FALL DETECTION USING WIFI CHANNEL

STATE INFORMATION

Project Report
Submitted to The APJ Abdul Kalam Technological University

in partial fulfillment of the requirements for the award of the Degree

of B.Tech. in Electronics and Communication Engineering

Submitted by

Amritha S S (TVE20AE014)

Anhajul S Rahman (TVE20AE015)

Mohammed Bin Saleem (TVE20AE040)

Vaishnav Sankar (TVE20AE059)

Guided by

Prof. Jerrin Thomas Panachakel

DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING

COLLEGE OF ENGINEERING TRIVANDRUM

KERALA

December 2023
 Declaration

I undersigned hereby declare that the project report FALL DETECTION

USING WIFI CHANNEL STATE INFORMATION, submitted for partial

fulfillment of the requirements for the award of degree of Bachelor of Technology of

the APJ Abdul Kalam Technological University, Kerala is a bonafide work done

by us under supervision of Prof. Jerrin Thomas Panachakel. This submission

represents our ideas in our own words and where ideas or words of others

have been included, we have adequately and accurately cited and referenced the

original sources. We also declare that we have adhered to ethics of academic

honesty and integrity and have not misrepresented or fabricated any data or idea

or fact or source in my submission. We understand that any violation of the above

will be a cause for disciplinary action by the institute and/or the University and

can also evoke penal action from the sources which have thus not been properly

cited or from whom proper permission has not been obtained. This report has

not been previously formed the basis for the award of any degree, diploma or

similar title of any other University.

Place : Thiruvananthapuram Amritha S S

Date : 05/12/2023 Anhajul S Rahman

Mohammed Bin Saleem

Vaishnav Sankar

2
 DEPT. OF ELECTRONICS & COMMUNICATION

ENGINEERING COLLEGE OF ENGINEERING TRIVANDRUM

2020 - 24

CERTIFICATE

This is to certify that the report entitled Fall Detection using WiFi

Channel State Information submitted by Amritha S S (TVE20AE014), An-

hajul S Rahman (TVE20AE015), Mohammed Bin Saleem (TVE20AE040)

& Vaishnav Sankar (TVE20AE059), to the APJ Abdul Kalam Technological

University in partial fulfillment of the B.Tech. degree in Applied Electronics and

Instrumentation Engineering is a bonafide record of the mini project work carried

out by them under our guidance and supervision. This report in any form has

not been submitted to any other University or Institute for any purpose.

Dr.Joseph Zacharias Dr.Alex Raj S.M

(Project Cordinator)
Professor and Head Professor
Dept.of ECE Dept.of ECE
College of Engineering Trivandrum College of Engineering Trivandrum
Trivandrum Trivandrum

Prof. Sindhu N Dr.Jerrin Thomas Panachakel

(Project Co-ordinator) (Project Guide)
Assistant Professor Assistant Professor
Dept.of ECE Dept.of ECE
College of Engineering Trivandrum College of Engineering Trivandrum
Trivandrum Trivandrum

External Examiner
 Acknowledgement

First of all, we would like to express our deepest appreciation and gratitude to

our guide Prof.Dr. Jerrin Thomas Panachakel (Guide), Department of Electronics

and Communication Engineering, College of Engineering Trivandrum, for all his

guidance and support. We could not have completed this project if it wasn’t for

his constant guidance and advice.

We also express our sincere gratitude to Dr Joseph Zacharias , Head of the

Department, Department of Electronics and Communication Engineering, for his

constant cooperation and guidance.

We gratefully acknowledge Dr. Sindu N , Assistant Professor and Dr. Alex

Raj SM, Professor, Department of Electronics and Communication Engineering,

our Project Coordinators for their wholehearted support and encouragement.

We also acknowledge our gratitude to all other members of faculty in the

Department of Electronics and Communication Engineering, our friends and

family members for their wholehearted cooperation and encouragement.

Amritha S S

Anhajul S Rahman

Mohammed Bin Saleem

Vaishnav Sankar

4
 Abstract

Falling is recognized as one of the most frequent accidents among elderly people.

The project aims to develop a non-contact fall detection system using Wi-Fi

channel state information (CSI) that is comfortable, reliable, and cost-effective,

particularly for elderly individuals. The use of Wi-Fi technology allows for non-

intrusive monitoring without the need for wearable devices or expensive camera-

based solutions. The system utilizes embedded devices, specifically ESP32

microcontrollers, to collect and analyze CSI data for accurate human activity

recognition and fall detection.

The proposed system overcomes the limitations of existing wearable and

camera-based solutions, which may be uncomfortable, prone to wear and tear,

and costly. By leveraging Wi-Fi CSI data and machine learning algorithms, the

system can accurately recognize different human activities, including falls, and

detect if a room is empty. The system’s effectiveness is demonstrated through

the analysis of CSI data collected from ESP32 microcontrollers, showcasing

its potential for real-world application in fall detection and human activity

recognition

The project exploits the Phase and Amplitude of the Fine-grained Channel

State Information (CSI) accessible in commodity Wi-Fi devices, and fulfils the

goal of segmenting and detecting the falls automatically in real-time, which allows

users to perform daily activities naturally and continuously without wearing any

devices on the body.

