JP2004095096A - Digital signal recording and reproducing device, recording and reproducing method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a digital signal recording and reproducing device in which influence of crosstalks caused when writing by a write-head and read-out by a read are performed simultaneously can be lightened. <P>SOLUTION: This device is provided with a write-head 101 writing data, a read-head 103 reading data, and a signal equalizing circuit to which a digitally reproduced signal reproduced by the read-head 103 and a recording signal by the write-head 101 are inputted. The signal equalizing circuit constitutes a non-linear filter 106 using a neural net, and can effectively reduce influence of crosstalks. <P>COPYRIGHT: (C)2004,JPO

Description

【００３６】
【数２】
ｆ（ｕ）＝１／（１＋ｅｘｐ（−Ｃ・ｕ））
の式が成立する。数２で示されるｆ（ｕ）は飽和型であるシグモイド関数であり、非線形性を持つ。さらに、定数Ｃはシグモイド関数の傾きを決定するパラメータであり、図３ではＣ＝１の場合を示している。
【００３７】
ここで、中間層ユニット１１６に非線形関数であるシグモイド関数を用いると、次のような効果が得られる。
【００３８】
即ち、入力層ユニット１１５にクロストークの影響をうけたディジタル再生信号Ｒ_ｉ（ｋ）が入力された場合、つまり、これは通常の振幅値から大きく外れたデータである可能性がある。この結果、中間層ユニット１１６に入力される入力値の絶対値が非常に大となっていた場合、従来の線形等化器では、このクロストークの影響を軽減することができず、理想値から大きく外れた値を出力してしまう。しかしながら、非線形の飽和関数であるシグモイド関数を使うことで、このような現象が起きたときに中間層ユニットの出力値を有限範囲内で抑えることができるため、クロストークの影響を効果的に軽減できる一因となる。
【００３９】
また、図示していないが、シグモイド関数の傾きＣを変化させると、関数の入力ｕに対して出力ｆ（ｕ）が飽和する範囲が変化する。そのため、実際に発生するクロストークによるディジタル再生信号Ｒ_ｉ（ｋ）の振幅の変化量に合わせて最適なクロストークの影響を抑える値にフィッティングすることが可能である。
【００４０】
中間層ユニット１１６の出力は中間層ｊ番目の中間層ユニット１１６と出力層ユニット１１７間の結合荷重Ｗ_ｊ（ｋ）が掛けられて、出力層ユニット１１７への入力の一つになる。出力層ユニット１１７では、入力された信号の総和を出力する。すなわち、出力層ユニット１１７の出力Ｚ（ｋ）は次式のようになる。
【００４１】
【数３】

【００４２】
結合荷重Ｖ_ｊｉ（ｋ）、Ｗ_ｊ（ｋ）は数４、数５を繰り返すことで修正される誤差逆伝播法によって学習が行われ決定される。
【００４３】
【数４】
Ｗ_ｊ（ｋ）＝Ｗ_ｊ（ｋ）＋ΔＷ_ｊ（ｋ）
【００４４】
【数５】
Ｖ_ｊｉ（ｋ）＝Ｖ_ｊｉ（ｋ）＋ΔＶ_ｊｉ（ｋ）
誤差逆伝播法とは、出力層のユニットが教師信号にできるだけ近い値を出力するように各結合荷重を調整するものである。すなわち、出力層ユニット出力と教師信号の２乗誤差を結合荷重で微分し、それに基づいて最急降下法を適用し、結合荷重の変化量を求める方法である。ここで、非線形フィルタ１０６の理想出力を教師信号として用いる。
【００４５】
まず全ての結合荷重の値に、初期値としてランダムな小さな値を設定し、ｋ番目の入力データが入力された時に期待される教師信号をＴ（ｋ）とすると、出力層ユニット１１７の出力Ｚ（ｋ）との２乗誤差Ｅは
【００４６】
【数６】
Ｅ＝１／２（Ｔ（ｋ）−Ｚ（ｋ））・（Ｔ（ｋ）−Ｚ（ｋ））
となる。数６の２乗誤差Ｅをもとに、出力層、中間層の結合荷重の変化量ΔＷ_ｊ（ｋ）、ΔＶ_ｊｉ（ｋ）は以下の式により求められる。
【００４７】
【数７】
ΔＷ_ｊ（ｋ）＝α（Ｔ（ｋ）−Ｚ（ｋ））・Ｙ_ｊ（ｋ）
【００４８】
【数８】
ΔＶ_ｊｉ（ｋ）＝β（Ｔ（ｋ）−Ｚ（ｋ））・Ｗ_ｊ（ｋ）・Ｙ_ｊ（ｋ）・（１−Ｙ_ｊ（ｋ））
