Integrated recurrent optical spectral slicer for equalization of 100-km C-band IM/DD transmission

A. Bogris; B. Charbonnier; C. Mesaritakis; C. Vigliar; F. Da Ros; G. Sarantoglou; H. Liu; I. Teofilovic; K. Sozos; P. Bienstman

REVIEW 3 major objections 4 minor 12 references

Integrated recurrent optical spectral slicer for equalization of 100-km C-band IM/DD transmission

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that a purpose-built silicon photonic spectral-slicing receiver can hold 32-GBd PAM-4 C-band transmission below forward-error-correction limits at 50 and 100 km, using only a simple linear equalizer after the optical…

desk verdict Custom low-loss ROSS filter is a genuine fabrication advance, but the 100-km two-node claim rests on an emulation that omits the 3-dB input-splitting penalty. read the letter →

arxiv 2507.11654 v1 pith:R2CPUK6Z submitted 2025-07-15 physics.optics

I. Teofilovic , K. Sozos , H. Liu , S. Malhouitre , S. Garcia , G. Sarantoglou , P. Bienstman , B. Charbonnier

show 5 more authors

C. Mesaritakis C. Vigliar P. Petropoulos A. Bogris F. Da Ros

This is my paper · ORCID

classification physics.optics PACS 42.79.Sz42.82.-m

keywords recurrentopticalspectralslicingsiliconphotonicsIM/DDtransmissionPAM-4C-bandequalizationfeed-forwardequalizerpowerfading

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper reports a silicon-photonics receiver front end that recovers dispersion-limited reach in short-haul intensity-modulation/direct-detection (IM/DD) links without coherent detection. The device, a recurrent optical spectral slicer (ROSS), is an asymmetric Mach-Zehnder interferometer with delayed feedback whose transfer function can be tuned by a thermo-optic heater to carve the received spectrum into non-overlapping slices. The paper claims that combining two such nodes followed by a simple linear feed-forward equalizer keeps a 32-GBd PAM-4 C-band signal below the KP4 hard-decision FEC threshold ($2.25\times10^{-4}$) at 50 km and below the 400ZR C-FEC threshold ($1.25\times10^{-2}$) at 100 km. In the reported measurements this is roughly two orders of magnitude lower bit-error ratio than a neural-network equalizer applied to the same signal without optical pre-processing. The reason this matters is that it extends the reach of low-cost direct-detection links by moving part of the equalization into the optical domain.

What carries the argument

The load-bearing element is the ROSS node, implemented here as an asymmetric Mach-Zehnder interferometer with a feedback loop from output to input containing about 19 ps of delay and a thermo-optic phase shifter. Feedback makes the filter recurrent (infinite impulse response), producing sharp spectral features with up to 15 dB extinction; tuning the heater voltage selects different spectral slices from the same device. On the receiver side, two independently tuned nodes' detected outputs are recombined by a 62-tap feed-forward equalizer spaced one symbol period apart, and the frequency diversity from non-overlapping slices is what suppresses the power-fading penalty.

What would settle it

Build or operate a chip with two physical ROSS nodes on the same die and measure 32-GBd PAM-4 at 50 and 100 km with both nodes running simultaneously; if the bit-error ratio at 100 km rises above the 400ZR C-FEC threshold or the 50-km result rises above KP4, the central claim fails. A cheaper check is to re-run the sequential-capture emulation with a different synchronization method and see whether the reported BER depends strongly on how the two traces are aligned.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a purpose-built recurrent spectral slicer, rather than a general-purpose programmable photonic circuit, makes the ROSS receiver practical. The fabricated filters show insertion losses of 9.8–10.4 dB, feedback-loop losses below 0.1 dB (versus more than 1 dB in the previous programmable implementation), and a rich family of sharp, high-extinction transfer functions that shift and reshape as the heater voltage changes. With two such nodes tuned to non-overlapping spectral slices, the paper reports BER below the KP4 threshold ($2.25\times10^{-4}$) at 50 km and below the 400ZR C-FEC threshold ($1.25\times10^{-2}$) at 100 km for 32-GBd PAM-4, with the short-delay (≈23 ps) design performing slightly better than the long-delay (≈44 ps) design.

