Labs CafeTele ACQUIRING · 193.1 THz · C-BAND
Optical Communication Professional · 4-Day Hands-On

Build, break & diagnose optical networks.

From total internal reflection to 800G coherent ZR pluggables — every concept in the syllabus is a live, physics-accurate simulator. Plan DWDM grids, size PON power budgets, hunt faults on an OTDR trace, and predict QoT with ML. Runs entirely in your browser.

4DAYS
16LIVE TOOLS
ITU-TACCURATE
16MISSIONS
Mission progress
0%

Fiber Optics Fundamentals

How light is trapped in glass, the fibers that carry it, and the three impairments that limit every link: attenuation, dispersion and nonlinearity.

Introduction to Optical Communication

basics · advantages over copper · the optical spectrum
Optical communication sends information as pulses of light down a hair-thin glass fiber instead of electrons down copper. Light’s enormous carrier frequency (~193 THz) gives a single fiber tens of terabits per second over thousands of kilometres, with far lower loss (~0.2 dB/km vs many dB per metre in copper), immunity to electromagnetic interference, security and light weight. Everything below is live — drag, hover and watch.
0
carrier frequency
0
fiber loss @1550nm
0
one fiber (96×400G)
0
reach (amplified)
Live from the real optical engines REAL
These figures are computed right now on the CafeTele server by GNPy (optical link) and OptiCommPy (bit-error rate) — not hard-coded numbers.
…
GSNR · GNPy
…
OSNR · GNPy
…
CD ps/nm · GNPy
…
Latency ms · GNPy
…
BER · OptiCommPy
How a photon link carries your bits
A 1 0 1 1… bit stream becomes light pulses, zig-zags down the core by total internal reflection, and lands on the photodiode. Each glowing packet is one ‘1’ bit.
The optical link, stage by stage — click any stage
Click a stage above to see exactly what it does.
Copper vs Fiber — signal survival
—
Copper Rx power
—
Fiber Rx power
—
Copper repeaters needed
Coax at GHz loses ~80 dB/km; fiber only ~0.2 dB/km. Drag the distance and watch copper pulses die out while the fiber pulses sail on.
Why 1550 nm? The fiber loss spectrum
Hover the curve. Silica is clearest in the C-band (1530–1565 nm) — the global long-haul window where loss bottoms at ~0.2 dB/km and the EDFA amplifies.

Total Internal Reflection & Numerical Aperture

Snell's law · critical angle · acceptance cone
Light is trapped in the fiber core by total internal reflection. Where the higher-index core (n₁) meets the lower-index cladding (n₂), any ray striking the boundary beyond the critical angle θc = arcsin(n₂/n₁) reflects perfectly instead of leaking out, so it zig-zags along the core. The numerical aperture NA = √(n₁²−n₂²) sets the acceptance cone — how steeply light can enter and still be guided.
θc = arcsin(n₂/n₁)NA = √(n₁²−n₂²)acceptance conestep-index core
—
Critical angle θc
—
Numerical aperture
—
Acceptance angle
—
Guided?
Δ (relative index): —. Light is guided only when the ray hits the core–cladding boundary at an angle greater than θc. Increase the launch angle until the ray escapes (refracts out) to see the acceptance limit.

SMF vs MMF & ITU-T Fiber Types

G.652 · G.655 · G.657 · step vs graded index
Single-mode fiber (SMF, ~9 µm core) carries exactly one spatial mode, removing modal dispersion and enabling long-haul DWDM. Multimode fiber (MMF, 50 µm graded core) carries hundreds of modes for cheap short data-center links. ITU-T standardises the key types: G.652 (standard SMF), G.655 (non-zero dispersion-shifted for DWDM) and G.657 (bend-insensitive for FTTH).
SMF ~9 µmMMF 50 µmG.652 / G.655 / G.657graded vs step
G.652.D (SMF) G.655 (NZDSF) G.657.A2 (Bend) OM4 (MMF)
ParameterSMF (G.652)MMF (OM4)
Core diameter~9 µm50 µm
Modes1 (single)Hundreds
Index profileStepGraded
Loss @1550nm0.20 dB/km~0.5 dB/km
Bandwidth-distance~Tbps·km4700 MHz·km
SourceLaser (DFB)VCSEL 850nm
Typical reach10s–1000s km≤ 400 m (DC)
Cost / useLong-haul, metro, FTTHData-center, LAN

