Optical Neural Network for High-Speed Signal Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current digital signal processors (DSPs) face limitations in processing high-speed optical signals due to operational speed and power consumption constraints, and they do not provide high-quality processing for optical telecommunication links with intensity modulation and direct detection systems, especially in ultra-high capacity networks.

Innovation Solution

An optical neural network (ONN) is used for processing optical signals, which includes layers of optical neurons with weighted interconnections controlled by phase shifters, allowing for optical convergence and enabling ultra-high data rate processing directly in the optical domain without the need for electronic conversion.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If digital signal processors (DSPs) are used to process optical signals, then data processing capability is provided, but operational speed is limited and power consumption is high

Engineering Contradiction:
Improvedata processing capabilityVSAvoidoperational speed
Core Design Contradiction:
ProductivityVSSpeed

Solution Approach 1:

The patent replaces electronic DSP processing with optical neural network processing. The optical neural network uses optical components (waveguides, interferometers, modulators) to perform signal processing directly in the optical domain, eliminating the need for electro-optic conversion and electronic processing. This substitution enables processing speeds matching the optical signal rates (100 GBauds and above) while reducing power consumption compared to electronic DSP systems.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If digital signal processors (DSPs) are used to process optical signals, then data processing capability is provided, but power consumption is greater than 20 Watts

Engineering Contradiction:
Improvedata processing capabilityVSAvoidpower consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent replaces power-hungry electronic DSP operations with low-power optical processing. The optical neural network performs matrix multiplications and nonlinear operations using optical interference and modulation, consuming only the power needed for optical modulation and detection rather than high-power electronic processing. This reduces power consumption from over 20 Watts to levels suitable for ultra-high capacity networks.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Productivity

If DSP-based processing is used for optical telecommunication links, then signal processing is performed, but processing quality is insufficient for intensity modulation and direct detection systems

Engineering Contradiction:
Improvesignal processing functionVSAvoidprocessing quality
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent replaces electronic DSP processing with optical neural network processing specifically designed for intensity modulation and direct detection (IM/DD) systems. The optical neural network performs equalization, signal recovery, and detection directly in the optical domain, maintaining signal integrity and achieving high processing quality for IM/DD systems where electronic DSP approaches fail.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

4Productivity

If optical signals are converted to electronic domain for processing, then DSP processing can be performed, but the conversion process adds complexity and limits speed

Engineering Contradiction:
Improveprocessing capabilityVSAvoidsystem complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent extracts the signal processing function from the electronic domain and implements it directly in the optical domain using an optical neural network. This eliminates the electro-optic conversion stage and its associated complexity (converters, synchronization, clock recovery), while maintaining full processing capability for equalization, signal recovery, and detection.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent substitutes electronic processing systems with optical processing systems, replacing complex electro-optic conversion infrastructure with integrated photonic circuits that perform all necessary signal processing operations directly on the optical signal.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach allows for efficient handling of ultra-high data rate signals in optical transmission systems, reducing the need for equalization at the receiver end and providing a simple and effective method for processing high-speed optical signals, while minimizing power consumption and operational complexity.

Implementation Method 1

The weightings are established by phase-shifting the input optical signals prior to the summation. The phase shift controller controls the phase-shifting so as to obtain required weightings for the input optical signals.

Methodology Applied
Scientific EffectPhase shifting:

Implementation Method 2

Yichen Shen et al: 'Deep learning with cohererent nanaphotonic circuits' describes an artificial neural network, which consists of a set of input artificial neurons connected to at least one hidden layer and an output layer. In particular an Optical Neural Network, ONN, architecture is described, where signals are encoded in the amplitude of optical pulses propagating in integrated photonic waveguides where they pass through an optical interference unit, OIU

Methodology Applied
Scientific EffectOptical interference: Interference

Implementation Method 3

optical signals are sampled by an optical sampler. The optical samples are processed by an optical neural network (ONN) which includes layers of optical neurons.

Methodology Applied
Scientific EffectOptical sampling:

Data Source

PatentEP3991325B1Photonic signal processing
Publication Date: 2023.08.30 HUAWEI TECH CO LTD
  • EP3991325B1 patent drawingFigure 1A
  • EP3991325B1 patent drawingFigure 1B~1C
  • EP3991325B1 patent drawingFigure 1D

AI summary

An apparatus for optical processing of a signal includes an optical sampler, an optical neural network and a phase shift controller. The optical sampler obtains parallel optical samples by sampling a continuous optical signal. The optical neural network optically processes the parallel optical samples and provides a final output optical signal resulting from the processing. The optical neural network includes interconnected optical neurons. Each of the optical neurons inputs multiple optical signals and outputs an optical signal which is a weighted sum of the input optical signals. The weightings are established by phase-shifting the input optical signals prior to the summation. The phase shift controller controls the phase-shifting so as to obtain required weightings for the input optical signals.