KR19990061966A - Nonlinear subchannel equalizer for partial response best sequence estimation - Google Patents

Abstract

The nonlinear equalizer for partial response maximum likelihood estimation includes subchannel generating means for generating a plurality of channel signals including a linear channel and a nonlinear channel by filtering an input signal from the channel and a plurality of channel signals generated by the subchannel generating means And filter means for performing a minimum pseudo regulated product (LMS) algorithm. Therefore, the present invention can effectively equalize a signal having a very large asymmetry phenomenon.

Description

Nonlinear subchannel equalizer for partial response best sequence estimation

The present invention relates to an equalizer for a signal supplied from a storage channel.

In the storage channel, a partial response maximum likelihood (PRML) scheme, which is a digital signal detection algorithm, is used to increase the recording capacity. The PRML system uses a linear equalizer to convert the waveform of the signal obtained from the storage channel into a desired shape (Partial Response).

In the conventional PRML system, a target polynomial P (D) close to the channel response is determined, and then waveform shaping to a desired P (D) is performed using a linear equalizer. That is, we make the total response signal of the channel response and the linear equalizer equal to the objective polynomial P (D), P (D) as the new channel response, and the controlled inter-ISI of P (D) , And performs sequential detection using a Viterbi algorithm.

Therefore, in order to reduce the burden on the equalizer, P (D), which is in good agreement with the channel response, must be obtained. In the case of a hard disk or a digital magnetic tape as a magnetic medium, since the recorded information is read by changing the magnetic field into an electrical signal, the objective polynomial P (D) = (1- D) 1 + D) ⁿ . From the frequency characteristics of P (D), it can be seen that null exists at frequencies 0 and π. Also, the value of n is chosen such that P (D) matches the characteristics of the channel, so that the noise rise in the equalizer is small. The most commonly used polynomials are PR-IV (1-D ² ) with n = 1 and EPR-IV (1 + DD ² -D ³ ) with n = 2 for hard disks. PR-Ⅰ (1 + D), PR-Ⅱ (1 + 2D + D ² ), and 1 + 2D + 2D ² + D ³ .

The existing PRML system has the structure shown in FIG. In the above structure, only the feed-forward filter performs waveform shaping, and the feedback feedback filter is fixed to produce only the error signal e (n) that updates the tap weight vector Wn. In general, it is known that the operation of the equalizer is not limited if the error rate is less than 10 ^-3 in the above structure. However, the above structure causes an invalid tap weight vector update for at least a time corresponding to the length of P (D) once an error occurs. In addition, this structure can not use the d condition of run-length limited codes, that is, RLL ( d, k ) code, which is a kind of line code used in the storage channel. In other words, an unacceptable decision sequence may be generated. For instance, "... 1 -1 ... 1", "1 ... 1 0 ..." and the like does not receive not less than d 1. For reference, in the case of RLL ( d, k ) Code, the number of consecutive data symbols, that is, the number of runs, must be in the range " d +1 to k +1". This improvement was shown in FIG. 2 below.

In the structure shown in FIG. 2, although the probability of occurrence of a decision error is increased due to an increase in the number of threshold levels, an error does not affect the subsequent tab update and the RLL ( d, k ) Can be used. Therefore, it is considered that this structure is advantageous in a storage channel in which a signal-to-noise ratio (SNR) is considerably high and is guaranteed to be about 20 dB. 2, when the RLL code and P (D) are given. A decision is made using the level of the signal that can occur and is used for tab update. When using P (D) = 1 + 2D + 2D ² + D ³ and the RLL (2,10) Code, the signal level is [-6 -4 0 4 6] or [-1 -2/3 0 2/3 1]. At this time, the threshold level is set to an intermediate value of each signal level.

Looking at the eye pattern of the reproduced signal from the optical disk, the asymmetry phenomenon occurs in which the signal forms a normal opening eye on one side and the eye shape is distorted on the other side. This is generally due to the saturation curve of the photodetector and the size of the depth or domain of the pit during recording is not optimized. The asymmetry of these signals is nonlinear and has a large effect on the performance degradation. FIG. 3 shows an example of the transfer characteristic modeling this asymmetry phenomenon.