ここで、α、βはそれぞれ学習のスピードを定める定数である。これらの式より、出力層ユニット１１７が教師信号にできるだけ近い値を出力するように出力層、中間層の結合荷重が更新される学習が行われる。
【００４９】
以上の学習アルゴリズムにより、学習用に作成したテストパターンを用いて事前学習を行うことで出力層、中間層の各結合荷重が更新される。ここで、本実施の形態のように第２の入力層を設け、隣接するライトヘッドの記録信号を入力とし学習を行うことで、第２の入力層ユニットと中間層ユニット間の結合荷重は、クロストークが発生する場合にはクロストークの影響を軽減するように更新される。この結果、クロストークが発生した場合においても上述した非線形フィルタ１０６は理想出力に近い値を出力し、クロストークの影響を軽減することが可能となる。
【００５０】
また、上述した非線形フィルタ１０６では、入力層Ａに入力されるディジタル再生信号と、入力層Ｂに入力される隣接するライトヘッドに用いる記録信号の同期が取れていない場合においても、入力層Ｂのユニットを複数にすることで、クロストークの影響を軽減することが可能となる。これは、入力層Ａに入力されるｋ番目のディジタル再生信号Ｒ_ｉ（ｋ）（ｉ＝０，・・・，ＮＡ）がクロストークの影響を受けたものであるとすると、クロストークの発生原因となる隣接するライトヘッドのｋ番目の記録信号Ｗ_ｉ（ｋ）（ｉ＝ＮＡ＋１、・・・、ＮＡ＋ＮＢ）が入力層Ｂのユニットのいずれかに入力されることで、クロストークの影響を軽減するように結合係数の更新がなされる学習が行われ、その結果、理想値に近い出力を得ることができる。即ち、入力層Ａの入力ユニット１１５に入力されるｋ番目のディジタル再生信号Ｒ_ｉ（ｋ）に対応するｋ番目の記録信号Ｗ_ｉ（ｋ）はＮＡ＋１≦ｉ≦ＮＡ＋ＮＢである入力層Ｂの入力層ユニット１１５に入力されればよい。
【００５１】
以上のように、本実施の形態によるディジタル信号記録再生装置によれば、クロストークの影響を含むディジタル再生信号と、クロストークの発生原因となる隣接するライトヘッドの記録信号を非線形フィルタ１０６に入力し、理想出力値と出力ユニットの誤差を利用して学習を行うことで各結合係数が決定され、そのためクロストークの影響が効果的に軽減される作用を有する。
【００５２】
なお、本実施の形態のディジタル磁気記録再生装置においては、非線形フィルタ１０６の中間層ユニット１１６における特性関数としてシグモイド関数を用いたが、一定以上の入力値に対して出力値を一定以内で抑えることのできる関数であれば他の関数を用いても良い。その場合においても、学習により各結合係数はクロストークの影響を軽減するように決定される。
【００５３】
また、本実施の形態の非線形フィルタ１０６は入力層、中間層、出力層からなる３層構造階層型ニューラルネットワークであったが、本発明はこれに限定されるものではなく、例えば中間層が２層以上（ｎ≧２）であってもよい。
【００５４】
学習についてはテストパターンを用いて事前学習を行い各結合係数を決定したが、本発明はこれに限定されるものではない。即ち、ヘッドや電気部品の経時変化による特性変化によってクロストークの状態が変わっても、例えば記録媒体の出し入れを行う度に学習を行うことで、最適な特性を保つことが可能である。また、入力層Ｂ１１１に入力される記録信号を隣接ヘッドで用いられるものとして説明を行ったが、これは隣接ヘッドのみに限定されるものではない。即ち、クロストークの影響を及ぼす可能性のあるライトヘッドからの記録信号であれば、それに起因するクロストークは本発明によって効果的に軽減することが可能である。複数以上の記録信号を入力層Ｂ１１１に入力する場合は、入力層Ｂの入力ユニットを増やしてそれぞれの記録信号を入力するか、または複数以上の記録信号の和を入力する等の構成にすればよい。
【００５５】
更に、上記実施の形態のディジタル信号記録再生装置では、回転ヘッドにヘッドを搭載した磁気記録再生装置の例で説明したが、本発明はこれに限定されるものではなく、固定ヘッドを用いたテープ記録装置や、ハードディスクのように回転媒体に記録する記録再生装置であってもよい。
【００５６】
また、本発明は、上述した本発明の記録再生装置の信号等化回路を、コンピュータを利用して実現するプログラムである。さらにはそのようなプログラムを格納したＣＤＲＯＭなどの、コンピュータでアクセス可能な媒体である。
【００５７】
このように、本発明の各構成要素は、ＡＮＤ，ＯＲ回路等のハードウェア回路で実現しても良いし、コンピュータを利用してソフトウェア的に実現してもかまわない。
【００５８】
【発明の効果】
以上のように本発明のディジタル信号記録再生装置によれば、クロストークの影響を効率よく軽減された出力を得ることが可能となるディジタル信号記録再生装置を得ることができる。
【図面の簡単な説明】
【図１】本発明の実施の形態におけるディジタル信号記録再生装置の回路図である。
【図２】同装置における非線形フィルタの回路図である。
【図３】同装置における非線形フィルタで用いられる非線形関数を説明するための図である。
【図４】従来のディジタル信号記録再生装置の回路図である。