Optical Spectrum & Bands

O · E · S · C · L · U — why the C-band rules long-haul
Silica fiber’s lowest loss sits in the infrared, divided into bands: O (1260–1360), E, S, C (1530–1565), L (1565–1625) and U. The C-band dominates long-haul because it combines the absolute loss minimum (~0.2 dB/km) with the gain window of the erbium amplifier (EDFA). The L-band adds capacity; the O-band serves short reach and PON.
O/E/S/C/L/UC-band ~0.2 dB/kmEDFA gain window1310 & 1550 nm
O 1260–1360 E 1360–1460 S 1460–1530 C 1530–1565 L 1565–1625 U 1625–1675

Fiber Impairment Calculator

Attenuation · chromatic dispersion · PMD · pulse broadening
Three impairments limit every link. Attenuation shrinks the signal (~0.2 dB/km at 1550 nm). Chromatic dispersion spreads a pulse because wavelengths travel at slightly different speeds (Δτ = D·L·Δλ), eventually causing inter-symbol interference. PMD adds random spreading that grows with √length, and nonlinear effects (SPM, XPM, FWM) distort high-power signals. Drive the calculator until dispersion closes the eye.
α ≈ 0.2 dB/kmΔτ = D·L·ΔλPMD ∝ √LSPM / XPM / FWM
Real CD / PMD / GSNR · GNPy REAL
GNPy computes the actual chromatic dispersion, PMD and GSNR accumulated over a real amplified fiber link — the same engine operators use.
…
CD ps/nm
…
PMD ps
…
GSNR
—
Total attenuation
—
Chromatic disp.
—
PMD (mean)
—
CD-limited reach
—
Top: launched pulse. Bottom: pulse after the span — chromatic dispersion broadens it. When broadening exceeds ~1 bit period, adjacent bits overlap (ISI) and the link fails. Watch the eye close as you raise length or bit rate.

Optical Components

The transmit/receive chain: sources, modulators, detectors and the amplifiers and passive components that build a real link.

Optical Sources

LED · Fabry-Pérot · DFB · tunable laser
The transmitter turns bits into light. LEDs are cheap and broad-spectrum (short MMF only). Laser diodes are coherent and narrow; the DFB laser uses an internal Bragg grating to emit one stable wavelength — the workhorse of DWDM. Tunable lasers sweep the whole C-band so a single part can serve any channel, essential for coherent transceivers.
LED broadDFB single-λtunable C-bandlow chirp
LED FP Laser DFB Laser Tunable
Optical spectrum of the source. A narrow line (DFB / tunable) means low chromatic-dispersion penalty and tight DWDM channel spacing; a broad LED spectrum spreads across many nm and disperses badly.

Modulation Formats — Live Constellation & BER

OOK · PAM4 · QPSK · 16-QAM · 64-QAM · Monte-Carlo over AWGN
Modulation maps bits onto the carrier’s amplitude and phase. OOK sends 1 bit/symbol, PAM4 2, QPSK 2 (phase), 16/64-QAM 4/6 bits/symbol. Higher orders carry more data per symbol but need far higher OSNR — the constellation clouds merge as noise rises. This tool runs a real Monte-Carlo BER in the browser and, on demand, on the OptiCommPy engine on the server.
bits/symbolOSNR wallFEC limit 3.8e-3real OptiCommPy
Constellation Analyzer REAL DSP
Acquire the real received IQ from the OptiCommPy engine — the constellation a coherent receiver actually sees, with measured EVM and BER.
…
EVM %
…
BER
…
RX symbols
Click acquire to run OptiCommPy.
—
Spectral eff.
—
BER (measured)
—
BER (theory)
—
OSNR @ FEC limit
—
Higher-order formats pack more bits/symbol (more capacity) but need far higher OSNR. Drag the slider down and watch the constellation clouds merge — that's the SNR wall every coherent link fights. FEC limit shown for BER 3.8×10⁻³ (7% HD-FEC).

