Introducing Combined Effects of Filtering and ASE Noise in Optical Links Supposing Different Equalization Algorithms
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.
Rosa: We're looking at the paper: "Introducing Combined Effects of Filtering and ASE Noise in Optical Links Supposing Different Equalization Algorithms".
Dev: Here's the gist of the paper. This paper develops and validates a discrete-time modeling framework for the joint impact of cascaded optical filtering, distributed ASE noise, and transceiver noise in coherent optical links because it provides a tractable basis for quality-of-transmission estimation and optical-network digitaltwin implementations.
Rosa: The core of the paper is the derivation and validation of post-equalization SNR models for coherent systems affected by optical filtering.
Dev: Modeling the System Components The framework begins with an optical-link abstraction where cascaded filtering, distributed ASE noise, transceiver impairment, and equalization are treated within a single SNRbased formulation.
Rosa: This involves introducing a transceiver model that maps experimental back-to-back BER measurements into an equivalent transceiver SNR and combining this with the ASE contribution in the optical-link abstraction for analytical derivations.
Paper summary: Rosa: So, wrapping up this paper, "Introducing Combined Effects of Filtering and ASE Noise in Optical Links Supposing Different Equalization Algorithms," the authors developed a discrete time modeling framework to study how cascaded optical filtering, distributed ASE noise, and transceiver noise interact in coherent optical links. Dev The main contribution is that it provides a tractable way to estimate the post-equalization signal to noise ratio when you're using various equalization algorithms.
Taro: It claims they derived analytical expressions for several strategies: matched filter bound, zero-forcing equalizer, minimum mean square error equalizer, fractionally spaced equalizer, and finite length equalizer. Rosa They validated this by comparing the formulas against time domain simulation and experimental measurements with commercial transceivers and filters. Dev The results show that the filtering penalties depend on the filter shape, noise source position, receiver noise, and equalizer architecture instead of just a scalar bandwidth loss.
Rosa: This framework gives us a tractable basis for quality of transmission estimation and optical network digital twin implementations because it lets you evaluate filtering penalties accurately without relying on full timedomain simulation for every lightpath. Dev The authors also flag that the finite length equalizer has a computational complexity of about L cubed, which is the most demanding among those tested models.
Taro: The paper shows that when comparing models, increasing the LMS taps to sixty-four makes all their results converge, and they suggest that for practical accuracy in simulation, you should prefer the full finite length equalizer because it tracks the simulation results with a maximum error of zero point two nine decibels. Rosa So, in simple terms, this work gives us a way to plan optical networks where we can estimate how much penalty we’re going to see from filtering and noise without needing a massive computation for every single link.
Conclusion: Rosa: So, this paper is about taking all those noise types—the filtering, the noise from amplifiers, and the receiver noise—and putting them together to see how they mess with your signal quality when you're using different ways to clean it up.
Dev: Right. It’s a modeling framework that lets you figure out exactly what happens at the output of an equalizer when all those things are fighting each other in a coherent optical link.
Rosa: The authors did this by taking the raw measurements from experimental setups and building these mathematical models for things like matched filter bound, zero-forcing, MMSE, and finite length equalizers.
Dev: And they validated these against real time domain simulations and actual hardware measurements with commercial gear. It shows that the penalty you get isn't just a simple number related to how narrow your filter is.
Rosa: It means the penalty depends on where the noise is coming from, what kind of filter you use, and how complicated your equalizer setup actually is.
Dev: That’s significant because it gives us a better handle on planning optical networks or even building digital twins for those networks because we stop assuming everything scales linearly.
Rosa: Exactly. It’s a way to estimate quality of transmission much more accurately without having to run super long simulations every single time you change one little parameter in the link.
Dev: But the paper also points out that while the finite length model is really accurate, it gets computationally heavy if you try to use it for everything.
Rosa: Yeah, so they give us a trade-off between being super accurate and being able to run calculations fast enough for real-time planning tools.
Dev: It’s a good starting point for understanding the actual noise floor in complex optical systems.
