JP2022061900A - Hologram recording/reproducing method and hologram recording/reproducing device - Google Patents

Abstract

To provide a hologram recording/reproducing method and a device capable of reducing network size of machine learning and significantly reducing recording capacity required for demodulation calculation processing, when demodulating page data reproduced from a hologram recording medium.SOLUTION: A hologram recording/reproducing method includes the steps of: reading page data recorded on a hologram recording medium for information recording, and dividing into per a modulation block (S6); demodulating first bit string data for bright spots arranged in the modulation block by a first CNN, and demodulating second bit string data for an amplitude-phase signal superimposed on the bright spots by a second CNN (S7); and associating the demodulated two bit string data with each other as one bit string data (S8).SELECTED DRAWING: Figure 1

Description

The present invention relates to a method and an apparatus for hologram recording and reproduction of page data, and in particular, when reproducing bit string information modulated into a two-dimensional modulation block and holographically recorded, it is demolished by using a machine learning method such as a neural network. The present invention relates to a hologram recording / reproducing method and a hologram recording / reproducing apparatus.

In the hologram memory, the light beam from the light source is divided into two paths, one of which is modulated by an SLM (Spatial Light Modulator) that displays page data to form signal light, and the other is used as unmodulated reference light. The data is recorded by simultaneously irradiating the hologram recording medium with these signal lights and the reference light and writing the interference fringes generated at that time on the hologram recording medium.
Further, at the time of reproduction, by irradiating the interference fringes with the reference light, the data carried by the original signal light can be reproduced as diffracted light, and this is captured and acquired by the camera. Since two-dimensional information can be collectively recorded or reproduced by a single light irradiation, there is an advantage that the transfer speed is high. In addition, the hologram recording device can record holograms at the same location in multiples by using various multiplexing methods such as angle multiplexing, phase code multiplexing, spherical reference optical shift multiplexing, etc., thereby recording information at high density. be able to.

In recent years, in order to further increase the capacity, the page data has been changed from the conventional binary code to a multi-value code to increase the amount of information, and at the same time, the feature of the hologram that can store not only the amplitude information but also the phase information. A recording / reproduction method combining amplitude modulation and phase modulation has been proposed (see Non-Patent Document 1).

FIG. 11 shows an optical system related to hologram recording / reproduction. At the time of recording, the light emitted from the laser light source 101 is magnified by the lens 102 and converted into p-polarized light by the 1/2 wave plate 103 whose optical axis direction is adjusted, so that the light is transmitted through the PBS 104.
The p-polarized light transmitted through the PBS 104 is obliquely (45 degrees) polarized by the 1/2 wavelength plate 105 whose optical axis direction is adjusted, and further divided into p-polarized and s-polarized by the PBS 106. The s-polarized light reflected by the PBS 106 is amplitude-modulated by the amplitude-modulated SLM107 and phase-modulated by the phase-modulated SLM108 to carry the amplitude-phase-modulated page data information and is arranged in the optical path of the lens group 109. After passing through the high-frequency blocking filter 110 for narrowing the recording volume, the hologram recording medium 114 is irradiated as signal light 121 at a predetermined position.
On the other hand, the p-polarization from the PBS 106 is reflected by the mirror 111, adjusted to s-polarization by the 1/2 wave plate 112, and the incident angle is adjusted by the mirror 113, and is referred to at the time of recording at the predetermined position of the hologram recording medium 114. It is irradiated as light 122.
The two lights of the signal light 121 and the reference light 122 interfere with each other, and the interference fringes generated thereby are recorded on the hologram recording medium 114 as a hologram.

Further, at the time of reproduction, the light emitted from the laser light source 101 is magnified by the lens 102 and then polarized obliquely (45 degrees) by the 1/2 wave plate 103 whose optical axis direction is adjusted, as in the case of recording, and the PBS 104 It is divided into p-polarized light and s-polarized light. As for the p-polarization transmitted through the PBS 104, the polarization state remains P-polarized in the 1/2 wavelength plate 105 whose optical axis direction is adjusted, so that the PBS 106 is also transmitted (the reflected light from the PBS 106 is transmitted). Does not occur). After that, the p-polarized light transmitted through the PBS 106 is reflected by the mirror 111, adjusted to s-polarized light by the 1/2 wave plate 112, and the incident angle is adjusted by the mirror 113, and the hologram recording medium 114 is used as the reference light 122 during reproduction. Is irradiated to the above-mentioned predetermined position. As a result, the s-polarized reproduction light 124 is emitted from the hologram recording medium 114.