Contents

Abstract i

List of Figures iii

1 Introduction 1

1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.4 Outline of Report . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Literature Review 4

3 Related Works 7

3.1 RT FALL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3.2 Deep Learning-Based Fall Detection Using WiFi Channel State

Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3.3 WiFall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3.4 FallDeFi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

4 Theoretical Overview 12

4.1 Hardware Specification . . . . . . . . . . . . . . . . . . . . . . . 12

4.2 Wi-Fi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4.3 Signals used for analysis . . . . . . . . . . . . . . . . . . . . . . . 20

i
 4.4 Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . 21

5 Methodology 22

5.1 Hardware Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

5.2 Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . 24

5.3 Dataset Preperation . . . . . . . . . . . . . . . . . . . . . . . . . 24

5.4 Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

6 Results and Discussion 26

7 Conclusions 28

ii
List of Figures

4.1 esp32 Microcontroller . . . . . . . . . . . . . . . . . . . . . . . . . 14

4.2 esp32 Pin-out Diagram . . . . . . . . . . . . . . . . . . . . . . . . 14

4.3 Block Diagram of a simplified OFDM System . . . . . . . . . . . 19

4.4 Block Diagram of MIMO System . . . . . . . . . . . . . . . . . . 20

5.1 Block Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

5.2 ESP32 using Broadcast Method . . . . . . . . . . . . . . . . . . . 23

6.1 CSI Data Obtained . . . . . . . . . . . . . . . . . . . . . . . . . . 26

iii
Chapter 1

Introduction

1.1 Background

The share of people over 65 years among the world population is on a steady

rise due to increases in human life expectancy. The size of the population of this

age group will be close to a quarter of the 20-64 age group by the year 2040.

Each year, about one third of people from this age group are vulnerable to falls

and about one out of five people who suffer falls are exposed to head injuries,

broken bones in wrists, arms, ankles and hip fractures. Many of those who are

not injured, are unable to get up on their own after a fall. This can result in

them developing a fear of falling, loss of confidence to live independently, low

physical activities, low social contacts, depression resulting in low quality of life

and even death. Falls are the leading cause of both fatal and nonfatal injuries for

the elderly. Therefore, a system that can automatically detect a fall and inform

caregivers or relatives within a short time can be an important health care tool.

A reliable fall detection system is helpful to the elderly community in two ways:

it reduces the time those who have fallen remain lying on the floor and reduces

the fear of falling. The development of fall detection systems is essential for this

reason. Although many fall detection systems exist today, most of them are based

1
on wearable devices or computer vision, which have some critical drawbacks,

caused by the fundamental nature of the systems. This project is aimed to provide

a non-contact fall detection system that is free of those drawbacks and is yet as

effective as the existing solutions. Furthermore, we use embedded devices as this

makes the system highly modifiable, deployable in many different environments,

and comparatively affordable.

1.2 Motivation

Sensor-based activity detection requires the user to wear a device containing the

sensors in many cases. Some methods use computer vision for the task, which

requires a camera to collect video or image data. Although these methods can

detect a fall incident quite accurately, some issues caused by these methods can

make implementing them in real-world scenario a challenge. The issues faced

while implementing solutions based on these methods are:

• A wearable device can be perceived as uncomfortable, causing unwillingness

to use them.

• A wearable device is more prone to wear and tear than a stationary device.

• A camera-based solution can be both computationally and monetarily

expensive.

Comfort is of utmost importance for the elderly and the sick. A user might be

understandably unwilling to wear a device if it is uncomfortable for them. Even

if a system is perfectly capable of performing its assigned task, implementing

it is challenging, if not impossible when the users are not willing to cooperate.

Also to have a widespread application of a system, cost and durability have to

be considered, especially in developing countries. So, in this project, we aim to

2
create a non-contact fall detection system using wifi channel state information

that is more comfortable, more reliable, and less costly.

1.3 Objective

We aim to build a non-contact fall detection system for monitoring the elderly

and the sick. The objectives we aim to achieve are as follows:

• Implementing a system that can detect if a person in its area of operation

has fallen down.

• Ensuring comfort by eliminating the need for wearing any device.

• Creating a dataset of channel state information recorded during different

activities including fall.

• Recognition of different human activities using Wi-Fi channel state infor-

mation to facilitate future improvement opportunities.

• Implementing a system to detect if the area of operation is empty.

• Ensuring that the system is easy to deploy and affordable by using

embedded devices.

