DWDM link with dispersion compensation
Four 40 GHz pulse trains on the ITU grid share 10 km of fibre: the fibre spreads every pulse past its bit slot, and a short spool of dispersion-compensating fibre gathers them back.
Open in the canvas →Background
Dense wavelength-division multiplexing (DWDM) is how one optical fibre carries many independent signals: each channel gets its own optical frequency, all of them travel together, and the receiving end separates them again. The frequencies are not chosen freely: ITU-T G.694.1 defines the grid they sit on, anchored to 193.1 THz, with fixed channel spacings from 12.5 GHz to 100 GHz and integer multiples of 100 GHz[1].
One classic way to combine and separate the channels is the thin-film filter: a stack of dielectric layers on a substrate that, by interference, transmits one channel and reflects all the others. A multichannel demultiplexer cascades such filters, each taking the light the previous one reflected[2]; run backwards, the same cascade adds channels onto a common fibre.
The fibre itself limits how fast each channel can go. For links of standard single-mode fibre (ITU-T G.652), the typical values given for the 1530–1565 nm band are an attenuation coefficient of 0.275 dB/km and a chromatic dispersion coefficient D of 17 ps/(nm·km)[3]. D is related to the group-velocity dispersion by D = −2πcβ₂/λ²[4], which gives β₂ ≈ −21.7 ps²/km at 1550 nm. A short pulse contains a band of wavelengths that travel at different speeds, so it spreads. A transform-limited Gaussian pulse of full width at half maximum Δt that picks up a group-delay dispersion φ₂ = β₂L leaves with a width Δt·√(1 + (4 ln2·φ₂/Δt²)²)[5]: the shorter the pulse, the faster it spreads.
At 40 Gb/s the bit slot is 25 ps. A return-to-zero pulse carver can make pulses with a 33% duty cycle, a full width at half maximum of a third of the bit slot[4], here about 8.3 ps. After 10 km of standard fibre a Gaussian pulse of that width is about 73 ps wide and smears across about three bit slots. The resulting intersymbol interference distorts the amplitude of the one-bits and raises the level within the zero-bits[4]. In Winzer and Essiambre's comparison of formats at 42.7 Gb/s, most tolerate an accumulated dispersion of only about 50 ps/nm, for a 2 dB OSNR penalty with no add/drop multiplexers and 85 GHz mux/demux bandwidths[4]; 10 km of this fibre accumulates 170 ps/nm. The distortion can be undone: fibre with negative dispersion (DCF) compensates the dispersion accumulated in the positive-dispersion transmission fibre. Its dispersion is made strongly negative, around −100 ps/(nm·km), so that a short length is enough and adds little loss[4].
What this setup demonstrates
Four pulsed lasers sit on the grid at a 200 GHz spacing: 193.4, 193.2, 193.0 and 192.8 THz, which are 1550.12, 1551.72, 1553.33 and 1554.94 nm. Each emits 1 mW of 8.3 ps transform-limited Gaussian pulses at 40 GHz. A train with a pulse in every slot is the clock of a 40 Gb/s return-to-zero signal, every bit a one, with no data pattern on it. Three band-pass filters, 1.2 nm wide, put the channels on one path. CH4 needs no filter of its own: the CH3 filter reflects it onto the path. The probe just after the multiplexer shows the four lines.
The link is 10 km of standard fibre: D = 17 ps/(nm·km), so β₂ = −21.7 ps²/km, and 0.275 dB/km. Then come 1.7 km of DCF: D = −100 ps/(nm·km), so β₂ = +127.6 ps²/km, with an assumed 0.5 dB/km. Together the two fibres add a group-delay dispersion of −217 ps² and +216.9 ps², which leaves −0.08 ps². The demultiplexer drops one channel at each filter. Every receiver reads 8.3 ps and 0.44 mW: 0 dBm launched, less 2.75 dB in the standard fibre and 0.85 dB in the DCF. The duration and power probes on the RX3 path show the pulse width and the power. The time probe on RX1 marks when pulses arrive, one per 25 ps slot; it shows timing, not pulse shape.
Things to try. Select the blue DCF and set its β₂ to 0. The duration probe, and the pulse width in any receiver's inspector, now read 73 ps: each pulse spans about three 25 ps bit slots, so neighbouring bits overlap. Set the DCF back and shorten it to 1.2 km: 64 ps² of dispersion remain and the pulses read 22.9 ps, nearly the whole bit slot. Delete a laser and only its receiver goes dark.
Each laser emits a regular train of Gaussian pulses, not data, and not the exact pulse shape of a modulator-carved RZ signal. So there are no bit patterns, no eye diagram and no bit-error rate. Noise from sources, amplifiers and receivers is not modelled. Neither are fibre nonlinearity, polarization-mode dispersion, or the variation of dispersion with wavelength. Every channel sees the same β₂ here, while the typical slope of standard fibre, 0.056 ps/(nm²·km)[3], makes D at the outer channels differ by about 0.27 ps/(nm·km). The DCF loss is an assumed value, not taken from a product.
The filters are ideal flat tops: complete transmission inside a 1.2 nm band, complete reflection outside it, and no roll-off, ripple or dispersion of their own. A real filter's passband, defined at 0.5 dB below the transmission at the centre wavelength, is typically 40–50% of the channel spacing[2]. The one here is wider (75%), so the ideal edges do not clip the pulse spectrum. The fibres are characterised by a fixed loss per metre and a constant β₂. Coupling at their ends is geometric and does not compute mode-field overlap.
Pulse widths are computed for each Gaussian pulse on its own. The time probe marks arrival times and does not draw the broadened pulses, so it does not show neighbouring bits running into each other; the duration readouts do.
Related components
References
- ITU-T Recommendation G.694.1, “Spectral grids for WDM applications: DWDM frequency grid” (10/2020), clauses 6–7
- C. He, D. Ma and W.-Z. Li, “Thin-film filters are the building blocks of multiplexing devices,” Lightwave (1 May 2001)
- ITU-T Recommendation G.652, “Characteristics of a single-mode optical fibre and cable” (08/2024), Table I.1
- P. J. Winzer and R.-J. Essiambre, “Advanced optical modulation formats,” Proceedings of the IEEE 94, 952–985 (2006): §IV-B, §V-A, §VI-E and Table 5
- MKS Newport, “The Effect of Dispersion on Ultrashort Pulses,” application note, Eq. 8