【図５】従来のディジタル信号記録再生装置において用いられているライトヘッド、リードヘッドを有する回転ドラムを説明するための図である。
【図６】次世代のディジタル信号記録再生装置において用いられる、従来例に比べて多数のライトヘッド及びリードヘッドを有する回転ドラムを説明するための図である。
【符号の説明】
１０１　　　　ライトヘッド
１０２　　　　記録媒体
１０３　　　　リードヘッド
１０４　　　　ローパスフィルタ
１０５　　　　Ａ／Ｄ変換器
１０６　　　　非線形フィルタ
１１０　　　　入力層Ａ
１１１　　　　入力層Ｂ
１１２　　　　中間層
１１３　　　　出力層
１１４　　　　遅延素子
１１５　　　　入力層ユニット
１１６　　　　中間層ユニット
１１７　　　　出力層ユニット
１１８　　　　乗算器[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a digital signal recording / reproducing apparatus (International Patent Classification G11B 20/10), and particularly to crosstalk from an adjacent write head included in a digital signal read from a magnetic recording medium such as a magnetic tape. It is about technology to reduce the impact.
[0002]
[Prior art]
The flow of signal processing when recording and reproducing data in a conventional magnetic tape device will be described with reference to FIG.
[0003]
Data recorded on the recording medium 102 by the write head 201 is read by the read head 203 and input to the analog pre-filter 204. The analog reproduction signal input to the analog pre-filter 204 is subjected to high-frequency noise cut and waveform equalization. Subsequently, the analog reproduced signal whose waveform has been equalized by the analog pre-filter 204 is sampled by the A / D converter 105 to be a digital reproduced signal, and further subjected to waveform equalization by a post-equalizer (not shown). The decoding is performed by a decoding means (not shown) to obtain binary data.
[0004]
Here, in a conventional magnetic tape device, during a normal write operation, data is written on a magnetic tape as a recording medium by a write head 201, and immediately after that, an operation called read-after-write is performed by a read head 203 to check data read. When a read error is detected at this time, a write retry operation of writing data again by the write head 201 is executed.
[0005]
An example of the operation of a conventional magnetic tape device will be described with reference to the drawings specifically for read-after-write. FIG. 6 shows a rotary drum 210 provided with a read head and a write head employed in a conventional digital signal recording / reproducing apparatus.
