JPH0620075A - Waveform processing device of neural network and its design method - Google Patents

Abstract

PURPOSE:To eliminate plural signal degradation factors in a recording/ reproducing system or a transmission system by switching a fixed weight in accordance with the detection value of an auxiliary detection means. CONSTITUTION:The output of a tilt sensor LC which detects the angle formed by the optical axis of the laser beam thrown to an optical disk 103 and the surface of the disk is inputted to a buffer 111, and the output of the buffer 111 is digitized by an A/D converter 112. The digitized extent of tilt is inputted to a controller 113, and data of a ROM 114 is designated by the controller 113, that is, a fixed weight corresponding to the extent of tilt is selected, and the value (fixed weight) of resistances Rji and Rj (R11, R21, R31, R41, R51, R12, R22, R32, R42, R52, R13, R23, R33, R43, R53, and R1 to R3) of a waveform processor WE is set.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a waveform processing apparatus for removing signal deterioration caused during recording / reproduction / transmission, and particularly, a waveform processing apparatus having a simple structure and high performance by using a neural network. An apparatus and a design method thereof are provided.

[0002]

2. Description of the Related Art Conventionally, signal processing in a transmission system (recording / reproducing) which is easily affected by non-linear distortion such as recording / reproducing by a high density optical recording medium / magnetic recording medium or transmission / reception in a communication path of a high information amount In that case, select the most suitable modulation method,
Efforts are being made to suppress signal deterioration by using various waveform equalizers (equalization / filtering), noise cancellers, emphasis / de-emphasis, and the like. For example, a transversal filter has been used as a waveform equalizer. The transversal type filter delays an input signal with a delay circuit, amplifies it with amplifiers having different amplification coefficients, adds and outputs it, and equalizes the waveform.

However, there are many causes of signal deterioration and it is not possible to quantitatively grasp which factor and how much the signal deterioration has occurred. Therefore, it is often the case that the conventional waveform equalizer cannot deal with the problem. That is, the signal (code)
Since different coefficient values (for example, the amplification coefficient of the amplifier in the above-mentioned transversal filter) are required depending on the degree of interference (degree of signal deterioration) and the cause of deterioration,
It is difficult to uniquely determine this, and the conventional linear waveform equalizer cannot sufficiently cope with this.

Under such a circumstance, a neural network having a non-linear conversion system is used to learn the characteristics of the recording / reproducing system using the above-mentioned optical / magnetic recording medium and the high-information-volume communication path system. It is possible to output the deteriorated digital signal from the transmission system after waveform equalization in real time.
The applicant has already applied for a patent (Japanese Patent Application No. 3-
No. 31952 The title of the invention: "A waveform equalizer by a neural network and its design method").

A waveform equalizer using this neural network delays an input signal (for example, a signal whose waveform has been deteriorated from a reproducing device) by a delay unit connected in series for a predetermined time, and the time is changed in time. A plurality of input signals (values) are input to the neural network and waveform equalization is performed. The neural network is, for example, a layered neural network (for example, three layers including an input layer, an intermediate layer [hidden layer], and an output layer) using a back propagation as a learning algorithm, and is connected with a large number of variable weights. It is a combination of units (body).
The unit is a conversion system with non-linear input / output characteristics,
The unit U receives the sum from the front layer obtained by multiplying the output value from the front layer by an independent weight, and the sum is nonlinearly converted and output to the rear layer. Using this neural network, the characteristics of the recording / reproducing system / communication path system described above are learned in advance, and the deteriorated digital signal from the transmission system is waveform-equalized and output.

[0006]

The waveform equalizer using the neural network has a certain effect on the waveform equalization of digital signals. However, in general, there are various unstable factors in the transmission / reproduction system. For example, in the case of optical disk reproduction, the tilt of the optical axis of the pickup and the disk, the aberration of the pickup itself, the aberration caused by the disk, the error at the time of adjustment, the deterioration of the parts of the reproducing device and the medium, etc. may cause the reproduced waveform to be distorted. Become. Further, the reproduced waveform also varies depending on the medium to be reproduced (reproduction-only disc having prepits, magneto-optical disc, etc.). Therefore, it is almost impossible to uniquely determine by learning the optimum weights (coefficients) considering all these factors, and it is not always easy to make a practical waveform equalization device using a neural network. There wasn't.

Therefore, the present invention provides a plurality of types of weights (or delay amounts) when an unstable element of an expected transmission system is changed discretely or continuously from an ideal state in a direction of degrading a signal. The optimum value group of) is obtained by learning on the workstation. That is, the auxiliary detecting means divides a part of the unstable element of the transmission system into a narrow range and makes it learn. At the time of reproduction, the connection weight (or delay amount) of the neural network is selected from a plurality of fixed values (or continuous values) according to the signal from the auxiliary detection means for detecting the unstable element, and set to the optimum value. It is something that is done.

