Reducing Computational Complexity of Neural Networks in Optical Channel Equalization: From Concepts to Implementation
This paper introduces a novel methodology for developing low-complexity neural network (NN) based equalizers to address impairments in high-speed coherent optical transmission systems. We present a comprehensive exploration and comparison of deep model compression techniques applied to feed-forward...
Gespeichert in:
Veröffentlicht in: | Journal of lightwave technology 2023-07, Vol.41 (14), p.4557-4581 |
---|---|
Hauptverfasser: | , , , , , , , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
container_end_page | 4581 |
---|---|
container_issue | 14 |
container_start_page | 4557 |
container_title | Journal of lightwave technology |
container_volume | 41 |
creator | Freire, Pedro J. Napoli, Antonio Spinnler, Bernhard Anderson, Michael Ron, Diego Arguello Schairer, Wolfgang Bex, Thomas Costa, Nelson Turitsyn, Sergei K. Prilepsky, Jaroslaw E. |
description | This paper introduces a novel methodology for developing low-complexity neural network (NN) based equalizers to address impairments in high-speed coherent optical transmission systems. We present a comprehensive exploration and comparison of deep model compression techniques applied to feed-forward and recurrent NN designs, assessing their impact on equalizer performance. Our investigation encompasses quantization, weight clustering, pruning, and other cutting-edge compression strategies. We propose and evaluate a Bayesian optimization-assisted compression approach that optimizes hyperparameters to simultaneously enhance performance and reduce complexity. Additionally, we introduce four distinct metrics (RMpS, BoP, NABS, and NLGs) to quantify computing complexity in various compression algorithms. These metrics serve as benchmarks for evaluating the relative effectiveness of NN equalizers when compression approaches are employed. The analysis is completed by evaluating the trade-off between compression complexity and performance using simulated and experimental data. By employing optimal compression techniques, we demonstrate the feasibility of designing a simplified NN-based equalizer surpassing the performance of conventional digital back-propagation (DBP) equalizers with only one step per span. This is achieved by reducing the number of multipliers through weighted clustering and pruning algorithms. Furthermore, we highlight that an NN-based equalizer can achieve better performance than the full electronic chromatic dispersion compensation block while maintaining a similar level of complexity. In conclusion, we outline remaining challenges, unanswered questions, and potential avenues for future research in this field. |
doi_str_mv | 10.1109/JLT.2023.3234327 |
format | Article |
fullrecord | <record><control><sourceid>proquest_ieee_</sourceid><recordid>TN_cdi_ieee_primary_10006884</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><ieee_id>10006884</ieee_id><sourcerecordid>2839516174</sourcerecordid><originalsourceid>FETCH-LOGICAL-c334t-4f73efa906dc375890cc930b18a3856ccc6ffb349e938396248c8671d14df7453</originalsourceid><addsrcrecordid>eNpNkMFLwzAUh4MoOKd3Dx4CnjuTvLRJvcnYdDI2kHkuXZpqZ9t0SYrOv9523cFTeOH7fo_3Q-iWkgmlJH54XW4mjDCYAAMOTJyhEQ1DGTBG4RyNiAAIpGD8El05tyOEci7FCLVvOmtVUX_gqama1qe-MHVaHqdS_xT-gE2OV7q13edK-29jvxwuarxufKF68DOta13i2b5Ny-L36D_iuTVVl1Er3XiHvcGLPq7S9bDgGl3kaen0zekdo_f5bDN9CZbr58X0aRkoAO4DngvQeRqTKFMgQhkTpWIgWypTkGGklIryfAs81jFIiCPGpZKRoBnlWS54CGN0P-Q21uxb7XyyM63t7nMJ64SQRlTwjiIDpaxxzuo8aWxRpfaQUJL05SZduUlfbnIqt1PuBqXQWv_DCYmk5PAHNAB23Q</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2839516174</pqid></control><display><type>article</type><title>Reducing Computational Complexity of Neural Networks in Optical Channel Equalization: From Concepts to Implementation</title><source>IEEE Electronic Library (IEL)</source><creator>Freire, Pedro J. ; Napoli, Antonio ; Spinnler, Bernhard ; Anderson, Michael ; Ron, Diego Arguello ; Schairer, Wolfgang ; Bex, Thomas ; Costa, Nelson ; Turitsyn, Sergei K. ; Prilepsky, Jaroslaw E.