Optical Receivers & OSNR → BER

PIN vs APD · responsivity · sensitivity · Q-factor
The receiver converts light back to current. A PIN photodiode is simple, fast and low-noise (sensitivity ~−18 dBm at 10G). An APD adds internal avalanche gain for ~8 dB better sensitivity at the cost of excess noise — used for longer reach. Link quality is captured by the Q-factor, which maps to BER through the steep ‘waterfall’ curve.
PIN low-noiseAPD avalanche gainQ → BERsensitivity
BER vs OSNR Waterfall REAL SWEEP
A real OptiCommPy sweep measures BER across OSNR — the classic waterfall that sets the required OSNR for each format.
…
Req OSNR @FEC
…
Sweep points
Click sweep to run the BER curve.
Eye Diagram Instrument REAL DSP
A real modulated, pulse-shaped, noisy signal is generated on the server (OptiCommPy) and folded into an eye diagram — the classic optical signal-quality instrument. A wide-open eye = clean signal; noise closes it.
…
Q-factor
…
Q (dB)
…
BER
…
Eye height
Click “Acquire” to run real DSP on the server.
Real receiver BER · OptiCommPy REAL
A genuine Monte-Carlo simulation on the server measures the bit-error rate an OOK receiver sees at 14 dB OSNR.
…
BER @14dB
…
bits/sym
PIN photodiode APD
—
Q-factor
—
Q (dB)
—
BER (OOK)
BER vs OSNR (OOK). The curve is brutally steep — a "waterfall". A 1–2 dB OSNR gain can drop BER by orders of magnitude. The red marker is your current OSNR; the dashed line is the 7% FEC threshold.

Amplifiers & Passive Components

EDFA · Raman · couplers · splitters · isolators
Optical amplifiers boost the signal without converting to electronics. The EDFA uses erbium-doped fiber pumped at 980/1480 nm to amplify the whole C-band at once (~3 dB quantum-limited noise figure). Raman amplification pumps the transmission fiber itself for distributed, low-noise gain. Passive parts — couplers, splitters, isolators, WSS/ROADM — route and combine wavelengths.
EDFA C-bandRaman distributedNF ≥ 3 dBROADM / WSS
Real OSNR after the EDFA chain · GNPy REAL
GNPy models the erbium-amplifier chain and returns the real OSNR/GSNR delivered to the receiver.
…
OSNR
…
GSNR
—
Output power
—
OSNR (1 span)
—
ASE penalty
PassiveFunctionTyp. loss
Coupler / splitter 1:2Split power3.0–3.5 dB
Splitter 1:32PON distribution~17 dB
IsolatorBlock reflections0.3–0.8 dB
CirculatorDirectional routing0.6–1.0 dB
FBGλ filter / DCMvaries
WSS (ROADM)λ add/drop/switch4–7 dB
Mux/Demux (AWG)Combine λ's3–6 dB
Connector (SC/LC)Mate fibers0.2–0.5 dB
Fusion spliceJoin fibers0.05–0.1 dB

Optical Networks

Multiplexing wavelengths, transporting them over OTN/SDH, distributing fiber to homes with PON, and feeding 5G radios.

