WO2020020991A1 - Procédé et système de multiplexage par répartition spatiale utilisant la reconnaissance de figures de speckle dans des fibres optiques multi-mode - Google Patents

Procédé et système de multiplexage par répartition spatiale utilisant la reconnaissance de figures de speckle dans des fibres optiques multi-mode Download PDF

Info

Publication number
WO2020020991A1
WO2020020991A1 PCT/EP2019/069998 EP2019069998W WO2020020991A1 WO 2020020991 A1 WO2020020991 A1 WO 2020020991A1 EP 2019069998 W EP2019069998 W EP 2019069998W WO 2020020991 A1 WO2020020991 A1 WO 2020020991A1
Authority
WO
WIPO (PCT)
Prior art keywords
coherent light
beams
fibre
light beams
multimode
Prior art date
Application number
PCT/EP2019/069998
Other languages
English (en)
Inventor
Guy Verschaffelt
Guy VAN DER SANDE
Jaël PAUWELS
Original Assignee
Vrije Universiteit Brussel
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from EP18185497.7A external-priority patent/EP3599730A1/fr
Application filed by Vrije Universiteit Brussel filed Critical Vrije Universiteit Brussel
Publication of WO2020020991A1 publication Critical patent/WO2020020991A1/fr

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04J—MULTIPLEX COMMUNICATION
    • H04J14/00—Optical multiplex systems
    • H04J14/04—Mode multiplex systems
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
    • H04B10/50—Transmitters
    • H04B10/516—Details of coding or modulation