</creator><creatorcontrib>Freire, Pedro J. ; Napoli, Antonio ; Spinnler, Bernhard ; Anderson, Michael ; Ron, Diego Arguello ; Schairer, Wolfgang ; Bex, Thomas ; Costa, Nelson ; Turitsyn, Sergei K. ; Prilepsky, Jaroslaw E.</creatorcontrib><description>This paper introduces a novel methodology for developing low-complexity neural network (NN) based equalizers to address impairments in high-speed coherent optical transmission systems. We present a comprehensive exploration and comparison of deep model compression techniques applied to feed-forward and recurrent NN designs, assessing their impact on equalizer performance. Our investigation encompasses quantization, weight clustering, pruning, and other cutting-edge compression strategies. We propose and evaluate a Bayesian optimization-assisted compression approach that optimizes hyperparameters to simultaneously enhance performance and reduce complexity. Additionally, we introduce four distinct metrics (RMpS, BoP, NABS, and NLGs) to quantify computing complexity in various compression algorithms. These metrics serve as benchmarks for evaluating the relative effectiveness of NN equalizers when compression approaches are employed. The analysis is completed by evaluating the trade-off between compression complexity and performance using simulated and experimental data. By employing optimal compression techniques, we demonstrate the feasibility of designing a simplified NN-based equalizer surpassing the performance of conventional digital back-propagation (DBP) equalizers with only one step per span. This is achieved by reducing the number of multipliers through weighted clustering and pruning algorithms. Furthermore, we highlight that an NN-based equalizer can achieve better performance than the full electronic chromatic dispersion compensation block while maintaining a similar level of complexity. In conclusion, we outline remaining challenges, unanswered questions, and potential avenues for future research in this field.</description><identifier>ISSN: 0733-8724</identifier><identifier>EISSN: 1558-2213</identifier><identifier>DOI: 10.1109/JLT.2023.3234327</identifier><identifier>CODEN: JLTEDG</identifier><language>eng</language><publisher>New York: IEEE</publisher><subject>Algorithms ; Artificial neural networks ; Back propagation networks ; Bayesian optimizer ; Clustering ; coherent detection ; Complexity ; computational complexity ; Computational modeling ; Equalizers ; Fiber nonlinear optics ; neural network ; Neural networks ; nonlinear equalizer ; Nonlinear optics ; Optical communication ; Optical fibers ; Optimization ; pruning ; quantization ; Symbols</subject><ispartof>Journal of lightwave technology, 2023-07, Vol.41 (14), p.4557-4581</ispartof><rights>Copyright The Institute of Electrical and Electronics Engineers, Inc. (IEEE) 2023</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><citedby>FETCH-LOGICAL-c334t-4f73efa906dc375890cc930b18a3856ccc6ffb349e938396248c8671d14df7453</citedby><cites>FETCH-LOGICAL-c334t-4f73efa906dc375890cc930b18a3856ccc6ffb349e938396248c8671d14df7453</cites><orcidid>0000-0001-7417-9398 ; 0000-0002-8678-5691 ; 0000-0002-9264-9274 ; 0000-0002-6004-385X ; 0000-0003-0101-3834 ; 0000-0002-3035-4112 ; 0000-0003-3145-1018 ; 0000-0001-9578-0297</orcidid></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://ieeexplore.ieee.org/document/10006884$$EHTML$$P50$$Gieee$$Hfree_for_read</linktohtml><link.rule.ids>314,780,784,796,27924,27925,54758</link.rule.ids></links><search><creatorcontrib>Freire, Pedro J.</creatorcontrib><creatorcontrib>Napoli, Antonio</creatorcontrib><creatorcontrib>Spinnler, Bernhard</creatorcontrib><creatorcontrib>Anderson, Michael</creatorcontrib><creatorcontrib>Ron, Diego Arguello</creatorcontrib><creatorcontrib>Schairer, Wolfgang</creatorcontrib><creatorcontrib>Bex, Thomas</creatorcontrib><creatorcontrib>Costa, Nelson</creatorcontrib><creatorcontrib>Turitsyn, Sergei K.