WDM Channel Planner — ITU-T Grid

DWDM 50/100 GHz (G.694.1) · CWDM 20 nm (G.694.2) · anchor 193.1 THz
WDM stacks many wavelengths on one fiber. DWDM packs 50/100 GHz-spaced channels (anchored at 193.1 THz, G.694.1) — 80–96 channels filling the C-band, each 100–800G, all amplified by one EDFA. CWDM (G.694.2) uses a coarse 20 nm grid with cheap uncooled lasers. Plan a real grid below.
193.1 THz anchor50 / 100 GHzCWDM 20 nm80–96 channels
Optical Spectrum Analyzer (OSA) REAL
A live OSA sweep of the C-band DWDM comb on the real ITU grid (193.1 THz anchor). The ASE noise floor / OSNR is anchored to a real GNPy link computation.
…
Channels
…
OSNR floor · GNPy
…
C-band span
Click sweep to acquire the spectrum.
Real per-channel GSNR · GNPy REAL
GNPy computes the real GSNR a DWDM channel achieves over an amplified C-band link.
…
GSNR
…
OSNR
…
CD ps/nm
—
Spacing
—
Capacity @100G/ch
—
Spectrum used
—
Band
DWDM frequencies are defined as 193.1 THz ± n×spacing. 50 GHz spacing doubles the channel count of 100 GHz in the same C-band, enabling 80–96 channels. CWDM uses cheap uncooled lasers on a coarse 20 nm grid (no EDFA across all channels).

Transport: SDH/SONET · OTN (G.709) · FEC

Container hierarchy · OTU wrapper · coding gain
Above the wavelength sits the digital transport layer. SDH/SONET is the legacy TDM hierarchy (STM-N / OC-N). OTN (ITU-T G.709) is the modern ‘digital wrapper’ that encapsulates any client (Ethernet, SDH) with overhead for monitoring plus powerful FEC. Forward error correction (Reed-Solomon, or soft-decision in coherent) adds coding gain so links run far below the raw error floor.
SDH STM-NOTN G.709digital wrapperFEC coding gain
Real FEC coding gain · CommPy REAL
A real convolutional encoder + Viterbi decoder on the server turns a 5×10⁻² channel BER into the post-FEC result.
…
pre-FEC BER
…
post-FEC BER
OTNLine rateClient
OTU12.666 Gb/sOC-48 / STM-16
OTU210.709 Gb/s10GbE / STM-64
OTU343.018 Gb/s40GbE / STM-256
OTU4111.81 Gb/s100GbE
OTUCnn×100 Gb/sflexible (Beyond 100G)
OTN frame: overhead (FAS, OTU/ODU/OPU) + client payload + FEC. The "digital wrapper" adds FEC and tandem-connection monitoring around any client. SDH/SONET (STM-N / OC-N) is the legacy TDM layer it often carries.
—
Net coding gain
—
Overhead
—
Post-FEC BER
—

PON / FTTx Power Budget

GPON · XGS-PON · NG-PON2 · OLT→ODN→ONT · split ratio
Fiber-to-the-home uses a passive optical network: one OLT in the exchange feeds many homes through a passive splitter (1:32, 1:64) to ONTs — no powered electronics in between. GPON (2.5/1.25G), XGS-PON (10/10G) and NG-PON2 differ in rate and power class. The whole design hinges on the power budget: fiber + splitter + connector loss must stay under the class budget.
OLT → splitter → ONTGPON / XGS-PON / NG-PON21:32 / 1:64power budget
—
Class budget
—
Total loss
—
Splitter loss
—
Margin
—
Loss = fiber (0.35 dB/km @1310 up / 0.21 @1577 down) + splitter (≈3.5·log₂N) + connectors (0.5 dB) + splices (0.1 dB). A passing link needs the total below the class budget with margin for aging/repairs.

