WO2012048776A1

WO2012048776A1 - Method and device for recovering digital data

Info

Publication number: WO2012048776A1
Application number: PCT/EP2011/004626
Authority: WO
Inventors: Tim Fingscheidt; Christoph Voges; Volker MÄRGNER; Florian Pflug
Original assignee: Technische Universität Carolo-Wilhelmina Zu Braunschweig
Priority date: 2010-09-27
Filing date: 2011-09-15
Publication date: 2012-04-19
Also published as: DE102010046540A1

Abstract

The invention relates to a method for recovering digital data that are stored on a storage medium (1) and/or are transmitted via a transmission medium (11), wherein stored and transmitted signal values (7, 17) are received and converted in each case into an analog or digital data value (8, 18) and a corresponding binary value (5, 15) and a reliability value (6, 16) are determined from the data value (8, 18), wherein the reliability value (6, 16) contains information about the reliability with which the specific binary value (5, 15) corresponds with the binary value represented by the received signal value (7, 17), characterised in that asymmetrical probability density functions are used to determine the binary value (5, 15) and the reliability value (6, 16). The invention further relates to a data recovery device for that purpose, and to a computer program that is suitable for carrying out such a method.

Description

Method and device for recovering digital data

The invention relates to a method for recovering digital data stored on a storage medium and / or transmitted via a transmission medium, according to the preamble of claim 1. The invention further relates to a data recovery device therefor according to claim 10 and a computer program.

For storage of digital data nowadays different storage media are used, for. As CDs, DVDs, hard drives or magnetic tapes. For the permanent archiving of data there is a need to use a long-term stable storage medium. After a few years, CDs and DVDs already show that such storage media are unlikely to be suitable for long-term data archiving, for example over a period of 500 years. In addition, the rapid development of computer storage media shows that previously used storage media, such as floppy disks, are now no longer easy to use due to the lack of suitable readers. Another problem of data archiving lies in the fact that readers must be permanently available for the storage media used for archiving. For these reasons, the possibilities for data archiving on photochemical media, eg microfilm, are being researched, as described for example in the publication by C. Voges, V. Märgner and T. Fingscheidt, "Digital Data Storage on Microfilm - Error Correction and Storage Capacity Issues,""in Proc. of IS & T Archiving Conference, Bern, Switzerland, June 2008, pages 212-215. Photochemical storage media have, in contrast z. As to microelectronic memory chips, no purely digital characteristics, ie digital data is not stored as pure binary information, but as a gray value. The invention has for its object to provide technical possibilities for the recovery of digital data, in which the readout reliability is increased.

This object is achieved by the invention specified in claims 1, 10, 11 and 12. The dependent claims indicate advantageous developments of the invention.

As recovery of digital data, all methods are included which allow reconstruction of stored or transmitted data into a purely digital form, e.g. Demodulation. For storing the data, digital or non-digital storage media may be used. For transmission of data, digital or non-digital transmission media may also be used, e.g. wireless or wired transmission media.

When storing digital data on a non-digital storage medium, e.g. As a photochemical medium or a magnetic tape, a conflict of objectives is on the one hand to accommodate as much data per unit area of the storage medium and thus to enable high specific storage capacity, on the other hand to be able to recover the data reliably with existing reading devices. The invention has the advantage specific memory capacity on a non-digital storage medium to increase the reliability of reading the digital data. If the reliability of the readout remains the same, the data density of the storage and thus the specific storage capacity of the storage medium can also be increased.