On the other hand, the s-polarized light reflected by the PBS 104 becomes the probe light 123 for reproduction. When the probe light 123 adjusted so that there is no in-plane phase unevenness by the correction phase modulation SLM116 and the reproduced light 124 are combined, the light is intensified in the in-phase region and the light is in the opposite-phase region. Is canceled. Further, the phase of the probe light 123 is changed to 0, π / 2, π, 3 / 2π by the piezo modulator 117, and each time the probe light 123 is combined with the s polarization from the PBS 104 described above. A total of four images of the obtained interference fringes are taken, and the phase information is converted into intensity information by performing an operation based on a four-step phase shift method (see, for example, Japanese Patent Application Laid-Open No. 2019-33467) with the arithmetic apparatus 120. Can be obtained.

By the way, since the page data has an area equivalent to 1 inch, low-frequency noise unevenness occurs in the plane due to the influence of expansion / contraction and positional deviation of the hologram recording medium, distortion generated by the optical system, various aberrations, and the like. For the same reason, image blurring occurs in the page data, and intersymbol interference such as light leaking from bright spots to dark spots in the page data occurs. Therefore, a technique has been proposed in which bright spots and dark spots are determined within a certain range (modulation block) in which page data is divided by a method using a difference code as a modulation code (for example, Patent No. 3209493). See issue). Even if there is deviation from various conditions during playback or uneven brightness in the in-plane direction of the laser beam, the magnitude of the brightness value can be determined relatively within the modulation block, so the effect of unevenness that occurs in the entire page data can be affected. It can be reduced to some extent, and the same effect can be obtained for both amplitude and phase signals.

On the other hand, as the number of multi-values increases, it becomes necessary to more accurately determine and demodulate the amplitude value and the phase value. Therefore, in addition to the use of the difference code, a noise compensation technique is required. For example, the reference page data in which all the pixel values are 255 (white pixels) in the case of 8-bit display is reproduced in advance in the optical system actually used, and the page data plane is stored from the image acquired by the camera. A method for compensating for amplitude noise, which specifies noise information superimposed on the image, has been proposed (see Patent Document 1). Since the luminance value decreases in the pixel on which noise is superimposed, compensation can be performed by dividing or subtracting the acquired noise information from the reproduction page data. The phase noise can be compensated in the same manner.
However, the method of Patent Document 1 cannot suppress intersymbol interference.

On the other hand, Non-Patent Document 2 states that intersymbol interference can be suppressed by performing signal processing with an equalizer based on a neural network model configured for each pixel. However, in the method of Non-Patent Document 2, since the neural network is constructed for each pixel of the page data, the amount of calculation and the time for creating the equalizer become enormous.

On the other hand, instead of the noise reduction method as described above, a technique for accurately extracting a signal from page data containing noise by machine learning has also been reported (see Patent Document 2 and Non-Patent Document 3). In this technique, the reproduced page data is demodulated as a bit string for each modulation block. That is, by repeatedly inputting page data mixed with noise into a convolutional neural network (hereinafter referred to as CNN) in advance, the characteristics of the signal buried in the noise are learned, so that the demodulation is accurate. can do.

Patent No. 4863913 Japanese Unexamined Patent Publication No. 2019-46520

Teruyoshi Nobukawa and Takanori Nomura, "Multilevel recording of complex amplitude data pages in a holographic data storage system using digital holography," Opt. Express, Vol. 24, No. 18, (2016), pp. 21001-21011 H. Osawa et al., "Neural Network Equalizer Matched to Recording Code in Holographic Data Storage," Jpn. J. Appl. Phys., 50, 09MB05, 2011. T. Shimobaba et al., "Convolutional neural network-based data page classification for holographic memory." Appl. Opt., 56 (26): 7327-7330 (2017)

