REVIEW 2 major objections 4 minor 15 references
A Comb-based Colorless Coherent WDM Transmitter
T0 review · 2 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read One unitary phase matrix, formed by fixed on-chip delays, lets a demultiplexer-free comb transmitter generate arbitrary independent signals on every comb line with only a single-tap MIMO pre-equalization.
desk verdict A credible proof-of-concept for demultiplexer-free comb-based WDM transmission via unitary time-delay pre-equalization, with the main open question being the stability margin of the free-running comb at higher channel counts. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The unitary phase matrix $\boldsymbol{\Phi} = \exp(j\boldsymbol{\varphi})$, whose entries are the phase shifts that the fixed time delays on each modulator branch imprint on every comb line. With delays $\tau_n = (n-1)T/N$ that evenly split the comb period, $\boldsymbol{\varphi}$ takes a discrete-Fourier-type form and $\boldsymbol{\Phi}$ becomes unitary, making the map from modulator drive signals to output WDM signals invertible without information loss. This matrix is what lets a bank of identical wideband modulators behave like a bank of per-wavelength modulators; its inverse, applied as a single-tap MIMO pre-equalizer, is the only complexity the scheme adds over a conventional transmitter.
What would settle it
Run the 3-channel 80-GBd link from the free-running microcomb with fixed MIMO-PEQ coefficients, then perturb the comb by changing pump power or chip temperature so that its free spectral range shifts by tens to hundreds of MHz or its line phases wander. If inter-channel crosstalk reappears and the monitored channel's SNR drops by more than about 1 dB before re-calibration, the load-bearing stability assumption fails in exactly the way the paper flags; the 26-hour test shows stability in one laboratory condition, not immunity to such perturbations.
Extended reading notes
Core claim
The paper's central claim is an equivalence: a demultiplexer-free transmitter that takes drive signals $\boldsymbol{X}$ as input is equivalent to a demultiplexer-based transmitter (one modulator per comb line) taking target WDM signals $\boldsymbol{Y}$ as input, with $\boldsymbol{Y} = \boldsymbol{\Phi}^T \boldsymbol{X}$. The matrix $\boldsymbol{\Phi} = \exp(j\boldsymbol{\varphi})$ is set by fixed on-chip delays: the phase of comb line $k$ on branch $n$ advances by $2\pi\Delta f\,\tau_n$ per line spacing, so each column of $\boldsymbol{\varphi}$ is an arithmetic phase sequence. When the delays $\tau_n$ evenly segment the comb repetition period $T = 1/\Delta f$, $\boldsymbol{\Phi}$ is unitary in the flat-power limit, so every target WDM signal can be generated exactly by driving the modulators with $\boldsymbol{X} = (\boldsymbol{\Phi}^T)^{-1}\boldsymbol{Y}$, a single-tap filter that is independent of the RF data. Because the construction is wavelength-agnostic, the same chip and DSP work in any comb window after re-calibrating $\boldsymbol{\Phi}$; because the equalizer only mixes static phase terms, no comb-to-RF locking or transmitter–receiver synchronization is needed. The demonstration delivers 80-GBd PS-64-QAM on four silicon I/Q modulators (4 × 633 Gb/s net under a rate-0.8262 soft-decision FEC), and a 3-channel 80-GBd 16-QAM link driven by a free-running 100-GHz-FSR Kerr microcomb, stable over 26 hours.
Load-bearing premise
The calibrated phase matrix $\boldsymbol{\Phi}$ stays accurate while data is being sent: the comb's relative line phases and powers, and the fixed on-chip delays, must not drift enough to break the single-tap equalizer's crosstalk cancellation, and the scheme has no active loop that re-measures or corrects $\boldsymbol{\Phi}$.
Editorial extensions
If this is right
- Colorless operation: the same modulator array and DSP work at any comb window after re-calibrating $\boldsymbol{\Phi}$ — shown at five windows across the C-band — so no per-channel wavelength control is needed.