Optical Transport for 5G

Fronthaul · midhaul · backhaul · CPRI vs eCPRI · PTP/SyncE
5G splits the radio into RU/DU/CU joined by optical fronthaul, midhaul and backhaul. Old CPRI carries raw time-domain IQ — huge, constant bandwidth. eCPRI (functional split 7-2x) sends frequency-domain IQ over Ethernet, ~4–5× less, and rides a shared optical network. Tight timing (PTP G.8275.1, SyncE) keeps TDD radios phase-aligned to ±1.5 µs.
FH / MH / BHCPRI vs eCPRIsplit 7-2xPTP / SyncE
—
Fronthaul rate
—
Max FH latency
—
Max FH reach
SegmentConnectsLatency budget
FronthaulRU ↔ DU~100 µs (one-way)
MidhaulDU ↔ CU~1–10 ms
BackhaulCU ↔ Core~10s ms
Timing & sync: 5G TDD needs tight phase alignment (±1.5 µs network limit). Delivered via PTP (IEEE 1588v2, G.8275.1) for phase/time and SyncE (G.8262) for frequency. eCPRI packetizes fronthaul over Ethernet so it can ride a shared optical network instead of dedicated CPRI fiber.

Coherent Optics, Design & AI/ML

The modern long-haul stack: coherent transceivers, end-to-end link design, field troubleshooting with OTDR, and machine learning for quality-of-transmission.

Coherent Optical Systems

100G / 400G / 800G · ZR pluggables · DSP chain
Coherent systems mix the incoming signal with a local-oscillator laser, recovering full amplitude, phase and both polarizations. A powerful DSP then compensates chromatic dispersion and PMD electronically and decodes high-order QAM. This is how 100G/400G/800G and pluggable 400ZR modules reach metro-to-long-haul distances on a single wavelength.
LO + DSPelectronic CD/PMD comp100–800G400ZR pluggable
Real nonlinear propagation · OptiCommPy SSFM REAL
OptiCommPy split-step propagates 16-QAM through 80 km at +2 dBm and returns the real EVM after CD compensation.
…
EVM %
…
regime
—
Modulation
—
Baud rate
—
Req. OSNR
—
Typical reach
Coherent DSP pipeline
Coherent detection mixes the signal with a local-oscillator laser, recovering full amplitude and phase on two polarizations. That lets DSP compensate CD and PMD electronically — no dispersion-compensating fiber needed — and decode high-order QAM.

Link & OSNR Budget Designer

Multi-span builder · end-to-end loss · OSNR · CD · margin
Designing a link balances two budgets. The power/loss budget ensures the signal arrives above receiver sensitivity. The OSNR budget ensures enough signal-to-noise after every EDFA: OSNR ≈ 58 + Pch − Lspan − NF − 10·log₁₀(N). Push launch power for OSNR — but too high triggers nonlinearity. The designer below builds a multi-span link and checks the margin.
power budgetOSNR ≈ 58+Pch−L−NF−10logNCD planningnonlinear ceiling
—
Total distance
—
End OSNR
—
OSNR margin
—
Accum. CD
—
OSNR(dB) ≈ 58 + Pch − Lspan − NF − 10·log₁₀(N). Push launch power up for OSNR — but too high triggers fiber nonlinearity (the simulator flags the nonlinear region). Coherent DSP handles the accumulated CD, so distance is OSNR-limited, not dispersion-limited.

OTDR Fault Simulator

Backscatter trace · connectors · splices · bends · breaks
The OTDR (optical time-domain reflectometer) fires pulses and times the back-scattered light to map the fiber: distance = c·t/(2n). A sloping line is normal Rayleigh backscatter; a downward step is a splice or bend loss; a reflective spike is a connector or break; the trace dropping to the noise floor marks the fiber end or a cut. It is the #1 field tool for locating faults.
d = c·t / 2nspike = connector/breakstep = splice/bendRayleigh slope
Live OTDR Instrument REAL .SOR
Real Telcordia .SOR field captures parsed by pyOTDR and rendered on an OTDR scope. Distance = c·t/2n; reflective spikes = connectors/breaks, steps = splices/bends.
…
Events
…
Wavelength
…
Total loss
Select a capture and acquire.
—
Events found
—
Fault distance
—
End-to-end loss
Distance = (c · Δt) / (2·n), n≈1.468. Reflective spikes = connectors/breaks (Fresnel reflection); a downward step with no spike = a splice or bend loss. The dashed slope is the fiber's intrinsic Rayleigh backscatter. Click "Shoot trace" after changing the scenario.

