Blind detection model optimization
Through optical power level modulation technology and detection model optimization, the bandwidth utilization and decoding complexity of the shared optical communication path of multiple light sources in optical communication is solved, and efficient fiber bandwidth utilization and low-cost decoding are achieved.
Patent Information
- Application Number
- CN202080029494.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2019-04-18
- Filing Date
- 2020-03-24
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2040-03-24
AI Technical Summary
Existing optical communication technologies, such as WDM and DDPDM, fail to fully utilize the bandwidth of optical fiber media, and there are power level mismatch and decoding complexity problems when multiple light sources share optical communication paths, resulting in high equipment costs and decoding time. long.
Optical power level modulation (OPLM) technology is used to transmit multiple data streams on the same wavelength through light sources of different power levels, and use Poisson distribution or machine learning models for detection model training and optimization to simplify the demultiplexing process. Avoid continuous interference cancellation.
The bandwidth improvement of transmitting multiple data streams simultaneously on a single optical link is achieved, simplifying the decoding and demultiplexing process, reducing equipment costs and delays, and supporting communications of multiple users at the same time.
Smart Images

Figure CN113728571B_ABST
Abstract
Description
Background Art
[0001] Optical communications, such as fiber optic communications, utilize a light source at one end that transmits one or more data streams by modulating the data streams into light signals. These light signals travel through a medium such as air or glass fibers (optical fibers) with internal reflective surfaces to a receiver that employs a photon detection module to detect the light signals. The detected light is then demodulated back into one or more data streams.
[0002] To efficiently utilize the available optical bandwidth, multiple different channels can be created by assigning a different wavelength of light to each channel. Different data streams can be placed on each channel and transmitted simultaneously over the same medium to the same receiver. This practice is often called wavelength division multiplexing (WDM). Some WDM systems allow up to 80 such channels per fiber, and the bandwidth per channel may be 40Gbit / sec to produce almost 3.1TB / sec of transmission on a single fiber (excluding losses due to overhead).
[0003] As a result of this large bandwidth, fiber optic systems are becoming increasingly popular with communications network providers, cloud service providers, and other entities that need to transmit large amounts of data very quickly. In addition to carrying large amounts of data, fiber optics offer other advantages, such as: less attenuation than cables—which provides the benefit of longer distances of communications cables using less network infrastructure; no electromagnetic interference; and various other benefits. BRIEF DESCRIPTION OF THE DRAWINGS
[0004] In the drawings, which are not necessarily drawn to scale, like numbers may describe similar parts in different views. Like numbers with different letter suffixes may represent different instances of similar components. The accompanying drawings generally illustrate various embodiments discussed in this document by way of example and not limitation.
[0005] Figure 1 Components of a simplified optical communication system according to some examples of the present disclosure are illustrated.
[0006] Figure 2 An optical power level modulation (OPLM) method performed by a receiver when decoding data according to some examples of the present disclosure is illustrated.
[0007] Figure 3 Bit errors and subsequent blind decision optimization for three detection models according to some examples of the present disclosure are illustrated.
[0008] Figure 4 An example of a method of blindly updating a detection model according to some examples of the present disclosure is illustrated.
[0009] Figure 5A method of updating a detection model according to some examples of the present disclosure is illustrated.
[0010] Figure 6 A schematic diagram of a system for increasing optical fiber bandwidth according to some examples of the present disclosure is illustrated.
[0011] Figure 7 A schematic diagram of a receiver according to some examples of the present disclosure is illustrated.
[0012] Figure 8 Illustrated are example machine learning components according to some examples of the present disclosure.
[0013] Fig. 9 is a block diagram illustrating an example of a machine upon which one or more embodiments may be implemented. DETAILED DESCRIPTION
[0014] Figure 1 Components of a simplified optical communication system in the form of an optical fiber system 100 according to some examples of the present disclosure are illustrated. A data stream 105 may include binary data generated by a higher network layer processed by a processing circuit 110. The processing circuit 110 may process the data of the data stream 105 in one or more ways to prepare for transmission. Example processing operations performed by the processing circuit 110 include applying one or more error correction codes (ECC), compression algorithms, encryption algorithms, etc. The data transformed by the processing circuit 110 is then passed to the light source 115 as a control signal. The light source 115 modulates the data by selectively turning on and off the light source according to the input data according to a modulation scheme. For example, in a simple modulation scheme, each bit can be transmitted during a predetermined time period (e.g., a time slot). During a particular time slot, if the current bit from the input data is '1', the light source can be turned on during the time slot, and if the current bit from the input data is '0', the light source can be turned off during the time slot. Other more complex modulation schemes, such as amplitude, phase, or polarization modulation, may be used. In some examples, light may be modulated on a sine wave.