[0008]

According to the present invention, in order to solve the above problems, a plurality of conversion systems having nonlinear input / output characteristics are connected by independent fixed weights, and the fixed weights are neural networks. In the waveform processing device by the neural network, which comprises a conversion means which is a weight previously learned with respect to the characteristics of the transmission / reproduction system by means of the neural network for waveform-processing the signal deteriorated by the transmission / reproduction system, Of auxiliary detection means (for example, a tilt sensor LC) for detecting the change of the auxiliary detection means, and a plurality of fixed weight groups (for example, ROM 114) according to the detection value of the auxiliary detection means. The present invention provides a waveform processing device by a neural network characterized in that a fixed weight is switched according to a value.

Further, the present invention is a method of designing a waveform processing device, which comprises a plurality of conversion systems having nonlinear input / output characteristics and which are connected by independent fixed weights, and which processes a signal deteriorated by a transmission reproduction system. Then, a neural network in which a plurality of conversion systems equivalent to the waveform processing device and having nonlinear input / output characteristics are connected by independent variable weights is used to variably learn the weights corresponding to the characteristics of the transmission reproduction system. In a method of designing a waveform processing device using a neural network for designing and determining a fixed weight of the conversion means based on the variable-learned weight, an assist for detecting a partial change in characteristics of the transmission / reproduction system. A detection means is provided, and a plurality of types of weights corresponding to the characteristics of the transmission / reproduction system are variably learned in accordance with the detection value of the auxiliary detection means, and supplemented based on the variably learned weights. There is provided a method of designing a waveform processing apparatus according to a neural network, wherein the weight of the fixed plurality of types of conversion means corresponding to the detected value of the detection means that it has to be determined design.

[0010]

The weight obtained by learning of the neural network learns the characteristics of the transmission / reproduction system in a narrow range divided by the auxiliary detecting means, so that practical learning is possible.
Further, the auxiliary detection means selects the optimum weight according to the characteristics of the transmission / reproduction system and performs the waveform processing.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of a waveform processing apparatus using a neural network and a design method thereof according to the present invention will be described in detail below with reference to the drawings. 1 to 4 are explanatory views showing respective embodiments of a waveform processing device by a neural network and a design method thereof. (Basic concept of waveform processing by the neural network N)
First, the waveform processing process by the neural network N, which is the basic concept of the waveform processing device by the neural network and the design method thereof, will be described. FIG. 5 is a conceptual diagram of waveform processing (signal processing) by the neural network. As shown in the figure, in the waveform processing process by the neural network, the input signal is given a predetermined delay time by the delay means DL connected in series, and a plurality of input signals (values) that are temporally before and after are provided. Are input to the neural network N.

The neural network N is a layered neural network (for example, three layers of an input layer, an intermediate layer [hidden layer], and an output layer) using a back propagation as a learning algorithm, and has a variable weight W. It is a connection (body) of a number of units (neurons, also called synapses) connected by. The unit U is a conversion system having a non-linear input / output characteristic, and the total sum from the front layer obtained by multiplying the output value from the front layer by an independent weight W is input to the unit U, and the sum is nonlinear. It is converted and output to the subsequent layer. In the neural network N, the output signal output with respect to the (known) input signal is compared with the teacher signal, and the variable weight W representing the strength of the coupling between the units is changed / converged, that is, learned and desired. This neural network is constructed to have a waveform processing function.

(Structure of Neural Net N and Unit U) FIG. 6 is a diagram showing a unit connection of the neural net shown in FIG. The unit U is a conversion system configured by a nonlinear function F (X) that outputs an output value Y = F (X) for an input value X, for example, a nonlinear input / output characteristic. As the input value, the sum obtained by multiplying the output value of the previous layer (or the input value to the neural network) by the independent variable weight W is input, and this input value is nonlinearly converted and It is output to the layer.

More specifically, FIG. 6 is a diagram showing the i-th unit combination of the n-th layer constituting the layered neural network, and the input value X (n , I) is a weight W that represents the strength of the coupling between the j-th unit of the (n−1) -th layer and the i-th unit of the n-th layer in the output value of each unit of the (n−1) -th layer that is the previous layer. (N, i, j)
It is configured to be the sum of those multiplied by. The input value X (n, i) is input to the non-linear function F (X), non-linearly converted to Y (n, i) = F {X (n, i)}, and output to the unit in the next layer. Is configured to.