1.4 Outline of Report

Chapter 1 gives a brief introduction to the project. In Chapter 2, the literature

review for the project is discussed. Chapter 3 gives an idea about the proposed

system used. Chapter 4 deals with the methodology of the system. Chapter 5

explain the system design. Results and discussions are conducted in chapter 6.

Finally, the report is concluded in chapter 7.

3
Chapter 2

Literature Review

Falls in older people: epidemiology, risk factors and strategies for

prevention.

LAURENCE Z. RUBENSTEIN UCLA School of Medicine and Geriatric Research

Education and Clinical Center (GRECC), VA Medical Center, Sepulveda, CA

91343, USA

Falls in older people are a common problem that can lead to morbidity, mortality,

and nursing home admissions. Many falls are associated with identifiable risk

factors, and addressing these factors can reduce falling rates. Effective fall

reduction programs involve systematic risk assessment, targeted interventions,

exercise programs, and environmental inspections. Interdisciplinary collaboration

is important for assessing and addressing fall risks. Unintentional falls are

a leading cause of death in older adults. Falls are a common and often

devastating problem among older people, causing a tremendous amount of

morbidity, mortality and use of health care services including premature nursing

home admission.

WiFi Sensing with Channel State Information: A Survey

YONGSEN MA, GANG ZHOU, and SHUANGQUAN WANG, Computer Science

Department, College of William Mary, USA

4
This article is a survey that provides a comprehensive review of WiFi sensing

technologies using Channel State Information (CSI). It discusses the signal

processing techniques, algorithms, applications, and performance results of

WiFi sensing with CSI. The survey categorizes WiFi sensing applications into

detection, recognition, and estimation, and highlights three challenges for WiFi

sensing: robustness and generalization, privacy and security, and coexistence

with networking. It also presents three future trends for WiFi sensing, including

integrating cross-layer network information, multi-device cooperation, and fusion

of different sensors.

Device-Free WiFi Human Sensing: From Pattern-Based to Model-

Based Approaches

Dan Wu, Daqing Zhang, Chenren Xu, Hao Wang, and Xiang Li

Recently, device-free WiFi CSI-based human behavior recognition has gained

interest for its potential as a ubiquitous sensing solution. This article categorizes

existing work into pattern-based and model-based recognition solutions, reviews

their approaches and applications, and highlights the advantages of model-based

approaches using human respiration detection as a case study. The proposed

Fresnel zone model is considered a promising model for device-free human sensing

using fine-grained WiFi CSI. Some potential applications of device-free WiFi CSI-

based human behavior recognition include indoor localization, gesture control,

gait identification, fall detection, tracking, activity recognition, and vital signs

monitoring.

Wi-ESP—A tool for CSI-based Device-Free Wi-Fi Sensing (DFWS

Muhammad Atif, Shapna Muralidharan, Heedong Ko, Byounghyun Yoo

Recent progress in Device-Free Wi-Fi Sensing (DFWS) has established the use

of wireless signals like Wi-Fi not only to communicate but also as a tool to

5
enable device-free sensing. As an emerging technique, DFWS has many capable

applications such as sensing activity and gesture and fall detection, monitoring

elderly, surveillance, and many more applications while waiving out the necessity

to mount devices on the object. A wide range of applications can use the channel

state information (CSI) from commercial Wi-Fi devices pervasively for ubiquitous

sensing. Existing CSI tools, such as the Intel 5300 network interface controller

tool or the Atheros 9390 tool, have limitations when deployed in large-scale

systems due to their high deployment costs and limitations in the resolution

of CSI measurements. Due to these shortcomings, DFWS applications need an

alternative CSI tool in order to perform efficiently. In this paper, we present

ESP32-based Wi-ESP as a CSI gathering tool that can report detailed CSI

measurements based on 802.11n standards. The proposed Wi-ESP tool works as a

complete device by collecting the CSI measurements as well as processing further

for DFWS applications. Wi-ESP can work as standalone device, unlike other CSI

tools, and can offer large-scale deployment to many DFWS applications. In this

paper, we have explored the options of Wi-ESP as a tool for CSI measurements

and processing and propose it as a tool for DFWS.

6
Chapter 3

Related Works

3.1 RT FALL

RT-Fall is its real-time detection capability. The system can detect falls within

a few milliseconds of their occurrence, which is critical for providing timely

assistance to the fallen person. RT-Fall exploits the phase and amplitude of

the fine-grained Channel State Information (CSI) accessible in commodity WiFi

devices, and for the first time fulfils the goal of segmenting and detecting the falls

automatically in real-time, which allows users to perform daily activities naturally

and continuously without wearing any devices on the body. Moreover, RT-

Fall can distinguish between falls and other activities, such as sitting, standing,

walking, and lying down, which reduces false alarms and improves the accuracy

of the system.

RT-Fall is also designed to be robust to various environmental factors, such

as interference, multipath, and mobility. The system can adapt to changes in the

WiFi signal caused by moving objects, changing furniture, or other factors that

may affect the CSI. This makes RT-Fall more reliable and resilient than other fall

detection systems that are sensitive to environmental changes.