When a machine learning method (neural network) is used for demodulation, there is a big problem that the network becomes enormous when the number of multiple values increases.
For example, in Patent Document 2, a modulation block is created using a 5: 9 modulation code as a modulation code for only an amplitude. This divides the bit string to be recorded into 5 bits and converts it into a 9-symbol modulation block composed of 3 × 3 symbols. On the contrary, the CNN that demodulates this inputs a modulation block of 9 symbols and outputs a bit string signal every 5 bits. If it is 5 bits, 2 to the 5th power, that is, 32 types of likelihoods are output, and the one with the highest likelihood is estimated to be a demodulation bit and demodulated. The CNN is generally configured so that after the convolution process in the convolution layer, the number of information determined by the product of the matrices can be output in the fully connected layer. For example, if 100 kinds of information (1, 100) are output in the convolution layer, the output becomes (1, 32) by multiplying the weight matrix (100, 32) in the fully connected layer, and 32 kinds of likelihoods. Is output.

That is, as the number of multi-values increases and the number of recordable data per modulation block increases, the number of CNN outputs increases and the weight matrix of the fully connected layer becomes enormous in size. In addition to slowing down the calculation speed, the weight matrix learned in advance must be saved in memory, which increases the network size of CNN and enormous recording capacity required for demodulation calculation processing. turn into.

The present invention has been made in view of the above circumstances, and when demodulating page data reproduced from a hologram recording medium, the network size of machine learning is reduced and the recording capacity required for demodulation calculation processing is significantly increased. It is an object of the present invention to provide a hologram recording / reproducing method and a hologram recording / reproducing apparatus capable of reducing the amount and speeding up the demodulation calculation process.

The present invention is a technique for reproducing the original signal from the recorded hologram, and the hologram recorded image of the subject obtained by using the interference effect of light is calculated by a troublesome calculation such as a phase step method. It relates to a signal reproduction technique capable of reproducing accurately and at high speed while significantly reducing processing. The present inventors have found a hologram recording / reproducing method and a hologram recording / reproducing apparatus having the following configurations that can achieve the above object as an extension of the hologram reproduction technique that has been studied for many years.

In the hologram recording / reproduction method of the present invention, page data recorded on a hologram recording medium for information recording is read out, divided into modulation blocks, and bright spots arranged in the modulation blocks are subjected to a first machine learning process. The first bit string data is demodulated, the second bit string data is demodulated by the second machine learning process for the amplitude phase signal superimposed on the bright spot, and the demodulated two bit string data are combined into one bit string data. It is characterized by associating as.

The first machine learning process can be a process by the first neural network, and the second machine learning process can be a second neural network process.
Further, the second bit string data and the label information for identifying the desired bright spot among the plurality of bright spots are input to the second neural network to demodulate the amplitude phase signal. Is preferable.

Further, the label information is preferably expressed in a one-hot format.
Further, by inputting a plurality of reproduced images obtained by phase-shifting the read page data into the second neural network, the amplitude phase signal can be obtained without using the phase calculation process by the phase shift method. It is preferable to demodulate.
Further, in the second neural network, a network unit for demodulating the amplitude phase signal is constructed for each of the plurality of reproduced images, and the constructed network portion is duplicated to obtain the number of the plurality of bright spots. It is preferable to construct a predetermined number of the network units corresponding to the above and perform demodulation processing of the predetermined number of the network units in parallel and at the same time.

Further, the hologram recording / reproducing device of the present invention includes a page data dividing means for dividing page data recorded in a hologram recording medium for information recording and read out into each modulation block, and a brightness arranged in the modulation block. A first machine learning means that demolishes the first bit string data based on the position of the point, and a second machine learning means that demolishes the second bit string data based on the amplitude phase signal superimposed on each of the bright spots. It is characterized by comprising a bit string data connecting means for associating the demodulated two bit string data as one bit string data.
In the description of the present application, a group of "symbols" in which a plurality of "symbols" are grouped together is referred to as a "block".

According to the hologram recording / reproduction method and apparatus of the present invention, a machine learning network is obtained by independently performing a first machine learning process for demodulating a bright spot position signal and a second machine learning process for demodulating an amplitude phase signal. The size can be reduced, the recording capacity required for demodulation calculation processing can be significantly reduced, and the high speed of demodulation processing can be maintained. In addition, the bright spot position signal demodulation process and the amplitude phase signal demodulation process can be executed in parallel, and the hyperparameters of machine learning can be set independently according to the noise characteristics of each signal. It is possible to increase the speed and improve the accuracy.