AI/ML for Optical Networks

QoT prediction · optical telemetry · soft-failure detection
Machine learning is now core to optical operations. QoT prediction estimates whether a new lightpath will work before turn-up, from features like OSNR margin, residual CD and nonlinearity. Soft-failure detection watches streaming telemetry to catch slow degradations (aging, dirt, bends) before they become outages. The tools here call a real scikit-learn model on the server.
QoT predictionsoft-failure detectionoptical telemetryreal scikit-learn
Quality-of-Transmission predictor
A lightweight in-browser classifier weighs link features to predict whether a new lightpath will work before you turn it up — the core of ML-assisted optical planning.
—
Soft-failure detection
Streaming OSNR telemetry. A slow degradation (aging, dirt, bending) is a "soft failure" — the anomaly detector flags drift before it becomes an outage.
Monitoring…

Real Lab — Live Open-Source Optical Engines

These panels don't fake anything — every "Run on server" button calls genuine open-source software running in a Python environment on the CafeTele lab server, and shows you the real output.

The stack running on the server

all genuine upstream projects · executed live · checking…
EngineWhat it really does
GNPy oopt-gnpyTelecom Infra Project's optical route & OSNR/GSNR planner — used by real operators
OptiCommPyCoherent DSP, fiber split-step, modulation & BER simulation
pyOTDRParses real Telcordia SR-4731 .SOR OTDR trace files
scikit-learnRandomForest QoT classifier trained live in your session
This is a real lab, not a video. The simulators in Days 1–4 teach the physics interactively; this tab proves it on production-grade tooling. The same GNPy you're calling here is the engine the open optical-networking community uses for vendor-agnostic planning.

A full Linux web-terminal with all four engines pre-installed is embedded at the bottom — run oclab-help to see the commands.

GNPy Continental Route Planner

real 5-node optical mesh network · computes GSNR / spans / distance per route
—
Path GSNR
—
Fiber spans
—
Distance
—
Accum. CD
—
Genuine GNPy routing over its reference optical mesh (5 cities). It finds the lightpath, counts EDFA-amplified spans and returns the end-to-end GSNR — the core go/no-go decision in optical network planning.

GNPy — Real Transmission

computes GSNR/OSNR/CD/PMD/latency over a path
—
GSNR (0.1nm)
—
OSNR ASE
—
CD ps/nm
—
PMD ps
—
Latency ms
click “Run” to execute the real GNPy CLI…

OptiCommPy — Real BER

Monte-Carlo over AWGN, server-side
—
BER (real)
—
SER
—
bits/sym

pyOTDR — Real .SOR Trace

parses an actual Telcordia OTDR capture

scikit-learn — Real QoT Model

RandomForest prediction, server-side
—

OptiCommPy SSFM — Real Fiber Propagation

split-step solver · Kerr nonlinearity · CD-compensated
—
EVM %
—
Regime
Raise launch power and watch the recovered constellation clouds spread — that is the real Kerr nonlinear penalty, computed by the split-step Fourier method (CD removed so only nonlinearity shows).

CommPy — Real FEC Coding Gain

convolutional code (rate 1/2, K=3) · Viterbi decode
—
Pre-FEC BER
—
Post-FEC BER
—
Improvement
—
A genuine Viterbi decoder runs on the server. Below the code's threshold the post-FEC BER collapses; above it, errors leak through — the same trade every optical transponder makes between coding gain and overhead.

Live Lab Terminal

real Linux shell · GNPy · OptiCommPy · pyOTDR · scikit-learn pre-installed
Requires the paid-access lab credential (same as the other CafeTele real labs). Once in, type oclab-help, then try gnpy-run 2, optical-ber qam64 22, otdr-parse, or qot-demo.

Missions — Prove You Can Operate

16 graded objectives across the four days. Complete the interaction on each tool and it auto-checks here. Progress is saved in your browser.