TECHNOLOGY EXPLAINER
Coherent Optical Communication and DSP
— capture the shape of the light wave itself, then compute away the fibre's distortion
Short links inside a data centre read only the brightness of the light. Links spanning tens of kilometres to over a hundred place a second laser at the receiver and interfere it with the incoming light to read phase and polarisation as well. That is coherent communication. The captured waveform is recomputed by a DSP (digital signal processor), which restores what the fibre has smeared out. OIF's 400ZR and 800ZR are the agreements for fitting all of this into a small module you can plug in and pull out.
- What coherent optical communication is (the short version)
- Direct versus coherent detection — what gets received is different
- Inside the receiver — a local laser and four outputs
- What the DSP does — rebuilding by computation the waveform the fibre broke
- Our calculation: how many symbols one symbol spreads across
- 400ZR and 800ZR — what the specifications define
- Our calculation: how many Tb/s fit on one fibre
- A materials engineer's view (1): computation replaced dispersion-compensating fibre
- A materials engineer's view (2): laser purity has become a specification
- What is still hard
- Glossary / References / Claim-to-source audit
Sourced = stated in a standard, a public document from a standards body, or a peer-reviewed paper (link given)
Our calculation = a figure this article derived, with the assumptions spelled out
Not yet confirmed = still in development or planned, with nothing yet settled or delivered
General explanations of the physics and materials-design interpretations are marked separately as Commentary.
1. What coherent optical communication is (the short version)
- What is different: a laser called the local oscillator (LO) sits at the receiver and is made to interfere with the incoming signal light, so that phase and polarisation information, not just brightness, can be extracted (commentary)
- Why it is powerful: Kikuchi writes that the digital coherent receiver makes spectrally efficient modulation formats such as PSK and QAM usable, and that because phase information is preserved, linear impairments from group-velocity dispersion (GVD) and polarisation-mode dispersion (PMD) in the fibre can be equalised in the digital domainSourced
- Where it is used: OIF's 400ZR targets uses such as amplified links up to 120 km, and 800ZR single-span amplified links of 80 to 120 km (such as data-centre interconnect)Sourced
Modulation formats such as QAM and polarisation multiplexing, and the principle of the Mach-Zehnder modulator that produces them, were covered in our explainer on optical modulation and modulation formats. This article deals with the side that receives the signal and restores it. The fibre itself (the physics of dispersion and loss) is covered in our optical fibre explainer, the idea of separating channels by wavelength in our explainer on wavelength-division multiplexing and microrings, and narrow-linewidth tunable lasers in our semiconductor lasers explainer.
2. Direct versus coherent detection — what gets received is different
In a PAM4 link inside a data centre, the photodetector simply turns the intensity of the light into current (direct detection). The phase of the light wave (the timing of its peaks and troughs) and the direction in which it oscillates (polarisation) are lost at the moment of detection (commentary).
The OIF 400ZR implementation agreement specifies that coded bits are grouped eight at a time into one DP-16QAM symbol, with two bits assigned to the I (in-phase) component of the X polarisation, two to the Q (quadrature) component of the X polarisation, and the rest to I and Q of the Y polarisationSourced. In other words, the receiver has to measure the optical field on four axes at once (our commentary).
3. Inside the receiver — a local laser and four outputs
The receiver splits the incoming light and the local oscillator by polarisation and interferes them in combinations offset by 90°, extracting four electrical signals: I and Q of the X polarisation and I and Q of the Y polarisation. A/D converters turn these into digital values, which the DSP then processes by computation (commentary).
4. What the DSP does — rebuilding by computation the waveform the fibre broke
The 400ZR and 800ZR implementation agreements set out in numbers the channel conditions a module must tolerate. Turned the other way round, that is a list of the impairments the DSP has to cancel by computation (our commentary).
| Impairment | 400ZR (100 GHz DWDM) | 800ZR | What the DSP does (commentary) |
|---|---|---|---|
| Chromatic dispersion (CD) | 0 to 2,400 ps/nm | 0 to 2,400 ps/nm | Works back and undoes the difference in arrival time between colours (frequencies) |
| Differential group delay (DGD) | Up to 28 ps | Up to 28 ps | Undoes the arrival-time difference between the two polarisations and separates polarisations that have mixed |
| Polarisation-dependent loss (PDL) | 2 dB | 2 dB | Corrects differences in strength between polarisations |
| Rate of polarisation rotation | 50 krad/s | 50 krad/s | Tracks a polarisation direction that changes from moment to moment |
| Frequency offset between transmitter and local oscillator | −3.6 to +3.6 GHz (capture range) | — | Estimates the offset and subtracts it |
| Laser phase noise | Equivalent to 500 kHz linewidth (transmit and receive) | Equivalent to 500 kHz linewidth (transmit and receive) | Estimates and follows the carrier phase |
| Noise tolerance at the receiver (OSNR tolerance) | 26 dB / 12.5 GHz | 27.0 dB / 12.5 GHz | Combined with error correction (C-FEC for 400ZR, OFEC for 800ZR) |
All figures are Sourced (OIF 400ZR implementation agreement Rev 3.0 [Ref. 1], Table 20 and elsewhere; 800ZR implementation agreement Rev 1.0 [Ref. 2], Table 22 and elsewhere). "—" marks a cell for which this article did not confirm a corresponding provision. The right-hand column is this article's commentary.