Load-bearing premise

The entire 100-km claim rests on treating two separately recorded, time-synchronized traces from the same chip as if they came from a real two-node receiver; if sequential captures are not equivalent to simultaneous operation of two physical ROSS nodes, the reach result is not yet demonstrated.

Editorial extensions

If this is right

A two-node ROSS receiver can hold 32-GBd PAM-4 C-band transmission below the 400ZR C-FEC threshold at 100 km, extending IM/DD reach without coherent detection.
Because the digital stage after slicing is a linear feed-forward equalizer, the scheme avoids complex nonlinear or neural-network equalizers while reportedly beating them by about two orders of magnitude in BER.
The same chip design realizes different ROSS nodes by voltage tuning, so a single fabrication can serve multiple spectral-slice configurations.
The comparison with the programmable-PIC implementation shows that reducing feedback-loop loss and insertion loss is what unlocks the added order-of-magnitude BER gain.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural next step the authors leave implicit is to put two physical ROSS nodes on one chip and operate them simultaneously; their two-node receiver was emulated by synchronizing sequentially recorded traces from a single chip, so an integrated multi-node version would be the direct test of the 100-km claim.
If the frequency-diversity mechanism is the real source of gain, adding more nodes should improve reach or support higher baud rates, since more non-overlapping slices should further mitigate dispersion-induced fading.
The same recurrent-slicing front end could be applied to other direct-detection formats or access-network links where chromatic-dispersion power fading, rather than noise, is the bottleneck.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

First thing to know: this is not the programmable-PIC ROSS demo repackaged. The authors designed and fabricated dedicated silicon photonic filters with feedback loss below 0.1 dB and free choice of MZI delay, and the transfer-function measurements show sharp, tunable, high-extinction spectral slices that the programmable platform couldn't give them. Measured BER at 50 km with two nodes is about two orders of magnitude below the no-preprocessing case and well under KP4 FEC; the trend with distance is consistent, and the best node pairs stay the same across distances, which is a good sign that the effect is real. The comparison against their own prior programmable-PIC result is useful and credibly isolates the benefit of lower loss and flexible delay. I take the result as a genuine engineering advance: a practical route to make ROSS receivers viable.

The soft spot is exactly where the reader and the stress-test note point. The two-node receiver was never built. They took sequential DSO captures from a single PIC at different heater voltages and synchronized them offline. The paper says this explicitly, which is honest, but it has real consequences. A physical two-node receiver needs an input splitter, so each node gets 3 dB less power than in the sequential full-power measurement. That OSNR penalty is not in the reported BER. Also, offline synchronization can pick a favorable timing offset that a real receiver with fixed electrical paths won't get. The 100-km BER is below the 400ZR C-FEC threshold, but without knowing the margin, I can't tell whether it survives the 3-dB hit. This should be flagged as a load-bearing caveat, not a cosmetic one.

Other complaints are minor and standard for this kind of proof-of-concept: no error bars, best voltage pairs chosen post hoc from the 50-km heatmaps, one launch power, one fiber length per point. None of these undercut the fabrication story. The self-citation to their own theory and prior experiments is not a problem because the BER benchmark against external FEC thresholds is independent.

Who it's for: people working on short-reach IM/DD, silicon photonics for communications, and optical pre-processing. It deserves a serious referee. A good referee should push for a true two-node measurement or at least a quantified account of the splitting loss and synchronization before the headline reach claim is treated as demonstrated. As is, I'd call it a solid proof-of-concept with an overstated headline.

Referee Report

3 major / 4 minor

Summary. The paper reports the design, fabrication, and system test of a silicon-photonics recurrent optical spectral slicing (ROSS) filter intended for IM/DD receivers. Two filter variants with different MZI arm delays (long and short ΔT) were fabricated and characterized, and a two-node ROSS receiver was emulated by sequentially capturing single-PIC DSO traces at different heater voltages and synchronizing them offline. For a 32-GBd PAM-4 C-band link, the authors report BER below the KP4 FEC threshold at 50 km and below the 400ZR C-FEC threshold at 100 km, with up to two orders of magnitude improvement over a neural-network equalizer without optical preprocessing.