Politecnico di Torino · LINKS Foundation · SM-Optics
eess.SP
Submitted: 2025-04-24
Updated: 2026-06-26
Journal ref: IEEE Open Journal of the Communications Society, vol. 7, pp. 9405-9430, 2026
DOI: 10.1109/OJCOMS.2026.3719514
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 66/100
The gist: This paper develops and validates a discrete-time modeling framework for the joint impact of cascaded optical filtering, distributed ASE noise…
Key concepts
- Optical Link Abstraction
- This is a simplified model that treats all impairments—like filtering, noise, and equalization—as part of a single Signal-to-Noise Ratio (SNR) calculation. It maps experimental measurements to an equivalent SNR to allow for analytical derivations of system performance under different conditions.
- Equalization Strategies
- The paper analyzes several types of equalizers, including Matched Filter Bound (MFB), Zero-Forcing Equalizer (ZFE), MMSE, and Fractionally Spaced Equalizer (FSE). These models are used to derive how different equalization methods handle the combined effects of filtering and noise in a coherent system.
- Filtering Penalties
- These penalties quantify the performance loss caused by optical filtering. The study shows that this penalty is not just a simple scalar bandwidth loss but depends on the filter shape, where noise sources are located, receiver noise, and the specific equalizer architecture used.
Terminology
Summary
This paper develops and validates a discrete-time modeling framework for the joint impact of cascaded optical filtering, distributed ASE noise, and transceiver noise in coherent optical links because it provides a tractable basis for quality-of-transmission estimation and optical-network digitaltwin implementations. p. 0 rosso@polito. it
3
rd Emanuele Virgillito
DET, Politecnico di Torino
Turin, Italy
4
th Stefano Straullu
LINKS Foundation
Turin, Italy
5
th Andrea Castoldi
SM-Optics
Cologno Monzese, Italy
6
th Andrea Bovio
SM-Optics
Cologno Monzese, Italy
7
th Francisco Martinez Rodriguez
SM-Optics
Cologno Monzese, Italy
8
th Vittorio Curri
DET, Politecnico di Torino
Turin, Italy
Abstract—This paper develops and validates a discrete-time
modeling framework for the joint impact of cascaded optical
filtering, distributed ASE noise, and transceiver noise in coherent
optical links. The objective of
this work is therefore to provide a modeling framework in
which cascaded filtering, distributed ASE noise, transceiver
impairment, and equalization are treated within a single SNRbased formulation. The core of the paper is the derivation and validation of
post-equalization SNR models for coherent systems affected
by optical filtering.Where the paper says it
Where the paper says it
Modeling the System Components
The framework begins with an optical-link abstraction where cascaded filtering, distributed ASE noise, transceiver impairment, and equalization are treated within a single SNRbased formulation. This involves introducing a transceiver model that maps experimental back-to-back BER measurements into an equivalent transceiver SNR and combining this with the ASE contribution in the optical-link abstraction for analytical derivations. p. 0 We first introduce a transceiver model
that maps experimental back-to-back BER measurements into
an equivalent transceiver SNR, following the approach of
[4], [5] and its use in [6]. This transceiver contribution is
then combined with the ASE contribution in the optical-link
abstraction used for the analytical derivations. A key point of the model is that ASE contributions injected
at different positions experience different downstream filters,
while the transceiver noise is added at the receiver. As a
consequence, the total noise at the equalizer input is generally
colored, and the equalizer model must account for both signal
distortion and noise enhancement.Where the paper says it
Where the paper says it
Derivation of Equalizer Models
The analytical derivations yield models for several equalization strategies starting from the optical link abstraction. These include:
-
Matched Filter Bound (MFB)
-
Zero-Forcing Equalizer (ZFE)
-
Minimum Mean Square Error Equalizer (MMSE)
-
Fractionally Spaced Equalizer (FSE)
-
Finite Length Equalizer (FLE)
The derivation is performed in discrete time, consistent with the operation of commercial coherent receivers after sampling. The model incorporates the transceiver impairment and ASE contribution into a unified SNR formulation used for evaluating filtering penalties. p. 1 The considered
equalizers, including ZFE, MMSE, FSE, and FLE, operate
after sampling, and the derivation admits arbitrary opticalfilter shapes, arbitrary placement of ASE-noise sources, and
colored transceiver noise. The transceiver impairment model
is embedded in the same SNR formulation used for the
optical link, so filtering penalties can be evaluated together
with receiver limitations.Where the paper says it
Analysis of Equalization Strategies
The paper examines the performance characteristics of these models, highlighting their trade-offs and complexity.
