Optical Feedback in Semiconductor Lasers: The Five Regimes and Coherence Collapse
What back-reflections actually do to a laser: the Tkach–Chraplyvy feedback regimes, why coherence collapse destroys linewidth and RIN, which devices are sensitive and why, and the isolation budget that keeps links stable.
Scope
Every fiber pigtail, connector, and chip facet reflects something, and some of it finds its way back into the laser that sent it. This article maps what happens as that feedback grows: the five classical regimes, the physics that makes semiconductor lasers uniquely touchy, the coherence-collapse state that wrecks links, and the practical isolation budget. The coherence collapse glossary entry is the summary version; this is the working reference.
Why semiconductor lasers care so much
Three properties conspire. The cavity is short (hundreds of microns), so even weak feedback is a meaningful fraction of the intracavity field after few round trips. The facet reflectivity is modest (~30% as cleaved), so the cavity is poorly protected from outside light. And most importantly, the gain medium couples amplitude to phase through the linewidth enhancement factor α: a feedback-induced intensity ripple modulates carrier density, which modulates index, which shifts phase. That turns simple reflections into delayed phase feedback. A gas or solid-state laser with α ≈ 0 shrugs off levels that send a DFB into chaos.
The external path adds the final ingredient: delay. A reflection from 2 m of fiber returns 20 ns later (thousands of optical periods, many relaxation-oscillation cycles). Delayed nonlinear feedback is the textbook recipe for instability and chaos.
The five regimes
Tkach and Chraplyvy's 1986 measurements on 1.5 µm DFBs organized feedback behavior into regimes by the fraction of emitted power returned. The boundaries below are their landmarks; exact values shift with device design, bias, distance, and feedback phase:
| Regime | Feedback level | What you observe |
|---|---|---|
| I | below ≈ −80 dB | Linewidth narrows or broadens with feedback phase; sub-ppm reflections already measurable in coherent systems |
| II | ≈ −80 to −45 dB | Mode hopping between external-cavity modes; line splitting, kHz–MHz jumps |
| III | ≈ −45 to −39 dB | A narrow window of stabilized, re-narrowed single-mode operation |
| IV | ≈ −39 to −8 dB | Coherence collapse: chaotic intensity and phase dynamics |
| V | above ≈ −8 dB | Strong-feedback regime: with an AR-coated facet, stable extended-cavity operation |
Regime IV earns its name. The field undergoes chaotic pulsations seeded by beating between cavity and external-cavity modes, amplified through the relaxation resonance: linewidth explodes from megahertz to gigahertz, RIN rises tens of dB with structure at the external-cavity free spectral range and , and a link that measured error-free develops a BER floor that no received-power increase fixes. In the time domain it is low-frequency fluctuations and picosecond pulsing; on an OSA it is a broad pedestal swallowing the line.
The collapse threshold is device-dependent in ways that follow directly from the physics: it falls steeply with increasing α (common approximations scale it as , the single biggest lever) and rises with relaxation-oscillation damping. Hence the pecking order of tolerance: quantum-dot lasers (small effective α, heavy damping) tolerate the most; strongly damped, low-α quantum-well DFBs are respectable; narrow-linewidth low-damping designs are the most fragile. Those are exactly the lasers whose linewidth most needs protecting.
The isolation budget
Translate the regime table into hardware. An open PC connector reflects −14 dB, deep into regime IV for any ordinary DFB if it's close enough. A mated UPC pair at −50 dB sits near the regime II/III neighborhood; an APC pair at −60 dB or better retreats toward regime I. This is the entire case for the return-loss discipline: APC connectors in analog and PON plants, no open connectors on a live narrow-linewidth source, and a built-in isolator (30–60 dB, sometimes dual-stage) inside virtually every telecom transmitter butterfly. Distance matters too: the same reflection is nastier from centimeters (within the coherence length, phase-coherent feedback) than from kilometers. The safe engineering posture is still to budget levels, not distances.
Diagnosis, when a system misbehaves: look for RIN structure at spacings, a linewidth or SMSR that changes when you flex a patch cord, or a BER floor that tracks connector events. An OTDR or OCWR finds the guilty interface.
The regimes you use on purpose
Feedback is a tool once controlled. Regime III/weak-feedback stabilization is the operating principle of self-injection locking: a high-Q microresonator's controlled backscatter narrows a DFB by orders of magnitude, the architecture behind current chip-scale ultra-narrow sources. Regime V is simply the external cavity laser: AR-coat the facet so the external mirror is the cavity, and the "instability" becomes the design. Even regime IV has users. Chaotic-laser random number generators and secure-communication experiments run coherence collapse as a feature.
The thread continues in the related entries: injection locking (feedback from a different master laser, a cousin phenomenon with the sign flipped to useful), coherence collapse, and RIN. The RIN measurement procedure is where feedback problems usually get caught first.