The variable weight W (n, i, j) represents the coupling strength of the unit (U).
A desired neural network is constructed by changing / converging (n, i, j), that is, learning. As a learning algorithm for constructing a neural network, there is known back propagation. That is,
In the neural network N shown in FIG. 5, when an input value (input pattern) is input to the input layer, this input value is transmitted / processed to the input layer → the intermediate layer → the output layer, and the output value (output pattern) is output from the output layer. ) Is output (feedforward processing). This output value is based on the weight obtained by the learning so far. On the other hand, when a desired output value (teacher signal) is given, it is transmitted and processed in the order of output layer → intermediate layer → input layer, and weights are learned (feedback processing). The weight learning is to change and converge the coupling strength between units, that is, the weight W (n, i, j) so as to reduce the difference between the actual output value and the desired output value.

Next, based on the basic concept of the waveform processing process by the neural network N described above, a concrete configuration of the waveform processing device WE by the neural network (a hardware execution part by the actual signal processing device, that is, a fixed part). The desired signal processing system based on the weight of 1) and the specific design method of the waveform processing device based on the neural network (software-executed portion by computer, that is, a pre-learning processing method based on variable weight) will be described in detail.

(Structure of Waveform Processing Device WE Based on Neural Network N) Next, the specific structure (hardware execution part) of the waveform processing device WE based on a neural network will be described in detail. FIG. 7 is a circuit diagram of a waveform processing device equivalent to the neural network N shown in FIGS. 5 and 6, specifically configured by an operational amplifier, a pair of diodes, a resistor, and the like. This waveform processing device is a substantial signal processing device having a fixed weight, which executes only desired signal processing (feedforward processing) at high speed, and is composed of a simple circuit having no learning function. It is a thing.

As shown in FIG. 7, the unit (U) is a general-purpose inverting operational amplifier 1a having a predetermined inverting amplification characteristic.
1b, 2a · 2b, 3a · 3b, 4a · 4b and buffer amplifiers 5, 6, 7, 8, 9 having predetermined amplification characteristics
And a diode pair (10, 12, 13) having non-linear input / output characteristics (junction characteristics). The diode pair is
The diode clip circuit is connected in parallel in the opposite direction between the signal line and the ground line and has a nonlinear input / output characteristic (junction characteristic).

Resistances between units (U) Rji, Rj (R1
1, R21, R31, R41, R51, R12, R22, R32, R42, R52, R13, R23, R33,
R43, R53, R1, R2, R3) function as a fixed weight that represents the strength of the bond between the units. That is, the weight W (2, i, j) corresponds to the resistance Rji, and the weight W (3,1, j) corresponds to the resistance Rji.
Corresponds to j. The resistance values of the fixed resistors Rji and Rj are values determined by the design method described later. The weight (coefficient) representing the strength of the coupling between the units has positive and negative values as described later, but for the negative weight, the inverting operational amplifier is used alone (for example, resistor R21 and The configuration of the inverting operational amplifier 1a) is inverted, and for positive weighting, two inverting operational amplifiers are used in series (for example, the configuration of the resistor R31, the inverting operational amplifier 1b, and the inverting operational amplifier 1a) to regenerate the signal. It corresponds by reversing. Reference numerals 14, 15, 16 and 17 are delay lines having a predetermined delay time.

The input signal (original signal) is given the predetermined delay time by delay means (delay lines) 14, 15, 16 and 17 and input to the input layer (of the neural network). A plurality of input waveforms (values) temporally before and after are amplified by the buffer amplifiers 5, 6, 7, 8, 9 which are units U (1,1) to U (1,5) of the input layer and then divided. Then, it is input to the unit of the intermediate layer through the resistor.

As the input value to the unit, the output value of the input layer which is the previous layer is the resistance Rji (R11, R21, R31, R41, R51, R12, R22, R32, R42 which functions as an independent fixed weight. , R5
2, R13, R23, R33, R43, R53), and is then weighted via the operational amplifier 1a, which constitutes the units U (2,1), U (2,2), U (2,3) of the intermediate layer. The signals are input to the inverting input terminals of 1b, 2a, 2b, 3a and 3b and added and combined. The non-inverting input terminal of the operational amplifier is grounded. The input value amplified (inverted or re-inverted) by the operational amplifier is non-linearly converted by the diode pairs 10, 11 and 12 and output to the output layer which is the rear layer. Since the diode pair has a non-linear input / output characteristic, the non-linear conversion is performed and the output is made. The output from the intermediate layer is input to the operational amplifier 4a (and 4b), which is the unit U (3,1) of the output layer, and the diode pair 13 through the resistor Rj (R1, R2, R3), and the above-mentioned intermediate Similar to the layer unit, it is non-linearly converted and output.