However, since the conventional method uses a linear equalization method, the performance degradation becomes conspicuous when there is a nonlinear distortion phenomenon such as asymmetry. That is, when there is asymmetry in the waveform detected in the channel, waveform shaping can not be effectively performed with such a linear equalizer, and therefore, performance of the PRML system is deteriorated.

It is therefore an object of the present invention to provide a new adaptive equalizer capable of performing waveform shaping even in the presence of such asymmetry of signals.

Figure 1 shows a conventional partial response maximum (PRML) receiver,

2 shows a conventional PRML receiver employing a multi-level decision scheme,

3 is a graph showing asymmetric transfer characteristics,

4 illustrates a non-linear subchannel equalizer according to a preferred embodiment of the present invention,

FIG. 5 (a) is a diagram showing the output of the PRML receiver of FIG. 2, and FIG. 5 (b)

FIG. 6 shows BER performance in channel asymmetry. FIG. 6 (a) shows the case of the existing PRML, and FIG. 6 (b) shows the case of the PRML according to the present invention.

DESCRIPTION OF THE REFERENCE NUMERALS

41, 42, 43, 44, 45, 46: filters 47, 48:

49: Multilevel decision unit 50:

In order to achieve the above object, a nonlinear equalizer for partial response maximum likelihood estimation includes subchannel generating means for generating a plurality of channel signals including a linear channel and a nonlinear channel by filtering an input signal from a channel; And filter means for performing a least squares regression product (LMS) algorithm on the generated plurality of channel signals.

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

The equalizer proposed in the present invention divides a signal obtained from a channel into a plurality of sub-channels, performs adaptive equalization in each of the sub-channels, and then performs waveform shaping by adding all of them. Accordingly, this equalizer is called a nonlinear sub-channel equalizer.

Referring to FIG. 4, a non-linear subchannel equalizer is an adaptive equalizer that divides an input signal X (n) from a channel into three subchannels and then obtains a tap coefficient for each channel, And a parallel structure is formed. The three channels consist of one linear channel and two nonlinear channels. This structure is proposed under the assumption that the asymmetry phenomenon of the channel changes only the slope in either the positive section or the negative section as shown in FIG. The criterion for dividing the nonlinear section is a 3T level value L3 as shown in FIG. This criterion is based on the assumption that the asymmetry of the reproduced signal does not appear in a section where the signal amplitude is -L3 to L3. This standard is also set considering the specifications and channel characteristics of the 2.6G-byte DVD. Therefore, when the present invention is applied to another PRML system or an equalization system, it is necessary to see characteristics of asymmetry and divide nonlinear sections accordingly.

When the input signal X (n) from the channel is applied to the equalizer of FIG. 4, the first filter 41 filters the input signal X (n) to generate the linear channel signal X ₁ (n). The second filter 42 filters the input signal X (n) to generate a first non-linear channel signal X ₂ (n). The third filter 43 performs filtering on the input signal X (n) to generate the second nonlinear channel signal X ₃ (n).

First finite response (FIR) filter (44) for receiving a linear channel signal X ₁ (n) is the filter tap weight vector (Wl), a linear channel signal X ₁ (n) received by the, and the resulting And supplies the filtered linear channel signal to the first adder 47. First second finite response filter 45 and filters the tap weight vector a first non-linear channel signal X ₂ (n) received by the (Wp), and as a result of receiving the non-linear channel signal X ₂ (n) And supplies the obtained filtered first nonlinear channel signal to the second adder 48. The third finite response filter 46 receiving the second nonlinear channel signal X ₃ (n) filters the received second nonlinear channel signal X ₂ (n) using the tap weight vector Wn, And supplies the obtained filtered second nonlinear channel signal to the second adder 48. The second adder 48 adds the filtered first nonlinear channel signal and the filtered second nonlinear channel signal and supplies the result of the addition to the first adder 47. The first adder 47 adds the filtered linear channel signal to the signal supplied from the second adder 48 and generates an estimate Y (n). The estimated value Y (n) is input to a multi-level decision unit (49) and a subtracter (50). The multi-level determination unit 49 performs a level determination operation on the input estimate value Y (n), and supplies the result to the subtractor 50. [ The subtracter 50 subtracts the estimated value Y (n) received from the first adder 47 from the signal applied from the multilevel decision unit 49 and outputs the result, that is, the error signal e (n) 3 finite response filters 44, 45, and 46, respectively. The first to third finite response filters 44, 45 and 46 update the tap-weight vector using the error signal e (n) supplied from the subtractor 50.