[0006]
Write heads 201a and 201b and read heads 203a and 203b are alternately arranged on the rotating drum 210, and two sets of read heads and write heads (a set of a write head 201a and a read head 203a, a set of a write head 201b and a read The head 203b) records and reproduces data on recording tracks having different azimuth angles. That is, recording and reproduction of different data (Ach, Bch) are alternately performed using two sets of read heads and write heads.
[0007]
In the rotary drum 210, when data is written on the magnetic tape 213 by the write head 201a, a read-out check of the data is performed by the read head 203a arranged behind the write head 201a by 回 転 rotation of the rotary drum. If a read error is detected at this time, the data is written again by the write head 201a.
[0008]
[Patent Document 1]
JP 9-245307 A
[Problems to be solved by the invention]
In a digital magnetic recording / reproducing system using a next-generation magnetic tape as a recording medium, three or more sets of different data are simultaneously signal-processed in order to perform recording / reproducing signal processing at a higher speed. As an example, when eight write heads and eight read heads are installed on the rotating drum 210, and the magnetic tape is wound around the center of the rotating drum by 180 degrees, a structure as shown in FIG. 4 is obtained.
[0010]
In FIG. 4, the write head 101a, the write head 101e, the read head 103a, and the read head 103e are Ach, the write head 101b, the write head 101f, the read head 103b, and the read head 103f are Bch, the write head 101c, the write head 101g, and the read head. A read head 103c and a read head 103g are a Cch, a write head 101d, a write head 101h, a read head 103d, and a read head 103h are a Dch write head and a read head.
[0011]
Here, as shown in FIG. 4, eight read heads and write heads are provided for four sets of mutually different data because a magnetic tape wound 180 degrees around the center of a rotating drum is used for four channels simultaneously. Performing recording / reproducing and performing read / write by alternately switching between two read heads and write heads provided for each channel every time the rotary drum makes a half rotation, thereby performing faster recording / reproducing signal processing. That's why.
[0012]
If read-after-write is performed in such a state, for example, when data is read by the read head 103a, another data is written in the adjacent write head 101b. In this case, the data read from the read head 103a is affected not only by the data recorded on the recording track of Ach, but also by the magnetic field generated by the operation of the write head 101b writing on Bch, which causes deterioration of the reproduction signal. I will.
[0013]
When crosstalk occurs due to the recording operation of such an adjacent write head, if the conventional digital signal recording / reproducing apparatus is used as it is, the reproduced signal is degraded, and the waveform equalization is not performed properly. However, there has been a problem that this leads to a decrease in
[0014]
Further, such crosstalk depends on a recording operation of an adjacent write head, and is a non-linear phenomenon caused by data on another recording track. Therefore, it is difficult to efficiently remove the influence of crosstalk even if a conventional linear equalizer is used.
[0015]
The present invention considers the problems of the conventional signal recording / reproducing apparatus and considers the influence of crosstalk that occurs when writing by the write head and reading by the read head occur simultaneously when performing the above-mentioned read-after-write. It is an object of the present invention to provide a digital signal recording / reproducing apparatus which can reduce the number of digital signals.
[0016]
[Means for Solving the Problems]
According to a first aspect of the present invention, there is provided a write head for writing data on a recording medium, a read head for reading data written on the recording medium, and a digital reproduction signal for digitizing an analog reproduction signal read by the read head. An A / D converter that outputs a digital reproduction signal, and a signal equalization circuit that receives the digital reproduction signal as an input.
A digital signal recording / reproducing apparatus for reducing the influence of crosstalk occurring when a write operation of the write head occurs simultaneously with a read operation of the read head by performing non-linear equalization by the signal equalization circuit.
[0017]
According to a second aspect of the present invention, the signal equalizing circuit comprises:
A second unit that receives the digital reproduction signal, delays the reproduction signal by a predetermined time by a plurality of first delay elements connected in cascade, and outputs each of the delayed digital reproduction signals. One input layer,
The write head inputs a recording signal used when writing data, delays the recording signal by a predetermined time by a plurality of cascade-connected second delay elements, and outputs each of the delayed recording signals. A second input layer having a plurality of output units;
The output of each unit of the first input layer and the second input layer is used as an input, and the product sum of the nth (n is a natural number) equalization coefficient determined by learning and the output of each unit of the preceding layer is calculated by a nonlinear function. An intermediate layer composed of n layers for converting and outputting;
An output layer that outputs a product sum of an output of each unit of the intermediate layer and an (n + 1) th equalization coefficient determined by the learning,
The first to (n + 1) th equalization coefficients are determined by learning performed to minimize an equalization error between the output of the output layer and a predetermined equalization target value. It is a recording and reproducing device.