(Method for Designing Waveform Processing Device by Neural Network) Next, a method for designing a waveform processing device by a neural network (software-executed portion by a workstation) will be described. The waveform processing device (WE) targeted by this design method is the circuit described in FIG. The method of designing a waveform processing device using a neural network is the technique shown in FIGS. 8A to 8C, and the resistances Rji, Rj (R11, R21, R31, R41, R51, R of the waveform processing device WE are used.
12, R22, R32, R42, R52, R13, R23, R33, R43, R53, R1, R2, R3
) Values (fixed weights) obtained by learning a neural network (neural network N shown in FIG. 5) equivalent to this waveform processing device WE, W (2, i, j), W (3, 1, j), that is, concretely determined by the variable weights that have changed / converged by learning. The weight representing the strength of the coupling between the units has positive and negative values as described above, but for the negative weight, the inverting operational amplifier is independently used to invert the signal and the positive weight is used. In this case, two inverting operational amplifiers are used in series to re-invert the signal to correspond.

As shown in FIG. 8A, the reproduction signal from the recording medium Mo reproduced by the reproducing device D is A / D converted and taken into the digital memory 20 at a predetermined sampling period. Known digital data used at the time of learning are recorded in advance on the recording medium Mo. The data taken in the digital memory 20 is transferred to the learning processing workstation 21 and subjected to the learning calculation processing described later. The sampling cycle is an interval of learning calculation processing (for example, a half time interval T / 2 of one cycle T of data).
Or shorter than the learning calculation processing interval. At this time, the reproduction signal is also input to the bit PLL (phase lock loop) circuit 22 and the clock of the reproduction signal (digital data) is extracted. The extracted clock is multiplied by the multiplying circuit 23 so as to correspond to the sampling period, and after the phase is adjusted by the phase circuit 24,
It is inputted as a sampling clock of the digital memory 20.

Next, as shown in FIG. 8B, the workstation 21 performs learning calculation processing. In the workstation 21, a neural network simulation program having a known back-propagation learning algorithm is prepared, and is equivalent to the specific (neural net) waveform processing device WE shown in FIG. Learning calculation processing is executed for the neural network N shown in FIG.

That is, the captured data reproduced by the reproducing device D, captured in the digital memory 20 and transferred to the workstation 21 is the waveform processing device W.
It is input to the neural network which is equivalent to E, and this input signal is transmitted / arithmetically processed in the input layer → intermediate layer → output layer,
An output signal is output from the output layer. This output signal is based on the weight obtained by the learning so far. The calculated output signal is a teacher signal (recording medium M
It is compared with the teacher data based on the known digital data pre-recorded in o), and is transmitted / arithmetically processed in the order of output layer → intermediate layer → input layer to learn.

The characteristics of the transmission / reproduction system to be learned (learned) include the characteristics of the data itself, the characteristics of the recording device, the characteristics of the reproducing device, the characteristics of the transmission line, etc., and depending on the size of the neural network, Select appropriately. In this way
The above-described learning is repeated based on various captured data until the desired correct answer rate (performance) is obtained, and the weight representing the strength of the connection between the units of the neural network N is changed. Then, the weights of the neural network N that have converged after learning the characteristics of the transmission / reproduction system are the resistances Rji, Rj (R11, R21, R31, R41, R51, R12, R22, R32, R4) of the waveform processing device WE.
2, R52, R13, R23, R33, R43, R53, R1, R2, R3).

Finally, as shown in FIG. 8C, a waveform processing device WE is manufactured based on the specifically determined resistance value and the presence / absence of the inverting operational amplifier, and reproduced by the reproducing device D. A reproduction signal from a general recording medium M is waveform-processed and output.

Based on the above description, the features of the present invention will be described in detail based on the embodiments shown in FIGS. Figure 1
FIG. 1 is a diagram showing a first embodiment of a waveform processing device and a designing method thereof according to the present invention, which is applied to a read-only optical disc device. This is because the learning region is divided and learned according to the tangential tilt amount (tilt amount) detected by the tilt sensor, that is, a plurality of weights corresponding to the tilt amount are obtained, and the weight is calculated according to the tilt amount. The weight is switched.

(First Waveform Processing Device) As shown in FIG. 1, in the first embodiment of the waveform processing device of the present invention, the reflected light from the optical disk 103 is condensed by the condensing lens 109 and the photodetector is used. It is incident on 110. The reproduction signal obtained from the output of this photodetector is input to the signal delay circuit DL that constitutes the waveform processing device WE, and the optical pickup 101
And a tilt sensor LC for detecting the tilt of the optical disc 103
The tilt signal detected by is generally configured to be input to the buffer 111.

In FIG. 1, the optical pickup 101 is
Information is recorded on the optical disc 103, and the optical disc 10
3 is used to read information. The laser light generated from the light source 104 in the optical pickup 101 is converted into a parallel light flux through the collimator lens 105, the beam splitter 106, and the quarter-wave plate 10 that changes the polarization state.
7, converged through the objective lens 108, and the disc 10
A beam spot P is formed on the reflecting surface of No. 3. The reflected light from the optical disc 103 is reflected by the objective lens 108, ¼
The light is incident on the beam splitter 106 through the wave plate 107, is changed in direction, and is separated from the incident light.
It is incident on 09. The light beam from the condenser lens 109 is incident on the photodetector 110, the output signal of the photodetector 110 is applied to the signal detection circuit 111A, and the reproduction signal is output from the signal detection circuit 111A.
It is configured to be input to E.