Significance. If the result holds, this is a meaningful advance for short-reach IM/DD links: a compact, purpose-designed integrated filter with low feedback loss provides optical frequency diversity that enables a simple FFE to extend the reach to 100 km. The paper provides measured transfer functions, BER results over distance, and a benchmark comparison including a programmable-PIC ROSS and a neural-network equalizer. The design builds transparently on the authors' prior theory, and the experimental benchmarking against external FEC thresholds is independent of fitting parameters. The main strength is the demonstrated potential of the custom filter, but the central claim relies on an emulated two-node receiver, which limits the strength of the headline conclusion.

major comments (3)

[System performance] The two-node ROSS receiver is emulated by sequentially measuring DSO traces from a single PIC at different heater voltages, as explicitly stated in the text. This emulation does not capture two essential behaviors of a real multi-node receiver: (i) the input signal must be split between the two nodes, incurring at least 3 dB of additional loss per node relative to the sequential full-power measurements, and (ii) the relative delay between the two nodes is fixed by hardware in a real receiver, whereas the offline synchronization of independently acquired traces may select a favorable relative delay. The paper should quantify the splitting penalty and repeat the key measurements with a real 1x2 splitter, or alternatively re-scope the abstract and conclusions to describe an emulated receiver. The synchronization procedure (e.g., cross-correlation, clock recovery) should also be described.
[Fig. 4 and associated text] The BER versus distance curves in Fig. 4 have no error bars or repeated trials, and the best node pairs are selected from the 50-km heatmaps in Fig. 3. As a result, the robustness of the 100-km margin over the 400ZR C-FEC threshold to thermal drift, polarization re-alignment, or heater-voltage variability cannot be assessed. Please provide repeated BER measurements for the best node pair at each distance (including re-alignment) and report the spread; also explicitly state whether the same heater voltage pair was used at all distances or re-optimized per distance.
[Abstract and Conclusions] The abstract and conclusions state as a fact that a two-node ROSS receiver achieves BER below FEC at 100 km, but the two-node receiver was never physically built. Given the importance of the splitting-loss and synchronization issues, the claims should be re-scoped to “emulated two-node receiver” unless the emulation is validated by including a splitter in the power budget or by a true multi-node measurement.

minor comments (4)

[System performance] The FFE description says “62 T-spaced-taps (31 symbols/filter)”; clarify whether this is 62 taps total or 62 taps per node, and how the two node outputs are combined.
[Fig. 4] The labels “long” and “short” are ambiguous in the figure and caption; add explicit legend entries or a statement identifying which curve corresponds to which ΔT design.
[Experimental setup] The reference path is described as “bypassing an EDFA and the PIC”, but the setup in Fig. 2 shows a pre-amplified receiver; clarify whether the reference measurement uses the same EDFA and BPF without the PIC or a completely different photodetection chain.
[System performance] The statement that “the best node positions are consistent over transmission distance” is not directly supported because the heatmaps in Fig. 3 are shown only at 50 km; either provide supporting data or soften the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported BER results are measured against external FEC thresholds, and the sequential two-node emulation is an experimental limitation rather than a derivation that reduces to its inputs.

full rationale

The paper's central result — BER below KP4 FEC at 50 km and below 400ZR C-FEC at 100 km for a two-node ROSS receiver — is an experimental measurement, not a quantity derived from fitted parameters or from the authors' prior work. The filter design is motivated by the authors' own previous theory [7] and programmable-PIC experiments [8], but this is design inheritance, not load-bearing circularity: the validation is carried out by direct BER counting against independent FEC thresholds, and the ROSS receiver is benchmarked against a neural-network equalizer without optical preprocessing. No equation in the paper reduces the claimed result to an input; the node voltages are chosen from measured BER heatmaps and then held fixed across distances, so the 100-km point is a new measurement rather than a fitted prediction. The explicit limitation that 'the two ROSS nodes were emulated by sequentially measuring DSO traces for different voltages applied to the PIC' is a real experimental caveat — a true simultaneous two-node receiver would incur input-splitting loss and fixed synchronization — but this affects external validity, not circularity. Self-citations to [7] and [8] are present, but they are not used to forbid alternatives or to assert uniqueness; they are contextual and are supplemented by the present independent measurements. Accordingly, no circular step can be exhibited, and the score is 0.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The main free parameters are the node operating voltages, the chosen arm delays, the equalizer tap count, and the fixed power levels. The critical ad-hoc assumption is that sequential single-node captures can be synchronized offline to emulate a true two-node ROSS receiver; the paper's central reach claim depends on that assumption. No new physical entities are introduced.

