Planned research · not yet implemented

FUTURE PROJECT D · FIBER-OPTIC SENSING

Intelligent Fiber-Optic Intrusion Detection System (iFIDS)

Detect, locate and classify physical disturbances along an optical fiber — and test, rather than assume, whether a lower-cost interrogator can do it.

iFIDS is planned research into distributed fiber-optic vibration sensing for perimeter and critical-infrastructure monitoring: Rayleigh backscattering, Distributed Acoustic Sensing (DAS), signal processing, event localisation along the fiber, machine-learning classification and an operator-oriented monitoring interface.

It applies a pattern familiar from radar and C4I work to a new sensor: extract signals from a physical sensor, estimate what is happening, classify events, and present actionable, honestly qualified information to an operator.

Understand the Physics The OTDR Research Question All Future Projects

Concept illustration: an optical interrogator sends a probe pulse into a fiber along a perimeter fence; Rayleigh backscatter returns from every point; a disturbance at an unknown distance appears in a distance-versus-time waterfall view as a bounded candidate event, next to a persistent stripe from machinery. — Concept illustration Distance from round-trip time, and how a localised event would appear in a waterfall view. Illustrative pattern, not measured data.

Status: Planned research; no interrogator, dataset or classifier has been built or tested
Builds on: Sensor integration, tracking & estimation, real-time C4I monitoring
First milestone: Software prototype on recorded DAS data: detect, locate, display
Key open question: Can a cost-optimised interrogator meet the dynamic-sensing requirements?

Problem & motivation

A Fiber Can Be Thousands of Sensors

Long perimeters — airports, industrial sites, pipelines, cable routes — are expensive to monitor with discrete sensors. Many already have optical fiber running along them. With a suitable interrogator, a single fiber can act as a continuous line of vibration sensors, reporting where along its length something is happening.

Commercial DAS interrogators exist but are typically costly, which limits adoption. iFIDS asks two questions in a disciplined order: first, can software reliably detect, locate and classify events from genuine distributed-sensing data; and second, can a lower-cost optical architecture provide data good enough for that software — measured against a DAS-capable reference.

Stage 1 · research & software prototypeAnalyse distributed-sensing data, detect disturbances, estimate fiber distance, classify events and present them to an operator.

Stage 2 · real fiber demonstratorAcquire vibration measurements with suitable optical hardware and validate the algorithms against controlled real events.

Research objectives

Understand Rayleigh backscattering and distributed acoustic sensing.
Investigate cost-effective interrogation architectures, including the limits of OTDR-based approaches.
Build a signal-processing pipeline to detect vibration events and estimate their fiber distance.
Investigate ML models for classifying events from measured (and clearly labelled simulated) features.
Develop a real-time interface: event location, class, confidence and alarm history.
Evaluate detection, false alarms, localisation error, latency and cost per monitored kilometre.

Sensing physics

Rayleigh Backscatter, OTDR and DAS

Understanding exactly what each instrument measures is the foundation of the project — and the reason the low-cost question cannot be answered by assumption.

A · Rayleigh backscattering
Light returns from everywhere
Microscopic refractive-index variations frozen into the glass scatter a tiny fraction of the transmitted light backward at every point. The return carries information about the fiber and its optical path.
B · Disturbance sensing
Strain changes the return
Vibration or strain alters the returned optical signal. In coherent DAS the interrogator analyses changes in the backscattered optical field — often its phase — to recover dynamic strain along the fiber.
C · OTDR vs. DAS
Different instruments, different jobs
A conventional OTDR measures return power versus distance to characterise attenuation, splices, connectors and breaks. A high-power pulse and a trace do not by themselves give acoustic sensitivity.

Concept diagram Schematic shapes only; no instrument data. Scroll sideways on narrow screens.