In the vicinity of the objective lens 108, a tracking actuator TA for causing the beam spot P to follow a track on the disk 103, and the objective lens is controlled so that the beam spot P forms an image on the disk surface. A focusing actuator FA, a tilt sensor LC for detecting the angle formed by the optical axis of the laser beam applied to the optical disc 3 and the disc surface, and the like are provided. The output of the tilt sensor LC is the buffer 11
1, and the output of the buffer 111 is digitized by the A / D converter 112. The digitized tilt amount is input to the controller 113, and the data in the ROM 114 is designated by the controller 113,
That is, a fixed weight corresponding to the tilt amount is selected, and the resistances Rji, Rj (R11, R21, R31, R41, R of the waveform processing device WE are selected.
51, R12, R22, R32, R42, R52, R13, R23, R33, R43, R53, R1, R2
, R3) (fixed weight) is set. The controller 113 divides the tilt amount into a plurality of areas, for example, four areas, and specifies data corresponding to the areas. As is clear from the above points, for example, 4
Four fixed weights Rji and Rj corresponding to one area are learned and determined in advance.

(First Waveform Processing Device Design Method) In order to determine the fixed weights Rji and Rj, learning is performed by simulation on a computer as described above, and variable weights W (2, i, j) are used. , W (3,1, j) makes a design decision. First, a disc on which a certain pattern is recorded is reproduced, and the reproduced waveform is digitized to be one of student signals. At the same time, the tilt amount (DC value) of the disc at that time is also captured in parallel. At this time, a disc having a minimum warp or surface wobbling (a glass substrate disc is preferable) is used for reproduction by a player capable of adjusting the tilt amount of the pickup, and the tilt amount is determined by a tilt sensor LC or an autocollimator. Use to measure. By repeating this, the tilt amount is changed within a range that can actually occur, and several sets (for example, four sets) of the student signal are prepared. A student signal and a teacher signal are necessary for designing a neural network by the back propagation method, and the student signal is prepared as described above. As the teacher signal, a waveform which is the original signal of the signal reproduced from the disc and which is blunted through a low-pass filter is used. In learning, a specific pattern for phase matching is prepared on the disc in order to correctly match the phases of the teacher signal and the student signal, and the same pattern is searched for in the reproduced signal to perform the phase matching. Learning is performed in this manner, and the optimum combination of variable weights W (2, i, j) and W (3,1, j) for each of the prepared tilt values (DC values) is obtained. When this is hardened, the variable weights W (2, i, j), W
To a fixed weight (resistors Rji, Rj) equivalent to (3,1, j),
An externally controllable variable resistor (for example, the one shown in FIG. 1B) is used.

According to the above configuration, the tilt value of the waveform processing device is A / D converted and enters the controller 113, and the controller 113 applies an appropriate fixed weight (resistance Rji, Rj) according to the value. It will be selected from the combinations stored in the ROM 114. The data in the ROM 114 is an optimum fixed weight (resistors Rji, Rj) for various tangential tilt values obtained as a result of learning.
Is a combination of. The speed required for switching the resistance value is about 1 μs, which is sufficient, and can be sufficiently realized by an analog switch.

In this neural network waveform processing circuit, it is possible to always perform optimal waveform processing for predictable tangential tilt, and the margin of the system for tangential tilt is greatly improved. Further, since the learning is divided according to the tilt value, the learning does not become difficult unlike the conventional method.
You can definitely make design decisions. In this embodiment, the tangential tilt is treated as an element that affects the transmission / reproduction system, but it affects the disk reproduction such as radial tilt, defocus, and off-track amount, and its size can be grasped quantitatively. Anything is fine. Further, these elements are not limited to the signals detected by the dedicated sensor as in the above embodiment, but may be signals extracted from the reproduction signal, the servo signal, or the like.

(Embodiment 2) Second Embodiment of Waveform Equalizer of the Present Invention
In the embodiment, the waveform equalizer of the present invention is applied to a magneto-optical device. In FIG. 2, the magneto-optical disk 103A
Is provided with a large number of circular information tracks. In this track, a read-only ROM section and a recording / playback RAM section are mixed. ROM section and RAM
The recording method and playback method of the part are different, therefore
Although the magneto-optical disc 103A and the disc device shown in FIG. 2 are different from the read-only optical disc 3 and the optical disc device shown in FIG. 1, the magneto-optical device shown in FIG. 2 has a general configuration. Detailed description is omitted.