The invention has the advantage of fully utilizing the analog values stored and read on the non-digital storage medium with regard to the information contained in them. In contrast to a hard threshold controlled conversion of the read analog values to binary values, the invention proposes efficient soft demodulation of the data stored on the non-digital storage medium. As a result, a high reproducibility of the read data is made possible. The invention is basically suitable for all digital and non-digital storage media, such as optical storage media and magnetic storage media. The invention is also suitable for the transmission of data over all digital and non-digital transmission media. On the storage medium and / or transmission medium, statistics and signal distributions based thereon can be measured on the readout / receive side, meaning that they are separated for eg stored / sent ones and zeros (in the case of binary data). From these signal distributions, in the case of the medium microfilm, for example, grayscale distributions, probability density functions can be determined separately for a stored one and a stored zero, respectively. In particular, the invention also relates to channels having non-Gaussian distributed probability density functions; in some media such as microfilm these functions are both finite (eg only values between black and white) and asymmetric. According to an advantageous development of the invention, the non-symmetrical probability density functions model noise or intersymbol functions. interference processes on the storage medium and / or the transmission medium.

According to an advantageous development of the invention, the non-symmetric probability density functions are modeled as a weighted sum of a number of Gaussian distributions. For this purpose, a Gaussian mixed distribution model can be used. The term "modeling" is understood here in the sense of "mathematically emulating". According to an advantageous embodiment of the invention, an optical memory is used as the storage medium, e.g. a photochemical storage medium. Advantageously, a film, in particular e.g. a microfilm can be used. The use of microfilm technology has the further advantage that, due to the past decades of use of microfilm technology in the field of analog document archiving, there are experiences in the scope with this storage medium and corresponding reading devices. In general, photochemical storage media have the advantage of being long-lasting. According to an advantageous development of the invention, the binary value and the reliability value combined are determined as the log-likelihood ratio. Here, the sign of the log-likelihood ratio reflects the binary value. The amount of the log likelihood ratio represents the reliability value. This allows a particularly efficient further data processing, which allows a robust decoding of the stored / transmitted data.

According to an advantageous embodiment of the invention, the parameters of the Gaussian distributions are determined by means of real data read from a non-digital storage medium or received by a transmission medium in a preceding training step. Advantageous is the training preceded the actual useful step of reading the digital data in time. In an advantageous embodiment of the invention, in the training step, the parameters of the Gaussian distributions are determined, for example, by using the expectation-maximization algorithm. This allows an efficient training of the Gaussian mixed distribution model used for the actual readout of the digital data.

The invention also relates to a data recovery device having a reading device and / or a receiving device, and having an evaluation device, wherein the evaluation device has a computer device that is set up to carry out a method of the previously described type.

The invention also relates to a data carrier having a computer program, which is set up to carry out a method of the previously described type. As a disk, all conventional data carriers come into question, such. Flash memory, DVD, CD, floppy or the Internet.

The invention will be explained in more detail using an exemplary embodiment using drawings.

Show it

Figures 1 and 2 a microfilm and Figure 3 shows a first embodiment of a data recovery device and

Figure 4 Distribution density functions Figure 5 shows a second embodiment of a data recovery device. FIGS. 1 and 2 show images of a part of a microfilm generated with a microscope, on which digital data are optically stored. As can be seen, the microfilm does not contain pure binary black and white information of the stored data, but has gray levels. Figures 1 and 2 are shown with the same scale. In the case of the microfilm according to FIG. 1, the data are stored in the 6 μm pitch, in the microfilm according to FIG. 2 in the 9 μm pitch.

3 shows a microfilm 1 and a data recovery device in the form of a data reading device 2, which is used for reading data from the microfilm 1. The data reading device 2 has a reading device 3 and an evaluation device 4. The reading device 3 may be formed, for example, as a film scanner. The evaluation device 4 can have, for example, as a computer means a microprocessor 9 executing a computer program. The reading device 3 generates data values 8 corresponding to the data 7 read from the microfilm 1. The data values 8 can be values with a continuous value range or discrete, digitized values, eg with 8 or 12 bit resolution. The data values 8 are processed in the evaluation device 4 as described below. In this case, an associated binary value 5 and an associated reliability value 6 are determined for each data bit read out on the storage medium from the corresponding data value 8 and output to further devices. By means of the reliability values, it can then be analyzed there whether the received binary values of the stored data satisfy certain safety requirements or not. Furthermore, the reliability values can be used, for example, for an effective equalization and / or a subsequent soft-input channel decoding (for soft-input channel decoding, see, for example: J. Hagenauer, "Rate-Compatible Conjugated Revolutionary Codes (RCPC Codes) and Their Applications". , IEEE TRANSACTIONS ON COMMUNICATIONS, Vol. 36, No. 4, April 1988, pages 389-400) Thus, if appropriate, it may be decided to read the storage medium again with another reading device with higher resolution, to ensure a higher security of the data read out. In addition, an analysis can be initiated as to whether there is a corruption of the data.