</creatorcontrib><creatorcontrib>Prilepsky, Jaroslaw E.</creatorcontrib><title>Reducing Computational Complexity of Neural Networks in Optical Channel Equalization: From Concepts to Implementation</title><title>Journal of lightwave technology</title><addtitle>JLT</addtitle><description>This paper introduces a novel methodology for developing low-complexity neural network (NN) based equalizers to address impairments in high-speed coherent optical transmission systems. We present a comprehensive exploration and comparison of deep model compression techniques applied to feed-forward and recurrent NN designs, assessing their impact on equalizer performance. Our investigation encompasses quantization, weight clustering, pruning, and other cutting-edge compression strategies. We propose and evaluate a Bayesian optimization-assisted compression approach that optimizes hyperparameters to simultaneously enhance performance and reduce complexity. Additionally, we introduce four distinct metrics (RMpS, BoP, NABS, and NLGs) to quantify computing complexity in various compression algorithms. These metrics serve as benchmarks for evaluating the relative effectiveness of NN equalizers when compression approaches are employed. The analysis is completed by evaluating the trade-off between compression complexity and performance using simulated and experimental data. By employing optimal compression techniques, we demonstrate the feasibility of designing a simplified NN-based equalizer surpassing the performance of conventional digital back-propagation (DBP) equalizers with only one step per span. This is achieved by reducing the number of multipliers through weighted clustering and pruning algorithms. Furthermore, we highlight that an NN-based equalizer can achieve better performance than the full electronic chromatic dispersion compensation block while maintaining a similar level of complexity. In conclusion, we outline remaining challenges, unanswered questions, and potential avenues for future research in this field.</description><subject>Algorithms</subject><subject>Artificial neural networks</subject><subject>Back propagation networks</subject><subject>Bayesian optimizer</subject><subject>Clustering</subject><subject>coherent detection</subject><subject>Complexity</subject><subject>computational complexity</subject><subject>Computational modeling</subject><subject>Equalizers</subject><subject>Fiber nonlinear optics</subject><subject>neural network</subject><subject>Neural networks</subject><subject>nonlinear equalizer</subject><subject>Nonlinear optics</subject><subject>Optical communication</subject><subject>Optical fibers</subject><subject>Optimization</subject><subject>pruning</subject><subject>quantization</subject><subject>Symbols</subject><issn>0733-8724</issn><issn>1558-2213</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>ESBDL</sourceid><sourceid>RIE</sourceid><recordid>eNpNkMFLwzAUh4MoOKd3Dx4CnjuTvLRJvcnYdDI2kHkuXZpqZ9t0SYrOv9523cFTeOH7fo_3Q-iWkgmlJH54XW4mjDCYAAMOTJyhEQ1DGTBG4RyNiAAIpGD8El05tyOEci7FCLVvOmtVUX_gqama1qe-MHVaHqdS_xT-gE2OV7q13edK-29jvxwuarxufKF68DOta13i2b5Ny-L36D_iuTVVl1Er3XiHvcGLPq7S9bDgGl3kaen0zekdo_f5bDN9CZbr58X0aRkoAO4DngvQeRqTKFMgQhkTpWIgWypTkGGklIryfAs81jFIiCPGpZKRoBnlWS54CGN0P-Q21uxb7XyyM63t7nMJ64SQRlTwjiIDpaxxzuo8aWxRpfaQUJL05SZduUlfbnIqt1PuBqXQWv_DCYmk5PAHNAB23Q</recordid><startdate>20230715</startdate><enddate>20230715</enddate><creator>Freire, Pedro J.</creator><creator>Napoli, Antonio</creator><creator>Spinnler, Bernhard</creator><creator>Anderson, Michael</creator><creator>Ron, Diego Arguello</creator><creator>Schairer, Wolfgang</creator><creator>Bex, Thomas</creator><creator>Costa, Nelson</creator><creator>Turitsyn, Sergei K.