Skip to content

Latest commit

 

History

History
98 lines (71 loc) · 3.58 KB

README.md

File metadata and controls

98 lines (71 loc) · 3.58 KB
Raw

Automatic Modulation Recognition (AMR) System

Overview

This project tackles the critical challenge of accurately classifying modulation types in telemetry, tracking, and command (TTC) systems, particularly in low carrier-to-noise density ratio (CNR) environments. By leveraging AI/ML techniques and FPGA-based hardware, the system achieves real-time signal recognition and efficient data processing, ensuring reliable communication.

Key Features

  • AI/ML-Based Modulation Classification: Combines CNN and LSTM layers for high accuracy (>92%) over a signal-to-noise ratio (SNR) range of [0,18].
  • Threshold-Based Logic: Implements predefined signal characteristics for fast, real-time decision-making.
  • FPGA Integration: Utilizes a PYNQ-Z2 FPGA for efficient, low-latency signal classification.

Problem Statement

Traditional AMR systems face challenges in low CNR environments due to:

  1. Low Accuracy: High noise levels degrade signal quality.
  2. High Computational Complexity: Resource-intensive algorithms hinder real-time applications.
  3. Limited Signal Identification: Inability to handle complex distortions.

Proposed Solution

Dual-Approach Strategy:

  1. AI/ML-Based Approach:

    • Model: CNN + LSTM with ~108k parameters.
    • Dataset: DeepSig RadioML 2016.10A dataset with 11 modulation types.
    • Accuracy: >92% for SNR [0,18].

  2. Threshold-Based Approach:

    • Utilizes lightweight logic for real-time classification.
    • Achieves >95% accuracy with minimal computational overhead.

FPGA Implementation

  • Advantages:
    • Parallel Processing: Enables real-time operation.
    • Flexibility: Supports reconfiguration for evolving requirements.
    • Low Latency: Ensures processing delays of <100ms.

Benefits

  • Enhanced Data Accuracy: Reliable modulation classification minimizes errors.
  • Real-Time Processing: Low-latency FPGA implementation ensures rapid decision-making.
  • Scalability: Adaptable to diverse modulation schemes and communication systems.
  • Energy Efficiency: Runs on 5V, consuming 40% less power than alternatives.

Applications

  • Aerospace & Defense: Real-time secure communication and signal intelligence.
  • Telecommunications: Supports 5G/6G networks with efficient spectrum usage.
  • IoT: Reliable communication in dense, connected environments.
  • Autonomous Systems: Ensures robust communication for vehicles and drones.

Implementation Details

Steps:

  1. Develop AI/ML algorithms using CNN and LSTM.
  2. Validate models on MATLAB simulations and the DeepSig dataset.
  3. Implement algorithms on FPGA using HDL.
  4. Test and benchmark the system for real-time performance.

Hardware:

  • Primary Device: PYNQ-Z2 FPGA
  • Possible Alternative: Nvidia Jetson Nano for hybrid deployment.

Results

  • ML Model Accuracy: >92% on SNR range [0,18].
  • Thresholding Approach Accuracy: >95% for selected modulation schemes.
  • Latency: <100ms processing time.
  • Power Efficiency: Operates smoothly on 5V.

Team

  • Team Members: Bharat Agarwal,Saurabh Singh, Yash Verma, Rudraksh Lande, Karan Seth, Mohit
  • Mentors: Dr. Robin Khosla, Nikhil Chauhan

Future Work

  • I/O Device Integration: Adding input and output interfaces for direct interaction with the FPGA.
  • Model Deployment: Optimizing the ML model for FPGA compatibility.
  • Testing & Benchmarking: Validating the system under diverse real-world conditions.

License

This project is licensed under WTFPL.

For any inquiries, feel free to contact us.