On the 2,400 ps/nm dispersion limit, the 800ZR implementation agreement notes that it is "based on G.652 fiber and a target reach of 80km"Sourced. ITU-T G.652 specifies the dispersion coefficient of G.652.D fibre at 1550 nm within the range 13.3 to 18.6 ps/(nm·km)Sourced.
5. Our calculation: how many symbols one symbol spreads across
How big an impairment is 2,400 ps/nm of dispersion? We estimate it as how many symbol periods a single symbol spreads over on its way through the fibre.
- Symbol duration: 1 / 59.84 GHz = 16.7 ps (400ZR) and 1 / 118.2 GHz = 8.46 ps (800ZR)Our calculation
- Signal spectral width: Δλ ≈ λ² × Δf / c = (1.55 µm)² × 59.84 GHz / (3.0 × 108 m/s) = 0.48 nm. For 800ZR, about 0.95 nmOur calculation
- Spread: 2,400 ps/nm × 0.48 nm = 1,150 ps → 1,150 / 16.7 = about 69 symbols. For 800ZR, 2,400 × 0.95 = 2,270 ps → 2,270 / 8.46 = about 270 symbolsOur calculation
- Consistency check: at G.652.D's upper value of 18.6 ps/(nm·km), 120 km gives 18.6 × 120 = 2,232 ps/nm and 80 km gives 1,488 ps/nm. Both fall within 2,400 ps/nmOur calculation
Assumptions and limits: this approximation treats the signal's frequency width as equal to the baud rate. The real spread depends on the shape of the spectrum. The calculation is meant to convey the order of magnitude: the DSP is untangling, symbol by symbol, a waveform in which tens to hundreds of symbols lie on top of one another.
6. 400ZR and 800ZR — what the specifications define
| Item | 400ZR (OIF-400ZR-03.0) | 800ZR (OIF-800ZR-01.0) |
|---|---|---|
| Modulation format | Single-carrier coherent DP-16QAM | DP-16QAM |
| Baud rate | 59.843750000 GBaud per polarisation (478.750 Gbps in total) | Nominal 118.203350603 Gbaud |
| Error correction | Concatenated FEC (C-FEC). Post-FEC error floor <1.0E-15 | OFEC |
| Applications | Amplified links up to 120 km (DWDM at 100 GHz or 75 GHz spacing), and unamplified single-wavelength links | Single-span amplified DWDM over 80 to 120 km |
| Channels | 48 × 100 GHz, or 64 × 75 GHz | 32 × 150 GHz (the defined plan) |
| Form factor | Not constrained | Aims at implementation in small pluggables with port density equivalent to client optical modules (form factor not constrained) |
All Sourced (OIF 400ZR implementation agreement Rev 3.0 [Ref. 1], 800ZR implementation agreement Rev 1.0 [Ref. 2]). Both versions are dated 8 October 2024.
The 400ZR implementation agreement lays out channel frequencies in 0.1 THz (100 GHz) steps referenced to 193.1 THz, with 400ZR modules using the range 191.3 to 196.1 THzSourced. Within that range, the table of 48 channels at 100 GHz spacing runs from 196.1 to 191.4 THz (n = 30 to −17)Sourced. In wavelength terms that is about 1,529 to 1,567 nm, the so-called C bandOur calculation.
The QSFP-DD MSA technical white paper (2023) cites coherent modules (1600ZR) as an example of the high-power modules QSFP-DD1600 will need to accommodate, and assumes coherent modules of 30 W and 50 W in its system cooling studiesSourced. The DSP's computational load shows up directly as heat (see also our explainer on optical transceivers).
7. Our calculation: how many Tb/s fit on one fibre
8. A materials engineer's view (1): computation replaced dispersion-compensating fibre
ITU-T G.652 says of using this fibre in the 1550 nm region that "some forms of chromatic dispersion compensation are often employed"Sourced.