Finally, RT-Fall has the potential to be integrated with other smart home

7
technologies, such as voice assistants, smart lighting, and security systems.

By combining fall detection with other services, RT-Fall can provide a more

comprehensive and personalized solution for elderly individuals and others at

risk of falling. For example, the system can automatically turn on the lights

or call for help when a fall is detected, or provide reminders for medication or

appointments.

3.2 Deep Learning-Based Fall Detection Using

WiFi Channel State Information

The Deep Learning-Based Fall Detection Using WiFi Channel State Information

(CSI) involves three-step operations to detect falls: CSI data pre-processing, CSI

feature extraction, and event classification using classifiers based on support-

vector machine (SVM), convolutional neural network (CNN), or long short-term

memory (LSTM).

The first step is CSI data pre-processing, which involves down-sampling and

reshaping the raw CSI data to reduce the dimensionality of the data and make

it suitable for feature extraction. The down-sampling reduces the number of CSI

samples per second, while the reshaping converts the 1D CSI data into a 2D

matrix format that can be used as an input to the deep learning-based classifiers.

The second step is CSI feature extraction, which involves extracting relevant

features from the pre-processed CSI data. The study uses three types of features:

time-domain features, frequency-domain features, and time-frequency domain

features. The frequency-domain features include the power spectral density

(PSD) of the CSI amplitude and phase. The time-frequency domain features

include the wavelet transform of the CSI amplitude and phase.

The third step is event classification, which involves using deep learning-

based classifiers to classify the extracted features into fall or non-fall events. The

8
study uses three types of classifiers: support-vector machine (SVM), convolutional

neural network (CNN), and long short-term memory (LSTM). The SVM classifier

uses a linear kernel to separate the feature space into two classes: fall and non-fall.

The CNN classifier uses a deep neural network with convolutional and pooling

layers to learn the spatial and temporal features of the CSI data. The LSTM

classifier uses a recurrent neural network with long short-term memory cells to

learn the temporal dependencies of the CSI data.

The study evaluates the performance of the proposed system using a

comprehensive dataset consisting of over 700 CSI samples including different types

of falls and other daily activities, performed in four different indoor environments

on and off the dominant paths. The study shows that the proposed system

achieves high accuracy in detecting falls and outperforms existing fall detection

systems. The study also discusses the impact of the size and sample variety of

the training dataset on the performance of deep learning-based classification and

makes the dataset available in the public domain for further research.

3.3 WiFall

The system works by analyzing the variations in the radio signals caused by

human activities. The researchers first analyze the radio propagation model

to understand the correlations between different radio signal variations and

activities. Based on their observations, they propose WiFall as a truly unobtrusive

fall detection system. WiFall uses the physical layer CSI as an indicator of

activities. It captures the CSI information from the wireless signals and applies

machine learning algorithms to classify different activities and detect falls. The

system is implemented on desktops equipped with commodity 802.11n NICs.

To achieve fall detection, WiFall employs an anomaly detection algorithm

for motion detection. It then uses a one-class Support Vector Machine (SVM)

9
classifier and Random Forest algorithm to classify different human activities and

identify falls. The system is trained using a dataset of labeled activities.

The performance of WiFall is evaluated in three typical indoor scenarios with

different transmitter-receiver (Tx-Rx) layouts. The experimental results show

that WiFall can achieve high accuracy in fall detection for a single person. It

yields 90% detection precision with a false alarm rate of 15% on average using the

one-class SVM classifier, and 94% detection precision with a 13% false alarm rate

using the Random Forest algorithm.Overall, WiFall demonstrates the feasibility

and effectiveness of using wireless networks for fall detection. It offers a low-cost

and unobtrusive solution that can be widely deployed in real-life environments.

3.4 FallDeFi

The Feature Extraction module extracts relevant features from the spectrogram

to distinguish falls from other activities. It includes a CSI noise filtering technique

for feature extraction in high frequencies. A rigorous feature selection procedure

is used to extract features that can classify a wide range of falls, such as trips,

slips, loss of consciousness, and loss of balance.

The Fall Classification module uses a trained Support Vector Machine (SVM)

classifier to classify the extracted features and determine whether an event is a fall

or a non-fall. The classifier is trained on a dataset of falls and non-falls, and it is

designed to be accurate and resilient to variations in the monitored environment.

While FallDeFi improves on the current state of the art in fall detection, it

is still a research prototype and requires further improvements before it can be

deployed for real-life applications. The system’s performance can be affected by

factors such as the density of WiFi access points and the coverage of WiFi signals

in different locations. Future work includes optimizing the deployment of WiFi

access points, mitigating false positives and false negatives, and exploring the

10
fusion of data from other fall detection technologies.