-
Infinite-length equalizers (ZFE, MMSE, FSE) are analyzed using a whitenoise equivalent link model which allows for the derivation of the SNR after equalization and the filtering penalty coefficient for all considered strategies. p. 1
Where the paper says it
The considered equalizers, including ZFE, MMSE, FSE, and FLE, operate after sampling, and the derivation admits arbitrary opticalfilter shapes, arbitrary placement of ASE-noise sources, and colored transceiver noise. The transceiver impairment model is embedded in the same SNR formulation used for the optical link, so filtering penalties can be evaluated together with receiver limitations.
-
The finite-length models capture the penalty observed when the LMS equalizer has a limited number of taps, and it is shown that the whitened finite-length approximation can be accurate in several practical cases while the full colored-noise FLE remains the most faithful model when implementation details must be represented more precisely. p. 1
Where the paper says it
Infinitelength models provide fast and useful estimates when the filtering condition is not extreme, while the finite-length formulation captures the limited-tap penalty observed in practical DSP. This makes the proposed approach suitable for qualityof-transmission estimation and for integration into opticalnetwork DT transmission models, where filtering penalties must be evaluated accurately without relying on full timedomain simulation for every lightpath.
Validation and Comparison
The analytical models are validated through two complementary ways and experimental measurements with commercial transceivers and accurately modeled optical filters.
-
Time-domain simulation is used to compare the proposed formulas with LMS-based receiver equalization under controlled and severe filtering scenarios, allowing variation of taps, equivalent bandwidth, and ASE-noise positions. p. 1
Where the paper says it
First, time-domain simulation is used to compare the proposed formulas with LMS-based receiver equalization under controlled and severe filtering scenarios. This validation exploits the simulation framework already used in [7] and allows the number of taps, the equivalent bandwidth of the filter cascade, and the position of the ASE-noise sources to be varied independently.
-
Experimental measurements with commercial transceivers and accurately modeled optical filters are used to validate the MMSE and FSE predictions in realistic conditions. p. 1
Where the paper says it
This validation exploits the simulation framework already used in [7] and allows the number of taps, the equivalent bandwidth of the filter cascade, and the position of the ASE-noise sources to be varied independently. Second, experimental measurements with commercial transceivers and accurately modeled optical filters, following the setup of [6], are used to validate the MMSE and FSE predictions in realistic conditions.
