using equation (1), each Xk(l) may be directly normalized using the square root of the estimated energy value gk(l). However, when the above formula is used, there is a trade-off point between the number of calculations and the convergence speed. α in Equation (5) is called a convergence coefficient, and in the circuit shown in FIG. 2, it is determined based on α=1/N.

Furthermore, when the input signal is a real number, the configuration becomes even simpler by using the property of the frequency sampling filter and the property that the first half and the second half of the DFT coefficient are complex conjugates.

As described above, by using the method of the present invention, it becomes possible to absorb timing phase shifts and remove out-of-band noise in a frequency sampling type automatic equalizer. Furthermore, deterioration in convergence speed that occurs in transversal type double sampling automatic equalizers can be completely prevented.

[Brief explanation of drawings]

FIGS. 1A and 1B are configuration examples of a frequency sampling circuit having N filter outputs used in the present invention, and this is also called a sliding DFT circuit. FIG. 2 shows that in the method of the present invention,
This is an implementation example of a circuit when a received signal in a QAM transmission system is a complex number.

Claims (1)

[Claims] 1. Consisting of a sampler that samples a received signal at a sampling interval shorter than the data symbol repetition period T, and a frequency sampling type automatic equalization means to which the output of this sampler is supplied,
An automatic equalizer characterized in that a filter coefficient of the automatic equalization means corresponding to a DFT coefficient corresponding to an out-of-band component of the first received signal is set to zero. 2. The automatic equalizer according to claim 1, wherein the sampling interval is T/2.