</creator><creator>Prilepsky, Jaroslaw E.</creator><general>IEEE</general><general>The Institute of Electrical and Electronics Engineers, Inc. (IEEE)</general><scope>97E</scope><scope>ESBDL</scope><scope>RIA</scope><scope>RIE</scope><scope>AAYXX</scope><scope>CITATION</scope><scope>7SP</scope><scope>7U5</scope><scope>8FD</scope><scope>H8D</scope><scope>L7M</scope><orcidid>https://orcid.org/0000-0001-7417-9398</orcidid><orcidid>https://orcid.org/0000-0002-8678-5691</orcidid><orcidid>https://orcid.org/0000-0002-9264-9274</orcidid><orcidid>https://orcid.org/0000-0002-6004-385X</orcidid><orcidid>https://orcid.org/0000-0003-0101-3834</orcidid><orcidid>https://orcid.org/0000-0002-3035-4112</orcidid><orcidid>https://orcid.org/0000-0003-3145-1018</orcidid><orcidid>https://orcid.org/0000-0001-9578-0297</orcidid></search><sort><creationdate>20230715</creationdate><title>Reducing Computational Complexity of Neural Networks in Optical Channel Equalization: From Concepts to Implementation</title><author>Freire, Pedro J. ; Napoli, Antonio ; Spinnler, Bernhard ; Anderson, Michael ; Ron, Diego Arguello ; Schairer, Wolfgang ; Bex, Thomas ; Costa, Nelson ; Turitsyn, Sergei K. ; Prilepsky, Jaroslaw E.</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c334t-4f73efa906dc375890cc930b18a3856ccc6ffb349e938396248c8671d14df7453</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Algorithms</topic><topic>Artificial neural networks</topic><topic>Back propagation networks</topic><topic>Bayesian optimizer</topic><topic>Clustering</topic><topic>coherent detection</topic><topic>Complexity</topic><topic>computational complexity</topic><topic>Computational modeling</topic><topic>Equalizers</topic><topic>Fiber nonlinear optics</topic><topic>neural network</topic><topic>Neural networks</topic><topic>nonlinear equalizer</topic><topic>Nonlinear optics</topic><topic>Optical communication</topic><topic>Optical fibers</topic><topic>Optimization</topic><topic>pruning</topic><topic>quantization</topic><topic>Symbols</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Freire, Pedro J.</creatorcontrib><creatorcontrib>Napoli, Antonio</creatorcontrib><creatorcontrib>Spinnler, Bernhard</creatorcontrib><creatorcontrib>Anderson, Michael</creatorcontrib><creatorcontrib>Ron, Diego Arguello</creatorcontrib><creatorcontrib>Schairer, Wolfgang</creatorcontrib><creatorcontrib>Bex, Thomas</creatorcontrib><creatorcontrib>Costa, Nelson</creatorcontrib><creatorcontrib>Turitsyn, Sergei K.</creatorcontrib><creatorcontrib>Prilepsky, Jaroslaw E.</creatorcontrib><collection>IEEE All-Society Periodicals Package (ASPP) 2005-present</collection><collection>IEEE Open Access Journals</collection><collection>IEEE All-Society Periodicals Package (ASPP) 1998-Present</collection><collection>IEEE Electronic Library (IEL)</collection><collection>CrossRef</collection><collection>Electronics & Communications Abstracts</collection><collection>Solid State and Superconductivity Abstracts</collection><collection>Technology Research Database</collection><collection>Aerospace Database</collection><collection>Advanced Technologies Database with Aerospace</collection><jtitle>Journal of lightwave technology</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Freire, Pedro J.</au><au>Napoli, Antonio</au><au>Spinnler, Bernhard</au><au>Anderson, Michael</au><au>Ron, Diego Arguello</au><au>Schairer, Wolfgang</au><au>Bex, Thomas</au><au>Costa, Nelson</au><au>Turitsyn, Sergei K.</au><au>Prilepsky, Jaroslaw E.