Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Quick Primer on Machine Learning in Wireless Communications

Faris B. Mismar

TL;DR

This paper tackles reproducibility challenges in ML-enabled wireless prototyping by introducing a quick-primer and the open-source deepwireless library for Python-based MIMO-OFDM simulations. It presents both model-driven and data-driven workflows around a single OFDM symbol, covering CSI estimation, equalization, symbol detection, and radio resource management with AI/ML use cases. The work demonstrates unsupervised, supervised, and reinforcement learning techniques (e.g., k-means, KDE, linear regression, CNNs, LSTMs, DQNs) and provides concrete scenarios and code to reproduce results on inexpensive hardware. By delivering accessible code, tutorials, and scalable workflows, it aims to accelerate AI-enabled air-interface prototyping for 4G/5G and beyond.

Abstract

This is our final issue of the quick primer on the use of Python to build a wireless communications prototype. This prototype simulates multiple-input and multiple-output (MIMO) systems for a single orthogonal frequency division multiplexing (OFDM) symbol. Further, it shows several artificial intelligence (AI) and machine learning (ML) use cases and introduces the deepwireless library for code implementation. The intent of this primer is to empower the reader with the means to efficiently create reproducible simulations related to AI and ML in wireless communications on inexpensive computing devices. This primer has sprung from a draft aligned with the syllabus of a graduate course (EESC 7v86) -- which we created to be first taught in Fall 2022 -- and has since evolved to where it stands today.

A Quick Primer on Machine Learning in Wireless Communications

TL;DR

This paper tackles reproducibility challenges in ML-enabled wireless prototyping by introducing a quick-primer and the open-source deepwireless library for Python-based MIMO-OFDM simulations. It presents both model-driven and data-driven workflows around a single OFDM symbol, covering CSI estimation, equalization, symbol detection, and radio resource management with AI/ML use cases. The work demonstrates unsupervised, supervised, and reinforcement learning techniques (e.g., k-means, KDE, linear regression, CNNs, LSTMs, DQNs) and provides concrete scenarios and code to reproduce results on inexpensive hardware. By delivering accessible code, tutorials, and scalable workflows, it aims to accelerate AI-enabled air-interface prototyping for 4G/5G and beyond.

Abstract

This is our final issue of the quick primer on the use of Python to build a wireless communications prototype. This prototype simulates multiple-input and multiple-output (MIMO) systems for a single orthogonal frequency division multiplexing (OFDM) symbol. Further, it shows several artificial intelligence (AI) and machine learning (ML) use cases and introduces the deepwireless library for code implementation. The intent of this primer is to empower the reader with the means to efficiently create reproducible simulations related to AI and ML in wireless communications on inexpensive computing devices. This primer has sprung from a draft aligned with the syllabus of a graduate course (EESC 7v86) -- which we created to be first taught in Fall 2022 -- and has since evolved to where it stands today.
Paper Structure (19 sections, 29 equations, 13 figures, 2 tables)

This paper contains 19 sections, 29 equations, 13 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Protocol Stack
  3. System Model
  4. Transmit power
  5. Signal to noise ratio
  6. Model-Driven Implementation
  7. Unsupervised Machine Learning Use Cases
  8. Clustering
  9. Kernel Density Estimation
  10. Supervised Machine Learning Use Cases
  11. Linear Regression
  12. Deep Learning
  13. Reinforcement Learning Use Cases
  14. Performance Measures
  15. Radio Performance Measures
  16. ...and 4 more sections

Figures (13)

  • Figure 1: The control plane air interface protocol stack.
  • Figure 2: A block diagram of the overall model of the wireless system (thicker boxes are optional blocks).
  • Figure 3: Fully connected deep neural network (depth $D=3$ and width $W=6$) of an input $N$ and an output $M$.
  • Figure 4: Actions, states, and rewards: reinforcement learning.
  • Figure 5: $16$-QAM constellation with Gray code which for any two adjacent symbols only a change of one bit is permissible.
  • ...and 8 more figures