A further alternative embodiment can provide that only one single signal forms the output from the evaluation device 4 instead of the two signals 5 and 6: In the aforementioned log-likelihood ratios, their signs unambiguously determine the binary value 5 and their magnitude the reliability value 6. In the following it is assumed that on the microfilm 1 binary values x are stored from the set {0, 1}. This results in read-out after a normalization step non-binary output values y (in Figure 3, the read data 7) from a range of values [0, 1], with which the most likely stored values x 'are recovered. In the context of a demodulation of the stored data, a log-likelihood ratio (LLR) L (x | y) (usable as an alternative output signal to the signals 5 and 6 in FIG. 3) can be calculated by

with the a-posteriori probabilities P (x = 1 | y) and P (x = 0 | y), the conditional probability density functions (PDFs) p (y | x = 1) and p (y | x = 0) and the A priori probabilities P (x = 1) and P (x = 0). In the case of equally distributed input values, P (x = 1) = P (x = 0) = 0.5. This leads to the simplification

In the case of digital data storage on a non-digital storage medium considered here, the PDFs p (y | x = 1) and p (y | x = 0) are asymmetrically distributed, see an exemplary distribution in FIG. 4. The PDFs can be combined with a Gaussian mixed distribution Model (Gaussian Mixture Model, GMM) as a weighted sum of / Gaussian distributions and

be approximated. Accordingly, results

and

with the standard deviations the averages and the

Weights where the weights are normalized according to:

The Gaussian distributions are defined by

and

The GMM parameters are obtained by real data extraction using the expectation-maximization algorithm.

Furthermore, in the above-described recovery of the digital data, it is possible to include further normalized non-binary outputs in the calculations described by formulas (1) to (7) in addition to the normalized non-binary output y, eg, the surrounding outputs. In equations (1) to (7), the scalar quantity y is to be replaced by a vectorial quantity Y which is composed of the normalized non-binary output y and of a certain number of further standardized output signals contained in the vector Ϋ:

The probability density functions (PDFs) p (Y | x = 1) and p (Y | x = 0) then have a correspondingly higher dimensionality, but can in principle be obtained in the same way as the PDFs in equations (1) to (7). , Their use leads to a reduction of intersymbol interference, which indicates a combined demodulator / equalizer. Its use can also be used, for example, with otherwise identical exposure parameters, a smaller grid size can be used, resulting in a higher storage density. Without limiting the generality, the further representation again uses the symbol y for scalar non-binary outputs. The Gaussian mixed distribution model described above is advantageously implemented in the evaluation device 4 in the form of software programming. At the output of the evaluation device 4, the log-likelihood ratio values L (x | y) are then output directly or in the form of the output data 5, 6. Here, the binary values 5 correspond to the sign of the log-likelihood ratio and the reliability values 6 correspond to the magnitude of the log-likelihood ratio.

As can be seen, the invention includes a determination of a Log Likelihood Ratio (LLRs) for asymmetric distribution density functions p (y | x = 1) and p (y | x = 0). These distribution density functions, as illustrated by way of example in FIG. 4, describe, for example, the brightness values or density values read in the case of a microfilm, the output signal in a storage medium or the output signal of a transmission channel. These values are distributed asymmetrically, which z. B. can be caused due to a limited range of y.