In the era of direct detection, compensating dispersion was the job of physical components. Dispersion-compensating fibre, with a refractive-index profile designed to give dispersion opposite to that of the transmission fibre, was inserted into the line to cancel it while the signal was still light — a problem, in effect, solved by the index profile of glass (our commentary).
With coherent reception, as Kikuchi notes, phase information is preserved, so group-velocity dispersion and polarisation-mode dispersion can be equalised in the digital domainSourced. That the 400ZR and 800ZR implementation agreements are written on the premise that the module accepts 0 to 2,400 ps/nm of dispersion means that the DSP takes on that compensationSourced.
- What went away: less need to design and insert dispersion-compensating parts for each link
- What moved: in exchange, the computation and power needed to untangle tens to hundreds of overlapping symbols moves inside a pluggable module (Section 5)
- What remains: the DSP handles linear impairments. Kikuchi's paper likewise describes what can be equalised digitally as linear transmission impairmentsSourced. Fibre loss, and nonlinear effects caused by strong light, remain the responsibility of the fibre's own materials and design
From a materials supplier's point of view, going coherent replaced part of the market for fixing the flaws of the transmission line with materials by semiconductor computation. And since that computation produces heat, the demand has now shifted to thermal materials (our commentary).
9. A materials engineer's view (2): laser purity has become a specification
For the transmit laser's phase noise, the 400ZR implementation agreement requires the high-frequency content above 100 MHz to be consistent with a 500 kHz linewidth, and states that the receiver's local oscillator has the same linewidthSourced. A direct-detection transceiver needs no laser on the receive side. Coherent operation needs lasers for both transmitting and receiving at each end of a link (our commentary).
What is more, those lasers must be able to tune to any of 48 frequencies in the C band (for 100 GHz DWDM)Sourced. Wide tunability and narrow linewidth are requirements that pull against each other in the cavity design of a semiconductor laser (our commentary).
- Narrow linewidth: calls for a longer, lower-loss cavity. Temperature fluctuation also turns into phase noise
- Wide tunability: calls for a structure whose refractive index or cavity length can be changed a great deal
- Small size and low power: all of this must fit inside a pluggable with limited heat dissipation
As our explainer on optical modulation and modulation formats showed, transmitting QAM requires a modulator that can control phase precisely. A coherent module packs a narrow-linewidth tunable laser, a high-precision modulator, four channels of photodetectors, high-speed A/D converters and a DSP into one box, making it one of the most materially diverse products among optical communication components (our commentary).
10. What is still hard
(1) DSP power and heat
Many of coherent's advantages rest on the DSP's computation. That the QSFP-DD MSA white paper includes 30 W and 50 W coherent modules in its cooling studiesSourced shows that the amount of computation becomes, directly, a heat-removal problem (our commentary). For 800ZR, with double the baud rate, the estimate in Section 5 put the overlap at about four times as largeOur calculation.
(2) The next generation
OIF has announced that it launched a 1600ZR+ project at its meeting in the first quarter of 2024 [Ref. 6]Sourced. The content and completion date of a coherent implementation agreement at the 1.6 Tb/s class could not be confirmed as of this article's researchNot yet confirmed.
(3) What this article does not cover
The power consumption of individual products, DSP circuit size, and the specific material make-up of photodetectors and modulators are not given, because the primary sources this article could open and check did not provide them in a comparable form. The physics of nonlinear effects and the engineering of fibre materials are left to our optical fibre explainer.
- Coherent optical communication receives the phase and polarisation of light as well, by interfering it with a local oscillator (commentary)
- Because phase is preserved, dispersion and polarisation-mode dispersion can be equalised by the DSPSourced
- 400ZR and 800ZR both use DP-16QAM and require tolerance of 2,400 ps/nm of dispersion, 28 ps of DGD and 50 krad/s of polarisation rotationSourced
- At 2,400 ps/nm, one symbol spreads over about 69 symbols (400ZR) or about 270 symbols (800ZR)Our calculation
- One fibre carries 25.6 Tb/s with either 400ZR (75 GHz) or 800ZR, at about 5.3 b/s/HzOur calculation
- Dispersion compensation has moved from being solved in glass to being solved by computation, and the heat from that computation is the new materials challenge (our commentary)
- Both transmit and receive need a laser with narrow linewidth (equivalent to 500 kHz) that can tune widely across the C bandSourced
11. Glossary
- Coherent detection
- A reception method that interferes the signal light with light from a laser at the receiver to extract the amplitude, phase and polarisation of the light.
- Local oscillator (LO)
- The reference laser light built into a coherent receiver.