It should be noted that the signal superimposed on a plurality of bright spots is sequentially demodulated by one network unit (CNN) for demodulating the amplitude phase without preparing the network unit (CNN) for the number of bright spots for demodulating the amplitude phase signal. If it is configured to do so, the recording area for storing the weight in the network unit (CNN) adjusted at the time of pre-learning in the arithmetic processing device can be reduced.
Further, by duplicating one network unit (CNN) for amplitude phase demodulation to construct a network unit (CNN) of another bright spot, and processing these a plurality of network units (CNN) in parallel, the process is performed in parallel. The recording area can be reduced by sharing data, and the demodulation process can be speeded up.

It is a flowchart which shows the recording / reproduction procedure in the hologram recording / reproduction method which concerns on embodiment of this invention. FIG. 3 is a schematic diagram (both (a) and (b)) for explaining modulation block generation by 20: 9 modulation used in the hologram recording / reproduction method shown in FIG. 1. FIG. 1A is a diagram showing symbol numbers in a modulation block by 20: 9 modulation used in the hologram recording / reproduction method shown in FIG. 1, and FIG. 3B is a diagram showing a bright spot position conversion table. It is a figure which shows the constellation map of the amplitude phase signal in 20: 9 modulation used in the hologram recording reproduction method shown in FIG. FIG. 3 is a diagram showing demodulation processing of 20: 9 modulation code reproduction data by two CNNs in the hologram recording / reproduction method shown in FIG. 1. In the hologram recording / reproduction method shown in FIG. 1, it is a figure which shows the label information which shows the target symbol which demodulates an amplitude phase signal ((a) is a scan example of a modulation block, (b) is the label information corresponding to each bright spot. Example of embodiment). In the hologram recording / reproduction method shown in FIG. 1, it is a figure which shows the input example to the amplitude phase signal demodulation CNN (all (a), (b), (c)). It is a flowchart which shows the recording / reproduction procedure in the hologram recording / reproduction method which concerns on the prior art (comparative example) which used the hard determination for demodulation. FIG. 8 is a diagram showing a constellation map (a) and demodulation examples ((b), (c)) of a bright spot position signal and an amplitude phase signal in 20: 9 modulation according to the prior art (comparative example) shown in FIG. It is a table which shows an example of the hyperparameters of CNN set in the hologram recording / reproduction method shown in FIG. It is a schematic diagram which shows the structural example of the optical system of the general hologram recording apparatus which records and reproduces an amplitude phase signal.

Hereinafter, the hologram recording / reproducing method and the hologram recording / reproducing apparatus according to the embodiment of the present invention will be described with reference to the drawings.

(Hologram recording / reproduction method according to an embodiment)
FIG. 1 shows a schematic flow of the hologram recording / reproducing method according to the embodiment of the present invention.
First, at the time of signal recording processing on the hologram recording medium, a bit string to be recorded is prepared (S1), and 20: 9 modulation processing is performed on this bit string (S2). Here, the 20: 9 modulation process is a process of modulating 20-bit bit string data into block data composed of 3 × 3 symbols.
Next, page data is created by arranging the modulation blocks 30 obtained by the above modulation process in a two-dimensional manner (S3).

After that, for example, the page data is hologram-recorded on the hologram recording medium 114 by using the optical system 10 of the hologram recording / reproducing device as shown in FIG. 11 (S4).
Further, at the time of signal reproduction processing, page data is reproduced from the hologram recording medium 114 by using the optical system 10 of the hologram recording medium 114 (S5). By this reproduction process, page data is reproduced from the hologram recording medium 114, and the modulation block 30 is taken out (S6).
Next, the block data of the extracted modulation block 30 is subjected to CNN demodulation processing, which is the point of the present embodiment (S7). As a result, 20-bit bit string data is obtained from the block data consisting of 3 × 3 symbols (S8).