The results show that filtering penalties depend on filter shape, noise-source position, receiver noise, and equalizer architecture rather than just a scalar bandwidth loss. p. 1 First, time-domain simulation is used to compare the
proposed formulas with LMS-based receiver equalization under controlled and severe filtering scenarios. This validation
exploits the simulation framework already used in [7] and
allows the number of taps, the equivalent bandwidth of the
filter cascade, and the position of the ASE-noise sources to
be varied independently.Where the paper says it
Conclusion on Model Selection
The framework provides a tractable way to include all these effects in system-level planning tools and optical-network digital twins. The ZFE provides a simpler and conservative estimate when the filtering cascade is not excessively tight, which is useful for low-complexity quality-of-transmission evaluation. p. 0 The resulting framework provides a tractable basis for
quality-of-transmission estimation and optical-network digitaltwin implementations. Index Terms—filtering penalty, equalization, MMSE, mathematical modeling
I. Beyond reviewing these models, we generalize them to
the case of arbitrary transceiver-noise PSD, with the goal of
providing a rigorous and practically relevant tool for ONDT
implementation. For each model, we highlight its main strengths, limitations,
and simplifying assumptions, in order to clarify when a
given abstraction should be preferred and how the appropriate
level of complexity can be selected according to the target
application, since in many cases a simpler model may provide
sufficient accuracy without requiring a more realistic and
computationally demanding description. Metro-access scenarios with a single ROADM and a single
noise source have also been investigated experimentally [12]. In addition, the influence of optical-filter frequency shifts and
bandwidth variations on the Q-factor has been studied through
experiments and simulations [13], showing that asymmetric
and bandwidth-dependent filtering effects must be represented
carefully when estimating system performance. Metro-access scenarios with a single ROADM and a single
noise source have also been investigated experimentally [12]. In addition, the influence of optical-filter frequency shifts and
bandwidth variations on the Q-factor has been studied through
experiments and simulations [13], showing that asymmetric
and bandwidth-dependent filtering effects must be represented
carefully when estimating system performance. Metro-access scenarios with a single ROADM and a single
noise source have also been investigated experimentally [12]. In addition, the influence of optical-filter frequency shifts and
bandwidth variations on the Q-factor has been studied through
experiments and simulations [13], showing that asymmetric
and bandwidth-dependent filtering effects must be represented
carefully when estimating system performance.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Experimental Findings
The experimental validation with commercial transceivers and ROADM filters further supports the proposed approach. The MMSE and FSE curves reproduce the measured Q-factor trends with good agreement and a conservative margin mainly associated with the filter-fitting strategy and with practical DSP implementation limits. p. 0 Timedomain simulations with LMS equalization validate the analytical
predictions over severe filtering conditions, different tap lengths,
and different ASE-noise positions along the link. Experimental results with commercial transceivers and ROADMs further confirm
the accuracy of the MMSE and FSE models, while highlighting
the role of realistic filter modeling and equalizer implementation
limits. Moreover, practical adaptive implementations of the FSE
based on LMS or constant modulus algorithms have been
shown to converge to MMSEE-equalizer performance [24]. In
the present analysis, timing-related effects are not considered;
therefore, the possible additional benefit of the FSE observed
in practical implementations, in particular its improved robustness to sampling-phase errors[8], [9], is beyond the scope of
this discussion. Moreover, practical adaptive implementations of the FSE
based on LMS or constant modulus algorithms have been
shown to converge to MMSEE-equalizer performance [24]. In
the present analysis, timing-related effects are not considered;
therefore, the possible additional benefit of the FSE observed
in practical implementations, in particular its improved robustness to sampling-phase errors[8], [9], is beyond the scope of
this discussion.Where the paper says it
Where the paper says it
Where the paper says it
Computational Complexity Summary
The computational complexity of the FLE is approximately O(M(Nf l)3), being O((Nf l)3), which makes it the most computationally demanding among the considered equalization models. On the other hand, the WFLE does not require matrix multiplication for all the noise sources, therefore its computational complexity is O((Nf l)3). p. 8 The expression of the noise variance after the equalizer in
(47) depends on the vector of the equalizer coefficients w. w = RxY · R
−1
YY (48)
To express σ
2
MMSEE-WFLE and w, it is necessary to compute the correlation matrices RYY and RxY, which are the
autocorrelation matrix of the output samples and the crosscorrelation matrix of the input-output samples, respectively. INFINITE LENGTH EQUALIZERS
In this section we examine the infinite length equalization
models, namely the ZFE, the MMSEE and the FSE. Regarding
ZFE and MMSEE, some additional assumptions and definitions are needed, since these models are based on matched
filter theory.Where the paper says it
Where the paper says it
Model Comparison Summary