Overall, FallDeFi is a low-cost, reliable, and environment-resilient fall de-

tection system that utilizes WiFi signals to monitor and classify falls in indoor

environments.

11
Chapter 4

Theoretical Overview

This chapter provides a theoretical overview of the project as well as specifications

for different hardware tools and technologies that are employed.

4.1 Hardware Specification

We use two ESP32 microcontrollers one as a transmitter, and one as a receiver.

The features of the device are given below:

• Processors

– CPU: Xtensa dual-core (or single-core) 32-bit LX6 microprocessor,

operating at 160 or 240 MHz and performing at up to 600 DMIPS

– Ultra-low power (ULP) co-processor

• Memory: 320 KiB RAM, 448 KiB ROM

• Wireless connectivity:

– Wi-Fi: 802.11 b/g/n

– Bluetooth: v4.2 BR/EDR and BLE (shares the radio with Wi-Fi)

• Peripheral interfaces:

12
– 34 × programmable GPIOs

– 12-bit SAR ADC up to 18 channels

– 2 × 8-bit DACs

– 10 × touch sensors (capacitive sensing GPIOs)

– 4 × SPI

– 2 × I²S interfaces

– 2 × I²C interfaces

– 3 × UART 4i) SD/SDIO/CE-ATA/MMC/eMMC host controller

– SDIO/SPI slave controller

– Ethernet MAC interface with dedicated DMA and planned IEEE 1588

Precision Time Protocol support.

– CAN bus 2.0 4m) Infrared remote controller (TX/RX, up to 8

channels)

– Motor PWM

– LED PWM (up to 16 channels)

– Hall effect sensor

– Ultra low power analog preamplifier Security: IEEE 802.11 stan-

dard security features all supported, including WPA, WPA2, WPA3

(depending on version) and WLAN Authentication and Privacy

Infrastructure (WAPI)

– Flash encryption

– 1024-bit OTP, up to 768-bit for customers

– Cryptographic hardware acceleration: AES, SHA-2, RSA, elliptic

curve cryptography (ECC), random number generator (RNG) Power

management:

13
 – Internal low-dropout regulator

– Individual power domain for RTC 4v) 5 µA deep sleep current

– Wake up from GPIO interrupt, timer, ADC measurements, capacitive

touch sensor interrupt

Figure 4.1: esp32 Microcontroller

Figure 4.2: esp32 Pin-out Diagram

The signal specification for esp32 microcontroller is given below:

• Bandwidth: 20 MHz

• Antenna: 1 RX and 1 TX

• Protocol: 802.11n

14
 • Modulation: OFDM (16 QAM)

• Subcarrier Number: 64

• Sampling Rate: 3.9 Hz

• Average RSSI: -77 dBm

• Guard Interval: 800 ns (MCS Index: 4)

• Technologies: MIMO, Frame Aggregation

The advantages of using esp32 are:

• esp32 can function as a stand-alone system or as a slave device to a

host MCU, eliminating communication stack overhead on the primary

application CPU

• Through its SPI / SDIO or I2C / UART interfaces, the esp32 may

communicate with other systems to provide Wi-Fi and Bluetooth capability.

• esp32 has a low-power processor designed for mobile devices, wearable

electronics, and IoT applications. It uses a combination of proprietary

software to achieve ultra-low power consumption.

• esp32 is highly-integrated with in-built antenna switches, RF balun,

power amplifier, low-noise receive amplifier, filters, and power management

modules.

In summary, esp32 is ideal for our project because is a power-efficient device

that is capable of using wifi communication, is easily integrable with other

systems, and has a fair amount of computing power

15
4.2 Wi-Fi

In this project, we exploited the capability of Wi-Fi technology to implement

fall detection. Wi-Fi, vastly used in high-speed internet access and wireless

communication, is a set of protocols governed by IEEE 802.11 standards .

IEEE 802.11 is part of the local area network (LAN) technical standards and

specifies the set of protocols for implementing wireless local area network (WLAN)

computer communication. These standards are maintained by the Institute of

Electrical and Electronics Engineers (IEEE). Though the first edition of these

standards was released in 1997, continuous development is being made and new

standards are coming with more capabilities to meet the ever-increasing demand

for high-speed wireless communication. The most notable standards of IEEE

802.11 are 802.11a, 802.11b, 802.11g, 802.11n, 802.11ac, and 802.11ax.

802.11a

This was the first standard to use the 5 GHz band for Wi-Fi which might seem

to be ahead of its time. But because of the higher frequency, its coverage areawas

much lower than the traditional 2.4 GHz band and suffered much from interference

problem. That is why 802.11a was not so popular compared to its 2.4 GHz

counterpart even though it had a higher data rate and went obsolete quickly.

But the main contribution of this standard was the introduction of Orthogonal

Frequency-Division Multiplexing (OFDM) which improved the data transmission

drastically. OFDM is based on the concept of orthogonal subcarriers with minimal

interference that makes it possible to cope with severe channel conditions without

complex equalization filters. OFDM is described in detail later in this section.