Three approaches to compare

The reference system validates the algorithms; the low-cost path is measured against it.
Approach	Role in the project	Main consideration
Conventional OTDR	Baseline for fiber characterisation and fault localisation	Usually not sufficient by itself for high-performance dynamic DAS
DAS-capable coherent interrogator	Reference system for acquiring distributed vibration data	Best reference for validating sensing and classification, but typically more expensive
Research-grade or custom low-cost interrogator	Experimental cost-reduction path	Must demonstrate dynamic sensitivity, range, repeatability and safe optical operation

Proposed architecture

Distributed Sensing, Centralised Decision Support

Optical acquisition, signal processing, classification and the operator interface are kept as separate modules, so the interrogator can change without rewriting the software.

iFIDS layered architecture

Proposed design

Physical sensing layer
Optical fiber installed along a monitored perimeter or infrastructure corridor.
Optical interrogation unit
- Laser source
- Optical receiver
- Timing
- Backscatter acquisition
DAS-capable reference instrument, or the experimental low-cost design under test.
Signal processing & event detection
- Preprocessing
- Noise suppression
- Change detection
- Event segmentation
- Distance estimation
ML classification & event location

Classification
Features, classifier, confidence, unknown-event handling.
Location
Fiber distance, zone, event timeline.
Monitoring, operator review & logging
- Alarm
- Event type & confidence
- Diagnostics
- Acknowledge
- Review evidence
- Record outcome

What happens when a vibration is detected

Deviation from background
The signal departs from the current, condition-specific background model.
Locate along the fiber
Estimate fiber distance; map it to a physical location only if the route has been surveyed.
Window & features
Extract a time window and its time, frequency and spatial features.
Classify — or say "unknown"
Assign a trained category with a confidence, or report the event as unknown.
Present & store
Show location, class, confidence and the supporting signal; keep evidence for review.

Not every disturbance is an intrusion. Wind, rain, vehicles, machinery and maintenance can produce similar signals. The system must report uncertainty and never present an unverified classification as fact.

Signal-processing pipeline

From Raw Backscatter to a Reviewed Event

Each stage is independently testable and the original measurements are retained, so every alarm can be traced back to the data that caused it.

Acquire & validate
Measurements, timestamps, instrument status and distance metadata; detect missing samples, invalid values and clock problems.
Establish a baseline
Model the normal background under different conditions — never a single static baseline for every hour and weather.
Condition the signal
Filtering, detrending, normalisation and noise suppression chosen from the actual instrument output and expected bandwidth.
Detect candidates
Thresholds, rolling statistics or change-point detection find regions and intervals that differ from background.
Estimate location
Map detected position to fiber distance, accounting for calibration and the instrument's spatial response.
Features & classification
Time-domain, frequency-domain and time-frequency features into a trained model with confidence.
Validate & record
Attach confidence, preserve a source-data segment and record whether an operator confirmed the event.
Health monitoring
Separate a sensing alarm from an instrument fault, fiber break or communications failure.

Candidate features

Candidate inputs, not guaranteed discriminators — usefulness must be shown on representative data from the selected interrogator and installation.
Feature family	Example measurements	Potential use
Time domain	RMS energy, peak amplitude, variance, crest factor	Distinguish stronger or more impulsive disturbances
Frequency domain	Spectral energy, dominant frequencies, band-energy ratios	Characterise different vibration patterns
Time-frequency	STFT, spectrogram features	Events whose frequency content changes over time
Spatial	Event extent, duration, energy across adjacent channels	Separate localised and extended disturbances
Temporal	Repetition rate, duration, periodicity	Distinguish recurring machinery from transient events

Machine-learning sequence: simple first

Model 1
Rule-based baseline
Thresholds and signal statistics. A transparent baseline that also reveals nuisance-alarm sources.
Model 2
Classical ML
Random Forest or SVM on extracted features: do the features separate the labelled classes?
Model 3
Time-frequency learning
A compact CNN on spectrograms or signal windows, used only if it improves generalisation.
Model 4
Confidence & unknowns
Calibration, out-of-distribution checks and a reject class, so unfamiliar events are not forced into known categories.

Dataset discipline. Overlapping windows from the same recording must never be split randomly across training and test sets — that inflates results. Separate sessions, locations or installation conditions are reserved for testing. Every recording keeps its label, location, time, installation, sensor settings, environment and procedure. Candidate classes: background noise, controlled footstep-like disturbances, vehicle or machinery vibration, controlled digging-like vibration in an authorised area, and maintenance activity.

Mathematical foundations

Ranging, Resolution, Bandwidth and Detection

These standard relations define what any interrogator — commercial or low-cost — must achieve. Worked numbers are arithmetic illustrations using typical silica-fiber values, not instrument specifications.