free parameters (4)

Node heater voltages (phase shifts) for the two ROSS nodes = Selected per node from 0-3 V sweep; best pairs vary by design
The operating point of each node is chosen from measured BER heatmaps at 50 km; the headline BER depends on this post-hoc selection.
MZI arm time delay ΔT = approx. 44 ps (long) and 23 ps (short)
Chosen based on the authors' previous theoretical analysis [7] and programmable-PIC experiments [8]; a hand-selected design parameter that determines the spectral slicing.
FFE tap count = 62 T-spaced taps (31 symbols per filter)
Equalizer complexity chosen by the authors; performance comparisons keep it fixed across ROSS and reference equalizers.
Launch power and pre-PIC EDFA gain = 0.5 dBm launch; 11 dBm into PIC
Operating point fixed without shown optimization; BER values depend on these choices.

assumptions (4)

domain assumption KP4 hard-decision FEC threshold (BER ≤ 2.25e-4) and 400ZR C-FEC threshold (BER ≤ 1.25e-2) are accepted benchmarks for judging transmission success.
The paper compares measured BER against these standards from refs [11,12]; the 'below FEC' claim depends on these external thresholds.
ad hoc to paper Sequentially captured DSO traces from a single PIC at different heater voltages, after synchronization, are equivalent to a true simultaneous two-node ROSS receiver for BER measurement.
Stated in System performance: 'the two ROSS nodes were emulated by sequentially measuring DSO traces... constructed by synchronizing the traces.' This equivalence is assumed and is load-bearing for the 100-km claim.
domain assumption The optical noise in each sequential capture is statistically independent and the transmitted pattern is repeatable, so offline combination does not bias BER.
Required for the emulation to be valid; not stated or verified in the paper.
domain assumption Dispersion-induced power fading in C-band IM/DD is the dominant impairment and intensity-only detection (square law) is the correct channel model.
Standard model inherited from refs [1-3]; not tested in this work.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Integrated recurrent optical spectral slicer for equalization of 100-km C-band IM/DD transmission." pith.science (2026). https://pith.science/paper/R2CPUK6Z

@misc{pith2026250711654,
  author       = {Pith},
  title        = {Pith review of: Integrated recurrent optical spectral slicer for equalization of 100-km C-band IM/DD transmission},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R2CPUK6Z}},
  note         = {Machine review of arXiv:2507.11654}
}

read the original abstract

A silicon-photonics recurrent spectral filter is designed, fabricated, and system tested to pro-vide optical pre-processing in a 32-GBd PAM-4 C-band transmission. Performance below FEC is re-ported for up to 100 km reach.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 7 canonical work pages

[1]

Digital signal processing for short-reach optical commu- nications: A review of current technologies and future trends

K. Zhong, X. Zhou, J. Huo, C. Yu, C. Lu, A.P.T. Lau "Digital signal processing for short-reach optical commu- nications: A review of current technologies and future trends." Journal of Lightwave Technology 36 377 (2018) https://doi.org/10.1109/JLT.2018.2793881

arXiv 2018
[2]

Modulation format and digital signal processing for IM-DD optics at post-200G era

D. Che, X. Chen. "Modulation format and digital signal processing for IM-DD optics at post-200G era" Journal of Lightwave Technology 42 588 (2023) https://doi.org/10.1109/JLT.2023.3311716

work page arXiv 2023
[3]

Recent Advances on Machine Learning-aided DSP for Short-reach and Long-haul Optical Communications