802.11a uses 52 subcarriers in OFDM, of which 48 subcarriers are used for data

transmission and the rest 4 subcarriers are used as pilot subcarriers.

802.11b

802.11b was the first widely accepted standard of Wi-Fi. Both 802.11a and

16
802.11b were released in 1999, with a major difference between them. Unlike

802.11a, 802.11b uses 2.4 GHz band. The 2.4 GHz band was not as crowded as

today and offered higher coverage and the capability to withstand interference.

These advantages made 802.11b popular despite having a much lower data rate

(up to 5.5 Mbit/s). This standard is still in use in some legacy devices.

802.11g

Introduced in 2003, 802.11g was a mixture of the previous two standards. It

operated in the 2.4 GHz band like 802.11b and utilizes the same OFDM-based

transmission scheme as 802.11a. This technical change gave a burst increase

in the data rate which could go up to 54 Mbit/s. 802.11g also uses a total of

52 subcarriers with a carrier separation of 0.3125 MHz. There are 14 partially

overlapping channels each of which has a separation of 20 or 25 MHz.

802.11n

This standard is also known as Wi-Fi Generation 4 (Wi-Fi 4). It includes several

new technologies that increased the capability of Wi-Fi further. The most notable

additions to this standard are:

• Multiple Input Multiple Output (MIMO)

• Frame aggregation

• WiFi Beamforming (Optional)

• 40 MHz channel bandwidth

• Security enhancement

MIMO technology is capable of conducting simultaneous data transmission

over multiple antennas. Frame aggregation allows sending two or more frames in

a single transmission. Beamforming improves the user experience by focusing the

Wi-Fi beams in the user’s direction. Thus these new features along with OFDM

increased the data rate from 72 Mbit/s to 600 Mvit/s. 802.11n has support for the

17
2.4 GHz band and optionally for the 5 GHz band. Most Wi-Fi-enabled devices

are still using this standard today.

Newer standards After 802.11n, a few major standards have come out that have

increased the data rate, reliability, and security further. 802.11ac (Wi-Fi 5) is

currently spreading in the consumer community which uses only the 5 GHz band.

It introduced a few new features, such as Multi-User MIMO (MU-MIMO), wider

80 MHz and 160 MHz channels, and Beamforming. 802.11ax (Wi-Fi 6E) is the

most recently approved standard adopted in 202 which uses three bands: 2.4

GHz, 5 GHz, and 6 GHz. The data rate can vary from 600 Mbit/s to 9608

Mbit/s. Currently, the development is being made for 802.11be standard or Wi-

Fi 7 which will provide even more data rate.The hardware used in our proposed

method, esp32 uses the popular IEEE 802.11n standard. It currently has the

largest user base and can utilize several recent technologies including OFDM,

MIMO, and frame aggregation.

OFDM

Orthogonal Frequency-Division Multiplexing (OFDM) is a sort of digital trans-

mission and a way of encoding digital data on multiple carrier frequencies

that is used in telecommunications. OFDM is a widely used wideband digital

communication technique, with applications including digital television and audio

broadcasting, DSL internet access, wireless networks, power line networks, and

4G/5G mobile communications. The capacity of OFDM to cope with severe

channel conditions without the use of sophisticated equalization filters is its

fundamental benefit over single-carrier methods. Because OFDM uses numerous

slowly modulated narrowband signals rather than a single rapidly modulated

wideband signal, channel equalization is simpler. This mechanism also makes

it easier to design single frequency networks (SFNs), in which multiple adjacent

transmitters send the same signal at the same frequency at the same time, because

the signals from multiple distant transmitters can be constructively recombined,

18
avoiding the interference that a traditional single-carrier system would face.

Figure 4.3: Block Diagram of a simplified OFDM System

MIMO

Multiple-Input MIMO (Many-Input, Multiple-Output) is a wireless technology

that uses multiple transmitters and receivers to carry more data at once. MIMO

is supported by all 802.11n wireless equipment. The technology enables 802.11n

to achieve faster rates than goods that do not have 802.11n. MIMO must

be supported by the station (mobile device) or the access point (AP) to be

implemented. Both the station and the access point must support MIMO for

the best performance and range. Multipath, a natural radio-wave phenomenon,

is used in MIMO technology. Multipath occurs when transmitted data bounces

off walls, ceilings, and other obstacles, arriving at the receiving antenna numerous

times at slightly varying angles and times. Multipath created interference and

hindered wireless communications in the past. MIMO technology with multipath

combines numerous, smart transmitters and receivers with an extra spatial

dimension to improve performance and range. By allowing antennas to mix data

streams arriving from diverse paths and at different times, MIMO boosts the

signal-capturing power of receivers. Smart antennas make use of spatial diversity

technology, which makes use of unused antennas. When the number of antennas

19
outnumbers the number of spatial streams, the antennas can boost receiver variety

and range.