FIG. 5 shows an application of the invention in data transmission technology. Recognizable is a transmission medium 11 and a data recovery device in the form of a data receiving device 12, which serves to receive data from the transmission medium 11. The data receiving device 12 has a receiving device 13 with a receiving antenna 10 and an evaluation device 14 with a computer means 19. The receiving device 13 may be formed, for example, as a radio receiver. Accordingly, the transmission medium 11 is then formed as a radio transmission path. The receiving device 13 generates for the data 17 received from the transmission medium 11 corresponding data values 18. The data values 18 may be values with continuous value range or discrete, digitized values, e.g. B. with a resolution of 8 or 2 bits. The data values 18 are processed in the evaluation device 14 in the same way as described above for the evaluation device 4 according to FIG. It For each data bit received by the transmission medium 11, an associated binary value 15 and an associated reliability value 16 are determined from the corresponding data value 18 and output to further devices. By means of the reliability values, it can then be analyzed there whether the received binary values of the stored data satisfy certain safety requirements or not. Furthermore, the reliability values z. B. be used for an effective equalization and / or a subsequent soft-input channel decoding. Thus, if appropriate, it may be decided to operate the transmission medium 11 at a lower transmission rate or with a different transmission frequency subjected to fewer interfering influences in order to ensure greater security of the data read out. In addition, an analysis can be initiated as to whether there is a corruption of the data.

Claims

claims:

Method for recovering digital data stored on a storage medium (1) and / or transmitted via a transmission medium (1 1), in which stored or transmitted signal values (7, 17) are recorded and respectively in the form of a analog value or digital data value (8, 18) and from the data value (8, 18) a corresponding binary value (5, 15) and a reliability value (6, 16) are determined, the reliability value (6, 16) being a There's about the reliability involved with that particular binary value

(5, 15) coincides with the binary value represented by the recorded signal value (7, 17), characterized in that non-symmetric probability density functions are used for the determination of the binary value (5, 15) and the reliability value (6, 16).

2. The method according to claim 1, characterized in that the non-symmetric probability density functions model noise or inter- symbol interference processes on the storage medium (1) and / or the transmission medium (11).

3. Method according to one of the preceding claims, characterized in that the non-symmetric probability density functions are modeled as the weighted sum of a number of Gaussian distributions.

4. The method according to any one of the preceding claims, characterized in that the data from a storage medium (1) by means of a reading device (3) are read, wherein the reading device (3) a quantized data value (8) for each read signal value (7 ) generated.

5. The method according to any one of the preceding claims, characterized in that as the storage medium (1), an optical memory, in particular a microfilm is used.

6. The method according to any one of the preceding claims, characterized in that the data from a transmission medium (1 1) by means of a receiving device (13) are received, wherein the receiving means (13) generates a data value (18) for each received signal value (17) ,

7. Method according to one of the preceding claims, characterized in that the binary value (5, 15) and the reliability value (6, 16) are combined in a log-likelihood ratio (LLR), the sign of the log-likelihood ratio (LLR) represents the binary value (5, 15) and the amount of the log-likelihood ratio (LLR) represents the reliability value (6, 16).

8. The method according to any one of the preceding claims, characterized in that the parameters of the Gaussian distributions are determined by means of real data read from a storage medium (1) or received by a transmission medium (11) in a preceding training step.

9. The method according to claim 8, characterized in that in the training step, the parameters of the Gaussian distributions are determined by using the expectation-maximization algorithm.

10. Data recovery device (2, 12) with a reading device (3) and / or a receiving device (13), and with an evaluation device (4, 14), wherein the evaluation device (4, 14) comprises a computer means (9, 19) adapted to carry out a method according to any one of the preceding claims.

11. Computer program with program code means set up for carrying out the method according to one of claims 1 to 9, when the computer program is executed on a computer.

12. Computer program with program code means which are stored on a machine-readable carrier, set up for carrying out the method according to one of claims 1 to 9, when the computer program is executed on a computer.