L. Schmalen, V. Lauinger, J. Ney, N. Wehn, P. Matalla, S. Randel, A. von Bank, E.-M. Edelmann "Recent Ad- vances on Machine Learning-aided DSP for Short-reach and Long-haul Optical Communications" OFC 2025 https://arxiv.org/abs/2411.10101

work page Pith review arXiv 2025
[4]

Advanced Neural Network-Based Equalization in Intensity-Modulated Direct-Detection Op- tical Systems: Current Status and Future Trends

Z. Xu, T. Ji, Q. Wu, W. Lu, H. Ji, Y. Yang, G. Qiao, J. Tang, C. Ceng, L. Liu, S. Wang, J. Liang, Z. Sun, L. Fan, J. Wei, W. Shieh "Advanced Neural Network-Based Equalization in Intensity-Modulated Direct-Detection Op- tical Systems: Current Status and Future Trends." Pho- tonic 11 702 (2024) https://doi.org/10.3390/photon- ics11080702

work page doi:10.3390/photon- 2024
[5]

A neuro- morphic silicon photonics nonlinear equalizer for optical communications with intensity modulation and direct de- tection

A. Katumba, X. Yin, J. Dambre, P. Bienstman, "A neuro- morphic silicon photonics nonlinear equalizer for optical communications with intensity modulation and direct de- tection" Journal of Lightwave Technology 37 2232 (2019) https://doi.org/10.1109/JLT.2019.2900568

arXiv 2019
[6]

Experimental investigation of optoelectronic receiver with reservoir computing in short reach optical fiber com- munications

S.M. Ranzini, R. Dischler, F. Da Ros, H. Bülow, D. Zibar "Experimental investigation of optoelectronic receiver with reservoir computing in short reach optical fiber com- munications" Journal of Lightwave Technology 39 2460 (2021) https://doi.org/10.1109/JLT.2021.3049473

work page arXiv 2021
[7]

High-speed photonic neu- romorphic computing using recurrent optical spectrum slicing neural networks

K. Sozos, A. Bogris, P. Bienstman, G. Sarantoglou, S. Deligiannidis, C. Mesaritakis "High-speed photonic neu- romorphic computing using recurrent optical spectrum slicing neural networks" Comm Eng. 1.1 24 (2022) https://doi.org/10.1038/s44172-022-00024-5

doi:10.1038/s44172-022-00024-5 2022
[8]

Experimental Investigation of a Recurrent Optical Spectrum Slicing Receiver for Intensity Modulation/Direct Detection sys- tems using Programmable Photonics

K. Sozos, F. Da Ros, M.P. Yankov, G. Sarantoglou, S. Deligiannidis, C. Mesaritakis, A. Bogris, "Experimental Investigation of a Recurrent Optical Spectrum Slicing Receiver for Intensity Modulation/Direct Detection sys- tems using Programmable Photonics" Journal of Light- wave Technology 42 7807 (2024) https://doi.org/10.1109/JLT.2024.3430489

work page arXiv 2024

Show all 12 references

[9]

Chromatic Distortion Precompensation in OFDM-based Optical Systems Through an Integrated Silicon Photonic Neural Network

P.R.N. Marciano, E. Staffoli, G. Maddinelli, M.S. Coelho, L.C.B. Silva, J.A.L. Silva, M.J. Pontes, M.E.V. Segatto, L. Pavesi "Chromatic Distortion Precompensation in OFDM-based Optical Systems Through an Integrated Silicon Photonic Neural Network" Journal of Lightwave Technolo...

2025
[10]

Multipurpose self-configuration of program- mable photonic circuits,

D. Pérez-López, A. López, P. DasMahapatra, and J. Capmany, “Multipurpose self-configuration of program- mable photonic circuits,” Nature Communications 11 6359, (2020), https://doi.org/10.1038/s41467-020- 19608-w

2020 doi
[11]

Information-theoretic tools for optical communications engineers

E. Agrell, and M. Secondini. "Information-theoretic tools for optical communications engineers." 2018 IEEE Pho- tonics Conference (IPC). IEEE, 2018. https://doi.org/10.1109/IPCon.2018.8527126