Figure 4.4: Block Diagram of MIMO System

4.3 Signals used for analysis

CSI

IEEE 802.11 a/g/n/ac/ax networks, data transmission and reception is done us-

ing OFDM. As discussed earlier, OFDM uses a number of orthogonal subcarriers

to transmit data in multiple spatial paths. While a transmitting packet is in the

medium, it is subjected to different obstructions, such as, fading, scattering and

power loss. As the subcarriers follow different spatial paths, these obstructions

affect each subcarrier differently. Thus, this physical layer information specific to

each subcarrier is known as Channel State Information (CSI). CSI is an overall

depiction of the channel state that includes scattering, fading, and multipath

effects in the signal’s propagation. In contrast to the received power strength

provided by RSSI, CSI statistics provide more information about the channel

degradation effects that the signal suffers due to its granularity of sub-carrier

frequencies and vector representation. Data is sent using MIMO and OFDM

20
systems. In narrow-band flat fading channel.

4.4 Machine Learning

In the 1950s, a branch of artificial intelligence known as machine learning was

discovered and developed. The earliest machine learning techniques date back to

the 1950s, however there have been very few notable studies and advancements

in this field. However, this field of study underwent a resurgence in the 1990s and

has continued to this day. Future advancements in this field of study are expected.

The complexity of analysing and interpreting the data, which is continually

expanding, is what has led to this development. The foundation of machine

learning is the idea that, with the help of this growing data, the best model for the

new data may be found among the old data. As a result, research into machine

learning will continue along with the growth in data. The actions performed

by computers, which are based on an algorithm and follow specific procedures,

have no margin for error. In some circumstances, computers make judgments

based on the current sample data, which is different from commands that are

created to produce an outcome depending on an input. In some circumstances,

computers may err in their decision-making just like people do. Putting it another

way, machine learning is the process of giving computers the capacity to learn

from data and experience just like a human brain. The primary goal of machine

learning is to develop models that can learn from previous data to become better,

recognize complicated patterns, and find answers to new problems.

21
Chapter 5

Methodology

In this chapter, we will discuss the implementation of our proposed system. At

first, we will describe the hardware setup used for collecting user data. Later, we

discuss how we intend to process the raw data to make them usable.

Figure 5.1: Block Diagram

5.1 Hardware Setup

The main hardware we used for this system is a pair of ESP32 MCU manufactured

by espressif. This is an ESP32 based module with Wi-Fi 802.11 b/g/n and

Bluetooth LE 4.2 connectivity and a dual-core processor. Traditional research on

22
Wi-Fi-based human activity recognition uses Intel 5300 or Atheros 9390 Network

Interface Card (NIC) of a laptop computer [7, 9, 12, 33, 34] which is not a realistic

choice for practical use because each node is a computer. The device we used is

small, low-cost, programmable, and deployment-friendly and the whole system

needs only one computer to process the data.

For our experiment, we need to send CSI data from one ESP32 device to

another. But ESP32 does not transmit CSI data with the initially provided

firmware. So, a customised firmware by Steven M Hernandez, “ESP32 CSI

Toolkit” is flashed to the devices to enable the transmission of CSI data. CSI

data is received using the broadcast method. In this method, one ESP32 device

acts as a transmitting device and all other devices are receiving device. The

transmitting ESP32 A sends CSI data using broadcasting. This method has the

highest detection accuracy and reliability and does not require any router device.

Figure 5.2: ESP32 using Broadcast Method

Each ESP32 module has one built-in PCB antenna that can transmit or receive

data. We used two ESP32 modules. Because of the movement of the subject,

the transmitting packets face multipath fading, scattering, reflection, and power

23
loss. The Channel State Information (CSI) of each packet can be analysed to find

patterns between the transmitting packets using machine learning algorithms and

thus recognize the activity performed by the subject.

5.2 Data Preprocessing

The data we collected includes the raw signal information and packets. We

need to extract the CSI data from the raw signal and isolate the phase and

amplitude information from it. As CSI data is inherently sensitive to different

environmental parameters, some cleaning process needs to be applied such as

calibration, denoising, and dimensionality reduction. There are various denoising

algorithms to choose from.

5.3 Dataset Preperation

The accuracy and effectiveness of any data-driven study depend much on a well

prepared dataset. But there are only a few open datasets available for activity

recognition using ESP32 CSI data. But these datasets do not have enough data

or provide the activities we need for this system. Hence, we need to prepare our

dataset using the hardware setup stated in the previous section.

5.4 Machine Learning

Machine learning models are trained on these datasets comprising CSI values

collected from the environment. The models learn to recognize distinct features

and variations associated with normal activities and falling incidents. These

features may include changes in signal strength, phase shifts, or frequency

alterations caused by a fall. Once trained, the machine learning model can classify

real-time CSI data, distinguishing between regular movements and potential fall

24
events.