Definitions

  • the present invention relates to methods, devices and systems for speckle pattern recognition and classification in multi-mode optical fibres and software or hardware.
  • SDM SDM-mode optical fibres.
  • Such fibres - even with a relatively small diameter of the order of 10 micrometer - support a large number of transverse modes, each of which can be used to transmit a data signal.
  • This approach has been discarded due to the unavoidable mixing between the many transverse modes that are supported in these multi-mode fibres. Because of this mode-mixing, a signal injected in one transverse mode will leak to many other modes after a short propagation distance, which lowers the signal-to-noise ratio significantly making efficient communication impossible.
  • Special mode selective filters are placed in between fibre splices such that the mode coupling is reduced. Still, mode mixing cannot be avoided resulting in a low signal-to-noise ratio and quite high bit-error rates, illustrating the complexity of this method.
  • United States Patent No. 5,136,666 by Anderson et al. discloses a fibre optic communication system, where on top of a standard temporal modulation, the input light beam’s spatial extent is also modulated with a certain pattern. This pattern is associated to an address. The authors describe that the light beam travelling through a multimode fibre will undergo mode mixing and at the output of a fibre a speckle image will appear. If two light beams are imprinted with an orthogonal spatial pattern (a different address), the speckle patterns will also be different. A detector set-up is described based on a photo refractive nonlinear optical crystal.
  • a diffraction pattern is written nonlinearly in this photorefractive crystal by using the interference of two beams which are specific to the spatial modulation corresponding to a certain address.
  • the goal is that if a light beam with a speckle image corresponding to the right address impinges on the crystal, the beam will be diffracted to another channel or detector corresponding to this address. A beam with an orthogonal speckle pattern would be left untouched.
  • EP 0 395 060 (A2) , (Nippon Telegraph & Telephone [JP], 1990-10-31) describes an optical device for wavelength-demultiplexing.
  • This device includes an optical conversion element [A] for converting a difference in wavelength between optical inputs to a difference in spatial power distribution at the output, by means of an optical multimode circuit or a diffraction grating.
  • This device also includes a second element [B] for pattern recognition and construction of the output signals, by means of a hologram element and/or a photo-detector array and neural network.
  • the system works when the plurality of optical inputs has different wavelengths. Output signals are analyzed by pattern recognition.
  • each spatial transverse mode of the multimode transmission medium carries an independent signal.
  • the transverse modes contain the information that is sent from the transmitter to the receiver. It is not easy to address and detect individual modes, especially when the number of modes is large.
  • This specific means of mode division multiplexing suffers from quasi-random mode-mixing, thereby mixing the information. This mixing can result in the output at the receiver being only partly or even no longer recovered at the receiver.
  • the results shown in this document demonstrate very weak mode -mixing and can be sensitive to unpredictable mode -mixing.
  • An aspect of embodiments of the present invention is an optical telecommunication system which employs space division multiplexing to transfer information from a transmitter to a receiver by means of transmission medium, for example a multimode optical waveguide.
  • space domain modulation is employed on a plurality of coherent light beams which are injected in the multimode transmission medium.
  • Space domain modulation is preferred compared to modulating using different wavelengths lt is not evident that space domain modulation of the optical inputs should lead to a plurality of spatial power distributions which can be used to extract meaningful output signals, e.g. by means of pattern recognition.
  • Embodiments of the present invention can recognize which one or ones of a spatially separated beams have been switched on.
  • Mode mixing in the transmission medium is preferably done with a large number of modes.
  • Space domain modulation is employed on a plurality of coherent optical inputs at the transmitter to obtain a plurality of spatial power distributions at the receiver which can be used to extract meaningful output signals by means of pattern recognition, given strong mode-mixing during propagation in the multimode transmission medium.
  • the inventors have found that even after strong mode-mixing, the spatial power distributions still contain the necessary information to construct meaningful output signals.
  • a plurality of coherent light beams are applied to a transmitting location by way of space domain modulation of said coherent light beams.
  • a complete optical telecommunication system is described in embodiments of the present invention.
  • This optical telecommunication system employs space division multiplexing to transfer information from the transmitter to the receiver by means of a multimode optical waveguide. Signals are preferably not split at different wavelengths at the output side while relying on a multimode component.
  • Embodiments of the present invention preferably do not include a telecommunication system using wavelength division multiplexing.
  • mode-mixing in the transfer medium results in degradation due to mode-mixing
  • embodiments of the present invention use mode mixing to separate signals having a different spatial modulation.
  • Embodiments of the present invention do not recognize a mode at the output of a multimode fibre. Instead, embodiments of the present invention relate spatial light distributions at the end facets, e.g. speckle, caused by interference of many modes, with spatial modulations at the input without targeting specific modes.
  • spatial light distributions at the end facets e.g. speckle
  • embodiments of the present invention provide a communication system for transmitting information from a transmitting location to a receiving location, comprising; a multimode connection having an input located at said transmitting location and having an output located at said receiving location,
  • coherent light generating means at said transmitting location operable to provide a plurality of coherent light beams to said transmitting location by way of space domain modulation of said coherent light beams, means providing space domain modulation of said coherent light beams to the input of said multimode connection, said multimode connection operating to provide a plurality of speckle patterns at the output thereof, means operable to receive said plurality of speckle patterns at the receiving location and to transmit the output of the multimode connection to an array of photodetectors, and means to process outputs of the array of photodetectors to classify one of the plurality of speckle patterns as from one of the coherent light beams.
  • the means to process can be a classifier.
  • the one of the plurality of speckle patterns can be co generated from one of the coherent light beams. This has the advantage of transmitting large amounts of data over one fibre without using special fibres, e.g. with hollow sections.
  • Means can be provided for applying time domain modulation to said coherent light beams.
  • the combination of this technique with the use of speckle classification allows an increase in data transmission.
  • These means can be provided by a time domain modulator.
  • the means to process can be operable to recover said time domain modulation from said plurality of speckle patterns.
  • the means to process may be provided by a demodulator.
  • means for setting a wavelength of a coherent light beam can be provided.
  • Providing a minimum difference in wavelength of each beam is a less preferred design of a means to generate independent speckle patterns at the output. Accordingly, space division multiplexing according to embodiments of the present invention is preferred. An important feature is the need for a minimum difference in the position of the input beams at the input.
  • the plurality of coherent lights beams can be a plurality of laser beams from different lasers or a single laser beam split into several laser beams or a combination of these two schemes. This provides a significant flexibility for arranging the laser beam input face.
  • Means to modulate the intensity or phase of one some or all of the plurality of coherent light beams can be provided.
  • the means to modulate can be a modulator. This allows a higher throughput of data.
  • the means to modulate may comprise means to modulate an injection current of the different lasers or to modulate the intensity with an electro-optical modulator to each of the several laser beams split from a single laser beam.
  • the multimode connection can comprise a multimode optical fibre that transmits the plurality of coherent beams from the input to the output.
  • the multimode fibre can support many transverse modes. This allows a higher data throughput.
  • an FMF can be used for the multimode connection.
  • multi-mode optical fibre can be sequentially coupled together in the multimode connection and/or that other fibre optic components such as optical amplifiers, optical splitters and add/drop multiplexers are placed in-between pieces of multi-mode optical fibre. This allows a higher data throughput.
  • pieces of FMF can be used for the multimode connection.
  • the plurality of coherent light beams can differ from one another in at least one of the following aspects: there is a minimum difference in the position of each beam at a front facet of the multimode connection and/or in the angle of incidence of each beam at the front facet of the multimodal connection and/or in the polarization of each beam. This can affect the way the speckle patterns are classified.
  • the plurality of coherent light beams can differ from one another such that there is a minimum difference in the wavelength of each beam.
  • means for setting a wavelength of a coherent light beam can be provided. Providing a minimum difference in wavelength of each beam is a less preferred design of generating independent speckle patterns at the output. Accordingly, space division multiplexing according to embodiments of the present invention is preferred. An important feature is the need for a minimum difference in the position of the input beams at the input.
  • the plurality of coherent light beams can differ from one another enough such that the speckle pattern generated by each of the coherent beams separately is uncorrelated with the speckle pattern generated by any of the other beams. This can affect the way the speckle patterns are classified.
  • the system can comprise a multi-mode fibre coupler. This can improve coupling and classification of speckle patterns.
  • an FMF coupler can be used.
  • the multi-mode fibre coupler or the FMF coupler can have a plurality of input fibres and an output, to which a part of the light of each exiting input fibre is coupled. This can improve classification of speckle patterns.
  • a cross-correlation between any two of the speckle patterns is preferably smaller than 0.2 or about 0.2. This can improve coupling and classification of speckle patterns.
  • the means to process outputs of the array of photodetectors is adapted to deduce from the speckle patterns which beams have been switched on at a particular moment in time.
  • Embodiments of the present invention can provide a method for transmitting information from a transmitting location to a receiving location, with a multimode connection having an input located at said transmitting location and having an output located at said receiving location, the method comprising:
  • the one of the plurality of speckle patterns can be co-generated from one of the coherent light beams.
  • Time domain modulation can be applied to said coherent light beams.
  • the processing is preferably operable to recover said time domain modulation from said plurality of speckle patterns.
  • the method can include setting any, some or all of: incident position on the input to the multimode connection, and
  • the method can include setting a wavelength of a coherent light beam, providing a minimum difference in wavelength of each beam is a less preferred design of methods to generating independent speckle patterns at the output. Accordingly, space division multiplexing according to embodiments of the present invention is preferred. An important feature is the need for a minimum difference in the position of the input beams at the input.
  • the plurality of coherent lights beams is preferably a plurality of laser beams from different lasers or a single laser beam split into several laser beams.
  • the method can include modulating the intensity or phase of one some or all of the plurality of coherent light beams.
  • the modulating can comprise modulating an injection current of the different lasers or modulating the intensity with an electro-optical modulator to each of the several laser beams split from a single laser beam.
  • the multimode connection can comprise a multimode optical fibre that transmits the plurality of coherent beams from the input to the output.
  • the multimode fibre can support many transverse modes.
  • an FMF can be used for the multimode connection.
  • multi-mode optical fibre can be sequentially coupled together in the multimode connection and/or that other fibre optic components such as optical amplifiers, optical splitters and add/drop multiplexers are placed in-between pieces of multi-mode optical fibre.
  • other fibre optic components such as optical amplifiers, optical splitters and add/drop multiplexers are placed in-between pieces of multi-mode optical fibre.
  • pieces of FMF can be used for the multimode connection.
  • the plurality of coherent light beams can differ from one another in at least one of the following aspects: there is a minimum difference in the position of each beam at a front facet of the multimode connection and/or in the angle of incidence of each beam at the front facet of the multimodal connection and/or in the polarization of each beam.