The comparison shows that increasing the LMS taps to 64 makes all the results converge. The WFLE is conservative with respect to timedomain simulation, with a maximum error of 0. p. 1 Infinitelength models provide fast and useful estimates when the
filtering condition is not extreme, while the finite-length formulation captures the limited-tap penalty observed in practical
DSP. This makes the proposed approach suitable for qualityof-transmission estimation and for integration into opticalnetwork DT transmission models, where filtering penalties
must be evaluated accurately without relying on full timedomain simulation for every lightpath. Infinitelength models provide fast and useful estimates when the
filtering condition is not extreme, while the finite-length formulation captures the limited-tap penalty observed in practical
DSP. This makes the proposed approach suitable for qualityof-transmission estimation and for integration into opticalnetwork DT transmission models, where filtering penalties
must be evaluated accurately without relying on full timedomain simulation for every lightpath. Beyond reviewing these models, we generalize them to
the case of arbitrary transceiver-noise PSD, with the goal of
providing a rigorous and practically relevant tool for ONDT
implementation. For each model, we highlight its main strengths, limitations,
and simplifying assumptions, in order to clarify when a
given abstraction should be preferred and how the appropriate
level of complexity can be selected according to the target
application, since in many cases a simpler model may provide
sufficient accuracy without requiring a more realistic and
computationally demanding description. SNR =
ROP
PASE + PTRX
=
4 · E¯
x · Rs · ||h||2
4 · Ntot · Rs
=
E¯
x · ||h||2
Ntot
(27)
where we point out that bandwidth terms cancel out since both
signal and noise power are computed with respect to the same
band equal to the symbol rate. INFINITE LENGTH EQUALIZERS
In this section we examine the infinite length equalization
models, namely the ZFE, the MMSEE and the FSE.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Experimental Performance Summary
The experimental validation of MMSE and FSE equalizers reproduce the measured Q-factor trends with good agreement and a conservative margin mainly associated with the filter-fitting strategy and with practical DSP implementation limits. The overlap between MMSE and FSE predictions in the absence of sampling-phase errors is consistent with the theoretical derivation, while the higher penalty observed in more demanding bitrate configurations is captured through the transceiver characterization. The results show that filtering penalties cannot be represented only by a scalar bandwidth loss.
Final Summary on Penalties
The resulting framework provides a tractable basis for quality-of-transmission estimation and optical-network digitaltwin implementations. p. 0 Experimental results with commercial transceivers and ROADMs further confirm
the accuracy of the MMSE and FSE models, while highlighting
the role of realistic filter modeling and equalizer implementation
limits. The resulting framework provides a tractable basis for
quality-of-transmission estimation and optical-network digitaltwin implementations. Experimental results with commercial transceivers and ROADMs further confirm
the accuracy of the MMSE and FSE models, while highlighting
the role of realistic filter modeling and equalizer implementation
limits. The resulting framework provides a tractable basis for
quality-of-transmission estimation and optical-network digitaltwin implementations. Experimental results with commercial transceivers and ROADMs further confirm
the accuracy of the MMSE and FSE models, while highlighting
the role of realistic filter modeling and equalizer implementation
limits. The resulting framework provides a tractable basis for
quality-of-transmission estimation and optical-network digitaltwin implementations. Experimental results with commercial transceivers and ROADMs further confirm
the accuracy of the MMSE and FSE models, while highlighting
the role of realistic filter modeling and equalizer implementation
limits. The resulting framework provides a tractable basis for
quality-of-transmission estimation and optical-network digitaltwin implementations. Experimental results with commercial transceivers and ROADMs further confirm
the accuracy of the MMSE and FSE models, while highlighting
the role of realistic filter modeling and equalizer implementation
limits. The resulting framework provides a tractable basis for
quality-of-transmission estimation and optical-network digitaltwin implementations. Experimental results with commercial transceivers and ROADMs further confirm
the accuracy of the MMSE and FSE models, while highlighting
the role of realistic filter modeling and equalizer implementation
limits. The resulting framework provides a tractable basis for
quality-of-transmission estimation and optical-network digitaltwin implementations.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Final Summary on Model Selection
The resulting framework provides a tractable basis for quality-of-transmission estimation and optical-network digitaltwin implementations. The resulting framework provides a tractable basis for quality-of-transmission estimation and optical-network digitaltwin implementations. The resulting framework provides a tractable basis for quality-of-transmission estimation and optical-network digitaltwin implementations. The resulting framework provides a tractable basis for quality-of-transmission estimation and optical-network digitaltwin implementations.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Develop optical network digital twins by incorporating filtering penalties into transmission models as described in
The resulting framework provides a tractable basis for quality-of-transmission estimation and optical-network digitaltwin implementations.