Throughout the entire methodology, document each phase’s progress, chal-

lenges faced, and solutions implemented, ensuring a transparent and replicable

approach to the project.

25
Chapter 6

Results and Discussion

In Phase 1 of our fall detection project, we successfully implemented the collection

of Channel State Information (CSI) values using an ESP32 microcontroller. The

ESP32 was configured to receive data from a WiFi router, enabling us to capture

the intricate details of the wireless channel. “ESP32 CSI Toolkit” by S. M.

Hernandez was used in the ESP32 to retrieve data.

Figure 6.1: CSI Data Obtained

This is the sample CSI Data Received with the ESP32 placed roughly 1m

from the router in a room thats approximately 3.5m to 7m. The ESP 32

26
Toolkit provides CSI values for all 64 subcarriers. Only the Amplitude value

from Subcarrier 44 is plotted.

The collected CSI values provide a rich source of information about the wire-

less environment, including signal strength, phase, and frequency information.

These details are crucial for understanding the unique characteristics of the

wireless signals in the area where the fall detection device will be deployed.

Moving forward, the dataset obtained in Phase 1 will serve as a foundation for

developing algorithms and models to detect falls based on changes in the wireless

environment. The precise and real-time nature of CSI data holds promise for

creating an effective and non-intrusive fall detection system.

In summary, Phase 1 lays the groundwork for the fall detection device by

successfully collecting WiFi CSI values. This data will be instrumental in the

subsequent phases, where we will focus on algorithm development and integration

of the fall detection system.

27
Chapter 7

Conclusions

In the culmination of Phase 1 of our fall detection project, we have successfully

achieved the foundational objective of collecting WiFi Channel State Information

(CSI) using an ESP32 microcontroller. The ESP32 served as a reliable interface,

facilitating seamless communication with a WiFi router and enabling the capture

of intricate details within the wireless environment. The gathered CSI dataset

provides a rich source of information, encompassing signal strength, phase, and

frequency characteristics critical for understanding the unique dynamics of the

wireless signals in our target area. The establishment of a robust data collection

process sets the stage for subsequent phases, where the focus will shift to

leveraging machine learning techniques for fall detection. The promising nature

of WiFi CSI data, coupled with the adaptability of machine learning algorithms,

holds great potential for the creation of an effective, non-intrusive fall detection

system. As we transition into the next phases, the insights gained in Phase 1 will

guide the development of algorithms and models, propelling our project towards

the realization of a sophisticated and reliable fall detection device.

28
References

[1] H. Wang, D. Zhang, Y. Wang, J. Ma, Y. Wang and S. Li, ”RT-Fall: A

Real-Time and Contactless Fall Detection System with Commodity WiFi

Devices,” in IEEE Transactions on Mobile Computing, vol. 16, no. 2, pp.

511-526, 1 Feb. 2017, doi: 10.1109/TMC.2016.2557795

[2] Muhammad Atif, Shapna Muralidharan, Heedong Ko, Byounghyun Yoo,

Wi-ESP—A tool for CSI-based Device-Free Wi-Fi Sensing (DFWS), Jour-

nal of Computational Design and Engineering, Volume 7, Issue 5, October

2020, Pages 644–656, https://doi.org/10.1093/jcde/qwaa048

[3] WiFi Sensing with Channel State Information: A Survey YONGSEN

MA, GANG ZHOU, and SHUANGQUAN WANG, Computer Science

Department, College of William Mary, USA

[4] Y. Wang, K. Wu and L. M. Ni, ”WiFall: Device-Free Fall Detection by

Wireless Networks,” in IEEE Transactions on Mobile Computing, vol. 16,

no. 2, pp. 581-594, 1 Feb. 2017, doi: 10.1109/TMC.2016.2557792. .

[5] D. Wu, D. Zhang, C. Xu, H. Wang and X. Li, ”Device-Free WiFi Human

Sensing: From Pattern-Based to Model-Based Approaches,” in IEEE

Communications Magazine, vol. 55, no. 10, pp. 91-97, Oct. 2017, doi:

10.1109/MCOM.2017.1700143. .

[6] Y. Chu, K. Cumanan, S. K. Sankarpandi, S. Smith and O. A. Dobre, ”Deep

29
 Learning-Based Fall Detection Using WiFi Channel State Information,”

in IEEE Access, vol. 11, pp. 83763-83780, 2023, doi: 10.1109/AC-

CESS.2023.3300726. .

[7] Sameera Palipana, David Rojas, Piyush Agrawal, and Dirk Pesch. 2018.

FallDeFi: Ubiquitous Fall Detection using Commodity Wi-Fi Devices.

Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 1, 4, Article

155 (December 2017), 25 pages. https://doi.org/10.1145/3161183

30