Localisation

Distance from round-trip time

z = \frac{c t}{2 n_{g}}

$z$: Distance along the fiber to the scattering point.
$t$: Time between launching the pulse and receiving that backscatter.
$c$: Speed of light in vacuum (≈ 3 × 10⁸ m/s).
$n_{g}$: Group index of the fiber (≈ 1.468 for standard single-mode); the factor 2 is the round trip.

This gives fiber distance. A map position requires a surveyed fiber route, including slack loops and cable routing.

Spatial resolution

Pulse width sets how close two events can be

Δ z \approx \frac{c τ}{2 n_{g}}

$τ$: Probe pulse duration. Shorter pulses resolve closer events but carry less energy, reducing signal-to-noise ratio.

Arithmetic example: $τ = 100$ ns gives $Δ z \approx$ 3×10⁸ × 10⁻⁷ / (2 × 1.468) ≈ 10.2 m.

Temporal bandwidth

Fiber length limits the sampling rate

f_{rep} \leq \frac{c}{2 n_{g} L}

f_{\max} \leq \frac{f_{rep}}{2}

$L$: Sensing fiber length: the next pulse can be launched only after backscatter from the far end has returned.
$f_{rep}$: Pulse repetition rate = sampling rate of every position along the fiber.
$f_{\max}$: Highest vibration frequency recoverable without aliasing (Nyquist).

Arithmetic example: for $L = 10$ km, $f_{rep} \leq$ ≈ 10.2 kHz, so vibrations up to ≈ 5.1 kHz are observable in principle.

Why a standard OTDR is not automatically DAS

Averaging trades dynamics for noise

P (z) \propto P_{0} e^{- 2 α z}

T_{trace} = \frac{N_{avg}}{f_{rep}}

$α$: Fiber attenuation coefficient: the steady downward slope of an OTDR trace.
$N_{avg}$: Number of pulse returns averaged to raise signal-to-noise for a usable power trace.

Arithmetic example: averaging 10,000 returns at 10 kHz takes ≈ 1 s per trace, so any vibration faster than about 1 Hz is averaged out. Incoherent power detection is also far less sensitive to tiny strains than coherent phase detection. Whether a modified OTDR-class design can avoid both limits is exactly the research question below.

Coherent DAS (φ-OTDR)

Phase change from dynamic strain

Δ φ \approx \frac{4 π n ξ L_{G}}{λ} ε

$ε$: Axial strain averaged over the gauge length.
$L_{G}$: Gauge length — the fiber section over which phase difference is measured.
$ξ$: Photo-elastic scaling factor (≈ 0.78 for silica fiber).
$λ, n$: Optical wavelength (e.g. 1550 nm) and fiber refractive index.

The linear phase–strain relation is why coherent interrogators can measure very small vibrations — and why laser coherence and phase stability become hardware requirements.

Detection & features

Adaptive background test with persistence

s (z, t) = \frac{e (z, t) - μ (z, t)}{σ (z, t)}

alarm if s > γ for \geq T_{p} windows

RMS = \sqrt{\frac{1}{N} \sum_{n = 1}^{N} x_{n}^{2}}

CF = \frac{{max}_{n} | x_{n} |}{RMS}

$e$: Band-limited energy at fiber position $z$ and time window $t$ .
$μ, σ$: Rolling per-position background mean and spread, updated so that weather and traffic shifts do not trigger alarms.
$γ, T_{p}$: Threshold and persistence; tuned on recorded data to trade detection probability against false alarms.

Proposed hardware & software

Laboratory Demonstrator First

A short, controlled fiber installation — no commercial perimeter system needs to be bought to begin the software.

01 · Fiber test section
Known length of single-mode fiber
Connectors and a documented route; bench first, outdoor section only once acquisition is validated.
02 · Optical acquisition
OTDR + DAS-capable reference
OTDR for baseline fiber characterisation; a DAS-capable instrument or a research partner's system for dynamic strain data.
03 · Controlled vibration source
Shaker / vibration platform
Safe, repeatable disturbances at known positions and times — the labelled ground truth.
04 · Processing computer
Existing Linux environment
Python signal processing, ML experiments, database and dashboard. No GPU needed for the first classical models.

Questions to resolve before buying equipment

Does the instrument provide raw time series, complex optical data, phase — or only a conventional OTDR trace?
Can it acquire repeated measurements fast enough for the target vibration bandwidth?
What are its range, spatial resolution, minimum detectable strain and noise floor?
Can it say where a disturbance occurred, not just that the total trace changed?
Can it export timestamped measurements for independent processing?
What optical power and laser-safety requirements apply?