In the figure, recording of information on the magneto-optical disk medium 103A, which is provided with a number of information tracks in a circular pattern, is performed by the magneto-optical recording method using the optical pickup 101A. Further, the reading of the information signal from the magneto-optical disk medium is also performed by using the optical pickup 101A. The optical pickup 101A includes a light source 104A such as a semiconductor laser, a collimator lens 105 for collimating the light flux from the light source 104A, an irradiation light flux from the light source 104A such as the semiconductor laser, and a reflected light flux from the optical disc medium 103A. The beam splitter 106 for separating is provided.

In FIG. 2, P is the objective lens 108.
With the light spot condensed by the optical pickup 1
01A also includes a tracking actuator TA and a focusing actuator FA. The reflected light from the magneto-optical disk medium 103A is reflected by the beam splitter 106 and enters the beam splitter 115. The light flux reflected by the beam splitter 115 is
The light enters the condenser lens 116, is condensed, and is detected by the photodetector 11.
It is incident on 7. A tracking error signal and a focusing error signal are detected from the output S3 of the photodetector.

The light flux that has passed through the beam splitter 115 is incident on the half-wave plate 119, whose polarization plane is changed, and then incident on an analyzer 123 such as a polarizing prism. The polarization prism 123 separates the light flux into two, and approximately the same amount of light flux enters the condenser lens 126 and the condenser lens 124. The light flux passing through the condenser lens 126 is incident on the photodetector 127, and the information signal S1 is output to the output terminal of the photodetector 127. The light flux passing through the condenser lens 124 enters a photodetector 125, and the information signal S2 is output to the output terminal of the photodetector 125.

The two outputs of the optical pickup 101A, that is, the information signal S1 and the information signal S2 are both outputs according to the information recorded on the magneto-optical disk medium 103A. These two signals are input to the differential amplifier 128, and the differential output thereof is taken out as the magneto-optical signal S4. This differential output S4 is the R of the optical disc 3A.
It is used as a reproduction information signal from the AM section. The two signals S1 and S2 are input to the summing amplifier 29,
The output of the summing amplifier 129 is taken out as the pre-pit signal S5. The output S5 of the summing amplifier is used as a reproduction information signal from the ROM section of the magneto-optical disk 3A.

The signals S4 and S5 are input to the switch 132. The magneto-optical / pre-pit detection circuit (signal discrimination circuit) 131 discriminates whether the signal currently read by the optical pickup 101A is the signal of the ROM section or the signal of the RAM section, and its output S8 is applied to the switch 132. Has been done. The output signal sia of the switch 132 is
Depending on the signal S8, it becomes the signal S4 or S5,
It becomes an input signal of the waveform equalizer WE. The waveform equalizer WE shown in FIG. 2 is similar to the waveform equalizer shown in FIG. 1, and is composed of a signal delay circuit and hardware. The signal sia corresponds to the signal si shown in FIG. Further, by the output S8 from the magneto-optical / pre-pit detection circuit (signal identification circuit) 131, fixed weights for magneto-optical and fixed weights Rji, Rj for pre-pit are provided.
Are configured to be switched. For example,
By a switching circuit as shown in FIG.

As described above, the second embodiment of the waveform equalizer of the present invention discriminates the type of signal to be reproduced when reproducing a disc having a different reproduction principle, and detects one of the input layers of the neural network. The operation of the neural network is switched by supplying a switching signal according to the type of disk to the section. When the reproduction principle is different as in the ROM section and the RAM section of this example, the state of intersymbol interference is also different, and the optimum weight between the neurons of the neural network is also different. In the present invention, the optimum weights are obtained by learning in advance, and suitable weights are switched according to the type of the disc during reproduction, and one neural network processing system is made compatible with a plurality of disc reproduction systems. Note that here, the magneto-optical signal /
Although the combination of pre-pit signals is taken as an example, the combination of input signals is not limited to the combination of magneto-optical signal / pre-pit signal.

(Third Embodiment) As shown in FIG. 3, this third embodiment is an example in which the delay amount of the delay element is variable instead of the weight according to the kind of the reproduction signal at the time of reproducing the optical disk. . In this embodiment, the player is compatible with two types of formats having different unit clocks. In the present invention, the delay amount is determined on the basis of the unit clock of the signal to be handled, so that for signals having different unit clocks, the delay amount must be changed accordingly. Therefore, the delay amount is switched according to the format based on the signal from the disk format detection circuit 200, for example, by a switching delay circuit DL 'shown in FIG. As a result, it is possible to provide a neural net waveform equalization device compatible with discs of a plurality of formats.

