Machine-learning optimization of data reading and writing

Patent metadata

Field	Value
Publication number	US-12019705-B2
Application number	US-202117301473-A
Country	US
Kind code	B2
Filing date	Apr 5, 2021
Priority date	Oct 17, 2017
Publication date	Jun 25, 2024
Grant date	Jun 25, 2024

Abstract

Examples are disclosed that relate to encoding data on a data-storage medium. The method comprises obtaining a representation of a measurement performed on the data-storage medium, the representation being based on a previously recorded pattern of data encoded in the data-storage medium in a layout that defines a plurality of data locations. The method further comprises inputting the representation into a data decoder comprising a trained machine-learning function, and obtaining from the data decoder, for each data location of the layout, a plurality of probability values, wherein each probability value is associated with a corresponding data value and represents the probability that the corresponding data value matches the actual data value in the previously recorded pattern of data at a same location in the layout.

First claim

The invention claimed is:

1. On a computing device, a method for encoding data on a data-storage medium, the method comprising:
obtaining a first training-data set to train a data decoder, the first training-data set corresponding to a first data-writing parameter configuration;
using the first training-data set and corresponding ground-truth data, training a machine-learning function of the data decoder to decode data written with the first data-writing parameter configuration, the machine-learning function outputting a first labelled probability array corresponding to the first training-data set;
obtaining a second training-data set to train the data decoder, the second training-data set corresponding to a second data-writing parameter configuration, wherein obtaining the first and second training-data sets includes probing the data-storage medium with a read beam and concurrently imaging the data-storage medium;
using the second training-data set and corresponding ground-truth data, training the machine-learning function of the data decoder to decode data written with the second data-writing parameter configuration, the machine-learning function outputting a second labelled probability array corresponding to the second training-data set;
assessing, based on the first and second labelled probability arrays, a reliability of decoding data written with the first data-writing parameter configuration and a reliability of decoding data written with the second data-writing parameter configuration;
selecting one of the first or second data-writing parameter configurations based on the reliabilities assessed of decoding data of at least the first and second training data sets; and
writing data to the data-storage medium using the selected one of the first or second data-writing parameter configurations.

2. The method of claim 1 wherein the one of the first or second data-writing parameter configurations is selected further based on a metric of data-writing performance, and wherein obtaining the second training-data set comprises obtaining a training-data set with a data-writing parameter configuration adjusted so as to increase the metric of data-writing performance.

3. The method of claim 1 wherein the first and second training data sets differ with respect to layout.

4. The method of claim 1 wherein the data-storage medium comprises an optical storage medium, the first and second training data sets comprise images, each element of the labelled probability array corresponds to a possible data value, and the data is encoded using an optical write-beam.

5. The method of claim 1 wherein the first and second training data sets differ with respect to one or more properties of a write-beam used to encode the data from which the respective first or second training data set was obtained.

6. The method of claim 1 wherein the first and second training data sets differ with respect to the read-beam properties.

7. The method of claim 1 wherein assessing the reliability comprises one or more of comparing the probability values of each element of the labeled probability array and comparing the decoded data to the corresponding ground-truth values.

8. On a computing device, a method for encoding data on an optical data-storage medium, the method comprising:
obtaining a first training-data set to train a data decoder, the first training-data set corresponding to a first data-writing parameter configuration and comprising images;
using the first training-data set and corresponding ground-truth data, training a machine-learning function of the data decoder to decode data written with the first data-writing parameter configuration, the machine-learning function outputting a first labelled probability array corresponding to the first training-data set;
obtaining a second training-data set to train the data decoder, the second training-data set corresponding to a second data-writing parameter configuration and comprising images;
using the second training-data set and corresponding ground-truth data, training the machine-learning function of the data decoder to decode data written with the second data-writing parameter configuration, the machine-learning function outputting a second labelled probability array corresponding to the second training-data set;
assessing, based on the first and second labelled probability arrays, a reliability of decoding data written with the first data-writing parameter configuration and a reliability of decoding data written with the second data-writing parameter configuration;
selecting one of the first or second data-writing parameter configurations based on the reliabilities assessed of decoding data of at least the first and second training data sets; and
writing data to the optical data-storage medium using the selected one of the first or second data-writing parameter configurations, wherein the data is encoded using an optical write beam and read using a beam of a predetermined polarization state, and wherein each corresponding data value differs with respect to a birefringence value, and
wherein each element of the first and second labelled probability arrays corresponds to a possible data value and represents a probability that the corresponding possible data value matches an actual data value encoded in the optical storage medium.

9. The method of claim 8 wherein the one of the first or second data-writing parameter configurations is selected further based on a metric of data-writing performance, and wherein obtaining the second training-data set comprises obtaining a training-data set with a data-writing parameter configuration adjusted so as to increase the metric of data-writing performance.

10. The method of claim 8 wherein the first and second training data sets differ with respect to the layout.

11. The method of claim 8 wherein the first and second training data sets differ with respect to one or more properties of the write-beam used to encode the data from which the respective first or second training data set was obtained.

12. The method of claim 8 wherein obtaining the first and second training data sets includes probing the optical data-storage medium with a read beam and concurrently imaging the optical data-storage medium, and wherein the first and second training data sets differ with respect to the read-beam properties.

13. The method of claim 8 wherein assessing the reliability comprises one or more of comparing the probability values of each element of the labeled probability array and comparing the decoded data to the corresponding ground-truth values.

14. The method of claim 8 wherein the machine-learning function comprises a convolutional neural network.

15. An optical data storage and retrieval system comprising:
an optical data-storage medium; and
one or more logic devices configured to execute instructions to operate data decoder including a machine-learning function, the machine-learning function being
trained (a) using a first training-data set corresponding to a first data-writing parameter configuration and corresponding ground-truth data, the machine-learning function being trained to decode data written with the first data-writing parameter configuration and configured to output a first labelled probability array corresponding to the first training-data set, the first training-data set comprising images; and
trained (b) using a second training-data set corresponding to a second data-writing parameter configuration and corresponding ground-truth data, the machine-learning function being further trained to decode data written with the second data-writing par

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Stefanovici Ioan Alexandru
• Thomsen Benn Charles
• Gaunt Alexander Lloyd
• Rowstron Antony Ian Taylor
• Nowozin Reinhard Sebastian Bernhard

Classifications

• H03M7/008

  using opto-electronic devices · CPC title

• H03M5/06

  Code representation, e.g. transition, for a given bit cell depending only on the information in that bit cell · CPC title

• H03M13/37

  Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35 · CPC title

• G11B20/10268

  bit detection or demodulation methods · CPC title

• G11B20/10009 Primary

  Improvement or modification of read or write signals · CPC title