</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Reducing Computational Complexity of Neural Networks in Optical Channel Equalization: From Concepts to Implementation</atitle><jtitle>Journal of lightwave technology</jtitle><stitle>JLT</stitle><date>2023-07-15</date><risdate>2023</risdate><volume>41</volume><issue>14</issue><spage>4557</spage><epage>4581</epage><pages>4557-4581</pages><issn>0733-8724</issn><eissn>1558-2213</eissn><coden>JLTEDG</coden><abstract>This paper introduces a novel methodology for developing low-complexity neural network (NN) based equalizers to address impairments in high-speed coherent optical transmission systems. We present a comprehensive exploration and comparison of deep model compression techniques applied to feed-forward and recurrent NN designs, assessing their impact on equalizer performance. Our investigation encompasses quantization, weight clustering, pruning, and other cutting-edge compression strategies. We propose and evaluate a Bayesian optimization-assisted compression approach that optimizes hyperparameters to simultaneously enhance performance and reduce complexity. Additionally, we introduce four distinct metrics (RMpS, BoP, NABS, and NLGs) to quantify computing complexity in various compression algorithms. These metrics serve as benchmarks for evaluating the relative effectiveness of NN equalizers when compression approaches are employed. The analysis is completed by evaluating the trade-off between compression complexity and performance using simulated and experimental data. By employing optimal compression techniques, we demonstrate the feasibility of designing a simplified NN-based equalizer surpassing the performance of conventional digital back-propagation (DBP) equalizers with only one step per span. This is achieved by reducing the number of multipliers through weighted clustering and pruning algorithms. Furthermore, we highlight that an NN-based equalizer can achieve better performance than the full electronic chromatic dispersion compensation block while maintaining a similar level of complexity. In conclusion, we outline remaining challenges, unanswered questions, and potential avenues for future research in this field.</abstract><cop>New York</cop><pub>IEEE</pub><doi>10.1109/JLT.2023.3234327</doi><tpages>25</tpages><orcidid>https://orcid.org/0000-0001-7417-9398</orcidid><orcidid>https://orcid.org/0000-0002-8678-5691</orcidid><orcidid>https://orcid.org/0000-0002-9264-9274</orcidid><orcidid>https://orcid.org/0000-0002-6004-385X</orcidid><orcidid>https://orcid.org/0000-0003-0101-3834</orcidid><orcidid>https://orcid.org/0000-0002-3035-4112</orcidid><orcidid>https://orcid.org/0000-0003-3145-1018</orcidid><orcidid>https://orcid.org/0000-0001-9578-0297</orcidid><oa>free_for_read</oa></addata></record> |
fulltext | fulltext |
identifier | ISSN: 0733-8724 |
ispartof | Journal of lightwave technology, 2023-07, Vol.41 (14), p.4557-4581 |
issn | 0733-8724 1558-2213 |
language | eng |
recordid | cdi_ieee_primary_10006884 |
source | IEEE Electronic Library (IEL) |
subjects | Algorithms Artificial neural networks Back propagation networks Bayesian optimizer Clustering coherent detection Complexity computational complexity Computational modeling Equalizers Fiber nonlinear optics neural network Neural networks nonlinear equalizer Nonlinear optics Optical communication Optical fibers Optimization pruning quantization Symbols |
title | Reducing Computational Complexity of Neural Networks in Optical Channel Equalization: From Concepts to Implementation |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-25T22%3A41%3A39IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_ieee_&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Reducing%20Computational%20Complexity%20of%20Neural%20Networks%20in%20Optical%20Channel%20Equalization:%20From%20Concepts%20to%20Implementation&rft.jtitle=Journal%20of%20lightwave%20technology&rft.au=Freire,%20Pedro%20J.&rft.date=2023-07-15&rft.volume=41&rft.issue=14&rft.spage=4557&rft.epage=4581&rft.pages=4557-4581&rft.issn=0733-8724&rft.eissn=1558-2213&rft.coden=JLTEDG&rft_id=info:doi/10.1109/JLT.2023.3234327&rft_dat=%3Cproquest_ieee_%3E2839516174%3C/proquest_ieee_%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2839516174&rft_id=info:pmid/&rft_ieee_id=10006884&rfr_iscdi=true |