These finite response filters 44, 45, and 46 update the tap weight vectors using a Least Mean Square (LMS) algorithm represented by the following Equation 1-5.

^{_{Y (n) = W l T}} (n) X 1 (n) + W p T (n) X 2 (n) + W n T (3) X 3 (n)

e (n) = d (n) - Y (n)

W _p (n + 1) = W _p (n) + e (n) X ₂ (n)

W _n (n + 1) = W _n (n) + μ e (n) X ₃ (n)

In the above equations, W (n) separated by its subscript is the tap weight vector of the corresponding finite response filter, and X (n) is a vector representing the input of each finite response filter. is the step-size parameter of the LMS algorithm. e (n) is an estimation error. And d (n) is a desired response.

Therefore, the first finite response filter 44 updates the tap weight vector Wl using equation (3), and the second finite response filter 45 updates the tap weight vector Wp using equation (4) do. Then, the third finite response filter 46 updates the tap weight vector Wn using Equation (5). As a result, the adder 47 generates the estimated value Y (n) expressed by Equation (1). The multi-level determination unit 49 and the subtractor 50 implementing the equation (2) subtract the estimated value Y (n) from the desired response d (n).

The estimated value Y (n) thus produced is transmitted to a Viterbi detector (not shown) for performing sequence detection.

A DVD system was selected as a system for verifying the equalizer of FIG. 4 according to the present invention and simulations were performed. The channel model used here was a trigonometric model used by Sony and Philips. The following Table 1 shows the parameters of the DVD used in the simulation and their specifications.

parameter details Laser wavelength 650 nm The numerical aperture of the objective lens 0.60 Modulation code 8/16, RLL (2,10) Reference linear scanning speed 6 m / s Track pitch 0.74 μm Minimum foot length 0.41 m

Simulation results are compared with the final bit error rate (BER) for the asymmetric transfer characteristic shown in FIG. The input SNR of the data obtained from the channel is 15 to 19 dB, and the performance change is observed while changing the value of the slope m in FIG. At this time, the number of taps of the FIR filter in the existing method and the proposed method is 27 tabs. FIG. 5 shows an equalizer output when the input SNR is 17 dB. FIG. 5 (a) shows the output of the apparatus of FIG. 2, and FIG. 5 (b) shows the output of the equalizer of FIG. . FIG. 6 shows the BER. FIG. 6A shows the BER of the PRML receiver of FIG. 2 with channel asymmetry, FIG. 6B shows the BER of the equalizer of FIG. 4 with channel asymmetry, Lt; / RTI > P (D) = 1 + 2D + 2D ² + D ³ was used as the target partial response. The asymmetric slopes in FIG. 6 represent the values of the slope (m) of the asymmetrical transfer characteristic shown in FIG. 3, and the case where the channel asymmetry phenomenon occurs in the 'negative region' as shown in FIG. The BER obtained from simulation shows that there is little difference in performance between m = 0.5 and 1.0 because the gain in the Viterbi detector is large. However, when the slope is smaller than 0.5, the difference in performance is remarkable.

As described above, the equalizer according to the present invention effectively reduces the magnitude of the error signal when there is an asymmetry in the signal obtained from the channel. That is, PR-shaping can be effectively performed even in the presence of signal asymmetry, and as a result, BER performance in a Viterbi detector can be improved. Therefore, it is very useful in a high-density optical disc system in which signal asymmetry phenomenon is very large.

Claims (1)

In a nonlinear equalizer for partial response maximum likelihood estimation, A subchannel generating means for generating a plurality of channel signals including a linear channel and a nonlinear channel by filtering an input signal from the channel; And And filter means for performing a minimum pseudo regulated product (LMS) algorithm on the generated plurality of channel signals.