irreq / tism

Software defined acoustic modem using deep-learning demodulation without a clock.

Warning some code is broken and the program will not work at the moment.

Modem

Time-independent softmodem - tism

This is a Python implementation of a Deep Learning Acoustic Modem for my third year final degree project in Engineering Science and Technology, (Teknikvetenskap). (2020-2021)

This program aims at reinventing the Acoustic coupler modem defined as [1]:

In telecommunications, an acoustic coupler is an interface device for coupling electrical signals by acoustical means—usually into and out of a telephone. https://en.wikipedia.org/wiki/Acoustic_coupler

By utilizing modern computational power, this modem will [SOON] be able to transfer data faster than previous semi hardware/software modems by approaching the demodulation process in a human-like manner.

Theoretical transfer speeds does not directly justify this pure software implementation, however it does indirectly justify the development of demodulation technologies that could be implemented for increased safety, transfer-speed and reduced error-rate.

Methods

Methods powering the modem are the following:

• Deep Demodulation using Tensorflow

• Deep Segmentation using Hidden Markov Models

• Sound Activity Segmentation using Support Vector Machines (SVM) utilizing an implemented version of silence removal from pyAudioAnalysis. Developed by Theodoros Giannakopoulos (https://github.com/tyiannak/pyAudioAnalysis/)

@article{giannakopoulos2015pyaudioanalysis, title={pyAudioAnalysis: An Open-Source Python Library for Audio Signal Analysis}, author={Giannakopoulos, Theodoros}, journal={PloS one}, volume={10}, number={12}, year={2015}, publisher={Public Library of Science} }

Background

Transfer speed and loss reduction finds itself in constant improvement in Digital Communications. The medium in which data used to be transfered in, air is now more or less obsolete for data transfer in favour of less lossier, and faster wireless radio communication, Wi-Fi, Li-Fi. My proposal to the growing bandwidth problem is to utilize deep adaptive technologies such as deep learning for data transfer optimization, on the fly. Protocols for Wi-FI are based on [3]:

IEEE 802.11 https://en.wikipedia.org/wiki/IEEE_802.11

However as a PoC (Proof of Concept), no initial human-like protocols are being set, the modem will itself define rules of communication during training and synchronization in production. Benefits are:

• Rules can change during transfer, not just change of rules to adapt for outside interference.

• Better suited rules I am not saying that other predefined rules or protocols are bad, however some are not as easely implemented in the acoustic spectrum.

Installation

Clone the modem using git clone:

$ git clone https://github.com/Irreq/gyarbete.git

Install requirements using pip:

$ pip3 install requirements.txt

Install the program using pip:

$ pip3 install .

Usage

The PoC Modem works in different ways:

$ python3 -m main.py [your_argument_here] [your_file_here] [extra_arguments_here]

Example :

$ python3 -m main.py Demodulate received_wave02.wav -s ~/Desktop/results.txt

Here, Demodulate tells the modem to process the wave file received_wave02.wav and save the results to the location ~/Desktop/results.txt

Dependencies

matplotlib, scipy, numpy, pyaudio, pydub, scikit-learn, keras, tensorflow

Credits

Special thanks to the following people for making this possible.

• David Masse (https://github.com/davidmasse/neural-net-1D-audio)

• Theodoros Giannakopoulos (https://github.com/tyiannak/pyAudioAnalysis/)

• Harrison Kinsley (https://pythonprogramming.net/)

• Florent Forest (https://github.com/FlorentF9/)

References

[1] In telecommunications, an acoustic coupler is an interface device for coupling electrical signals by acoustical means—usually into and out of a telephone. https://en.wikipedia.org/wiki/Acoustic_coupler

[3] IEEE 802.11 https://en.wikipedia.org/wiki/IEEE_802.11