  • the plurality of coherent light beams can differ from one another such that there is a minimum difference in the wavelength of each beam.
  • Providing a minimum difference in wavelength of each beam is a less preferred design of methods to generating independent speckle patterns at the output. Accordingly, space division multiplexing according to embodiments of the present invention is preferred.
  • An important feature is the need for a minimum difference in the position of the input beams at the input.
  • the plurality of coherent light beams can differ from one another enough such that the speckle pattern generated by each of the coherent beams separately is uncorrelated with the speckle pattern generated by any of the other beams.
  • a cross-correlation between any two of the speckle patterns is preferably smaller than 0.2 or about 0.2.
  • Processing of outputs of the array of photodetectors can be adapted to deduce from the speckle patterns which beams have been switched on at a particular moment in time.
  • Embodiments of present invention present a classifier for use with a system for transmitting information from a transmitting location to a receiving location, over a multimode connection having an input located at said transmitting location and having an output located at said receiving location, coherent light generating means at said transmitting location operable to provide a plurality of coherent light beams to said transmitting location by way of space domain modulation of said coherent light beams, means providing space domain modulation of said coherent light beams to the input of said multimode connection, said multimode connection operating to provide a plurality of speckle patterns at the output thereof, the classifier comprising: means operable to receive said plurality of speckle patterns at the receiving location and to transmit the output of the multimode connection to an array of photodetectors, and means to process outputs of the array of photodetectors to classify one of the plurality of speckle patterns as from one of the coherent light beams.
  • the classifier can be a linear classifier or a correlation classifier.
  • the means to process can be operable to recover said time domain modulation from said plurality of speckle patterns.
  • the means to process can be a demodulator.
  • a cross-correlation between any two of the speckle patterns is preferably smaller than 0.2 or about 0.2.
  • the means to process outputs of the array of photodetectors is preferably adapted to deduce from the speckle patterns which beams have been switched on at a particular moment in time.
  • the classifier can be adapted to receive intensity modulated or phase modulated beams of one some or all of the plurality of coherent light beams.
  • the multimode connection can comprise a multimode optical fibre that transmits the plurality of coherent beams to the output.
  • the multimode fibre can support many transverse modes.
  • an FMF can be used for the multimode connection.
  • the classifier can be adapted to receive a plurality of coherent light beams which differ from one another in at least one of the following aspects: in the position of each beam at a front facet of the multimode connection and/or in the angle of incidence of each beam at the front facet of the multimodal connection and/or in the polarization of each beam.
  • the plurality of coherent light beams can differ from one another such that there is a minimum difference in the wavelength of each beam.
  • Providing a minimum difference in wavelength of each beam is a less preferred design of methods to generating independent speckle patterns at the output. Accordingly, space division multiplexing according to embodiments of the present invention is preferred.
  • An important feature is the need for a minimum difference in the position of the input beams at the input.
  • the classifier can be adapted to receive the plurality of coherent light beams which differ from one another enough such that the speckle pattern generated by each of the coherent beams separately is uncorrelated with the speckle pattern generated by any of the other beams.
  • the classifier can include or can work with a multi-mode fibre coupler.
  • an FMF coupler can be used for the multimode connection.
  • the multi-mode fibre coupler or the FMF coupler can have a plurality of input fibres and an output, to which a part of the light of each exiting input fibre is coupled for coupling to the classifier.
  • the classifier can be adapted to receive speckle patterns whereby a cross-correlation between any two of the speckle patterns can be smaller than about 0.2.
  • the classifier can work with or include the means to process outputs of the array of photodetectors which means can be adapted to deduce from the speckle patterns which beams have been switched on at a particular moment in time.
  • Embodiments of the present invention provide a computer program product that executes any of the methods according to embodiments of the present invention when loaded onto a processing engine or can implement the system according to embodiments of the present invention.
  • a non-transitory signal storage medium can be provided for storing the computer program product.
  • Figure 1 shows speckle patterns measured at the end of a multimode fibre with 200 micron diameter and using a single mode semiconductor laser as source. Left and right images are recorded for slightly different positions of the source in the plane transverse to the laser beam’s propagation direction.
  • Figure 2 is a schematic illustration of a communication link according to an embodiment of the present invention.
  • Figure 3 illustrates a set-up that is used to couple four lasers into a fibre according to an embodiment of the present invention.
  • the differences in focal lengths are not of that much importance, although they will influence the spot sizes.
  • Figure 4 illustrates the reflection of the four laser beams onto the front fibre facet, imaged with a CCD. They are preferably separated as much as possible within the extent of the fibre core to get a low correlation between the images at the output of the fibre when different lasers are turned on.
  • Figures 5A to 5D illustrate methods in accordance with embodiments of the present invention.
  • Figure 6 illustrates speckle correlation between the first signal 10 and the signal that is sent through the fibre in accordance with an embodiment of the present invention.
  • Figure 7 illustrates a speckle correlation between the first signal 01 and the signal that is sent through the fibre in accordance with an embodiment of the present invention.
  • Figure 8 illustrates the success rate for different threshold decision levels for the correlation in accordance with an embodiment of the present invention.
  • Figure 9 illustrates a speckle correlation between the first signal 1000 and the signal that is sent through the fibre in accordance with an embodiment of the present invention.
  • Figure 10 illustrates the threshold ranges for different measurements for which the success rate is optimal with the corresponding maximal success rate SR when using one, two, three and four lasers in accordance with an embodiment of the present invention.
  • Figure 11 illustrates a representation how one can achieve the densest fill factor, where spots are separated with D from each other in accordance with an embodiment of the present invention.
  • Figure 12 illustrates the relation between the success rate and the number of neurons that are present in the classifier when 16 kernels with sizes of one by one pixel are used as observations in accordance with an embodiment of the present invention.
  • Figure 13 illustrates the relation between the success rate and the number of kernels with sizes of one pixel and 20 neurons in accordance with an embodiment of the present invention.
  • Figure l4a shows a plot the total classifier error rate against the number of speckle intensities sampled from each of the recorded images.
  • the dashed horizontal line represents the decision threshold used to recover the input states: 1 for classifier outputs above threshold, and 0 below threshold.
  • a mechanical and optical arrangement of lasers is shown schematically in Figure 2 and comprise an input sub-system ISS, a multi-mode optical fibre based connection sub-system MMC and a receiver sub-system RSS.
  • multiple data-signals are transmitted spatially in parallel with one another over a multi-mode optical fibre MMF.
  • the data-signals are spatially and temporally coherent optical beams that are intensity/phase modulated in time.
  • the multi-mode fibre MMF is such that it supports several transverse modes. The field injected in these modes mixes during propagation along the fibre. At the end-facet of the fibre, the data-signals are no longer spatially separated nor restricted in size. Rather, these signals have spread over the different propagation modes of the multi-mode fibre MMF.
  • the resulting speckle pattern is spatially sampled at the end facet of the fibre, and these patterns are classified by detection electronics DE.
  • the detection electronics DE can include or can be adapted to use a classifier. During classification the measured pattern is compared with a training set of patterns. These can be regularly sent through the data-link in order to compensate for variations in the mode-mixing in the fibre.
  • N optical fields (labelled B 1 , B2 ....BN) are coupled into the multi- mode connection system MMC.
  • Each of these N beams can originate from a different laser, or a single laser can be used whose output beam is split into N beams by means of optical beam splitters or a number of lasers can be used and a number of beam splitters as well.
  • N different input data signals are added to this input sub-system by modulating the intensity or phase of the fields Bl, B2 ...BN. This can e.g.
  • each laser should be temporally and spatially coherent such that the resulting speckle pattern at the output facet has a sufficiently large speckle contrast, i.e. the speckle contrast should be larger than 0.1. Multiple coherently related lasers are not required.
  • Each of the beams Bl-BN is coupled into the multi-mode optical fibre based connection sub system MMC.
  • This sub-system consists - at least partly - of a multi-mode optical fibre MMF that transmits the beams Bl-BN from the input sub-system ISS to the receiver sub-system RSS.
  • the multi-mode optical fibre MMF should support many transverse modes (e.g. a typical multi- mode fibre with a core diameter of 50 pm supports about 400 modes). It is also possible that several pieces of multi-mode optical fibre MMF are sequentially coupled together in the connection sub-system MMC and/or that other fibre optic components (such as optical amplifiers, optical splitters and add/drop multiplexers) are placed in-between pieces of multi- mode optical fibre MMF.
  • the quasi-random interference between these different transverse modes will lead to a speckle pattern at the output facet of the connection sub-system MMC even when a single beam is used in the input sub system ISS.
  • a speckle pattern consists of a collection of bright and dark regions. The exact position and shape of these spots /patterns depends on several parameters of both the input beam Bl- BN and of the connection sub-system MMC: the pattern will change when changing the wavelength, or‘coupling’ or‘incidence’ and/or polarization of the input beam and when changing the fibre bends, temperature and/or mechanical stress in the multi-mode fibres.
  • Coupling or incidence includes other variations such as spot size, beam convergence and incidence angle.
  • the N beams B 1 -BN that are coupled into the connection sub-system MMC should differ from one another so as to create different speckle patterns for example in at least one of the following aspects: there should be a minimum difference in the wavelength of each beam Bl-BN and/or in the position of each beam at the front facet of the connection sub-system MMC and/or in the angle of incidence of each beam Bl-BN at the front facet of the connection sub-system MCC and/or in the polarization of each beam Bl-BN. These differences should be large enough such that the speckle pattern generated by each of the beams separately is uncorrelated with the speckle pattern generated by any of the other beams, i.e.
  • the cross-correlation between any two of these speckle patterns should be smaller than 0.2 or about 0.2.
  • the numerical value of the minimum difference in several parameters can be measured and depends on the specifications of the multi-mode fibre MMF that is used. For example, for a 50 pm core multi-mode step index fibre, the measured minimum wavelength difference is 0.15 nm and the measured minimum difference in position is 4 pm.
  • Scrambling of the injected polarization by a MMF means that the polarised light exiting the fibre always has both s and p components.
  • the receiver can be arranged to detect both components or be switched to filter out a single component or any combination of these.
  • the speckle patterns for different combinations of input polarization (s or p) and output polarizations (s, p, or both s+p) have been investigated. It has been found that s and p polarized speckle patterns are always uncorrelated, and that they are each partially correlated to the unpolarized speckle patterns (e.g. s+p) . Speckle patterns generated by orthogonal input polarizations are also uncorrelated, regardless of their own polarization state.
  • a multi-mode fibre coupler can be used (not shown).
  • Such a coupler has X input fibres and 1 output, to which (part of) the light of each input fibre is coupled. With such a coupler, the speckle patterns corresponding to each of the inputs are uncorrelated.
  • An FMF coupler can also be used with an FMF.
  • data is received by identifying, from the speckle pattern, which beam has been switched on at a particular moment in time. This can be done by classifying, (using a classifier NxC) each speckle pattern at the output of the connection sub system MCC. Thereby, no attempt needs be made to control the exact speckle pattern originating from each beam Bl-BN.
  • the speckle patterns originating from different beams Bl- BN are preferably uncorrelated, such that the classification algorithm of the classifier NxC at the output can make a distinction between e.g. beam 1 being switched on and a different beam 2 being switched on.