Hereinafter, detailed explanations will be supplemented for each of the above-mentioned procedures.
First, the above-mentioned 20: 9 modulation will be described.
As shown in FIG. 2, the modulation block 30 is composed of 3 × 3 symbols (in this embodiment, the reference symbol 23 is arranged in the central cell). The data string to be recorded is divided into 20 bits, and the modulation blocks 30 configured by modulation are two-dimensionally arranged to form page data. The amplitude signal of the page data is input to the amplitude modulation SLM107 and the phase signal is input to the phase modulation SLM108 to be displayed and carried on the signal light 121.
In the 20: 9 modulation of the present embodiment, the first 4 bits are superimposed on the four bright spot position signals in the modulation block 30, and the remaining 16 bits are superimposed on each bright spot (I to IV). Modulated into a signal.
FIG. 3 shows a conversion table showing the relationship between the first 4 bits and the position of the bright spot in the modulation block 30.

As shown in FIG. 3A, it is prepared so that 0 or 1 is set in each symbol number of 8 squares excluding the central square for the reference symbol 23 among the 9 squares in the modulation block 30. The symbol.
Then, according to the first 4 bit values of the input bit string, according to the table of FIG. 3 (b), the corresponding positions of each symbol number (the above eight cells) are set to 0 in the case of dark spots and bright spots (dark spots). In the case of a point of brightness other than), 1 is set. For example, when the input 4-bit value is "1101", the cells of the symbol numbers 1 to 8 are set so as to be "00011101" in order.

Next, FIG. 4 shows a constellation map representing the bit string corresponding to the amplitude phase signal. This constellation map is expressed in polar coordinates, the radius is set to the luminance value, and the declination is set to the phase value. The reference symbol 23 located in the central cell of the modulation block 30 has, for example, a luminance value fixed at 255 and a phase value fixed at 0, and serves as a reference for eight symbol values located in the remaining cells in the modulation block 30. Functions as an indicator symbol.
The brightness value and phase value of the reproduced data fluctuate due to noise during recording and reproduction. However, as a preprocessing for demodulating the above eight symbols, reference symbol 23 having a known brightness value and phase value is referred to, and each of them is used. By standardizing the brightness value and phase value of the symbol by the known brightness value and phase value of the reference symbol 23, the influence of noise generated in the entire modulation block 30 can be compensated.

Further, at the time of signal reproduction processing, the page data reproduced from the hologram recording medium 114 is acquired by the camera 119. The modulation block 30 is sequentially taken out from the acquired reproduction page data, and demodulation processing is performed using the CNN built in the arithmetic unit 120.
When the method shown in Patent Document 2 or Non-Patent Document 3 described above is applied to this demodulation process, the modulation block 30 having nine symbols is input and a 20-bit signal is output. The fully connected layer of CNN requires a huge matrix to output 2 to the 20th power, that is, 1,048,576 kinds of likelihoods, which causes a decrease in calculation speed and a memory shortage depending on a general arithmetic unit. , An error occurs in the arithmetic processing.

On the other hand, in the method of this embodiment, as shown in FIG. 5, 4 bits having the bright spot position information and 16 bits having the amplitude phase information are demodulated by two independent CNNs. The bright spot position signal demodulation CNN outputs 2 to the 4th power, that is, 16 types of likelihoods in order to demodulate the 4-bit bright spot position signal. By inputting the four modulation blocks 30 extracted from the four reproduction page data obtained by sequentially changing the phase of the probe light 123 with the piezo modulator 117 to the bright spot position signal demodulation CNN. It is a 4-channel input.

Similarly, for one amplitude phase signal demodulation CNN, four modulation blocks 30 can be input to form a network configuration for demodulating 16 bits. As a result, the network configuration for demodulation can be made smaller than that of the above-mentioned conventional technique.
However, even with such a network configuration that demodulates 16 bits, it is inevitable that a large network that outputs 2 to the 16th power, that is, 65,536 types of likelihoods will be output. Therefore, more preferably, the amplitude phase signal is divided into bright spots of 4 bits, and each is demodulated sequentially. At that time, in order to specify which bright point the amplitude phase signal superimposed is demodulated, the label information 40 of 4 channels is input to the amplitude phase signal demodulation CNN in addition to the 4 modulation blocks 30. A total of 8 channels are input.