Laser safety. Any experimental optical transmitter requires rated components and appropriate laser-safety procedures; high-power optical pulses can be hazardous.

Proposed software

ifids/
├── acquisition/        data_import.py · instrument_interface.py
├── signal_processing/  preprocessing.py · event_detection.py · localization.py
├── machine_learning/   feature_extraction.py · train_model.py · event_classifier.py
├── monitoring/         event_manager.py · alarm_rules.py
├── dashboard/
├── database/
├── datasets/
├── tests/
├── results/
└── README.md

High-rate raw samples go to files (HDF5); the database holds events and metadata.
Component	Starting point
Operating system	Ubuntu Linux
Signal processing	Python, NumPy, SciPy
Visualisation	Matplotlib and PyQt5, or a web dashboard
Machine learning	scikit-learn; PyTorch only if a CNN is justified
Data storage	CSV/HDF5 for experiments; SQLite first, MariaDB or PostgreSQL for multi-user
Event interface	REST API or message-based when required
Evaluation	Reproducible scripts, confusion matrices, latency and false-alarm reports

The central research question

Can a Cost-Optimised Interrogator Do the Job?

Research objective: evaluate the feasibility of a cost-optimised optical interrogation architecture for detecting and classifying distributed vibration events, using a DAS-capable system as the reference.

A conventional OTDR may remain valuable for fiber diagnostics even if it cannot meet the dynamic-sensing requirements; a modified or alternative architecture must be designed and tested, not assumed to work.
Evaluation dimension	Question
Dynamic sensitivity	Can the low-cost architecture reliably detect the relevant disturbances?
Spatial resolution	Can it distinguish events at different positions?
Temporal bandwidth	Can it capture the time-varying signals needed for discrimination?
Repeatability	Does the same controlled disturbance produce consistent measurements?
Detection performance	Does detection remain reliable across test conditions?
Cost	Equipment, installation, maintenance and computing cost per monitored kilometre?

Strategy

First obtain DAS-capable data — through a research partner, a commercial system or a documented dataset — and develop the software against it. Then evaluate independently whether a lower-cost optical implementation can reproduce the performance the application needs. This validates the processing and classification pipeline without investing prematurely in an interrogator that may not meet the sensing requirements.

Engineering challenges

Nuisance alarms from wind, rain, traffic and machinery.
Environmental drift of the background over hours and seasons.
Labelled data scarcity for rare real events.
Route mapping from optical distance to physical location.
Data volume of continuous high-rate acquisition.
Cybersecurity of configuration changes and event logs.

Experimental validation & performance metrics

Reporting What Matters Operationally

For critical infrastructure, false alarms are operationally important: a model with high overall accuracy can still be unsuitable if it generates frequent nuisance alarms or misses rare events.

Planned metrics; simulated and real-optical results will always be reported separately.
Metric	Definition / purpose
Precision, recall, F1 per class	Classification quality per event type, not one blended accuracy
False-alarm rate	False alarms per hour, or per monitored km per hour
Detection probability	Share of controlled events detected under defined test conditions
Localisation error	Estimated minus true fiber position, reported as a distribution
Latency	Time from disturbance to detection and to classification
Generalisation	Performance across environmental conditions and separate test sessions
Cost per km	Equipment, installation, maintenance and compute per monitored kilometre

Deployment requirements (later stage)

Fiber-route mapping — surveyed route translating optical distance into physical location.
Environmental adaptation — background models for weather, traffic and normal site activity.
Alarm management — group repeated detections into events, suppress duplicates carefully, preserve evidence.
Health monitoring — distinguish sensing alarms from instrument faults, breaks and communication failures.
Cybersecurity — operator authentication, protected configuration, restricted access, auditable logs.
Human oversight — classifications stay decision support until validated for the actual site.

Phased roadmap

About Six Months Part-Time — Hardware Access Permitting

The timeline depends heavily on access to suitable optical hardware and labelled measurements. If hardware access is delayed, software work continues on documented recorded datasets.