(Embodiment 4) The fourth embodiment is an example in which the individual difference after the optical disk pickup is assembled can be adjusted within a certain range by configuring a part of the fixed weight with a semi-fixed resistor. . At the time of assembling the pickup, a deviation of the optical axis and an aberration occur due to the accuracy of the optical parts themselves, the assembling and adjusting accuracy of the parts. Normally, the design is performed in consideration of these expected accuracies to allow a margin for the entire pickup. However, when demands for pickup accuracy become severe due to downsizing of pickups, high density of discs, etc., installation of a mechanism for adjustment and demands for parts and assembling precision become strict, and a margin is lost.

In the present embodiment, a reproduction signal obtained by shifting a portion of the student signal at the time of learning, which influences the individual difference at the time of assembling the pickup, from the optimum position within the expected range,
Prepare several types. Next, learning is performed for each student signal, and an optimal combination of weights (coupling coefficients) is obtained. At this time, the coupling coefficient between the neurons varies depending on how the parts are displaced. Looking at this result, we selected several coupling parts where the coupling coefficient greatly changes due to differences in student signals from the ones with large changes, and when making them hard, it is possible to cover the range of changes obtained by simulation. Use a semi-fixed resistor.

An example of such a configuration is shown in FIG. In this example, from the result of the simulation, adjusting the coupling coefficients Rj of the intermediate layer and the output layer is most effective in correcting the pickup assembly accuracy error, and the semi-fixed resistor Rj is used in that portion. The assembled and adjusted pickup is placed on the player to reproduce a disc, and the semi-fixed resistor Rj is adjusted while observing the reproduced waveform to reproduce the reproduced waveform due to the deviation from the ideal state of the optical system that could not be adjusted during the pickup assembly. Disturbance can be corrected. As a result, it is possible to further suppress the individual difference that occurs when the pickup is mass-produced.

Although the pickup has been described in the present embodiment, the present embodiment can be applied to a portion other than the pickup where individual differences occurring during assembly and adjustment pose a problem for the reproduced waveform. Also, using the waveform that has deteriorated due to aging of the device as one of the student signals, determine the coupling part to which the semi-fixed resistor is applied in the neural network as described above, and determine the deterioration of the reproduced waveform due to aging with this waveform etc. It is also possible to make corrections with a digitalization device.

Although the optical disk reproducing system has been described in the above-mentioned embodiments, it can be applied to a transmission system other than the optical disk. Moreover, it is also possible to use the first to fourth embodiments in combination.

As described in detail above, according to the present invention, a plurality of types of expected unstable elements of the transmission system are obtained when the elements are changed discretely or continuously from the ideal state in the direction of degrading the signal. The optimum value group of the weight (or the delay amount) is obtained by learning on the workstation,
That is, since the auxiliary detecting means divides a part of the unstable element of the transmission system into a narrow range for learning, the learning decision of the neural network becomes easy. Furthermore, since the optimum weight is selected, the performance as a waveform equalizer using a neural network is improved.

[0049]

According to the present invention, it is possible to eliminate a plurality of signal deterioration factors in a recording / reproducing system or a transmission system, which could not be realized by a conventional waveform equalizer using a neural network, and to reduce the deterioration of signal quality at low cost. It can be improved very well.

[Brief description of drawings]

FIG. 1 is a diagram showing a first embodiment of a waveform equalizer of the present invention.

FIG. 2 is a diagram showing a second embodiment of the waveform equalizer of the present invention.

FIG. 3 is a diagram showing a third embodiment of the waveform equalizer of the present invention.

FIG. 4 is a diagram showing a fourth embodiment of the waveform equalizer of the present invention.

FIG. 5 is a basic conceptual diagram showing an example of waveform equalization by a neural network.

6 is a diagram showing a unit connection of the neural network shown in FIG.

FIG. 7 is a specific circuit diagram of a waveform equalization device using a neural network, showing an example of a physical signal device that is executed by hardware.

FIG. 8 is a diagram showing a method of designing a waveform equalizer using a neural network, and is for explaining software execution.

[Explanation of symbols]

WE Waveform equalizer Mo Known recording medium (transmission system) M Recording medium (transmission system) D Playback (reception) device DL Delay means N Neural network U unit W Variable weight WE Waveform equalizer LC ... Tilt sensor (auxiliary) Detecting means) 1a • 1b, 2a • 2b, 3a • 3b, 4a • 4b Inverting operational amplifier 10, 11, 12, 13 Diode pair 14, 15, 16, 17 Delay line 20 Digital memory 21 Workstation 22 bit PLL circuit 113 Controller 114 ROM 131 Magneto-optical / pre-pit detection circuit (auxiliary detection means) 200 Format detection circuit (auxiliary detection means) r1, r2, r3 Resistance to (the unit of) the final layer (semi-fixed weight) R11, R21, R31, R41, R51 Resistance of the middle layer to the first unit (fixed weight) R12, R22, R32, R42, R52 Resistance of the middle layer to the second unit Resistance (fixed weight) R13, R23, R33, R43, R53 Resistance to the third unit of the intermediate layer (fixed weight) R1, R2, R3 Resistance to (the unit of) the final layer (fixed weight)

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁵ Identification number Internal reference number FI Technical indication H03H 17/00 B 7037-5J (72) Inventor Hirokazu Shimada 3-chome Moriya-cho, Kanagawa-ku, Yokohama-shi, Kanagawa Address 12 Victor Company of Japan, Ltd.