- Free-running combs become practical sources: with no RF reference and no transmitter–receiver synchronization, the comb FSR never has to be locked to a clock, and any comb whose spacing matches the on-chip delays can be used.
- Scaling adds almost nothing: the extra complexity is a single-tap $N \times N$ matrix multiply per channel, and the estimated sub-10-femtosecond fabrication deviation in delays keeps $\boldsymbol{\Phi}$ unitary even for tens of modulators.
- The architecture transfers to other platforms: the same delay-and-equalize idea applies to thin-film lithium niobate modulators and, combined with on-chip amplification, points toward WDM transmitters with Tb/s/mm$^2$-class density.
Reading between the lines
- A tracking loop is the natural missing piece: nothing prevents a slow pilot-tone or dither-based re-estimation of $\boldsymbol{\Phi}$ during operation, which would extend the 26-hour laboratory stability result toward field conditions with thermal drift and vibration — the paper does not propose this.
- Because $\boldsymbol{\Phi}$ is a discrete-Fourier-type matrix, the transmitter is the spatial dual of a DFT-spread link; the same unitary-delay construction could, in principle, be run in reverse at a receiver as a comb-based demultiplexer for point-to-point links, though the paper claims only the transmitter side.
- The per-channel bandwidth cap is the comb FSR, so a 100-GHz comb supports roughly 100 GBd per line; a stress test the paper does not run is subcarrier multiplexing within each slot to find where band-edge crosstalk breaks the single-tap equalizer.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a comb-based coherent WDM transmitter that can modulate independent signals onto each comb line without optically demultiplexing the comb. The key idea is to add a series of time delays to the branches feeding an array of I/Q modulators, so that the mixing of signals at each wavelength is described by a phase matrix Φ. If the delays evenly segment the comb repetition period, Φ is a DFT-like unitary matrix, and a single-tap MIMO pre-equalization can generate arbitrary target WDM signals from the drive signals. The authors experimentally demonstrate the idea with a silicon-photonic 4-modulator array and a 4×633 Gb/s WDM signal using a 4-line E/O comb, and with a 3-channel free-running Kerr microcomb showing colorless operation and 26-hour SNR stability. They argue that this removes the need for on-chip demultiplexers, transmitter–receiver synchronization, and locking of the comb FSR, and that the approach scales to tens of channels with only modest DSP complexity.
Significance. The central proposal is elegant and potentially impactful: it replaces a difficult integrated demultiplexer with a fixed delay network and a small MIMO pre-equalization step, which is a genuine simplification for chip-scale coherent WDM transmitters. The paper provides real experimental evidence: a 4×633 Gb/s WDM transmission with a silicon-photonics modulator array, and a free-running microcomb demonstration with colorless operation across the C-band and a 26-hour stability test. These are meaningful proof-of-concept results, and the underlying linear algebra is standard and transparent. The main gaps are the lack of a self-contained proof of the unitary-matrix lemma (delegated to a prior same-group conference paper) and, more importantly, the absence of a quantitative analysis of the tolerance of the scheme to comb FSR drift, which is load-bearing for the stated 'free-running comb' and scalability advantages. Overall, the idea is sound and well worth publishing after the stability and self-containment issues are addressed.
major comments (2)
- [Section 4, Fig. 3d] The central advantage of using a free-running comb depends on the calibrated phase matrix Φ remaining valid during operation. The paper states that the MIMO-PEQ coefficients would change if the relative power or phase among comb lines fluctuates, but it does not quantify the allowed FSR drift or report the FSR stability over the 26-hour test. For an N-channel scheme with delays spanning the comb period, a fractional FSR error ε after calibration produces a phase error of roughly 2πNε on the highest comb line, so the tolerable drift shrinks linearly with N. For N=40, a 0.1% FSR drift already gives a phase error of about 0.25 rad, which can seriously degrade crosstalk cancellation. The current evidence does not establish that a free-running microcomb under realistic thermal/environmental conditions can meet this tolerance at the scales claimed. Please add a detuning experiment (e.g., intentionally perturbing the comb FSR and measuring SNR) or an analytical tolerance bound, and qualify the 'no locking' claim accordingly.
