Generic Timeseries Classification

Table of Contents

• Introduction
• Supported Combinations
• Dataset and Model Details
  • Dataset
  • Model Architecture
  • Input Features
  • Output
• Feature Extraction Configuration
• Steps to Run the Example
• See Also
• Sample Output

Introduction

Generic Timeseries Classification is a hello world example for understanding usage of AI models on TI MCU. This project demonstrates implementation of an AI-based waveform detection system on AM26x microcontrollers. It showcases how to deploy machine learning models for real-time waveform classification in embedded systems, serving as a hello world example. TI provides a complete development ecosystem with toolchains and SDKs that significantly streamline all stages of Edge AI solution development.

The example uses a simple classification dataset with three waveform classes: sine, sawtooth, and square waveforms.

Supported Combinations

Parameter	Value
CPU + OS	r5fss0-0 nortos
Toolchain	ti-arm-clang
Board	am263px-lp
Example folder	examples/ai/generic_timeseries_classification/

Dataset and Model Details

Dataset

TI has created a specialized waveform dataset containing signal measurements. The dataset is divided into three classes: sine waveform, sawtooth waveform and square waveform.

Parameter	Value
Sensor	Signal readings
Sampling Rate	40 Hz
Channels	1 (signal magnitude)
Samples per File	25,600 samples
Total Files	75 files (~25 files in each of 3 classes)

Model Architecture

This lightweight classification model CLS_1k_NPU contains approximately 1,000 parameters and follows a streamlined architecture consisting of four convolutional layers (each enhanced with BatchNorm and ReLU activation functions) followed by a single linear layer.

Input Features

The model takes 4D input (N,C,H,W):

• N (1): batch size which is restricted to 1
• C (1): channels which is 1 for signal
• H (256): samples of timeseries signals which is 256 in this example
• W (1): width of samples is restricted to 1 for timeseries applications

Output

This model produces a 1D output representing the three possible classes. The position of the highest value in this output indicates the classification result:

• Class 0: Sine waveform
• Class 1: Sawtooth waveform
• Class 2: Square waveform

Feature Extraction Configuration

Feature extraction transforms raw data into meaningful inputs for the AI model. For this waveform detection system, applying FFT to identify frequencies, followed by binning and logarithmic scaling, produces superior results.

The feature extraction pipeline is configured in user_input_config.h:

• FE_FFT: Performs Fast Fourier Transform on the raw frame, calculating magnitude values from complex outputs
• FE_DC_REM: Removes the DC component from Fourier transform
• FE_BIN: Groups frequencies into bins, reducing dimensionality while preserving frequency distribution patterns
• FE_LOG: Applies logarithmic scaling to the binned frequency data
• FE_CONCAT: Combines scaled outputs from multiple data frames for temporal context

Key parameters:

• FE_FRAME_SIZE: 1024 samples per FFT frame
• FE_FEATURE_SIZE_PER_FRAME: 64 features per frame
• FE_NUM_FRAME_CONCAT: 8 frames concatenated
• FE_FFT_BIN_SIZE: 8 (bin width)
• FE_NN_OUT_SIZE: 3 (number of output classes)

Steps to Run the Example

• When using CCS projects to build, import the CCS project for the required combination and build it using the CCS project menu (see Using SDK with CCS Projects).
• When using makefiles to build, note the required combination and build using make command (see Using SDK with Makefiles)
• Launch a CCS debug session and run the executable, see CCS Launch, Load and Run
• The application will run inference on test vectors and display results in the console

See Also

AI Examples

Sample Output

Generic Timeseries Classification Example Started ...
Feature extraction mismatches 0
All tests have passed!!
Golden vectors matched: 3 not matched: 0