Claims (6)

[Claims] 1. A plurality of conversion systems having non-linear input / output characteristics are connected by independent fixed weights, and the fixed weights are weights previously learned by a neural network with respect to characteristics of a transmission / reproduction system. In a waveform processing device using a neural network, which comprises a certain conversion means and performs waveform processing of a signal deteriorated by the transmission / reproduction system, an auxiliary detection means for detecting a partial change in characteristics of the transmission / reproduction system, and an auxiliary detection means of the auxiliary detection means. A waveform processing device using a neural network, comprising: a plurality of types of fixed weight groups corresponding to detected values, wherein the fixed weights are switched according to the detected values of the auxiliary detecting means. 2. A method of designing a waveform processing apparatus, which comprises a plurality of conversion systems each having a non-linear input / output characteristic, which are connected by independent fixed weights, and which processes a signal deteriorated by a transmission / reproduction system. Then, a neural network in which a plurality of conversion systems equivalent to the waveform processing device and having nonlinear input / output characteristics are connected by independent variable weights is used to variably learn the weights corresponding to the characteristics of the transmission reproduction system. In a method of designing a waveform processing device by a neural network for designing and determining a fixed weight of the conversion means based on the variable-learned weight, an assist for detecting a partial change in characteristics of the transmission / reproduction system. The auxiliary detection means has a detection means, and variably learns a plurality of types of weights corresponding to the characteristics of the transmission / reproduction system according to the detection value of the auxiliary detection means, and based on the variably learned weights. A method for designing a waveform processing device using a neural network, wherein fixed weights of a plurality of types of conversion means are designed and determined according to detected values of stages. 3. A delay means for delaying a deteriorated signal from a transmission reproduction system and a plurality of conversion systems having nonlinear input / output characteristics are connected by independent fixed weights, and the fixed weights are neural networks. In the waveform processing device by a neural network, which comprises a conversion means which is a weight learned in advance with respect to the characteristics of the transmission / reproduction system by the above, and inputs the deteriorated signal temporally moved forward and backward into the conversion means to perform waveform processing, Auxiliary detection means for detecting a partial change in characteristics of the transmission / reproduction system, and the delay means have a plurality of types of delay amounts according to the detection value of the auxiliary detection means. A waveform processing device using a neural network, wherein the delay amount of the delay means is switched according to the delay amount. 4. A delay means for delaying a deteriorated signal from a transmission / reproduction system, and a conversion means in which a plurality of conversion systems having nonlinear input / output characteristics are connected by independent fixed weights. A method of designing a waveform processing device for performing waveform processing on a deteriorated signal that has been moved forward and backward, wherein a plurality of conversion systems equivalent to the waveform processing device and having nonlinear input / output characteristics are connected by independent variable weights. Of a waveform processing device by a neural network in which the weight corresponding to the characteristics of the transmission and reproduction system is variably learned by the generated neural network, and the fixed weight of the conversion means is designed and determined based on the variably learned weight. In the design method, an auxiliary detection unit for detecting a partial change in the characteristics of the transmission / reproduction system is provided, and the transmission re-transmission is performed while switching the delay amount of the delay unit according to the detection value of the auxiliary detection unit. A weight processing corresponding to a characteristic of a living system is variably learned, and a fixed weight of the converting means is designed and determined based on the variably learned weight. Design method. 5. A plurality of conversion systems having non-linear input / output characteristics are connected by independent fixed weights, and the fixed weights are weights learned in advance for the characteristics of the transmission / reproduction system by a neural network. In a waveform processing device based on a neural network, which comprises a certain conversion means and performs waveform processing of a signal deteriorated by the transmission reproduction system, a part of independent weights of the plurality of conversion systems is set as a weight adjustable in a constant width, and after learning, A waveform processing device using a neural network, which is capable of fine adjustment. 6. A waveform for waveform-processing a signal deteriorated by a transmission / reproduction system, comprising a conversion means in which a plurality of conversion systems having nonlinear input / output characteristics are connected independently with fixed and fixed widths and adjustable weights. A method of designing a processing device, wherein the characteristics of the transmission reproduction system are equivalent to the waveform processing device, and a neural network in which a plurality of conversion systems having nonlinear input / output characteristics are connected by independent variable weights Waveforms by a neural network, wherein the weight corresponding to the variable weight is variably learned, and the widths of the fixed weight and the adjustable weight of the converting means are designed and determined based on the variably learned weight. Processing equipment.