Leaky Spiking Neuron Interface for Direct PDM Signal Processing

Solution Overview

Problem

Existing sound processing systems require pre-processing and conversion logic, leading to larger silicon area and higher power consumption, as well as separate hardware for data conversion and signal feature-extraction, which are not efficiently integrated with neural networks.

Innovation Solution

A direct hardware interface between sensors and neural networks using leaky spiking neurons that perform both data conversion and signal feature-extraction, eliminating the need for pre-processing hardware and allowing the neural network to handle PDM signals directly, thereby reducing silicon area and power consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If pre-processing hardware stage is used to convert PDM signal to MFCC coefficients, then signal processing capability is improved, but silicon area and power consumption increase

Engineering Contradiction:

Improvesignal processing capabilityVSAvoidsilicon area

Core Design Contradiction:

ProductivityVSArea of stationary object

Solution Approach 1:

The patent merges the pre-processing functions (PDM to PCM conversion, Mel-filterbank, MFCC calculation) directly into the neural network hardware. The neural network is designed to accept PDM signals as direct input and perform feature extraction internally, eliminating the need for separate pre-processing hardware stages. This integration reduces silicon area while maintaining signal processing capability.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The neural network is designed with multi-functionality to handle both signal conversion and feature extraction tasks. The same hardware components perform multiple functions: PDM signal reception, temporal integration, Mel-filterbank processing, and neural network computation, all within a unified architecture that reduces overall hardware requirements.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Productivity

If pre-processing hardware stage is used to convert PDM signal to MFCC coefficients, then signal processing capability is improved, but power consumption increases

Engineering Contradiction:

Improvesignal processing capabilityVSAvoidpower consumption

Core Design Contradiction:

ProductivityVSUse of energy by stationary object

Solution Approach 1:

The patent merges the pre-processing functions (PDM to PCM conversion, Mel-filterbank, MFCC calculation) directly into the neural network hardware. The neural network is designed to accept PDM signals as direct input and perform feature extraction internally, eliminating the need for separate pre-processing hardware stages. This integration reduces silicon area while maintaining signal processing capability.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The neural network is designed with multi-functionality to handle both signal conversion and feature extraction tasks. The same hardware components perform multiple functions: PDM signal reception, temporal integration, Mel-filterbank processing, and neural network computation, all within a unified architecture that reduces overall hardware requirements.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Productivity

If separate hardware for data conversion and feature-extraction is used, then processing functions are improved, but device complexity increases

Engineering Contradiction:

Improveprocessing functionsVSAvoiddevice complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent merges the pre-processing functions (PDM to PCM conversion, Mel-filterbank, MFCC calculation) directly into the neural network hardware. The neural network is designed to accept PDM signals as direct input and perform feature extraction internally, eliminating the need for separate pre-processing hardware stages. This integration reduces silicon area while maintaining signal processing capability.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS11238887B2Interface to leaky spiking neurons

Publication Date: 2022.02.01 DSP GROUP

AI summary

A processor, that may include at least one neural network that comprises at least one leaky spiking neuron; wherein the at least one leaky spiking neuron is configured to directly receive an input pulse density modulation (PDM) signal from a sensor; wherein the input PDM signal represents a detected signal that was detected by the sensor; and wherein the at least one neural network is configured to process the input PDM signal to provide an indication about the detected input signal.