FIG. 6 shows a method for generating the label information 40. Of the eight symbols excluding the reference symbol 23, the symbol positions of the four bright spots (I to IV) are numbered in raster scan order (see (a)), and the one-hot format (only the target is "1", that is. Everything except is expressed by "0") (see (b)).
As illustrated in FIG. 7, the lower left symbol in the modulation block 30 is the third bright spot III in the scan order except for the central reference symbol 23, so the label information 40 is shown in FIG. 6 (b). It can be expressed as (black, black, white, black). Therefore, the 5 to 8 channels corresponding to the label information 40 are all black or all white images composed of 3 × 3 symbols having the same size as the modulation block 30, respectively.
Therefore, in order to demodulate a total of 16 bits of data from the amplitude phase signal superimposed on each of the four bright spots (I to IV), in addition to the four modulation blocks 30, (white, black, black, black). ), (Black, white, black, black), (black, black, white, black) and (black, black, black, white) can be sequentially changed and input to the amplitude phase signal demodulation CNN. good.

The input size of the label information 40 is the same as the size of the modulation block 30 input to the amplitude phase signal demodulation CNN. For example, when another modulation code different from that of the above embodiment is used, or when a symbol adjacent to the periphery of the modulation block 30 is included and input to the amplitude phase signal demodulation CNN in order to improve noise immunity. If the size of the modulation block 30 input to the amplitude phase signal demodulation CNN is, for example, 5 × 5, the label information 40 is also composed of 5 × 5 all-white or all-black symbols, respectively.

(Hologram recording / playback device according to the embodiment)
The hologram recording / reproducing device according to the present embodiment is an apparatus for carrying out the hologram recording / reproducing method according to the present embodiment described above, and the optical system for recording / reproducing the hologram may be the one shown in FIG. can. However, each means constituting the arithmetic unit 120 is required to have at least the following configuration.
That is, the arithmetic unit 120 has a page data dividing means for dividing the page data (input from the camera 119) recorded on the hologram recording medium 114 into each modulation block 30, and bright spots arranged in the modulation block 30. A first machine learning means (CNN) that demodulates the first bit string data based on the position of, and a second machine learning means (CNN) that demodulates the second bit string data based on the amplitude phase signal superimposed on the bright spot. CNN) and a bit string data connecting means for connecting these two demodulated bit string data as one bit string data.
The hologram recording / reproducing device according to this embodiment can also exert the same effect as the hologram recording / reproducing method according to the present embodiment.

As described above, the bright spot position signal and the amplitude phase signal superimposed on each bright spot are independently demodulated by CNN by 4 bits each, so that the network size of the CNN is recorded without enlarging it. The data in the hologram memory can be demodulated. In particular, when the modulation block 30 is a multi-value modulation code, the effect produced by the method of this embodiment is great.

For each CNN learning, a data set in which the data reproduced from the hologram recording / reproducing device and the original recording bit string are paired is used. The weights in the CNN are adjusted so that the data can be demodulated accurately according to the noise characteristics of the hologram recording / playback device. It is necessary to save the weights of the CNN adjusted by pre-learning, but if the number of CNNs that duplicate the weights in the amplitude phase signal demodulation CNN is prepared according to each bright spot, the bright spot position signal All CNN demodulation processes including the demodulation CNN (in this embodiment, five CNN demodulation processes consisting of one bright spot position signal demodulation CNN and four amplitude phase signal demodulation CNNs) can be executed in parallel. The demodulation process can be further speeded up.

Here, the verification result of the simulation to which the noise generated during recording / reproduction is added will be described by comparing this embodiment with the comparative example.
The noise sources include fixed pattern noise in the laser light source 101 in FIG. 11, intercode interference noise due to the high frequency cutoff filter 110, fixed pattern noise in the probe light 123, amplitude modulation SLM107, phase modulation SLM108, and pixels of the camera 119. We assumed sampling noise due to pitch difference, amplitude-modulated SLM107, phase-modulated SLM108, and random noise generated in the camera 119.