  • the field emitted at the output of the connection sub system MMC is sampled at M different locations. Additionally, some of the feature extraction can be done in optics (not shown) after the output of the connection sub-system MMC but before the sampling of the field. Sampling of the field can e.g. be done by an array of waveguides e.g. optical fibres, each coupled to a high-speed photo-detector that converts the intensity of one read-out waveguide (which plays the role of one detection location) into the electrical domain. In conjunction with the classification algorithm, the number M and the location size are determined.
  • the classifier NxC can be adapted for, and the classification algorithm can be based on, calculating correlations between stored test speckle images corresponding to each of the lasers being turned on and defining a decision level for each calculated correlation.
  • the classification algorithm can be based on a learning system, which can be trained either offline or online. The learning can be done offline in software, by recording a sequence of speckle images for known combinations of switched-on lasers and using this data in the training phase. This can be done, for example, with an artificial neural network-based learning system.
  • the mode mixing in the multi-mode optical connection sub-system MMC can change, e.g. due to fluctuations in temperature, fibre bends and mechanical stress, and the training images will need to be sent again and the detection sub-system DE retrained.
  • these test images do not need to be sent again and a learning rule can be defined based on e.g. the amount of bit errors occurring in the channel ln this way, the learning system such as an artificial neural network will constantly be adapted to compensate for the evolution in the mode mixing and mode interference.
  • the learning system can be based on a deep learning approach, but other approaches such as extreme learning machines, reservoir computing, a correlation classifier or a simple linear classifier are included within the scope of the invention.
  • Fast electronic circuits can be used to implement the learning system such as the neural network that processes the detector signals at very high speed, such that the training and operation can be done online
  • the laser or lasers is/are arranged so that the laser beams Bl-BN fall onto the input facet of a fibre MMF.
  • the intensity/phase of each beam can be modulated in time.
  • a separate laser B 1 -BN can be used for each beam, or a single laser beam can be split into several output beams e.g. using beam splitters.
  • the advantage of the latter approach is that the speckle contrast at the output fibre facet will remain large, even when a large number of beams is used. Because of the scrambling of the polarization in the fibre MMF, the contrast will be close to l/sqrt(2) if one starts from a single coherent source.
  • the speckle contrast at the output will be l/sqrt(2N) and, thus, it becomes small if N is large. A small contrast will make the detection system more difficult and is therefore less preferred.
  • the laser beams Bl-BN are arranged to strike the input facet of the fibre MMF at different positions and/or angles, and these differences can be set large enough such that the speckle pattern corresponding to each individual beam has a low correlation with the patterns of the other beams.
  • the output patterns should be orthogonal.
  • the minimally required difference in position or angle in order to achieve this de-correlation depends on the size and refractive index profile of the fibre MMF.
  • the minimum difference in position of incidence of the laser beams at the input facet can be, for example, 4 pm for a 50 pm core step-index fibre, and 10 pm for a 200 pm core step-index fibre.
  • transmitted can be defined by the ratio— j where d is the diameter of the fibre core.
  • the wavelength of the lasers is preferably within the transmission band of the fibre. Fibres for datacommunication and telecommunication applications typically work around 680nm, 850nm, 980nm, 1300nm or l550nm.
  • the power of the lasers is that needed for longer propagation distances in order to compensate for losses in the fibre. Power levels below 10 pW are probably too low to be useful. Above lOOmW, eye laser safety becomes an issue. Hence, the most useful range of laser power levels is from about 100 pW to lOOmW.
  • the laser beam should have a sufficiently large temporal and spatial coherence, otherwise the speckle contrast at the fibre’s output facet will be low and it will be difficult to identify/ classify different speckle patterns.
  • the required temporal and spatial coherence will depend on the type of fibre considered and its core diameter. For example, it is preferred if the linewidth of each beam is less than 0.2nm.
  • Low spatial coherence laser sources are less useful. This also includes multi-mode lasers, random lasers, chaotic microcavity lasers and large-area YCSELs. LEDs and supercontinuum sources which are less useful.
  • the speckle contrast is high, e.g. close to 1, as this will make it easier for the detection system DE to classify different patterns.
  • the classification system NxC should preferably be able to deal with lower contrasts, e.g. down to contrast values greater than 0.1.
  • the type of multi-mode fibre MMF that can be used with embodiments of the present invention supports more than a few modes, e.g. more than 10.
  • This number of modes supported by a multi-mode fibre depends on the fibre geometry (such as its numerical aperture and diameter), but can easily amount to several thousands of modes e.g. up to 100 000 modes can be supported by the fibre.
  • the number of modes is not the number of parallel channels that can be used. The latter is determined by the minimum shift between beams at the input facet and will typically be lower than the number of modes supported by the fibre.
  • Fibres with a core diameter of 50 pm or 62.5 pm have been found to be suitable and these are standard sizes used in datacommunication.
  • gradient-index fibres are preferred as this type of fibre minimizes modal dispersion (such that the temporal data rate will be maximal in this fibre type), but other refractive index profiles will also work, e.g. at lower data rates.
  • Fibre length can be between O.lm and lOkm or longer.
  • the lower limit is due to the fact that some fibre length is needed for the modes to mix and hence generate a speckle pattern.
  • the upper limit is due to modal dispersion which makes the temporal data rate inversely proportional to the fibre length, e.g. the minimum duration Dt of one temporal bit in a gradient index fibre is
  • L is the length of the fibre
  • nl is the refractive index of the fibre core
  • c is the speed of light in vacuum
  • NA is the numerical aperture of the fibre.
  • the coherence properties of the laser source can be selected based on the resulting speckle contrast at the output facet of the multi-mode fibre.
  • the structure of the fibre will influence the size of the speckle spots, and the minimum difference in the position, angle or wavelength of the input beams that leads to uncorrelated speckle patterns.
  • the intensity of the speckle pattern will be sampled at a number of locations across the output facet of the multi-mode fibre MMF.
  • This can be done using a detector array.
  • a lens or lens system can be used to image the output facet of the multi-mode fibre onto the detector array.
  • One way to implement the detector array is to use a fibre array, of which each fibre is connected to a fast photo-detector. The speed of the photo-detectors is required to be matched to the minimum duration Dt of one temporal bit.
  • Each of the photo detectors will generate an electrical signal proportional to the optical intensity that falls onto the detector. These electrical signals are than processed electronically to classify the corresponding speckle pattern. This processing can e.g. be done using an FPGA or a dedicated electronic circuit.
  • a calibration set of speckle patterns can be transmitted through the multi-mode fibre.
  • a set of known bit patterns, by turning some lasers on and others off, are sent through the system and these patterns are used either to calculate the cross-correlation with data patterns, or to train the detection system.
  • An adaptive bit error rate iterative controller can be used to adjust for changes in speckle.
  • the observed speckle patterns will not be fixed in time and changes will occur due to changes in fibre bends, temperature, ... and slight shifts in operational wavelength. These will lead to considerable changes in the spatial mode mixing and, thus, in the measured patterns.
  • Embodiments of the present invention can deal with this by regularly recalibrating the communication link. This can be done for example by a supervised learning procedure. While supervised learning can be important for a space division multiplexing link, the classification algorithm is preferably able to self-leam in real and evolving operation conditions. A suitable cost function is used such that the classification algorithm is able to detect deviations from optimal performance.
  • One such suitable cost function is the bit error rate of transmissions in the system. This cost function needs to be minimize, e.g. by self learning and can be readily measured from the bit error correction bits that are part of every data packet.
  • the detection system is adapted to progressively adjust its parameters to reduce or minimise the cost function.
  • a Monte Carlo method can be used whereby random changes are made and the quality achieved with these changes is measured (e.g. bit error rate). The system accepts any improvement and moves then to see if an improvement can be achieved by further operation of Monte Carlo methods. If the read-out protocol is based on an artificial neural network, the connection weights in the artificial neural network will constantly be adjusted to keep the cost function minimized.
  • Various online learning algorithms can be used such as gradient descent, recursive least squares or reward-modulated Hebbian learning.
  • Embodiments of the present invention use another degree of freedom, namely the data capacity can be further increased using the spatial dimension.
  • Embodiments of the present invention implement space-division multiplexing using standard multimode fibres and standard optical telecommunication components.
  • Embodiments of the present invention are able to to identify from the speckle pattern which laser beam of a plurality of laser beams, e.g. from an array of sources, has been switched on at a particular moment in time. This way, each laser beam carries in parallel a signal through the fibre, and the data capacity is increased proportional to the number of laser bundles that are coupled to the fibre in parallel.
  • Embodiments of the present invention reconstruct the emitted data by classification of the observed speckle pattern, even though the target classes will shift in time because of minor changes in e.g. temperature or bending of the fibres.
  • SDM optical fibre components
  • the implementation of SDM according to embodiments of the present invention has the advantage that it can use standard optical fibre components while increasing data transmission capacity in different networks, because several spatial modes carry data over one fibre simultaneously.
  • the ultimate telecommunication speed of an (optical) system is usually expressed using the spectral efficiency in bits/(Hz s). This number expresses the number of bits that can be transmitted per second and per Hz of signal bandwidth.
  • Current state-of-the-art telecom over long-distance optical fibres reaches about 10 bits/(Hz s).
  • Embodiments of the present invention may reach a maximum spectral efficiency of about 1000 bits/(Hz s).
  • Embodiments of the present invention allow for the observed speckle patterns not being fixed in time.
  • the speckle patterns can change, e.g. in fibre bends, due to temperature changes, due to slight shifts in operational wavelength. These changes can be considerable in the spatial mode mixing and, thus, in the measured patterns.
  • Embodiments of the present invention can make use of supervised learning during installment of a space division multiplexing telecomm link.
  • a classification algorithm can be used such that it is able to self-leam in real and evolving operation conditions.
  • Embodiments of the present invention include a test procedure to adapt the read-out during operation, e.g. a test pattern can be sent at intervals to retrain the output.
  • Embodiments of the present invention can reconstruct the emitted data by classification of the observed speckle patterns. This can be done without requiring expensive specialty fibres.
  • Embodiments of the present invention can use multiple lasers or a single laser split into more light beams that emit light through a fibre.
  • the mode combination and mixing in the fibre are not controlled but rather a detection system recognizes from an output pattern a pattern of input lasers being switched on or a detection system is taught how to recognize from output patterns a pattern of input lasers being switched on.
  • a speckle pattern will appear, which is imaged onto an image acquiring device such as a CCD sensor or camera, CMOS camera etc.
  • a fan-out fibre bundle for example can be used for this purpose, whereby each fibre from the bundle plays the role of a “detection pixel” and can be coupled to a high-speed photodetector.
  • -A correlation between different ones of these speckle images is calculated to detect the states of the lasers by only looking at the output speckle pattern. Hence which lasers are turned on is detected from the speckle pattern when multiple lasers are turned on.
  • a set-up according to an embodiment is shown in Figure 3.
  • Four laser diodes are shown 1-4, e.g. all the same type such as HL6358MG. All of the lasers 1-4 can be coupled into the fibre by using beam-splitters 5, 6, 7.
  • a 50 mhi diameter step-index fibre 14 with a length of 2 m is used for experiments.
  • An image sensor such as a CCD sensor 10, a CMOS camera, a fan-out fibre bundle for example, whereby each fibre from the bundle plays the role of a“detection pixel” and is coupled to a high-speed photodetector is used to image the speckle pattern at the end facet of the fibre 14.
