Neural Network SMPS Control for Dynamic Switching Frequency

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing switch mode power converters using PID regulation struggle to react quickly to fast-changing load currents, leading to output voltage dips and efficiency issues at low loads due to high switching losses.

Innovation Solution

A controller with an artificial neural network (ANN) that adjusts the switching frequency based on output current and voltage signals, allowing for dynamic changes in switching period and duty cycle to improve responsiveness and efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If PID regulation with fixed switching frequency is used, then the control structure is simple, but the response to fast-changing load currents is slow causing output voltage dips

Engineering Contradiction:
Improveresponse speed to load changesVSAvoidoutput voltage stability
Core Design Contradiction:
SpeedVSReliability

Solution Approach 1:

The patent implements dynamic switching frequency adjustment where the switching frequency varies based on operating conditions. During transient events or light load conditions, the frequency is increased to improve response speed and reduce voltage dips. During steady-state heavy load conditions, the frequency is reduced to minimize switching losses. This dynamic adaptation resolves the contradiction between fast response and voltage stability.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the switching frequency parameter dynamically based on system state. By monitoring load conditions and adjusting the switching frequency accordingly, the system achieves both fast response during transients (higher frequency) and stable output voltage during steady-state (optimized frequency), thereby resolving the technical contradiction.

Inventive Principle:
Principle #35Parameter changes

2Speed

If high switching frequency is maintained for fast response, then the response to transients is fast, but switching losses increase at low loads

Engineering Contradiction:
Improveresponse speed to transientsVSAvoidswitching losses
Core Design Contradiction:
SpeedVSLoss of energy

Solution Approach 1:

The system dynamically adjusts switching frequency based on load conditions. During transient events, the frequency is increased to ensure fast response. During light load steady-state conditions, the frequency is reduced to minimize switching losses. This dynamic behavior resolves the contradiction between maintaining fast response capability and reducing energy losses.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent employs periodic monitoring of load conditions and adjusts switching frequency accordingly. The controller periodically evaluates whether transient conditions exist and modifies the switching frequency in response, achieving fast response when needed while conserving energy during normal operation.

Inventive Principle:
Principle #19Periodic action

3Loss of energy

If low switching frequency is used to reduce switching losses, then energy efficiency improves at light loads, but the response capability to sudden load changes deteriorates

Engineering Contradiction:
Improveswitching lossesVSAvoidresponse speed to load changes
Core Design Contradiction:
Loss of energyVSSpeed

Solution Approach 1:

The switching frequency is dynamically adjusted based on real-time detection of transient conditions. During light load steady-state, a lower frequency reduces switching losses. Upon detecting a transient event or light load condition, the frequency is immediately increased to restore fast response capability. This dynamic adjustment resolves the contradiction between energy efficiency and response speed.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system uses feedback from load current monitoring to adjust switching frequency. When the controller detects light load conditions or transient events through feedback signals, it modifies the switching frequency accordingly, achieving both energy efficiency during normal operation and fast response during transients.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4503411A1A neural network architecture for control of power converters
Publication Date: 2025.02.05 INFINEON TECH AUSTRIA AG
  • EP4503411A1 patent drawingFigure 1
  • EP4503411A1 patent drawingFigure 2~4
  • EP4503411A1 patent drawingFigure 5~6

AI summary

A controller is presented. The controller may be configured to receive a voltage signal indicative of an output voltage of a switched-mode power supply SMPS, receive a current signal indicative of an output current of the SMPS, and generate a switching signal for controlling a switching element of the SMPS. The controller may comprise a duty cycle controller configured to determine a duty cycle signal based on the voltage signal. The controller may comprise a frequency modifier configured to generate a frequency signal based on at least one of: the voltage signal, the current signal, the duty cycle signal, or another output signal of the duty cycle controller. The controller may comprise a modulator configured to determine the switching signal based on the duty cycle signal and the frequency signal. The frequency modifier may comprise an artificial neural network.