- I / Q
- The in-phase (I) and quadrature (Q) components. The two axes that represent the optical field as a complex number.
- DP-16QAM
- A format that puts 16QAM on each of two polarisations to send eight bits per symbol.
- DSP
- Digital signal processing. The circuit that corrects and demodulates the digitised signal by computation.
- Chromatic dispersion (CD)
- The phenomenon in which different colours (frequencies) of light travel at different speeds in fibre, spreading pulses. Measured in ps/nm.
- DGD / PMD
- The difference in arrival time between the two polarisations (DGD), and the polarisation-mode dispersion (PMD) that causes it.
- PDL
- Polarisation-dependent loss. Loss that differs with the direction of polarisation.
- OSNR
- Optical signal-to-noise ratio. The strength of the signal relative to noise from optical amplifiers.
- DWDM
- Dense wavelength-division multiplexing. Packing many wavelengths at close spacing onto one fibre.
- C band
- The wavelength band of roughly 1,530 to 1,565 nm, the core band for long-distance communication.
- 400ZR / 800ZR
- Implementation agreements for coherent optical interfaces defined by the OIF.
- C-FEC / OFEC
- The forward error correction schemes used in 400ZR and 800ZR.
- Linewidth
- The spread in frequency of laser light. The narrower it is, the more stable the phase.
12. References (primary sources)
- OIF "Implementation Agreement 400ZR (OIF-400ZR-03.0)", 8 October 2024 (PDF) — oiforum.com
- OIF "Implementation Agreement for 800ZR Coherent Interfaces (OIF-800ZR-01.0)", 8 October 2024 (PDF) — oiforum.com
- Kikuchi K. (IEICE Electronics Express, 2011) "Digital coherent optical communication systems: fundamentals and future prospects", Vol. 8, No. 20, pp. 1642–1662. DOI: 10.1587/elex.8.1642 — jstage.jst.go.jp
- ITU-T "G.652 (08/2024) Characteristics of a single-mode optical fibre and cable" — itu.int
- QSFP-DD MSA "Enabling QSFP-DD1600 Ecosystem With Performance-Driven Thermal Innovations", technical white paper, September 2023 (PDF) — qsfp-dd.com
- OIF "OIF Publishes Implementation Agreement for 112 Gb/s Retimed Transmitter Linear Receiver (RTLR) ...", 18 November 2025 (its related-articles list includes "OIF Launches 1600ZR+ Coherent Optical ... Projects ... at Q1 2024 Technical and MA&E Meeting") — oiforum.com
13. Claim-to-source audit
| Claim in the text | Basis | Label |
|---|---|---|
| That 400ZR is single-carrier coherent DP-16QAM with C-FEC (post-FEC error floor <1.0E-15), for amplified links up to 120 km (100 GHz / 75 GHz DWDM) and unamplified single-wavelength links. 59.843750000 GBaud per polarisation, 478.750 Gbps in total. That the 8 bits are assigned two each to X-polarisation I and Q and Y-polarisation I and Q. The optical channel specification for 100 GHz DWDM (191.3 to 196.1 THz, 193.1 + n × 0.1 THz, 48 channels for n = 30 to −17, G.652, CD 0 to 2,400 ps/nm, DGD up to 28 ps, PDL 2 dB, polarisation rotation 50 krad/s). The 64 × 75 GHz plan. That the high-frequency phase noise of the transmit laser is consistent with a 500 kHz linewidth and the LO has the same linewidth. A carrier-to-LO frequency offset capture range of −3.6 to +3.6 GHz. OSNR tolerance of 26 dB / 12.5 GHz. That the form factor is not constrained. That the version is dated 8 October 2024 | OIF 400ZR implementation agreement Rev 3.0Reference 1 https://www.oiforum.com/wp-content/uploads/OIF-400ZR-03.0.pdf | Sourced |
| That 800ZR targets single-span amplified DWDM over 80 to 120 km and aims at low-power implementation in small pluggables with port density equivalent to client optics (form factor not constrained). DP-16QAM and OFEC. Nominal 118.203350603 Gbaud. The defined 32 × 150 GHz plan. CD 0 to 2,400 ps/nm ("based on G.652 fiber and a target reach of 80km"), DGD up to 28 ps, PDL 2 dB, polarisation rotation 50 krad/s. OSNR tolerance of 27.0 dB / 12.5 GHz. Phase noise consistent with a 500 kHz linewidth and the same linewidth for the LO. The note that for part of the transmit power range, filling all channels may need case-by-case engineering | OIF 800ZR implementation agreement Rev 1.0Reference 2 https://www.oiforum.com/wp-content/uploads/OIF-800ZR-01.0.pdf | Sourced |
| That the digital coherent receiver makes spectrally efficient modulation formats such as PSK and QAM usable. That because phase information is preserved, linear transmission impairments from group-velocity dispersion and polarisation-mode dispersion can be equalised in the digital domain. That the paper discusses mitigating transmission impairments with DSP | Kikuchi, IEICE Electronics Express 8(20), 1642 (2011)Reference 3 https://www.jstage.jst.go.jp/article/elex/8/20/8_20_1642/_article | Sourced |