  • a further image sensor such as a CCD sensor 12, a CMOS camera, a fan-out fibre bundle for example, whereby each fibre from the bundle plays the role of a“detection pixel” and is coupled to a high-speed photodetector, can be used to image the reflections of the laser light on the front fibre facet (see Figure 4).
  • Each bright spot in this figure represents one of the four lasers that is coupled into the fibre 14.
  • the purpose of these images of the reflections of the laser light at the front facet is to position these spots far enough from one another such that different lasers lead to un-correlated speckle patterns at the exit.
  • the spot sizes are not the same for all lasers. This is because of the different distances and different lenses that are being used in the set-up.
  • the spots are also brighter than others. This is because not all laser beams are coupled as well into the fibre as others. At the exit of the fibre 14, it is preferred if the different laser beams have the same power. If a laser signal is not coupled well, the input power of the laser can be increased by increasing the operating current, resulting in a brighter spot at the entrance of the fibre.
  • Embodiments are now described for detecting signals created by using only two laser diodes and then with four laser diodes.
  • the lasers 1-4 are to be characterized, since they will probably not have the same characteristics, even if these lasers are from the same type and manufacturer. For example, characteristics of four lasers are shown in Table 1.
  • Table 1 shows important characteristics of the four lasers, including the threshold current 1th, the operating current lop at which the experiment is executed, the operating temperature Top, the power Pop and wavelength, V?/ measured at that operating current and temperature and the range of operating currents at which the laser is operating single- mode.
  • the temperature is chosen as a constant at 22 °C.
  • the operating currents are chosen such that the power at the exit of the fibre is approximately the same.
  • the laser is operated in single- mode. The system should stay within the minimum and maximum of the range in which the laser is operating single-mode, to avoid multi-mode operation. The exact value of these parameters is of less importance, compared to all lasers being operated single mode.
  • the wavelengths are the same or within a range plus or minus 5nm, preferably plus or minus 3nm.
  • the wavelength difference is important when it is desired to characterise the effect of only the spatial position, otherwise the wavelength difference will increase the decorrelation of the speckle patterns.
  • both lasers can be on at the same time (represented with 11), one or the other can be turned on (represented with 10 for laser 1 on and 01 for laser 2 on) and both can be turned off (represented with 00).
  • step 110 the power of both beams is preferably measured and by using neutral density filters if necessary, the power is preferably kept approximately the same (with a difference that is preferably smaller than 50% of the average power of the different lasers) for both lasers at the exit of the fibre.
  • the results are shown in Table 2. Note that the power PI is not the same as the operating power Pop from Table 1.
  • the laser beam with power Pop first passes through a ND-filter and a few beam splitters before it arrives at the fibre entrance with a power of Pi. The power level is of minor importance.
  • Table 2 shows power output for each laser in front and behind the fibre.
  • PI represents the power measured in front of the fibre and P2 is the power that is measured at the exit of the fibre.
  • a power of approximately 0. lmW at the end of the fibre for each individual laser diode is achieved preferably. It is preferred if the power should be such that each pixel can be measured with a Signal-to-Noise Ratio of minimally 2.
  • step 112 different signals are sent through the fibre and imaged onto the image sensor or image sensors such as CCD image sensors 10, CMOS cameras a fan-out fibre bundle for example, whereby each fibre from the bundle plays the role of a “detection pixel” and is coupled to a high-speed photodetector.
  • image sensor or image sensors such as CCD image sensors 10, CMOS cameras a fan-out fibre bundle for example, whereby each fibre from the bundle plays the role of a “detection pixel” and is coupled to a high-speed photodetector.
  • these different signals followed the sequence: 10-11-01-00 which is repeated multiple times in a fixed sequence.
  • step 114 Each time when the signal is changed, the corresponding speckle pattern is recorded in step 114 in non-volatile memory such as a memory disk or solid state memory such as a flash memory.
  • step 116 the images are then filtered such that only the fast varying spatial intensity fluctuations remain. From these images, the correlation is calculated in step 118 with respect to the first signal of 10 (see Figure 6) and again with respect to the first signal of 01 (see Figure 7). From these correlations, the system is able to determine automatically in step 124 which lasers are turned on.
  • a decision threshold (DT) can be defined as shown in Figures 6 and 7, such that a correlation value above (or below) the decision threshold (DT) signifies that the corresponding laser at the input was turned on (or off).
  • a success rate is calculated in step 126, i.e. the amount of signals that are classified correctly by the computer for a given decision threshold (DT) on the correlation of Figures 6 and 7.
  • DT decision threshold
  • the success rate can be determined by dividing it with the total amount of signals that were sent. This provides a success rate in step 126 for both laser 1 and laser 2 and this can be used to set up the telecommunications network.
  • the success rate calculated in function of this threshold value for the different lasers is shown in Figure 8. The success rate is one if, for that given decision threshold, it can always correctly detect the state of the particular laser and decreases for each wrong detection.
  • the decision threshold is close to zero, some errors will occur because of false positive estimations, i.e. the corresponding laser is detected to be switched on while it actually was switched off. If the decision is too high, errors will occur because of false negatives, i.e. the laser is detected as being switched off while it was actually on. But from the results presented there exists quite a broad range of decision threshold values for which the classification of the speckle patterns based on their correlation works very well. By averaging over both success rates, the range of thresholds can be determined for which the success rate is optimal, meaning that the presence of both lasers can be detected correctly.
  • each laser diode is turned on for a moment and the speckle image that is created by that particular laser diode is recorded and saved with the image sensor such as CCD 10, a CMOS camera, a fan-out fibre bundle for example, whereby each fibre from the bundle plays the role of a“detection pixel” and is coupled to a high-speed photodetector.
  • the bit codes for these images are: 1000, 0100, 0010 and 0001, each corresponding with one laser.
  • all 16 possible patterns that can be created by combining those four lasers can be sent through the fibre in any order with any number of repeats.
  • a decision threshold is defined as shown in Figure 9 such that a correlation value above (below) the decision threshold signifies that the corresponding laser (Laser 1 in Figure 9) at the input was switched on (off). From these correlations, the success rate is calculated for different lasers in step 210. By averaging over these success rates, a total success rate is calculated in step 212. From this success rate, one can determine in step 214 the range of decision thresholds for which the success rate is optimal. Figure 10 shows the range of optimal decision thresholds for experiments with different numbers of lasers.
  • the present invention is not limited to two or four lasers.
  • a larger or maximum number of laser spots can be used with uncorrelated speckle patterns at the output.
  • the maximum number will be related to the densest possible stacking of spots on the fibre core.
  • the highest fill factor can be achieved if the spots are arranged as in Figure 11.
  • D is the separation between neighbouring spots and is related to the vertical distance H as:
  • the surface area of the parallelogram abed is equal to:
  • the area that can be filled is now calculated.
  • a problem with this classification method is that the correlation between entire images has to be calculated. This needs a significant computation time. This might cause increasing problems when sending different patterns through the fibre at a higher bit rate. Further difficulties might occur in recognising the signal at the exit of the fibre immediately despite capturing full-frame images such as CCD -images at high-speed or CMOS camera images or a fan-out fibre bundle for example, whereby each fibre from the bundle plays the role of a“detection pixel” and is coupled to a high-speed photodetector.
  • a linear classifier is used to classify the different patterns. This has the advantage of using only a limited amount of pixels from the speckle images.
  • a linear classifier can be used because the pattern created by a combination of lasers appears as if it were the superposition of the separated patterns of each individual laser, according to the calculations of the correlations from above.
  • a non-linear classifier can also work well and is not excluded by the present invention. Linear classification has two purposes. When given a series of observations, different groups or classes can be created to which these observations are allocated. Therefore, it is required to allocate each new observation into one of these classes. A distinction between a linear classifier and a non-linear classifier can be made.
  • a linear classifier is directly applied on the data of the input space.
  • a linear classifier fails, however, when the data is not linearly separable and the prediction accuracy of non-linear classifiers is in general higher than that of linear classifiers. But, when it is linearly separable, linear classifiers are simpler and show more efficient training and testing procedures.
  • Embodiments of the present invention can make use of a linear classifier, since it has been found that the correlations described in the previous section showed that a pattern where multiple lasers are turned on (e.g. 1010), can be recognized from the patterns where the individual lasers are turned on (1000 and 0010 in this example). For example, 16 different classes can be used, one for each combinations of the four lasers. It is required to determine how the pattern for each class will look like. Therefore, the classifier is trained in order to recognize the different patterns. And after the learning procedure, for unknown patterns the linear classifier can be used to recognize them correctly, while using only a limited amount of pixels during the learning and recognizing procedures.
  • a linear classifier since it has been found that the correlations described in the previous section showed that a pattern where multiple lasers are turned on (e.g. 1010), can be recognized from the patterns where the individual lasers are turned on (1000 and 0010 in this example). For example, 16 different classes can be used, one for each combinations of the four lasers. It is required to determine
  • the signal states of 4 independent SDM channels through the MMF link were recovered using linear classifiers.
  • the speckle patterns were sparsely sampled by only retaining small subsets of pixels from the recorded images.
  • the classifier outputs were constructed as weighted sums of input features, namely the sampled speckle intensities. For each SDM channel, a set of read
  • Figure 14 plots the total classifier error rate against the number of speckle intensities sampled from each of the recorded images.
  • Fig. 14(a) the total error rate (cumulative over all channels) is plotted against the number of speckle intensities sampled from the recorded images.
  • the error bars correspond with performance fluctuations over iterations with different sampling locations. It was found that at least 4 intensity samples are needed to obtain error- free recovery of all 4 signal states. This corresponds to just one sample per SDM channel.
  • An error rate is calculated by dividing the number of misclassified bits by the total number of bits sent.
  • a drawback of using MMFs for long-haul data transmission can be intermodal dispersion. This can limit the maximal modulation rate of the input signal, and, thus, limits the channel capacity.
  • linear classification method An advantage of a linear classification method is its bandwidth potential. Contrary to the correlation-based scheme described below, the linear classification method can not only work with MMF but can also benefit from switching to few mode fibres (FMFs) where intermodal dispersion can at least partially be mitigated. Some any or all of the eembodiments of the present invention can make use of an FMF for the multimode connection. In an ideal case, the number of fibre modes is matched to the number of parallel space division multiplexed channels to be implemented.
  • Embodiments of the present invention can make use of a correlation-based classifier.
  • a correlation-based classifier operates by calculating the correlation between the recorded speckle patterns and the pre-recorded patterns generated by individual lasers. Training of the correlation classifier therefore comprises recording the number of speckle patterns corresponding with signal states of the individual lasers. These patterns are reference patterns for the correlation- based classifier.
  • the individual signal states are recovered by calculating the cross- correlation between the speckle pattern generated by a signal state and the pre-recorded reference patterns generated by the individual signal beams.
  • the corresponding speckle pattern is partially correlated to the reference speckle patterns generated by the each of active input signals individually.
  • a decision threshold is defined such that cross-correlations which exceed this threshold can identify active inputs and cross correlations below threshold identify inactive inputs.