- [Section 2, Eq. (4.1) and Eq. (3)] The unitary property of Φ is a load-bearing mathematical claim, but it is delegated to Ref. [5] rather than derived in this paper. The proof is simple (for delays evenly segmenting the repetition period, Φ is a DFT matrix and is unitary up to the normalization factor √N), and including it would make the paper self-contained. In addition, the paper should clarify the normalization: the matrix Φ defined in Eq. (2) with τ_n satisfying Eq. (3) satisfies ΦΦ^H = N·I, not ΦΦ^H = I, so strictly speaking it is √N times a unitary matrix. This does not affect the invertibility argument, but the wording 'Φ would be unitary' is imprecise and should be corrected.
minor comments (4)
- [Section 4, reliability test] The text says the 26-hour test monitors SNR, but the caption of Fig. 3d notes that the RF signal was turned off for a period (the RF probe was raised for safety). Please specify the exact on/off intervals and state whether the comb stayed in the same soliton state for the entire test.
- [Section 2, Eq. (1)] The phase vector φ⃑_n is defined with the first comb line as a reference, but the notation in Eq. (1) is not fully explicit. It would be clearer to write φ_{n,m} = 2π Δf (m−1) τ_n for m=1,...,N and explain that the constant common phase is dropped.
- [Section 3, paragraph 1] The claim that 'the delay deviation due to fabrication in waveguide width is estimated to lower than 10 fs' is important for scalability, but no reference or measurement is given. Please provide either a citation or a brief derivation of this estimate.
- [General] The abstract and introduction could define 'colorless' more precisely; in this context it appears to mean that the modulator branches are wavelength-agnostic and the scheme can be recalibrated for any comb center and FSR, but the term is not standard in this exact usage.
Circularity Check
No significant circularity: the unitary-matrix lemma is a parameter-free mathematical fact cited to prior work, and the experimental validation is independent of the calibration.
full rationale
The derivation chain is self-contained once the phase-matrix model is stated. Eq. (4.1), Y = Phi^T X, is a physical forward model based on time delays and the Fourier time-shift property, not a definition of Y in terms of X that would make the subsequent inversion trivial. The unitary property is delegated to Ref. [5], a same-group citation, but it is a parameter-free mathematical lemma with stated assumptions (delays evenly segment the repetition period) and does not assume the transmitter equivalence being claimed. The proof-of-concept is not circular because the MIMO-PEQ coefficients are obtained by a disclosed calibration of Phi and then evaluated on measured NGMI/SNR of transmitted signals; success is not defined by refitting the calibration data. The amplitude/phase calibration is a system-identification step, not a fitted parameter renamed as a prediction. Therefore no step in the paper reduces by construction to its own inputs, and the central claim retains independent experimental content.
Assumptions & free parameters
free parameters (2)
- On-chip branch delays =
0, 3.06, 6.12, 9.18 ps for 4 branches; 0, 3.06, 6.12 ps for 3 branches
- Calibrated Phi matrix entries =
Measured complex phase/amplitude values for each comb line and branch
assumptions (6)
- standard math Time-shifting property of the Fourier transform: a time delay tau introduces a phase exp(j 2 pi f tau) on each frequency component.
- standard math The phase matrix Phi is unitary when the delays evenly segment the comb repetition period.
- domain assumption Each I/Q modulator produces identical replicas of its drive signal on every comb line.
- domain assumption Channel signal bandwidth does not exceed the comb FSR.
- domain assumption Comb FSR and relative line phases are stable over time.
- domain assumption Fabrication delay deviation is below 10 fs.