[0018]
According to a third aspect of the present invention, when the signal equalization circuit is not synchronized between the digital reproduction signal input to the first input layer and the recording signal input to the second input layer, Also, there is provided a digital signal recording / reproducing apparatus according to the first aspect of the present invention which performs the learning so as to reduce the influence of the crosstalk.
[0019]
According to a fourth aspect of the present invention, there is provided a write head for writing data on a recording medium, a read head for reading data written on the recording medium, and a digital reproduction signal for digitizing an analog reproduction signal read by the read head. A digital signal recording / reproducing apparatus, comprising: an A / D converter that outputs the digital reproduction signal; and a signal equalization circuit that receives the digital reproduction signal.
A digital signal recording / reproducing method for reducing the influence of crosstalk generated when the write operation of the write head occurs simultaneously with the read operation of the read head by performing nonlinear equalization by the signal equalization circuit.
[0020]
According to the above-described configuration of the present invention, the influence of crosstalk that occurs when writing by the write head and reading by the read head occur simultaneously when performing read-after-write is reduced.
[0021]
According to a fifth aspect of the present invention, there is provided a program for causing a computer to function as the signal equalizing circuit of the digital signal recording / reproducing apparatus according to the first aspect of the present invention.
[0022]
A sixth aspect of the present invention is a medium carrying the program of the fifth aspect of the present invention, wherein the medium can be processed by a computer.
[0023]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to FIGS. 1, 2, and 3. FIG. FIG. 1 shows a configuration example of a digital signal recording / reproducing apparatus according to an embodiment of the present invention.
[0024]
In FIG. 1, data recorded on a recording medium 102 by a write head 101 is read by a read head 103 and input to a low-pass filter 104. The analog reproduction signal input to the low-pass filter 104 is cut in high-frequency noise by limiting high-frequency passage.
[0025]
Here, since the low-pass filter 104 only needs to have a function of limiting the high-pass frequency, it may be constituted by a filter having a first-order or second-order. That is, in the conventional example, a high-order filter having a complicated structure was used to play a role of cutting high-frequency noise and performing waveform equalization, but in the present invention, the waveform equalization of the digital reproduction signal is performed. Since the waveform equalization is performed by the nonlinear filter 106 that reduces the influence of crosstalk by using the recording signal from the write head 101, the low-pass filter 104 does not need to have a complicated structure. For this reason, when the number of analog components is reduced, variations in individual performance when mounted are reduced.
[0026]
The analog reproduced signal whose high-pass frequency is restricted is sampled by the A / D converter 105 and converted into a digital reproduced signal.
[0027]
The digital reproduction signal sampled by the A / D converter 105 is input to the nonlinear filter 106 together with a recording signal used when the write head 101 performs recording. A non-linear filter 106, which is an example of the signal equalization circuit of the present invention, performs waveform equalization of a digital reproduction signal and outputs a signal in which the influence of crosstalk is reduced.
[0028]
The signal whose waveform has been equalized by the non-linear filter 106 and the influence of the crosstalk has been reduced is subjected to further waveform equalization by a post-equalizer (not shown), and then decoded by a decoding means (not shown). A value conversion process is performed.
[0029]
Here, the operation of the nonlinear filter 106 will be described with reference to FIGS. In the present invention, a non-linear filter is configured using a neural network in which functions of nerve cells constituting the brain of a higher organism are engineered.
[0030]
FIG. 2 shows a neural network type nonlinear filter according to the present invention in which the number of intermediate layers composed of n layers is one (n = 1), that is, a three-layer structure of an input layer, an intermediate layer, and an output layer. In FIG. 2, the nonlinear filter 106 includes an input layer A 110 and an input layer B 111 having a plurality of delay elements 114 and a plurality of input layer units 115 connected in cascade, and an intermediate filter having a multiplier 118 and a plurality of intermediate layer units 116. It comprises a layer 112 and an output layer 113 having a multiplier 118 and one output layer unit 117.