  • a good value for the decision threshold can be obtained a posteriori from the sampled correlation coefficients (i.e. after recording all possible sequences). However, based on the probability density functions of the sampling correlation coefficients the probability of a given threshold value yielding correct classification of all possible sequences can be calculated. Thus, a priori an estimate of the optimal decision threshold can be made by numerically optimizing this likelihood.
  • a method 300 will now be described. From the speckle images recorded with the image sensor such as the CCD, a small subset of size N pixels pixels is chosen in step 302 on Figure 5C.
  • the present invention is not limited to CCD imagers but can include any suitable, preferably high speed imagers such as CMOS cameras, a fan-out fibre bundle for example, whereby each fibre from the bundle plays the role of a“detection pixel” and is coupled to a high-speed photodetector. These chosen pixels are preferably evenly distributed over the image in a 2D grid.
  • the corresponding pixel intensity values are measured in step 304. These represent an observation u(m), with in representing the image index, ranging from 1 to M.
  • the matrix U of size Npixeis x M combines all observations, with M representing the number of observations. These observations will serve as inputs to a small feed- forward artificial neural network in step 306.
  • An input mask Win of size n xNpixels is constructed in step 308 to couple these inputs to the first and only layer of n neurons. This mask is constructed with random values uniformly distributed between -1 and +1. Such a random mask diversifies the neural responses as it allows different neurons to respond to different features in the speckle images.
  • a state matrix X of size M x n, containing the states of the neurons in response to the presented images, is this found by multiplying the input mask with the input matrix U in step 310.
  • the coupling matrix or readout weights Wout of size n ®/ 1, describes how the neurons couple to an output classifier.
  • the output of the i-th classifier is (), with size 1 x M
  • One classifier will be constructed per laser, so that four sets of readout weights Wout,i, / are required ranging from 1 to 4. Instep 314 of a learning procedure these readout weights can be optimized to ensure that all laser states (in a training set) are correctly predicted by their corresponding classifier. Since for each speckle image in the training set, the corresponding laser states are known, a target output Yi, of size M x 1 , can be constructed for each of the classifiers in step 316. These target signals consist of zeros and ones. A zero represents that for a given pattern, the specific laser signal being tested is turned off (or alternatively on), while a one represents that the laser signal being tested is turned on (or alternatively off). E.g. for Figure 6, only two lasers are used and the first four images of this sequence are given target values 1,1,0 and 0 for the first laser and is repeated for the rest of the sequence.
  • the training procedure can work as follows.
  • the M observations in the training set are presented to the network in step 318 and the corresponding neuron states are recorded in the state matrix X in step 320.
  • the error is minimized between the classifier output Oi and the target output Y, by optimizing out, i.
  • Step 324 comprises multiplying the pseudo-inverse of X with the target signal Yi from which the optimal readout weights Wout.i are found in step 326:
  • the present invention is not limited to such a method, e.g. algorithms exist, where also the values of the input mask are altered to minimize the output error. However, these approaches can be selected if more complex training procedures are used.
  • the communication procedure can work as follows.
  • step 402 of a method 400 shown in Figure 5D readout weights are available that allow allocation of the different patterns into their correct class and they can now be used to process previously unseen patterns and classify them correctly.
  • the network is presented in step 404 with observations in the communication set and the corresponding neuron states are recorded in step 406 in a new state matrix Xcomm. These are calculated using Eq. 6, using the same input mask Win as before in step 408.
  • the new outputs Ocomm.i are then calculated using Eq. 7 in step 410.
  • a classifiers output Ocomm,i will not be an array of zeros and ones. Rather, assuming a successful training, the output values will be close to zero if the corresponding laser is turned off for a given pattern and close to one if it is turned on. Therefore, a threshold needs to be drawn in step 412, for example here chosen at 0.5. In step 414 for everything above this value, the linear classifier will recognize the laser as turned on for a given pattern and everything below this value is recognized as turned off. Success rates are calculated as above (i.e. this is the ratio between correctly classified signals and the total amount of signal send) in step 416. This is done for all four lasers and the total success rate is calculated from the mean of the individual success rates.
  • whcrc /A/. is a non-linear transfer function, in this embodiment a hyperbolic tangent. However, it did not improve the results.
  • a signal 1010 for example, is recognized as the superposition of signals 1000 and 0010 using the correlations between the different images.
  • to be certain neurons are included in the classifier, just in case there is a non-linearity.
  • the success rate increases with the number of neurons and reaches the maximum success rate when the number of neurons is approximately the same as the number of SDM channels (i.e. optical sources) by a small margin, i.e. 4 in this experiment shown in figure 12.
  • the number of neurons needs to be at least as much as the number of sources, in order to gather enough information from the recorded images to reconstruct the input signal. Otherwise, not every bit of the input signal can be successfully reconstructed and the success rate will go down.
  • the number of pixels is swept and the success rate is calculated for different values of the number of pixels.
  • the size of each pixel can be optimized. When only choosing one pixel in the image, the success rate is low, this is because there is only information of one pixel for each image in order to reconstruct the states of 4 input lasers, and thereby recognize 16 different classes. If the pixel value of two images that belong in different classes is approximately the same, the linear classifier will not be able to recognize them always correctly and the success rate will go down. The success rate goes up as more pixels are chosen in the image, because there is more information available for the classifier in order to recognize the patterns more correctly. The success rate reaches one when using 9 pixels in the image.
  • SDM using specialty fibres is based on the development of multiple cores and/or a waveguide structure within fibres that support a few transverse modes, whereby each core/mode carries an optical data signal
  • embodiments of the present invention make use of many transverse modes which are accessed and which recombine randomly and uncontrollably into a speckle pattern which is used for classification of the output.
  • Embodiments of the present invention can be used with standard multimode fibre and can be inserted in the existing telecom architecture.
  • bit pattern at the end of the fibre is detected at the receiver side.
  • the present invention can be implemented in local area networks (LANs) and data centers, e.g. connection lengths ranging up to lOOOm.
  • LANs local area networks
  • data centers e.g. connection lengths ranging up to lOOOm.
  • multi-mode fibres are used presently and often new data links are configured and expanded.
  • the present invention can be implemented in long distance telecomm links.
  • Embodiments of the present invention do not need to make use of photo -refractivity which is typically a slow nonlinear process, and hence the calibration process in embodiments of the present invention are quick. Embodiments of the present invention do not impose high power requirements of the laser systems used.
  • Electronic devices e.g. classifiers, such as described with respect to embodiments of the present invention can be standalone digital processing devices or can be embedded in another device. Such devices or methods can use a digital processing engine to carry out functions.
  • the processing engine preferably has processing capability such as provided by one or more microprocessors, FPGA’s, or a central processing unit (CPU) and/or a Graphics Processing Unit (GPU), and is adapted to carry out the respective functions by being programmed with software, i.e. one or more computer programs.
  • References to software can encompass any type of programs in any language executable directly or indirectly by a processor, either via a compiled or interpretative language such as Java.
  • any of the methods of the present invention can be performed by or assisted by logic circuits, electronic hardware, processors or circuitry which can encompass any kind of logic or analog circuitry, integrated to any degree, and not limited to general purpose processors, digital signal processors, ASICs, FPGAs, discrete components or transistor logic gates and similar.
  • Such devices may have memory (such as non-volatile memory, non-transitory computer readable medium, RAM and/or ROM), an operating system, optionally a display such as a fixed format display, ports for data entry devices such as a keyboard, a pointer device such as a “mouse”, serial or parallel ports to communicate other devices, network cards and connections to connect to any of the networks.
  • the software can be embodied in a computer program product adapted to carry out the functions itemised below when the software is loaded onto the controller and executed on one or more processing engines such as microprocessors, ASIC’s, FPGA’s etc.
  • devices for use with any of the embodiments of the present invention can incorporate a computer system capable of running one or more computer applications in the form of computer software.
  • the electronic device can include a main memory, preferably random access memory (RAM), and may also include a non-transitory hard disk drive and/or a removable non-transitory memory, and/or a non-transitory solid state memory.
  • RAM random access memory
  • Non-transitory removable memory can be an optical disk such as a compact disc (CD-ROM or DVD-ROM), a magnetic tape, which is read by and written to by a suitable reader.
  • the removable non-transitory memory can be a computer readable medium having stored therein computer software and/or data.
  • the non-volatile storage memory can be used to store persistent information that should not be lost if the computer system is powered down.
  • the application programs may use and store information in the non volatile memory.
  • Computer program products and digital processing hardware can be adapted to transmit information from a transmitting location to a receiving location, with a multimode connection having an input located at said transmitting location and having an output located at said receiving location.
  • processing engines such as microprocessors, ASIC’s, FPGA’s the following functions are carried out: generating a plurality of coherent light beams at said transmitting location and providing the plurality of coherent light beams to said transmitting location by way of space domain modulation of said coherent light beams at the input of said multimode connection, and/or generating a plurality of speckle patterns at the output thereof, receiving said plurality of speckle patterns at the receiving location and transmitting the output of the multimode connection to an array of photodetectors, and processing outputs of the array of photodetectors to classify one of the plurality of speckle patterns as from one of the coherent light beams.
  • the one of the plurality of speckle patterns can be co-generated from one of the coherent light
  • processing engines such as microprocessors, ASIC’s, FPGA’s the following functions are carried out: applying time domain modulation to said coherent light beams, processing can be operable to recover said time domain modulation from said plurality of speckle patterns.
  • the plurality of coherent lights beams are a plurality of laser beams from different lasers or a single laser beam split into several laser beams.
  • modulating comprises modulating an injection current of the different lasers or modulating the intensity with an electro-optical modulator to each of the several laser beams split from a single laser beam.
  • the multimode connection can comprise a multimode optical fibre that transmits the plurality of coherent beams from the input to the output, or the multimode fibre can support many transverse modes, or the multimode connection comprises a few mode optical fibre.
  • the plurality of coherent light beams differ from one another in at least one of the following aspects: there is a minimum difference in the wavelength of each beam and/or in the position of each beam at a front facet of the multimode connection and/or in the angle of incidence of each beam at the front facet of the multimodal connection and/or in the polarization of each beam, or the plurality of coherent light beams differ from one another enough such that the speckle pattern generated by each of the coherent beams separately is uncorrelated with the speckle pattern generated by any of the other beams, or cross-correlating between any two of the speckle patterns being smaller than 0.2 or about 0.2, or processing of outputs of the array of photodetectors is adapted to deduce from the speckle patterns which beams have been switched on at a particular moment in time.
  • a computer program product can be provided which when executed on a processing engine can perform any of the methods according to the present invention.
  • a non- transitory storage means can be used to store the computer program product.
  • the computer program product can be compiled for a processing engine in any of the electronic devices or is written in an interpretive language such as Java that runs on a virtual machine such as the Java virtual machine.
  • the non-transitory signal storage medium can be for example, an optical disk (CD-ROM or DVD-ROM), a digital magnetic tape, a magnetic disk, a solid state memory such as a USB flash memory, a ROM, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Electromagnetism (AREA)
  • Optical Communication System (AREA)