Cite this review
Pith. "Pith review of A Comb-based Colorless Coherent WDM Transmitter." pith.science (2026). https://pith.science/paper/QCP4MR5G
@misc{pith2026250521607,
author = {Pith},
title = {Pith review of: A Comb-based Colorless Coherent WDM Transmitter},
year = {2026},
howpublished = {\url{https://pith.science/paper/QCP4MR5G}},
note = {Machine review of arXiv:2505.21607}
}
read the original abstract
We propose a comb-based WDM transmitter capable of modulating independent signals to comb lines without demultiplexing them and prove its concept and potential scalability in a WDM transmitter consisting of a Kerr microcomb and a silicon I/Q modulator array.
Figures
Reference graph
Works this paper leans on
-
[5]
Conclusion We propose a new chip-scale solution for highly-parallel coherent transmitters. It simplifies the monolithic integration of an OFC and a modulator array with any integration platform like silicon or thin-film lithium niobate [9]. The color- less nature allows for scaling to a larger number of integrated modulators without requiring individual w...
-
[1]
Introduction Chip-scale optical frequency comb s (OFC) [1,2] have been actively developed over the years as a promising multi- channel light source for highly parallel wavelength-division multiplexing (WDM) transmissions. The co-integration of an OFC and parallel modulators onto one photonic integrated circuit (PIC) requires a wavelength demultiplexer in ...
-
[2]
1, using an OFC with 𝑁 lines followed by an array of 𝑁 I/Q modulators (IQM)
Principle We assume an 𝑁-channel transmitter in Fig. 1, using an OFC with 𝑁 lines followed by an array of 𝑁 I/Q modulators (IQM). Without a demultiplexer, each modulator generates 𝑁 identical replicas of its input signal evenly spaced on the spectrum. Each wavelength mixes all the 𝑁 signals after colorless power combining all branches. A series of time Fi...
-
[3]
2a) with an array of 4 IQMs on 220-nm SOI wafers
Proof-of-concept of MIMO-PEQ using a SiPh modulator array with 4×633-Gb/s net data rate We fabricated a SiPh chip (Fig. 2a) with an array of 4 IQMs on 220-nm SOI wafers. Each IQM consists of 2 Mach- Zehnder modulators (MZM) [7] and 3 thermal phase shifters for bias control. The measured 3-dB E/O bandwidth is about 34 GHz. The input is equally split into 4...
-
[4]
Proof of feasibility of combining a Si3N4 microcomb and the SiPh modulator array We use a dissipative Kerr soliton (DKS) microcomb to verify the feasibility of using a free-running OFC (i.e., no need of an RF reference) with the SiPh modulator array. The combination of the wideband microresonator-generated OFC and the SiPh modulators reveal the potential ...
-
[6]
Marin-Palomo et al., Nature 546, 274-279 (2017)
P. Marin-Palomo et al., Nature 546, 274-279 (2017)
work page 2017
- [7]
-
[8]
Rizzo et al., Nature Photonics 17, 781-790 (2023)
A. Rizzo et al., Nature Photonics 17, 781-790 (2023)
work page 2023
Show all 15 references
-
[9]
Yamazaki et al., J
H. Yamazaki et al., J. Lightw. Technol. 43, 1550-1564 (2025)
2025
-
[10]
Che et al., ECOC’2024, PDP Th3A.6
D. Che et al., ECOC’2024, PDP Th3A.6
2024
-
[11]
Nagarajan et al., J
R. Nagarajan et al., J. Lightw. Technol. 39, 5221-5231 (2021)
2021
-
[12]
S. Y. Siew et al., J. Lightwave Technol. 39, 4374–4389 (2021)
2021
-
[13]
J. M. Gené et al., OFC’2020, paper M3G.3
2020
-
[14]
Wang et al., Nature Communications 10, 978 (2019)
C. Wang et al., Nature Communications 10, 978 (2019)
2019
-
[15]
Liu et al., Science 376, 1309-1313 (2022)
Y. Liu et al., Science 376, 1309-1313 (2022). Fig. 3. A demultiplexer-free 3-channel WDM transmitter using a DKS microcomb (100-GHz FSR). (a) System SNR as a function of symbol rates. (b) Microcomb spectrum at C-band. (c) Demonstration of the colorless nature of the scheme: th...
2022
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Sign in to comment.