[0031]
The digital reproduction signal output from the A / D converter 105 is delayed by the delay element 114 having one data sampling clock time delay amount, and is output to the next-stage delay element 114. The digital reproduction signal delayed by the delay element 114 is input to each of the input layer units 115 including the NA input layers A110.
[0032]
As in the case of the input layer A, a recording signal, which is digital data to be recorded by an adjacent write head, is delayed and input to each of the NB input layer units 115 of the input layer B. .
[0033]
Each input layer unit 115 in the input layer A110 and the input layer B111 outputs the input signal as it is. That is, assuming that the input layer A110 and the input layer B111 are collectively considered as an input layer composed of NA + NB input layer units, the k-th digital reproduction signal R _i (k) or the recording signal _Wi (k) is input. Here, 0 <digital reproduction signal described above in the case of i ≦ NA _R i (k) is, in the case of NA + 1 ≦ i ≦ NA + NB recording signal _W i described above (k) is the input of the input layer unit 115.
[0034]
The output X _i (k) from the input layer unit 115 is multiplied by a multiplier 118 by a coupling weight V _ji (k) between the i-th input layer unit 115 of the input layer and the j-th intermediate layer unit 116 of the intermediate layer. , One of the inputs to the intermediate layer unit 116. The output Y _j (k) of the intermediate layer unit 116 is obtained by transforming the sum U _j (k) of the inputs of the intermediate layer unit 116 by a sigmoid function as shown in FIG. That is,
[0035]
(Equation 1)

[0036]
(Equation 2)
f (u) = 1 / (1 + exp (−C · u))
Is satisfied. F (u) shown in Expression 2 is a sigmoid function of a saturation type and has nonlinearity. Further, the constant C is a parameter for determining the slope of the sigmoid function, and FIG. 3 shows a case where C = 1.
[0037]
Here, when a sigmoid function that is a nonlinear function is used for the intermediate layer unit 116, the following effects can be obtained.
[0038]
That is, when the digital reproduction signal R _i (k) affected by the crosstalk is input to the input layer unit 115, that is, there is a possibility that the digital reproduction signal R _i (k) is data greatly deviating from a normal amplitude value. As a result, if the absolute value of the input value input to the intermediate layer unit 116 is very large, the conventional linear equalizer cannot reduce the influence of the crosstalk, and It outputs a value that deviates greatly. However, by using the sigmoid function, which is a nonlinear saturation function, when such a phenomenon occurs, the output value of the intermediate layer unit can be suppressed within a finite range, thus effectively reducing the effect of crosstalk. It can be a factor.
[0039]
Although not shown, when the slope C of the sigmoid function is changed, the range in which the output f (u) is saturated with respect to the input u of the function changes. Therefore, it is possible to fit an optimum value for suppressing the influence of crosstalk in accordance with the amount of change in the amplitude of the digital reproduction signal R _i (k) due to crosstalk actually occurring.
[0040]
The output of the intermediate layer unit 116 is multiplied by the coupling load W _j (k) between the jth intermediate layer unit 116 of the intermediate layer and the output layer unit 117, and becomes one of the inputs to the output layer unit 117. The output layer unit 117 outputs the sum of the input signals. That is, the output Z (k) of the output layer unit 117 is as follows.
[0041]
[Equation 3]

[0042]
The connection weights V _ji (k) and W _j (k) are determined by learning by an error backpropagation method modified by repeating Equations 4 and 5.
[0043]
(Equation 4)
W _j (k) = W _j (k) + ΔW _j (k)
[0044]
(Equation 5)
V _ji (k) = V _ji (k) + ΔV _ji (k)
The error backpropagation method adjusts each connection weight so that the output layer unit outputs a value as close as possible to the teacher signal. That is, the square error between the output of the output layer unit and the teacher signal is differentiated by the connection weight, and the steepest descent method is applied based thereon, thereby obtaining the change amount of the connection weight. Here, the ideal output of the nonlinear filter 106 is used as a teacher signal.
