Channel Circuit with Trained Neural Network Noise Mixture Estimator

Patent metadata

Field	Value
Publication number	US-2024170056-A1
Application number	US-202318354178-A
Country	US
Kind code	A1
Filing date	Jul 18, 2023
Priority date	Nov 21, 2022
Publication date	May 23, 2024
Grant date	—

Abstract

Example channel circuits, data storage devices, and methods for using a trained neural network to estimate the noise mixture in a read signal are described. Samples are determined from a digital read signal, such as the read signal from the non-volatile storage medium of a data storage device. The samples are processed through one or more instances of a neural network comprised of trained coefficients and outputting a set of estimate values for a noise mixture of the read signal. The set of estimate values may then be used to adjust parameters of the read channel for processing the read signal to detect and decode data.

First claim

What is claimed is:

1 . A channel circuit, comprising:
an interval selector configured to determine at least one sample from a digital read signal; a trained noise estimator configured to:
process the at least one sample through an instance of a neural network comprised of trained coefficients; and
output a set of estimate values corresponding to a noise mixture in the digital read signal; and
adjustment logic configured to adjust at least one parameter of a read channel based on the set of estimate values.

2 . The channel circuit of claim 1 , wherein the set of estimate values include estimate values for a plurality of noise types selected from:
jitter; electronic noise; and color noise.

3 . The channel circuit of claim 1 , wherein:
the set of estimate values include a jitter value, an electronic noise value, and a color noise value; and combination of the jitter value, the electronic noise value, and the color noise value correspond to a total noise value.

4 . The channel circuit of claim 1 , further comprising:
an analog-to-digital converter configured to:
receive an analog read signal determined from a non-volatile storage medium;
determine the digital read signal from the analog read signal; and
buffer the digital read signal to a buffer memory, wherein the interval selector determines the at least one sample from the buffer memory.

5 . The channel circuit of claim 1 , wherein:
the at least one sample determined by the interval selector includes a plurality of samples; each sample of the plurality of samples is configured with a sample size and a sample interval from the digital read signal; the trained noise estimator comprises a plurality of instances of the neural network: each sample of the plurality of samples is processed through a different instance of the plurality of instances of the neural network; and the set of estimate values are based on outputs from the plurality of samples and the plurality of instances of the neural network.

6 . The channel circuit of claim 5 , further comprising:
estimate combination logic configured to:
receive a corresponding set of estimate values from each sample of the plurality of samples; and
determine, based on the corresponding sets of estimate values, the set of estimate values provided to the adjustment logic.

7 . The channel circuit of claim 1 , further comprising:
an iterative detector configured to:
detect data bits from the digital read signal; and
decode detected bit patterns to determine data units stored to a non-volatile storage medium, wherein the at least one parameter of the read channel is configured to adjust values in the digital read signal for processing by the iterative detector.

8 . The channel circuit of claim 1 , wherein:
the interval selector is further configured to determine the at least one sample for a predefined adjustment interval; the adjustment logic is further configured to adjust the at least one parameter of the read channel during the predefined adjustment interval; and the predefined adjustment interval is each data sector of a non-volatile storage medium.

9 . The channel circuit of claim 1 , further comprising:
estimator training logic configured to:
receive a training data signal comprised of known noise components and known data components;
sample the training data signal to determine a set of training samples; and
process, using the read channel and a neural network trainer, the set of training samples to determine the trained coefficients by separating a plurality of noise components from data components in the set of training samples.

10 . A data storage device comprising the channel circuit of claim 1 and further comprising a non-volatile storage medium configured to store data.

11 . A method comprising:
determining at least one sample from a digital read signal; processing the at least one sample through an instance of a neural network comprised of trained coefficients; outputting, from the instance of the neural network, a set of estimate values corresponding to a noise mixture in the digital read signal; and adjusting at least one parameter of a read channel based on the set of estimate values.

12 . The method of claim 11 , wherein the set of estimate values include estimate values for a plurality of noise types selected from:
jitter; electronic noise; and color noise.

13 . The method of claim 11 , wherein:
the set of estimate values include a jitter value, an electronic noise value, and a color noise value; and combination of the jitter value, the electronic noise value, and the color noise value corresponds to a total noise value.

14 . The method of claim 11 , further comprising:
receiving, by an analog-to-digital converter, an analog read signal determined from a non-volatile storage medium; determining, by the analog-to-digital converter, the digital read signal from the analog read signal; buffering the digital read signal to a buffer memory; and determining the at least one sample from the buffer memory.

15 . The method of claim 11 , further comprising:
determining a plurality of samples from the digital read signal, wherein each sample of the plurality of samples is configured with a sample size and a sample interval from the digital read signal; and processing each sample of the plurality of samples through a different instance of a plurality of instances of the neural network, wherein the set of estimate values are based on outputs from the plurality of samples and the plurality of instances of the neural network.

16 . The method of claim 15 , further comprising:
receiving a corresponding set of estimate values from each sample of the plurality of samples; and determining, based on the corresponding sets of estimate values, the set of estimate values used for adjusting the at least one parameter of the read channel.

17 . The method of claim 11 , further comprising:
detecting, by an iterative detector, data bits from the digital read signal; and decoding, by the iterative detector, detected bit patterns to determine data units stored in a non-volatile storage medium, wherein adjusting the at least one parameter of the read channel includes adjusting values in the digital read signal for processing by the iterative detector.

18 . The method of claim 11 , further comprising:
determining the at least one sample for a predefined adjustment interval; and adjusting the at least one parameter of the read channel during the predefined adjustment interval, wherein the predefined adjustment interval is each data sector of a non-volatile storage medium.

19 . The method of claim 11 , further comprising:
receiving a training data signal comprised of known noise components and known data components; sampling the training data signal to determine a set of training samples; and training, using the read channel and the set of training samples, the neural network to determine the trained coefficients for separating a plurality of noise components from data components in the set of training samples.

20 . A data storage device comprising:
a non-volatile storage medium; a channel circuit including a read channel; means for determining at least one sample from a digital read signal; means for processing the at least one sample through an instance of a neural network comprised of trained coefficients; means for outputting, from the instance of the neural network, a set of estimate valu

Assignees

• Western Digital Tech Inc

Inventors

• Qin Minghai
• Sun Chao
• Vucinic Dejan
• Yip Henry
• Goode Jonas
• Galbraith Richard
• Oboukhov Iouri

Classifications

• G06N3/084

  Backpropagation, e.g. using gradient descent · CPC title

• G06N3/08

  Learning methods · CPC title

• G11B20/10046

  filtering or equalising, e.g. setting the tap weights of an FIR filter · CPC title

• G11C11/54 Primary

  using elements simulating biological cells, e.g. neuron · CPC title

• G11C11/416

  Read-write [R-W] circuits · CPC title