[0015] The light generated by the light source then propagates to the receiver via an optical communication path. The optical communication path is the path taken by the light source from the transmitting light source to the receiving sensor. The path can be through one or more media, such as a single optical fiber, air, etc. Figure 1 In the example of , the optical communication path passes through a single optical fiber 120. In the example where the medium is air, the optical communication path can be an alignment of a transmitting light source and a sensor at a receiver.
[0016] The receiver includes a photodetector 125 and a processing circuit 130. The photodetector 125 collects counts of the number of photons detected during a detection period, which corresponds to the amount of time that a single bit of the data stream 105 is transmitted. Based on the photon counts, the photodetector produces a data stream, which is then input to the processing circuit 130, which applies the inverse of the operations that the processing circuit 110 applied to produce the data stream 135. The goal is to transmit the data stream 105 to the receiver as quickly as possible while making the data stream 135 match the data stream 105.
[0017] As already noted, WDM can be used to optimize the use of optical communication paths. Other techniques such as amplitude modulation (AM) and digital domain power division multiplexing (DDPDM) also increase the bandwidth of the medium. It will be clear that none of these techniques fully utilizes the entire bandwidth available in the medium, and in some cases, these techniques have undesirable disadvantages. For example, none of these techniques allow two different transmitters with two different light sources to transmit simultaneously at the same wavelength and through the same communication path (e.g., optical fiber) as the receiver. In addition, the decision area used to allocate bits to each stream is equal for each bit combination in these techniques (corresponding to the photon counting area of the detected bit combination). These symmetrical decision areas present difficulties when different transmitters have slightly different power levels. Finally, for DDPDM, the decoding, demodulation, and interference cancellation used are very complex and require a lot of processing resources. For example, DDPDM demodulates and remodulates the same signal multiple times at the receiver. This increases equipment cost and / or decoding time.
[0018] Another technique, optical power level modulation (OPLM), described by Amer Hassan and Edward Giaimo (Attorney Docket No. 1777.C02US1) in a co-pending application, “Power Switching for Systems Implementing Throughput Improvements for Optical Communications,” filed on the same date as the present application, the entire terms of which are incorporated herein by reference, allows multiple data streams to be transmitted over the same optical communication path (e.g., the same optical fiber) using different light sources transmitting at different power levels having the same wavelength. The OPLM technique can be combined with a variety of modulation schemes, such as simple modulation schemes in which the light source is turned on to send a ‘1’ and turned off to send a ‘0’, as well as more complex schemes such as AM, DDPDM, WDM, etc. In an OPLM system, each light source corresponding to each stream transmits at the same frequency and using different power levels on the same optical communication path. The receiver demultiplexes the data for each stream by applying one or more detection models to the photon counts observed at the receiver to determine the possible bit allocations for each stream.
[0019] Since each light source transmits at a different power level, a first light source activated alone, a second light source activated alone, or both activated together produce different photon counts detected at the receiver. A detection model in the form of a Poisson distribution (or other detection model) can calculate the probability of each light source being activated based on these photon counts. This can be used to assign bit values to each stream. Therefore, a detection model such as a Poisson distribution can be used to demultiplex the stream. For example, consider a system with two data streams and two light sources, using a modulation scheme in which the light source activation is signaled with a '1' in the bit stream, and the light source is turned off with a '0' in the bit stream. In this system, a first model can indicate the probability that the first light source is turned on and the second light source is turned off - this corresponds to a bit combination of '1' and '0'. A second model can indicate the probability that the second light source is turned on and the first light source is turned off - this corresponds to a bit combination of '0' and '1'. A third model can indicate the probability that both light sources are activated and therefore corresponds to a bit combination of '1' and '1'.