[0045]
First, a random small value is set as an initial value for all connection weight values, and an expected teacher signal when the k-th input data is input is T (k), and an output Z of the output layer unit 117 is obtained. The square error E with (k) is
(Equation 6)
E = １ ／ (T (k) −Z (k)) · (T (k) −Z (k))
It becomes. Based on the squared error E of Equation 6, the change amounts ΔW _j (k) and ΔV _ji (k) of the connection load of the output layer and the intermediate layer are obtained by the following equations.
[0047]
(Equation 7)
ΔW _j (k) = α (T (k) −Z (k)) · Y _j (k)
[0048]
(Equation 8)
ΔV _ji (k) = β (T (k) −Z (k)) · W _j (k) · Y _j (k) · (1−Y _j (k))
Here, α and β are constants that determine the learning speed. From these equations, learning is performed in which the output weights of the output layer and the intermediate layer are updated so that the output layer unit 117 outputs a value as close as possible to the teacher signal.
[0049]
According to the learning algorithm described above, the connection weights of the output layer and the intermediate layer are updated by performing pre-learning using the test pattern created for learning. Here, the coupling load between the second input layer unit and the intermediate layer unit is reduced by providing the second input layer as in the present embodiment and performing learning with the recording signal of the adjacent write head as an input. When crosstalk occurs, it is updated so as to reduce the influence of crosstalk. As a result, even when crosstalk occurs, the above-described nonlinear filter 106 outputs a value close to the ideal output, and the influence of crosstalk can be reduced.
[0050]
Further, in the above-described nonlinear filter 106, even when the digital reproduction signal input to the input layer A is not synchronized with the recording signal input to the input layer B and used for the adjacent write head, the input layer B is not synchronized. By using a plurality of units, it is possible to reduce the influence of crosstalk. This is because if the k-th digital reproduction signal R _i (k) (i = 0,..., NA) input to the input layer A is affected by the crosstalk, the crosstalk occurs. The k-th recording signal W _i (k) (i = NA + 1,..., NA + NB) of the adjacent write head causing the input is input to any of the units of the input layer B, thereby reducing the influence of crosstalk. Learning in which the coupling coefficient is updated so as to reduce it is performed, and as a result, an output close to the ideal value can be obtained. That is, the k-th recording signal W _i (k) corresponding to the k-th digital reproduction signal R _i (k) input to the input unit 115 of the input layer A is input to the input layer B where NA + 1 ≦ i ≦ NA + NB. What is necessary is just to input to the layer unit 115.
[0051]
As described above, according to the digital signal recording / reproducing apparatus of the present embodiment, the digital reproduced signal including the influence of the crosstalk and the recording signal of the adjacent write head which causes the crosstalk are input to the nonlinear filter 106. Then, by performing learning using the error between the ideal output value and the output unit, each coupling coefficient is determined, and thus the effect of crosstalk is effectively reduced.
[0052]
In the digital magnetic recording / reproducing apparatus according to the present embodiment, the sigmoid function is used as the characteristic function in the intermediate layer unit 116 of the non-linear filter 106. Other functions may be used as long as the function can be used. Even in that case, each coupling coefficient is determined by learning so as to reduce the influence of crosstalk.
[0053]
Although the nonlinear filter 106 of the present embodiment is a three-layer hierarchical neural network having an input layer, an intermediate layer, and an output layer, the present invention is not limited to this. The number of layers may be more than (n ≧ 2).
[0054]
As for learning, each coupling coefficient is determined by performing preliminary learning using a test pattern, but the present invention is not limited to this. That is, even if the state of the crosstalk changes due to a change in characteristics of the head or the electric component due to a change over time, optimal characteristics can be maintained by learning, for example, each time a recording medium is inserted and removed. Further, the description has been made assuming that the recording signal input to the input layer B111 is used by the adjacent head, but this is not limited to only the adjacent head. That is, if the recording signal is from a write head that may affect the crosstalk, the crosstalk caused by the recording signal can be effectively reduced by the present invention. When a plurality of recording signals are input to the input layer B111, the input units of the input layer B may be increased to input the respective recording signals, or the sum of the plurality of recording signals may be input. Good.