This allows DTs to evaluate filtering effects accurately without relying on full timedomain simulation for every lightpath. -
Implement a unified SNRbased formulation that couples cascaded optical filtering, distributed ASE noise, and transceiver noise into a single metric. This enables the evaluation of optical-link penalties and receiver impairments within the same mathematical framework.
-
Design equalization algorithms with explicit colored-noise treatment to derive analytical expressions for Matched Filter Bound, Zero-Forcing Equalization, Minimum Mean Square Error Equalization, Fractionally Spaced Equalization, and Finite Length Equalizer models. This provides a tractable basis for quality-of-transmission estimation and optical network digital twin implementations.
-
Utilize time-domain simulations with LMS equalization to validate analytical predictions over severe filtering conditions and different tap lengths, noise positions along the link, and filter cascade bandwidths. This allows for testing the developed models on a scale unachievable in a laboratory environment.
-
Create a measurement-based transceiver SNR characterization methodology that maps experimental back-to-back BER measurements into an equivalent transceiver SNR using fitting formulas like
SNRTRX = N · ROP / (ROP + D)
and combines it with the Gaussian Channel model for total system SNR. This allows for the accurate modeling of receiver impairments in coherent systems. -
Integrate noise whitening techniques to derive models for ZFE, MMSEE, and FSE equalizers without loss of information, exploiting the
white-noise equivalent channel representation.
This simplifies modeling by deriving quantities likeH(f), Stot(f) and Ntot
which represent the white-noise equivalent channel frequency response. -
Develop a flexible framework for Quality of Transmission estimation that allows for comparison between different equalizer models by considering their computational complexity and accuracy tradeoffs, such as comparing
infinite-length models
withfinite-length formulations.
Abstract
This paper develops and validates a discrete-time modeling framework for the joint impact of cascaded optical filtering, distributed ASE noise, and transceiver noise in coherent optical links. The work focuses on the post-equalization signal-to-noise ratio, which is the central quantity used to quantify filtering penalty under receiver DSP. Starting from an optical-link abstraction with arbitrary filter transfer functions and colored noise spectra, we derive analytical expressions for the matched-filter bound, Zero-Forcing Equalization, Minimum Mean Square Error Equalization, Fractionally Spaced Equalization, and finite-length equalizers with and without explicit colored-noise treatment. The model is coupled to a measurement-based transceiver SNR characterization, so that optical-link penalties and receiver impairments can be evaluated within the same formulation. Time-domain simulations with LMS equalization validate the analytical predictions over severe filtering conditions, different tap lengths, and different ASE-noise positions along the link. Experimental results with commercial transceivers and ROADMs further confirm the accuracy of the MMSE and FSE models, while highlighting the role of realistic filter modeling and equalizer implementation limits. The resulting framework provides a tractable basis for quality-of-transmission estimation and optical-network digital-twin implementations.
Related papers
- Open-H-RF: A Large-Scale Ultrasound Raw Channel Dataset Initiative
- Scalable Attention for 5G NR Channel Estimation
- Low-Altitude Wireless Networks: A Comprehensive Survey
- Sensing-Assisted OFDM with Index Modulation: A Waveform for Optical Wireless Integrated Communication and Localisation
- Iteration-Free Candidate Pruning and an Index-Domain VLSI Architecture for Sort-Free MIMO Box Decoding up to 1024-QAM
- A Foundation Model for Large-Scale Wireless Network Planning, Operation and Optimization