[0020] As a result, multiple data streams from multiple light sources can be sent over a single optical link that can have double, triple, quadruple, or more times the bandwidth of a single channel on a single link. In addition, problems with other techniques can be avoided. For example, the decision area for each bit combination (e.g., Poisson distribution) may be different from the decision area for other bit combinations (e.g., Figure 3The area under the curve in ( ) is different. This allows light sources with different power levels. Also, unlike DDPDM, detection is very simple. In addition, since multiple light sources can be used, the system can support multiple users on the same communication path (e.g., the same optical fiber).
[0021] Figure 2 An OPLM method 200 performed by a receiver in demultiplexing data according to some examples of the present disclosure is illustrated. At operation 210, the receiver may determine a photon count of photons observed during a predetermined time period. The predetermined time period may be a period of time (e.g., a time slot) during which the transmitter and the receiver are synchronized to transmit one or more bits of a bit stream (e.g., packetized bits). At operation 215, the receiver determines a first probability that a first light source corresponding to a first stream of data is turned on at a first power level and a second light source corresponding to a second stream of data is turned off using the photon counts and a first detection model. At operation 220, the receiver determines a second probability that a first light source corresponding to the first stream of data is turned off and a second light source corresponding to the second stream of data is turned on at a second power level using the photon counts and a second detection model. At operation 225, the receiver determines a third probability that the first light source is turned on at a first power level and the second light source is turned on at a second power level using the photon counts and a third detection model.
[0022] The detection model can be a Poisson probability distribution, a machine learning model (e.g., a neural network, a regression model, a clustering model, a decision tree), etc. For example, as described above, a first light source alone transmits at a first power level, a second light source alone transmits at a second power level, and the first and second light sources together transmit can produce different average numbers of photons that strike a receiver. Therefore, the probability that a particular combination of light sources causes those photons to strike a receiver can be modeled using a Poisson distribution. These Poisson distributions can be used as detection models.
[0023] In operation 230, the system can determine the bit values of the first stream of data and the second stream of data based on the first probability, the second probability, and the third probability. For example, the model that produces the highest probability value can be selected and the bit value corresponding to the model can be assigned to the bit stream. As described above, the detection model can correspond to the bit values of various data streams. In some examples, the value of '0' of the two bit streams can be determined by comparing the photon count with a predetermined minimum threshold (e.g., before operations 215, 220, and 225 or during operation 230). In other examples, a separate model can be used for the '0' value of the two bit streams. Method 200 is used for two light sources and two bit streams, but can be applied to additional light sources with additional bit streams. In other examples, a single model can output a bit allocation.
[0024] The OPLM system solves the technical problem of efficient bandwidth utilization in optical communications without the disadvantages of the above-mentioned previous methods. For example, the method allows the transmission of multiple data streams using a single light source or the transmission of multiple data streams using multiple light sources. In the present disclosure, any interference from multiple light sources is explained by the detection model trained using any such interference. In addition, since the model may have unequal probability of decision regions, the use of different light sources with different power levels does not cause problems as with AM and DDPDM. In addition, these models may be adjusted over time to take into account aging transmitter circuits. In contrast to DDPDM, the present disclosure does not require the remodulation of the received signal by performing continuous interference cancellation. Instead, the present disclosure utilizes the average photon count of a specific bit combination. Since the disclosed detection model is a relatively simple probability distribution, the process of decoding and demultiplexing the data stream can use relatively simple, cheap and fast hardware and / or software to demultiplex the input without the need for more complex hardware, such as those required for the method using continuous interference cancellation.
[0025] The OPLM uses a training sequence to determine the detection model. For example, by instructing each light source to activate individually, and combining with other light sources to determine the average photon count for each combination. The average photon count can be used to build a detection model, such as a Poisson probability model, or to train a supervised or unsupervised machine learning model. During operation, the photon counts observed during the transmission time slot are submitted to each detection model and the detection model with the highest probability can be selected. The bit allocation corresponding to this detection model can be assigned to the bits for each specific stream.
[0026] While this works well for the initial setup model, changes in the medium, such as changes in the fiber, changes in the power output of the light source over time, or other factors, can cause errors in demultiplexing the bits. Rerunning the training sequence periodically can correct these errors, but may cause data transmission delays while the training sequence is rerun.