[0055]
Further, in the digital signal recording / reproducing apparatus of the above-described embodiment, an example of a magnetic recording / reproducing apparatus in which a head is mounted on a rotary head has been described. However, the present invention is not limited to this, and a tape using a fixed head is used. It may be a recording device or a recording / reproducing device that records on a rotating medium such as a hard disk.
[0056]
Further, the present invention is a program for realizing the above-described signal equalization circuit of the recording / reproducing apparatus of the present invention using a computer. Further, it is a computer-accessible medium such as a CDROM storing such a program.
[0057]
Thus, each component of the present invention may be realized by a hardware circuit such as an AND or OR circuit, or may be realized by software using a computer.
[0058]
【The invention's effect】
As described above, according to the digital signal recording / reproducing apparatus of the present invention, it is possible to obtain a digital signal recording / reproducing apparatus capable of obtaining an output in which the influence of crosstalk is efficiently reduced.
[Brief description of the drawings]
FIG. 1 is a circuit diagram of a digital signal recording / reproducing apparatus according to an embodiment of the present invention.
FIG. 2 is a circuit diagram of a nonlinear filter in the device.
FIG. 3 is a diagram for explaining a nonlinear function used in a nonlinear filter in the device.
FIG. 4 is a circuit diagram of a conventional digital signal recording / reproducing apparatus.
FIG. 5 is a diagram for explaining a rotary drum having a write head and a read head used in a conventional digital signal recording / reproducing apparatus.
FIG. 6 is a diagram for explaining a rotating drum having a larger number of write heads and read heads than a conventional example, which is used in a next-generation digital signal recording / reproducing apparatus.
[Explanation of symbols]
101 write head 102 recording medium 103 read head 104 low pass filter 105 A / D converter 106 nonlinear filter 110 input layer A
111 Input layer B
112 Middle layer 113 Output layer 114 Delay element 115 Input layer unit 116 Middle layer unit 117 Output layer unit 118 Multiplier

Claims (6)

A write head for writing data on a recording medium, a read head for reading data written on the recording medium, and an A / D converter for digitizing an analog reproduction signal read by the read head and outputting a digital reproduction signal Device, a signal equalization circuit that receives the digital reproduction signal as input,
A digital signal recording / reproducing apparatus for reducing the influence of crosstalk occurring when a write operation of the write head occurs simultaneously with a read operation of the read head by performing non-linear equalization by the signal equalization circuit. The signal equalization circuit,
A second unit that receives the digital reproduction signal, delays the reproduction signal by a predetermined time by a plurality of first delay elements connected in cascade, and outputs each of the delayed digital reproduction signals. One input layer,
The write head inputs a recording signal used when writing data, delays the recording signal by a predetermined time by a plurality of cascade-connected second delay elements, and outputs each of the delayed recording signals. A second input layer having a plurality of output units;
The output of each unit of the first input layer and the second input layer is used as an input, and the product sum of the nth (n is a natural number) equalization coefficient determined by learning and the output of each unit of the preceding layer is calculated by a nonlinear function. An intermediate layer composed of n layers for converting and outputting;
An output layer that outputs a product sum of an output of each unit of the intermediate layer and an (n + 1) th equalization coefficient determined by the learning,
The digital signal according to claim 1, wherein the first to (n + 1) th equalization coefficients are determined by learning performed to minimize an equalization error between an output of the output layer and a predetermined equalization target value. Recording and playback device. The signal equalizing circuit is capable of controlling the influence of the crosstalk even when the digital reproduction signal input to the first input layer and the recording signal input to the second input layer are not synchronized. 2. The digital signal recording / reproducing apparatus according to claim 1, wherein the learning is performed so as to reduce the learning. A write head for writing data on a recording medium, a read head for reading data written on the recording medium, and an A / D converter for digitizing an analog reproduction signal read by the read head and outputting a digital reproduction signal Recording and reproducing method of a digital signal recording and reproducing apparatus, comprising:
A digital signal recording / reproducing method for reducing the influence of crosstalk occurring when a write operation of the write head occurs simultaneously with a read operation of the read head by performing non-linear equalization by the signal equalization circuit. A program for causing a computer to function as the signal equalization circuit of the digital signal recording / reproducing apparatus according to claim 1. A medium carrying the program according to claim 5, wherein the medium can be processed by a computer.

Images