2018
[12]

https://www.oiforum.com/wp-content/up- loads/OIF-400ZR-03.0.pdf

OIF ”OIF-400ZR-03.0 – Implementation Agreement 400ZR” (2024). https://www.oiforum.com/wp-content/up- loads/OIF-400ZR-03.0.pdf

2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Digital signal processing for short-reach optical commu- nications: A review of current technologies and future trends

K. Zhong, X. Zhou, J. Huo, C. Yu, C. Lu, A.P.T. Lau "Digital signal processing for short-reach optical commu- nications: A review of current technologies and future trends." Journal of Lightwave Technology 36 377 (2018) https://doi.org/10.1109/JLT.2018.2793881

arXiv 2018

[2] [2]

Modulation format and digital signal processing for IM-DD optics at post-200G era

D. Che, X. Chen. "Modulation format and digital signal processing for IM-DD optics at post-200G era" Journal of Lightwave Technology 42 588 (2023) https://doi.org/10.1109/JLT.2023.3311716

work page arXiv 2023

[3] [3]

Recent Advances on Machine Learning-aided DSP for Short-reach and Long-haul Optical Communications

L. Schmalen, V. Lauinger, J. Ney, N. Wehn, P. Matalla, S. Randel, A. von Bank, E.-M. Edelmann "Recent Ad- vances on Machine Learning-aided DSP for Short-reach and Long-haul Optical Communications" OFC 2025 https://arxiv.org/abs/2411.10101

work page Pith review arXiv 2025

[4] [4]

Advanced Neural Network-Based Equalization in Intensity-Modulated Direct-Detection Op- tical Systems: Current Status and Future Trends

Z. Xu, T. Ji, Q. Wu, W. Lu, H. Ji, Y. Yang, G. Qiao, J. Tang, C. Ceng, L. Liu, S. Wang, J. Liang, Z. Sun, L. Fan, J. Wei, W. Shieh "Advanced Neural Network-Based Equalization in Intensity-Modulated Direct-Detection Op- tical Systems: Current Status and Future Trends." Pho- tonic 11 702 (2024) https://doi.org/10.3390/photon- ics11080702

work page doi:10.3390/photon- 2024

[5] [5]

A neuro- morphic silicon photonics nonlinear equalizer for optical communications with intensity modulation and direct de- tection

A. Katumba, X. Yin, J. Dambre, P. Bienstman, "A neuro- morphic silicon photonics nonlinear equalizer for optical communications with intensity modulation and direct de- tection" Journal of Lightwave Technology 37 2232 (2019) https://doi.org/10.1109/JLT.2019.2900568

arXiv 2019

[6] [6]

Experimental investigation of optoelectronic receiver with reservoir computing in short reach optical fiber com- munications

S.M. Ranzini, R. Dischler, F. Da Ros, H. Bülow, D. Zibar "Experimental investigation of optoelectronic receiver with reservoir computing in short reach optical fiber com- munications" Journal of Lightwave Technology 39 2460 (2021) https://doi.org/10.1109/JLT.2021.3049473

work page arXiv 2021

[7] [7]

High-speed photonic neu- romorphic computing using recurrent optical spectrum slicing neural networks

K. Sozos, A. Bogris, P. Bienstman, G. Sarantoglou, S. Deligiannidis, C. Mesaritakis "High-speed photonic neu- romorphic computing using recurrent optical spectrum slicing neural networks" Comm Eng. 1.1 24 (2022) https://doi.org/10.1038/s44172-022-00024-5

doi:10.1038/s44172-022-00024-5 2022

[8] [8]

Experimental Investigation of a Recurrent Optical Spectrum Slicing Receiver for Intensity Modulation/Direct Detection sys- tems using Programmable Photonics

K. Sozos, F. Da Ros, M.P. Yankov, G. Sarantoglou, S. Deligiannidis, C. Mesaritakis, A. Bogris, "Experimental Investigation of a Recurrent Optical Spectrum Slicing Receiver for Intensity Modulation/Direct Detection sys- tems using Programmable Photonics" Journal of Light- wave Technology 42 7807 (2024) https://doi.org/10.1109/JLT.2024.3430489

work page arXiv 2024