1 of 6

Spectrum Classification Application

Avanish Mishra

Brendan Bruce

Noah Paladino

2 of 6

Project Summary

Plan: <See ML Modem Plan>

  • Creating a program that recognizes modulation scheme of input signals and demodulates them
  • Will also choose the best modulation scheme to modulate and transmit signals based on the SNR of the given frequency band

​

Demodulator System Example

3 of 6

This Week

  • Finished generation of modulated data using QPSK/16QAM
    • Finalized the settings of the modulation blocks
    • Able to input a SNR and generate data that has added AWGN (Additive White Gaussian Noise) depending on that SNR
  • Working on packaging/saving the data to be loaded into machine learning program

4 of 6

Dataset Generation Flow Chart:

Random Source: Generates random numbers, set to [0,256) to generate bytes

Throttle: Sets the sample rate of the byte stream

Constellation Modulator:

  • Modulates the byte stream depending on input mod scheme (bytes -> complex numbers)
  • Upsamples based on samples/symbol rate
  • Filters the upsampled signal to scale the values from basic values such as -1, 0, 1

Channel Model: Adds additive white gaussian noise to the upsampled, filtered signal

Vector Sink: Loads stream of complex numbers into an array

5 of 6

Determining Noise Level using SNR

​

  • Need to calculate level of noise to add to signal
  • Using an inputted SNR to calculate the noise voltage

​

Asignal is RMS of signal before noise is added, Anoise is noise level as a voltage

  • Above formula translates to:

Anoise = Asignal / (10^(SNRdB/20))

6 of 6

Next Week

  • Load training data into CNN program for a few modulation schemes
  • Start testing demodulator by testing between these few schemes
  • Adjust SNR used for dataset generation based on error rate of demodulator

​