Abstract

L'invention concerne un procédé et un système de transmission d'informations à partir d'un emplacement de transmission, comprenant : une connexion multimode ayant une entrée située au niveau dudit emplacement de transmission et ayant une sortie située au niveau dudit emplacement de réception, un moyen de génération de lumière cohérente au niveau dudit emplacement de transmission utilisable pour fournir une pluralité de faisceaux de lumière cohérente audit emplacement de transmission au moyen d'une modulation de domaine spatial desdits faisceaux de lumière cohérente, un moyen fournissant une modulation de domaine spatial desdits faisceaux de lumière cohérente à l'entrée de ladite connexion multimode, ladite connexion multimode fonctionnant pour fournir une pluralité de figures de speckle à la sortie de celle-ci, un moyen utilisable pour recevoir ladite pluralité de figures de speckle au niveau d'un emplacement de réception et transmettre la sortie de la connexion multimode à un réseau de photodétecteurs, un moyen pour traiter des sorties du réseau de photodétecteurs de sorte à classifier l'une de la pluralité de figures de speckle en tant que provenant de l'un des faisceaux de lumière cohérente.
PCT/EP2019/069998 2018-07-25 2019-07-25 Procédé et système de multiplexage par répartition spatiale utilisant la reconnaissance de figures de speckle dans des fibres optiques multi-mode WO2020020991A1 (fr)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
EP18185497.7 2018-07-25
EP18185497.7A EP3599730A1 (fr) 2018-07-25 2018-07-25 Procédé de multiplexage par répartition dans l'espace et système au moyen de la reconnaissance de motifs de chatoiement dans des fibres optiques multimode
GB201900210 2019-01-07
GB1900210.4 2019-01-07

Publications (1)

Publication Number Publication Date
WO2020020991A1 true WO2020020991A1 (fr) 2020-01-30

Family

ID=67587722

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/EP2019/069998 WO2020020991A1 (fr) 2018-07-25 2019-07-25 Procédé et système de multiplexage par répartition spatiale utilisant la reconnaissance de figures de speckle dans des fibres optiques multi-mode

Country Status (1)