(Verification result of comparative example)
The comparative example is the case of demodulation using the hardness determination. FIG. 8 shows a recording / reproducing procedure in the hologram recording / reproducing method according to the comparative example. In the procedure shown in FIG. 8, the steps (S1) in the above-mentioned procedure shown in FIG. 1 up to the steps (S101 to S105) relating to the process of recording the page data and optically reading the page data. Since it is the same as in S5), only the steps (S106 to S111) after the processing step using the 4-step phase shift method will be described here.

In the case of hard determination, both the amplitude (luminance) information and the phase information of the reproduced page data are obtained by using the operation of the 4-step phase shift method based on the reproduced four page data (S106). The modulation block 30 is sequentially extracted from the page data and demodulated (S107). At that time, the luminance value and the phase value of each symbol (each symbol other than the reference symbol 23) of the extracted modulation block 30 are standardized by the known luminance value and phase value of the reference symbol 23, and the whole modulation block 30 is covered. The influence of the generated noise is compensated (S108).

For 8 symbols other than the reference symbol 23, a constellation map is created based on the standardized luminance value and phase value. An example of this constellation map is shown in FIG. 9 (a).
As shown in FIG. 9A, among the positions of the above 8 symbols, 4 points selected in order from the outer peripheral side (that is, 4 points selected in descending order of luminance value: 5 in the circle in FIG. 9A). Points with numbers 8, 4, and 6) can be estimated to be bright spots. As a result, the symbol numbers 1 to 8 are sequentially determined to be "00011101" (1 is assigned to the bright spots and 0 is assigned to the other dark spots), so that the symbol numbers 1 to 8 are based on the conversion table shown in FIG. 3 (b). , The reproduction 4-bit value is determined to be "1101", and as shown in FIGS. 9 (b) and 9 (c), first, the first 4 bits of the 20 bits are demodulated (S109).

Next, the amplitude phase signal superimposed on these four points is demodulated. For four points, demodulate the closest amplitude phase signal point in the map of FIG. 9 (a) in raster scan order (similar to FIG. 6 (a)) from the upper left symbol in the modulation block 30. Judge as a bit. This is performed for each of the four bright spots (I to IV), and as shown in FIGS. 9 (b) and 9 (c), the obtained demodulation bits are combined to form a 16-bit (“00100011100001100”) bit string. (S110). By continuously combining this bit string with 4 bits (“1101”) demodulated from the bright spot position signal, a 20-bit demodulated bit string (reproduction bit string) as shown in FIG. 9 (c) can be obtained (reproduction bit string). S111).

In this comparative example, an error occurs in the demodulation of the amplitude phase signal superimposed on the bright spot of the symbol number 4 due to the noise during recording / reproduction. The correct bit string at the time of recording was "0001", but the amplitude noise and phase noise cause a positional shift from the amplitude phase signal point "0001" at the time of recording on the constellation map, and the amplitude phase signal point closest to the distance is. Since it became "0010", a 2-bit demodulation error occurred.
In the case of the comparative example of demodulation using the hardness determination, the bit error rate after demodulation was 4.1 × 10 ^-3 . The demodulation errors of the bright spot position signal and the amplitude phase signal were 2.8 × 10 ^-3 and 5.1 × 10 ^-3 , respectively.

(Verification result of the example)
In this embodiment, it is referred from the hardness determination result that more errors occurred during demodulation of the amplitude phase signal than the bright spot position signal, and by increasing the number of layers and the number of weights of the amplitude phase signal demodulation CNN, noise was generated. On the other hand, it has a robust network configuration. For the bright spot position signal demodulation CNN, the number of layers and the number of weights were reduced, the demodulation speed was increased, and the recording area for storing the weights after learning was reduced. FIG. 10 shows the configuration of the CNN optimized in this way and the conditions at the time of learning.

As described above, in the CNN demodulation according to the present embodiment, the output of both the bright spot position signal demodulation CNN and the amplitude phase signal demodulation CNN is 4 bits, but the network configuration can be changed according to the characteristics. , Demodulation speed and demodulation accuracy can be optimized.
As described above, when the bright spot position signal can be demodulated accurately but the amplitude phase signal has a demodulation error due to the difference in the amount of noise and the noise immunity of the signal in each signal. In the comparative example, it is difficult to improve this. This example is significantly different from the comparative example in this respect.
Further, in the case of this embodiment in which demodulation is performed using the CNN generated as described above, the bit error rate after demodulation is 2.1 × 10 ^-3 , and the bit error rate is higher than that in the comparative example. It was reduced to about half. The demodulation errors of the bright spot position signal and the amplitude phase signal were 1.3 × 10 ^-4 and 2.6 × 10 ^-3 , and both signals could be demodulated more accurately than in the comparative example.