| That the dispersion coefficient of G.652.D at 1550 nm is specified as 13.3 to 18.6 ps/(nm·km) (linear approximation, 1460 to 1625 nm). That when used in the 1550 nm region, "some forms of chromatic dispersion compensation are often employed" | ITU-T G.652 (08/2024)Reference 4 https://www.itu.int/rec/T-REC-G.652-202408-I/en | Sourced |
| That coherent modules (1600ZR) are cited as an example of the high-power modules QSFP-DD1600 must accommodate, and that 30 W and 50 W coherent modules are included in the cooling studies | QSFP-DD MSA technical white paper (2023)Reference 5 http://www.qsfp-dd.com/wp-content/uploads/2023/09/2023-QSFP-DD%20MSA-ThermalWhitepaper.pdf | Sourced |
| That OIF has announced launching a 1600ZR+ project at its first-quarter 2024 meeting (a related-article headline) | Related-articles list on the OIF websiteReference 6 https://www.oiforum.com/oif-publishes-implementation-agreement-for-112-gb-s-retimed-transmitter-linear-receiver-rtlr-electrical-and-optical-interface-advancing-energy-efficiency/ | Sourced |
| Symbol durations (about 16.7 ps and about 8.46 ps), spectral widths (about 0.48 nm and about 0.95 nm), spread from dispersion (about 1,150 ps and about 2,270 ps) and the number of overlapping symbols (about 69 and about 270), and DGD of 28 ps in symbols (about 1.7 and about 3.3). 18.6 × 120 = 2,232 ps/nm and 18.6 × 80 = 1,488 ps/nm. Conversion of 191.3 to 196.1 THz into wavelength (about 1,529 to 1,567 nm). 48 × 400 = 19.2 Tb/s, 64 × 400 = 25.6 Tb/s, 32 × 800 = 25.6 Tb/s, and b/s/Hz. That the overlap grows about fourfold | Our calculation, assuming a wavelength of 1550 nm, an approximation treating the signal's frequency width as equal to the baud rate, and capacity counted at the client rate | Our calculation |
| The content and completion date of a coherent implementation agreement at the 1.6 Tb/s class | The project launch announcement was confirmed, but content and completion are an outlook that could not be confirmed as of this article's research | Not yet confirmed |
| General explanations of the difference in what direct and coherent detection receive, of the receiver structure (polarisation splitting, 90° hybrid, balanced detection, A/D) and of the DSP processing steps. The framing of the implementation agreements' tolerance values as a list of impairments the DSP cancels. The reading that DSP replaced dispersion-compensating fibre, and the framing that remaining nonlinearity and loss stay with the materials side. The framing that coherent needs lasers for both transmit and receive, and the reading that narrow linewidth and wide tunability pull against each other. The framing of coherent modules as products with a particularly wide range of materials | General explanations of the physics, and this article's framing and commentary based on public documents. Not views expressed by any of the bodies or researchers | Commentary |
| Power consumption of individual products, DSP circuit size, and the material make-up of photodetectors and modulators | Not stated in this article, because the primary sources it confirmed did not provide them in a comparable form | Commentary |
| That Figs. 1, 2 and 5 are explanatory drawings, that Figs. 3 and 4 are drawings that include our calculation, and that the hero image is AI-generated | A note by this article | Commentary |
Last updated 26 September 2026. Sources are limited to primary material (OIF implementation agreements, an ITU-T Recommendation, a peer-reviewed paper and the QSFP-DD MSA technical white paper). Because the article includes general explanations of the physics and materials-design interpretations, those are marked as Commentary and kept separate from sourced fact. The power consumption of individual products, DSP circuit size and the material make-up of each component are not given, because comparable primary sources could not be confirmed. The figures in Sections 5 and 7 are estimates based on the stated assumptions. All figures are for explanation. Figs. 1, 2 and 5 are vector drawings, Figs. 3 and 4 are vector drawings that include our calculation, and the hero image is AI-generated; none of them shows the structure of a real receiver, a real spectrum or the appearance of a real product.