Country Link
WO (1) WO2020020991A1 (fr)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112862081A (zh) * 2021-03-18 2021-05-28 清华大学 基于傅立叶变换的人工神经网络的多模光纤成像方法
DE102020118871B3 (de) 2020-07-16 2021-07-15 Technische Universität Dresden Empfangsvorrichtung und Verfahren zum Ermitteln von Übertragungseigenschaften eines optischen Wellenleiters sowie Verfahren zum computergestützten Trainieren eines neuronalen Netzwerks
CN114358270A (zh) * 2022-01-07 2022-04-15 南京鼎芯光电科技有限公司 一种多维复用技术的光子神经网络及实现方法
CN117147024A (zh) * 2023-09-01 2023-12-01 天津大学 一种力传感触头及光纤末端力传感装置与三维力解算方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0395060A2 (fr) 1989-04-28 1990-10-31 Nippon Telegraph and Telephone Corporation Récepteur optique
US5136666A (en) 1991-08-06 1992-08-04 The University Of Colorado Foundation, Inc. Fiber optic communication method and apparatus providing mode multiplexing and holographic demultiplexing
US20100329671A1 (en) 2009-06-26 2010-12-30 Alcatel-Lucent Usa Inc. Transverse-mode multiplexing for optical communication systems

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0395060A2 (fr) 1989-04-28 1990-10-31 Nippon Telegraph and Telephone Corporation Récepteur optique
US5136666A (en) 1991-08-06 1992-08-04 The University Of Colorado Foundation, Inc. Fiber optic communication method and apparatus providing mode multiplexing and holographic demultiplexing
US20100329671A1 (en) 2009-06-26 2010-12-30 Alcatel-Lucent Usa Inc. Transverse-mode multiplexing for optical communication systems

Non-Patent Citations (12)

* Cited by examiner, † Cited by third party
Title
A. SANO ET AL.: "409-Tb/s + 409-Tb/s crosstalk suppressed bidirectional MCF transmission over 450 km using propagation-direction interleaving", OPTICS EXPRESS, vol. 21, 2013, pages 16777 - 16783
D. J. RICHARDSON: "New optical fibres for high-capacity optical communications", PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY A, vol. 374, 2016
D.J. RICHARDSONJ.M. FINIL.E. NELSON: "Space Division Multiplexing in Optical Fibres", NATURE PHOTONICS, vol. 7, 2013, pages 354 - 362
FLAMM D ET AL: "All-Digital Holographic Tool for Mode Excitation and Analysis in Optical Fibers", JOURNAL OF LIGHTWAVE TECHNOLOGY,, vol. 31, no. 7, 1 April 2013 (2013-04-01), pages 1023 - 1032, XP011496818, ISSN: 0733-8724, DOI: 10.1109/JLT.2013.2240258 *
FLAMM, DANIEL ET AL.: "All-digital holographic tool for mode excitation and analysis in optical fibres", JOURNAL OF LIGHTWAVE TECHNOLOGY, vol. 31.7, 2013, pages 1023 - 1032, XP011493808, doi:10.1109/JLT.2013.2240258
H. CHENC. JINB. HUANGN. K. FONTAINER. RYFK. SHANGN. GREGOIRES. MORENCYR.-J. ESSIAMBREG. LI: "Integrated cladding-pumped multicore few-mode erbium-doped fibre amplifier for space-division-multiplexed communications", NATURE PHOTONICS, vol. 10, 2016, pages 529 - 533
K. IGARASHID. SOMAY. WAKAYAMAK. TAKESHIMAY. KAWAGUCHIN. YOSHIKANET. TSURITANII. MORITAM. SUZUKI: "Ultra-dense spatial-division-multiplexed optical fibre transmission over 6-mode 19-core fibres", OPTICS EXPRESS, vol. 24, 2016, pages 10213 - 10231
K. SHIBAHARA ET AL.: "Dense SDM (12 core x 3 mode) transmission over 527 km with 33.2-ns mode-dispersion employing low complexity parallel MIMO frequency-domain equalization", PROCEEDINGS OF THE OPTICAL FIBRE COMMUNICATIONS CONFERENCE OFC 2015, PAPER TH5C.3, 2015
N. BAI ET AL.: "Mode-division multiplexed transmission with inline few-mode fibre amplifier", OPTICS EXPRESS, vol. 20, 2012, pages 2668 - 2680
R. RYFN. K. FONTAINEH. CHENB. GUANB. HUANGM. ESMAEELPOURA. H. GNAUCKS. RANDELS.J.B. YOOA.M.J. KOONEN: "Mode-multiplexed transmission over conventional graded-index multimode fibres", OPTICS EXPRESS, vol. 23, 2015, pages 235 - 246
R.G.H. VAN UDENR. AMEZCUA CORREAE. ANTONIO LOPEZF.M. HUIJSKENSC. XIAG. LI, A. SCHULZGENH. DE WAARDTA.M.J. KOONENC.M. OKONKWO: "Ultra-high-density spatial division multiplexing with a few-mode multicore fibre", NATURE PHOTONICS, vol. 8, 2014, pages 865 - 870, XP055266612, doi:10.1038/nphoton.2014.243
S. BERDAGUEP. FACQ: "Mode division multiplexing in optical fibres", APPLIED OPTICS, vol. 21, 1982, pages 1950 - 1955

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102020118871B3 (de) 2020-07-16 2021-07-15 Technische Universität Dresden Empfangsvorrichtung und Verfahren zum Ermitteln von Übertragungseigenschaften eines optischen Wellenleiters sowie Verfahren zum computergestützten Trainieren eines neuronalen Netzwerks
EP3940973A1 (fr) * 2020-07-16 2022-01-19 Technische Universität Dresden Dispositif de réception et procédé permettant de déterminer les propriétés de transmission d'un conducteur d'onde optique, ainsi que procédé d'apprentissage assisté par ordinateur d'un réseau neuronal
US11262502B2 (en) 2020-07-16 2022-03-01 Technische Universitat Dresden Receiving device and method for determining transmission characteristics of an optical waveguide
CN112862081A (zh) * 2021-03-18 2021-05-28 清华大学 基于傅立叶变换的人工神经网络的多模光纤成像方法
CN114358270A (zh) * 2022-01-07 2022-04-15 南京鼎芯光电科技有限公司 一种多维复用技术的光子神经网络及实现方法
CN117147024A (zh) * 2023-09-01 2023-12-01 天津大学 一种力传感触头及光纤末端力传感装置与三维力解算方法

Similar Documents

Publication Publication Date Title
WO2020020991A1 (fr) Procédé et système de multiplexage par répartition spatiale utilisant la reconnaissance de figures de speckle dans des fibres optiques multi-mode
Katumba et al. Neuromorphic computing based on silicon photonics and reservoir computing
Natalino et al. Experimental study of machine-learning-based detection and identification of physical-layer attacks in optical networks
US10581529B2 (en) Rapid out-of-band signal communication in optical components
US20170070289A1 (en) Free space optical receiver and free space optical receiving method
EP2495611A1 (fr) Système de communication optique
CN107947929A (zh) 基于k‑近邻处理的连续变量量子密钥分发系统及实现方法
Tanaka et al. Intelligent monitoring of optical fiber bend using artificial neural networks trained with constellation data
US11808659B2 (en) Parallel optics based optical time domain reflectometer acquisition
US6965736B1 (en) Method for monitoring the transmission quality of an optical transmission system, in particular of an optical wavelength-division multiplex network
Cho et al. Constellation-based identification of linear and nonlinear OSNR using machine learning: a study of link-agnostic performance
EP3599730A1 (fr) Procédé de multiplexage par répartition dans l'espace et système au moyen de la reconnaissance de motifs de chatoiement dans des fibres optiques multimode
Liu et al. Experimental demonstration of optical fiber eavesdropping detection based on deep learning
Peng et al. A photonics-inspired compact network: toward real-time AI processing in communication systems
US11047766B2 (en) Systems and methods for identification and testing of optical fibers
Usmani et al. Road Traffic Detection with a LSTM Autoencoder using State of Polarization on Deployed Metropolitan Fiber Cable
EP3786606A1 (fr) Acquisition d'un réflectomètre de domaine temporel optique à base d'optiques parallèles
RU2668856C1 (ru) Система для получения изображения банкноты
US10020888B1 (en) Transmitter and receiver for direct communication of multiple optical wavelengths via an optical link
Hornsteiner Fiber Optic Technology Trends in Data Transmission: Digitalization of data advance the need for constant upgrading of data networks
CN109150297A (zh) 一种光模块灵敏度测试方法和系统
Esteves et al. Deep learning for BER prediction in optical connections impaired by inter-core crosstalk
EP1004954B9 (fr) Dispositif optique de traítement de signaux optiques numériques
Matsumoto et al. Proposal for neural-network applications to fiber-optic transmission
US20210376949A1 (en) Polarization division multiplexed (pdm) communication systems and devices and methods of use thereof

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19752428

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19752428

Country of ref document: EP

Kind code of ref document: A1