Further, as can be seen from the comparison between FIGS. 1 and 8, in the hardness determination according to the comparative example, the four page data reproduced by changing the phase of the probe light 123 are subjected to arithmetic processing by the 4-step phase shift method and standardization. However, in the CNN demodulation according to the present embodiment, the four page data images can be directly input to the CNN without prior processing, so that the calculation process can be greatly simplified. In addition, in the rigid determination according to the comparative example, in order to demodulate the amplitude phase signal, it is necessary to demodulate the bright spot position signal first, but in the CNN demodulation according to this embodiment, both signals are demodulated in parallel at the same time. Because it can be done, it is possible to increase the speed.

The hologram recording / reproducing method and the hologram recording / reproducing apparatus of the present invention are not limited to those of the above-described embodiment, and various other modifications can be taken. For example, in the present embodiment, CNN having excellent image recognition is used as a machine learning method, but instead of this, a neural network such as a multi-layer perceptron that does not include a convolution layer or another machine learning method may be used. Further, in the present embodiment, the input to the CNN is set to four modulation blocks in consideration of the four-step phase method, but the present invention is not limited to this, and the modulation block is set to any plurality of modulation blocks other than the four. It is possible.

10, 110 Optical system 23 Reference symbol 30 Modulation block 40 Label information 101 Laser light source 102, 115 Lens 103, 105, 112 1/2 wave plate 104, 106 PBS
107 Amplitude Modulation SLM
108 Phase Modulation SLM
109 Lens group 110 High frequency cutoff filter 111, 113 Mirror 114 Hologram recording medium 116 Phase modulation SLM for correction
117 Piezo Modulator 118 Beam Splitter (BS)
119 Camera 120 Arithmetic Logic Unit 121 Signal Light 122 Reference Light 123 Probe Light 124 Reproduction Light

Claims (7)

The page data recorded on the hologram recording medium for information recording is read out and divided into modulation blocks, and the bright spots arranged in the modulation block are demodulated by the first machine learning process and the first bit string data is demodulated. The hologram recording is characterized in that the second bit string data is demodulated by the second machine learning process for the amplitude phase signal superimposed on the bright spot, and the demodulated two bit string data are associated with each other as one bit string data. Playback method. The hologram recording / reproduction method according to claim 1, wherein the first machine learning process is a process by a first neural network, and the second machine learning process is a process by a second neural network. .. It is characterized in that the second bit string data and label information for identifying a desired bright spot among the plurality of bright spots are input to the second neural network to demodulate the amplitude phase signal. The hologram recording / reproducing method according to claim 2. The hologram recording / reproduction method according to claim 3, wherein the label information is expressed in a one-hot format. By inputting a plurality of reproduced images obtained by phase-shifting the read page data into the second neural network, the amplitude phase signal is demodulated without using phase calculation processing by the phase shift method. The hologram recording / reproducing method according to any one of claims 2 to 4, wherein the hologram recording / reproducing method is characterized. The second neural network constructs a network unit that demodulates the amplitude phase signal for each of the plurality of reproduced images, duplicates the constructed network unit, and corresponds to the number of the plurality of bright spots. The hologram according to claim 5, wherein the predetermined number of the network units are constructed, and the demodulation processing of the predetermined number of the network units is performed in parallel and simultaneously. Recording / playback method. The page data dividing means for dividing the page data recorded and read out on the hologram recording medium for information recording into each modulation block, and the first bit string data based on the positions of the bright spots arranged in the modulation block. The first machine learning means for demodulating, the second machine learning means for demodulating the second bit string data based on the amplitude phase signal superimposed on each of the bright spots, and the demodulated two bit string data are 1 A hologram recording / playback device characterized in that it is provided with a bit string data connection means associated with two bit string data.

Images