01Weeks 1–3
Research & requirements
- Document DAS and OTDR principles
- Compare interrogation approaches
- Define fiber length and event requirements
- Identify a source of raw DAS data
Deliverable: requirements and technical feasibility assessment.
02Weeks 4–7
Offline signal-processing prototype
- Import recorded measurements
- Traces and spectrograms
- Candidate-event detection
- Distance estimation where supported
Deliverable: Python analysis application on recorded data.
03Weeks 8–11
Labelled dataset & ML baseline
- Controlled vibration tests or documented data
- Rule-based detector
- One classical ML classifier
Deliverable: dataset description, baseline model, performance report.
04Weeks 12–16
Optical hardware validation
- DAS-capable instrument or lab partner
- Sensitivity, spatial response, repeatability
- Optical measurement vs. applied disturbance
Deliverable: real-data validation and hardware feasibility report.
05Weeks 17–20
Monitoring interface
- Fiber route, event distance, class, confidence
- Waveform / spectrogram view
- Alarm history, data-quality and health
Deliverable: integrated iFIDS monitoring prototype.
06Weeks 21–24
Controlled field evaluation
- Documented fiber route
- Authorised test events + normal activity
- Detection, false alarms, location error, latency
Deliverable: demo video, results, limitations, next-stage plan.

First milestone must show four things: genuine sensing data is processed (not only simulated waveforms); controlled events are detected against background activity; false alarms and missed events are measured, not just accuracy; and there is a credible path to testing the proposed low-cost acquisition architecture.

Expected contribution

Expected Contribution & Research Direction

The potential differentiator is the combination of cost-conscious sensing research, signal-processing expertise, intelligent event classification and real-time C2-style monitoring — supported by measured results rather than claims of field readiness. A scalable design could later monitor multiple fiber segments and feed a central security or command-and-control interface, connecting naturally with my experience in real-time tracking, sensor integration and operator-oriented C2 systems.

Potential applications

Airport and facility perimeters
Critical infrastructure and industrial installations
Pipeline and cable-route corridors
Other environments needing distributed disturbance monitoring

Proposed project lead

Mohammed Ismail Ali Attala

Senior Systems Engineer · C4I/C2 Integration · Tracking & Navigation

Seeking research partners with DAS-capable instruments, datasets or authorised test sites.

Contact & Collaboration

Working with Fiber Sensing or Perimeter Security?

I welcome discussions with research groups, instrument suppliers and infrastructure operators about distributed sensing, signal processing and event classification.

Send an Email LinkedIn Back to Future Projects

Intelligent Fiber-Optic Intrusion Detection System (iFIDS)

A Fiber Can Be Thousands of Sensors

Research objectives

Rayleigh Backscatter, OTDR and DAS

Light returns from everywhere

Strain changes the return

Different instruments, different jobs

Three approaches to compare

Distributed Sensing, Centralised Decision Support

iFIDS layered architecture

Physical sensing layer

Optical interrogation unit

Signal processing & event detection

ML classification & event location

Classification

Location

Monitoring, operator review & logging

What happens when a vibration is detected

Deviation from background

Locate along the fiber

Window & features

Classify — or say "unknown"

Present & store

From Raw Backscatter to a Reviewed Event

Acquire & validate

Establish a baseline

Condition the signal

Detect candidates

Estimate location

Features & classification

Validate & record

Health monitoring

Candidate features

Machine-learning sequence: simple first

Rule-based baseline

Classical ML

Time-frequency learning

Confidence & unknowns

Ranging, Resolution, Bandwidth and Detection

Distance from round-trip time

Pulse width sets how close two events can be

Fiber length limits the sampling rate

Averaging trades dynamics for noise

Phase change from dynamic strain

Adaptive background test with persistence

Laboratory Demonstrator First

Known length of single-mode fiber

OTDR + DAS-capable reference

Shaker / vibration platform

Existing Linux environment

Questions to resolve before buying equipment

Proposed software

Can a Cost-Optimised Interrogator Do the Job?

Strategy

Engineering challenges

Reporting What Matters Operationally

Deployment requirements (later stage)

About Six Months Part-Time — Hardware Access Permitting

Research & requirements

Offline signal-processing prototype

Labelled dataset & ML baseline

Optical hardware validation

Monitoring interface

Controlled field evaluation

Expected Contribution & Research Direction

Potential applications

Mohammed Ismail Ali Attala

Working with Fiber Sensing or Perimeter Security?