Software Configuration and Setup

This section provides guidance on configuring and running the software components of the High-Performance Analog Meets AI demonstration.

Overview

The demonstration consists of two main software components:

  1. Holoscan Modulation Classification Application - A real-time AI-powered modulation classification system built with NVIDIA Holoscan SDK

  2. Data Visualization Dashboard - A Python-based Dash application for monitoring and analyzing classification results

Holoscan Modulation Classification Application

Description

This application performs real-time modulation classification on incoming data using MATLAB-generated CUDA code. It processes signals from ADRV9009 devices and classifies them into one of eight modulation schemes:

  • BPSK (Binary Phase Shift Keying)

  • QPSK (Quadrature Phase Shift Keying)

  • 8PSK (8-Phase Shift Keying)

  • 16QAM (16-Quadrature Amplitude Modulation)

  • 64QAM (64-Quadrature Amplitude Modulation)

  • PAM4 (4-Pulse Amplitude Modulation)

  • GFSK (Gaussian Frequency Shift Keying)

  • CPFSK (Continuous Phase Frequency Shift Keying)

Key Components

Holohub application: The Holohub application and IIO operators

There are 5 operators (each with its own Python binding) that can read/write IIO attributes, read/write IIO buffers and configure the initial setup for a device/system. This demonstration uses the buffer read operator in order to read from the 2 ADRV9009 devices. After that, the matlab module (that needs to be compiled beforehand) will call the classification function that returns the modulation and confidence level. This will be printed to the output file from where the python dash app will read and display the information.

Configuration

  1. Network Configuration

    Update IP addresses in main.cpp for your ADRV9009 devices:

    // Configure IIO buffer read for Talise device at IP 10.43.1.10 auto
iio_rx_1 = make_operator<ops::IIOBufferRead>("iio_rx_1",
                                  Arg("ctx") =
                                  std::string("ip:10.43.1.10"), Arg("dev")
                                  = std::string("axi-adrv9009-rx-hpc"), //
                                  ... other parameters

  2. Output Configuration

    Modify the YAML file to set output file paths:

    matlab:
  out_file: "modulation_results.txt"

  3. Channel Configuration

    The application is configured for 8 channels (4 I/Q pairs):

    std::vector<std::string> channel_names = {
    "voltage0_i", "voltage0_q", "voltage1_i", "voltage1_q", "voltage2_i",
    "voltage2_q", "voltage3_i", "voltage3_q"
};

Building and Running

  1. Build the Application

  2. Run the Application

The application will:

  • Connect to the specified ADRV9009 devices

  • Continuously read data samples (8192 samples per buffer, but it is configurable from the code)

  • Process data through the MATLAB classification model

  • Output results to modulation_results.txt and modulation_results1.txt

Data Visualization Dashboard

Location

The visualization script is located at: jupiter_modulation/examples/plot_identification_data.py

Description

This Python application provides a web-based dashboard for monitoring and analyzing the modulation classification results in real-time. It features:

  • Real-time confusion matrices

  • Constellation diagrams

  • Time-domain waveform plots

  • Classification accuracy metrics

  • Interactive modulation selection

Key Features

Real-time Monitoring

  • Reads results from Holoscan output files every 3 seconds

  • Updates visualizations dynamically

  • Tracks classification performance over time

Modulation Control

  • Manual modulation selection or automatic random switching

  • Real-time transmission to connected SDR devices

  • Supports all 8 modulation schemes

Signal Visualization

  • Constellation diagrams showing I/Q relationships

  • Time-domain waveforms (I and Q components)

  • Interactive plots with hover information

Prerequisites

Install required Python packages:

~$

pip3 install dash plotly pandas scipy scikit-learn numpy pyadi-iio

Hardware Requirements

The dashboard expects the following hardware setup:

  • 4 ADRV9002 devices for signal transmission (IPs: 192.168.0.15-18)

  • 2 ADRV9009ZU11eg devices for signal reception (configured in Holoscan app)

  • Modulated data files in modulated_data/ directory

Configuration

  1. File Paths

    Update file paths to match your Holoscan output location:

    file_path =
'/home/analog/git/holohub/build/matlab_classify_modulator/modulation_results.txt'
file_path1 =
'/home/analog/git/holohub/build/matlab_classify_modulator/modulation_results1.txt'

  2. SDR Configuration

    Configure SDR devices with appropriate IP addresses and stream profiles:

    sdr = adi.adrv9002(uri="ip:192.168.0.15")
sdr.write_stream_profile("lte_40_lvds_api_68_14_10.stream",
"lte_40_lvds_api_68_14_10.json")

  3. Modulated Data

    Ensure modulated data files are available in the modulated_data/ directory:

    • mod_BPSK.mat

    • mod_QPSK.mat

    • mod_8PSK.mat

    • mod_16QAM.mat

    • mod_64QAM.mat

    • mod_PAM4.mat

    • mod_GFSK.mat

    • mod_CPFSK.mat

Running the Dashboard

  1. Start the Holoscan Application

    Ensure the Holoscan modulation classification application is running and generating output files.

  2. Launch the Dashboard

    ~$

    python3 plot_identification_data.py

  3. Access the Web Interface

    Open a web browser and navigate to http://localhost:8050

Usage Workflow

  1. Select Modulation: Use the dropdown to choose a specific modulation or select “Random” for automatic switching

  2. Monitor Classification: Observe real-time classification results for both ADRV9009 devices

  3. Analyze Performance: Review confusion matrices and accuracy metrics

  4. Examine Signals: Study constellation diagrams and time-domain waveforms

Support and Resources

For additional support and documentation: