# US 2025/0165778 A1 · Signal denoising based on adaptable deep neural networks

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0165778 A1
- Document ID: US20250165778A1
- Status: Published application
- Kind: A1
- Application number: 18/808,653
- Filed: 2024-08-19
- Published: 2025-05-22
- Earliest claimed: 2023-11-17
- Assignees: Sony Group Corporation; Sony Corporation of America
- Applicants: Sony Group Corporation; Sony Corporation of America
- Inventors: WANXIN XU; KO-KAI ALBERT HUANG
- Web page: https://patentgrep.com/patent/US20250165778A1

## CPC

- G06N 3/08: Learning methods
- A61B 5/243: specially adapted for magnetocardiographic [MCG] signals
- A61B 5/318: Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B 5/346: Analysis of electrocardiograms
- A61B 5/7203: {for noise prevention, reduction or removal}
- A61B 5/7257: {using Fourier transforms}
- A61B 5/7267: {involving training the classification device}
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)

## Abstract

An electronic device and a method for implementation for signal denoising based on adaptable deep neural networks. The electronic device receives training data comprising a first bio-signal and a second bio-signal that is different from the first bio-signal. The electronic device computes a weighted sum of the first bio-signal and the second bio-signal. The electronic device generates a mixed signal based on the weighted sum. The electronic device generates an output signal based on application of a denoising neural network (DNN) on the mixed signal. Further, the electronic device computes a loss based on a comparison of the output signal with the first bio-signal and trains the DNN for a number of epochs until the computed loss is below a threshold.

## Claims

### Claim 1

An electronic device, comprising:
a processor configured to:
receive training data comprising a first bio-signal and a second bio-signal that is different from the first bio-signal;
compute a weighted sum of the first bio-signal and the second bio-signal;
generate a mixed signal based on the weighted sum;
generate an output signal based on application of a denoising neural network (DNN) on the mixed signal;
compute a loss based on a comparison of the output signal with the first bio-signal; and
train the DNN for a number of epochs until the computed loss is below a threshold.

### Claim 2 (depends on 1)

The electronic device according to claim 1, wherein the first bio-signal is a magneto cardiogram (MCG) signal, and the second bio-signal is an electrocardiogram (ECG) signal.

### Claim 3 (depends on 1)

The electronic device according to claim 1, wherein the first bio-signal is one of an electrocardiogram (ECG) signal, and the second bio-signal is a magneto cardiogram (MCG) signal.

### Claim 4 (depends on 1)

The electronic device according to claim 1, wherein the loss between the output signal and the first bio-signal is computed based on a spectral loss function.

### Claim 5 (depends on 4)

The electronic device according to claim 4, wherein the spectral loss function is based on Short-Time Fourier Transform (STFT) or Gabor transform.

### Claim 6 (depends on 1)

The electronic device according to claim 1, wherein the computation of the weighted sum is based on a first weight for the first bio-signal and a second weight for the second bio-signal, and the second weight is more than the first weight.

### Claim 7 (depends on 6)

The electronic device according to claim 6, wherein the processor is further configured to set each of the first weight and the second weight based on a user input or a random function.

### Claim 8 (depends on 1)

The electronic device according to claim 1, wherein the processor is further configured to:
receive, from a sensor associated with a user, a bio-signal comprising a noise component via a single channel input or a multi-channel input; and
generate a denoised bio-signal based on application of the trained denoising neural network on the received bio-signal,
wherein the denoised bio-signal is generated after a removal of the noise component from the received bio-signal.

### Claim 9 (depends on 1)

The electronic device according to claim 1, wherein the DNN is an encoder-decoder network comprising an encoder, a decoder coupled to an output of the encoder.

### Claim 10 (depends on 1)

The electronic device according to claim 1, wherein the DNN is a denoising autoencoder or a variant of the denoising autoencoder.

### Claim 11 (depends on 1)

The electronic device according to claim 1, wherein the DNN is a U-net or a variant of the U-net.

### Claim 12

A method, comprising:
in an electronic device:
receiving training data comprising a first bio-signal and a second bio-signal that is different from the first bio-signal;
computing a weighted sum of the first bio-signal and the second bio-signal;
generating a mixed signal based on the weighted sum;
generating an output signal based on application of a denoising neural network (DNN) on the mixed signal;
computing a loss based on a comparison of the output signal with the first bio-signal; and
training the DNN for a number of epochs until the computed loss is below a threshold.

### Claim 13 (depends on 12)

The method according to claim 12, wherein the first bio-signal is a magneto cardiogram (MCG) signal, and the second bio-signal is an electrocardiogram (ECG) signal.

### Claim 14 (depends on 12)

The method according to claim 12, wherein the first bio-signal is an electrocardiogram (ECG) signal, and the second bio-signal is a magneto cardiogram (MCG) signal.

### Claim 15 (depends on 12)

The method according to claim 12, wherein the loss between the output signal and the first bio-signal is computed based on a spectral loss function.

### Claim 16 (depends on 15)

The method according to claim 15, wherein the spectral loss function is based on Short-Time Fourier Transform (STFT) or Gabor transform.

### Claim 17 (depends on 12)

The method according to claim 12, wherein the computation of the weighted sum is based on a first weight for the first bio-signal and a second weight for the second bio-signal, and the second weight is more than the first weight.

### Claim 18 (depends on 17)

The method according to claim 17, further comprising setting each of the first weight and the second weight based on a user input or a random function.

### Claim 19 (depends on 12)

The method according to claim 12, further comprising:
receiving, from a sensor associated with a user, a bio-signal comprising a noise component via a single channel input or a multi-channel input; and
generating a denoised bio-signal based on application of the trained denoising neural network on the received bio-signal,
wherein the denoised bio-signal is generated after a removal of the noise component from the received bio-signal.

### Claim 20

A non-transitory computer-readable medium having stored thereon, computer-executable instructions that when executed by an electronic device, causes the electronic device to execute operations, the operations comprising:
receiving training data comprising a first bio-signal and a second bio-signal that is different from the first bio-signal;
computing a weighted sum of the first bio-signal and the second bio-signal;
generating a mixed signal based on the weighted sum;
generating an output signal based on application of a denoising neural network (DNN) on the mixed signal;
computing a loss based on a comparison of the output signal with the first bio-signal; and
training the DNN for a number of epochs until the computed loss is below a threshold.