[0027] In some examples, methods, systems, optical devices, and machine-readable media for blind detection model optimization are disclosed. A receiver may monitor one or more optimization metrics to determine whether the optimization metrics satisfy an optimization condition. Once the optimization condition is observed, the system may perform blind detection model optimization without using a training sequence with a transmitter. That is, the detection model is optimized using normal data sent by the transmitter. For example, if the optimization condition is an error threshold and if the error rate determined by the ECC exceeds the error threshold, the system may start detection model optimization. The system may save an observed data set, the observed data set including multiple groups of photon counts and demultiplexed bit combinations and / or correction bit combinations determined by the ECC. The observed data set may include both correctly demultiplexed and incorrectly demultiplexed data. The observation set may be used to adjust the detection model to reduce the error rate. For example, the median and / or range of the detection model may be adjusted to produce the lowest error rate in the observation set.
[0028] Figure 3 3 illustrates bit errors and subsequent blind decision optimization 300 for three detection models according to some examples of the present disclosure. Figure 3 In the example shown in , the detection model is a Poisson probability curve with photon counts on the x-axis and composite probability on the y-axis. Figure 3 In the example of FIG. 3 , a simple modulation scheme is used in conjunction with OPLM whereby the light source is turned off to transmit a zero and activated to transmit a 1 during a particular time slot. Other modulation schemes may be used in conjunction with OPLM to achieve higher data rates. The bit combinations for each of the original detection patterns 320 , 325 , and 330 are shown at the top of the vertical line bisecting each pattern.
[0029] exist Figure 3 In the example of FIG. 1 , a photon count 340 is received (represented by the vertical line). According to the original detection models 320, 325, and 330, this is demultiplexed into a bit combination of 0, 1 because the original detection model 325 produces the highest probability (shown by the vertical line representing the photon count 340 intersecting the probability model). If a subsequent ECC check indicates that an error exists, the error metric can be updated. If the error metric exceeds a threshold (e.g., an optimization condition is observed), the system can optimize the model based on the observed data set.
[0030] exist Figure 3In the example of FIG. 1 , assuming that there is only one sample photon count 340 in the observed data set, with incorrect bit assignments 0 1 and correct bit assignments (determined by ECC) 1,1, the system may try and optimize the original detection models 320, 325, and 330 to correct the error. For example, the original detection models 320, 325, and 330 may be moved to the left on the diagram to produce updated detection models 321, 326, and 331. Using the updated detection models 321, 326, and 331, model 331 is now the model with the highest probability of returning photon count 340, and therefore the new bit assignment will be 1,1, which is correct according to the ECC. Subsequent demodulation attempts will utilize the updated detection model.
[0031] For a detection model based on a Poisson distribution, to update the detection model, the system can calculate a new detection model based on the photon counts of the data in the observed data set. For example, for each bit combination determined by the ECC data, the observed photon counts can be averaged to determine a new Poisson distribution for that bit combination. For example, if the observed data is as shown in the following table:
[0032] Photon counting Actual bit value (as determined by ECC) 0 (0,0) 30 (0,1) 45 (0,1) 17 (0,1) 33 (0,1) 50 (1,0) 55 (1,0) 67 (1,0) 88 (1,0) 115 (1,1) 122 (1,1) 137 (1,1)
[0033] The new average photon count per bit combination is given by:
[0034] Bit combination Average photon count (0,0) 0 (0,1) 31.25 (1,0) 65 (1,1) 124.6
[0035] In some examples, the model can be replaced entirely by the observed photon count data from the observed data. In other examples, the photon counts from the training process run when the model was initially built can be combined with the observed data to form a new model. In these examples, the training data can be combined in a way that weighs its contribution to the new detection model based on the recency of the observed training data. For example, for a Poisson distribution model, the receiver can use the following formula:
[0036] New average photon count
[0037] =((1-x)*training average photon count)+((x)
[0038] *Average observed photon count)
[0039] Where 0 <x<1
[0040] As noted, x can be based on the time that has passed since training occurred. In other examples, x can be based on an optimization metric. For example, x may depend on the error rate, such that the higher the error rate, the higher x is, and the less the receiver relies on the old model when computing the new model (because it indicates that the training data is less reliable).
[0041] For non-Poisson distributed detection models, such as machine learning models, the system can apply observed data to update the model by retraining the model using both previously observed data and training data or by utilizing incremental learning methods to acquire new knowledge without forgetting previously acquired knowledge. An example algorithm of the latter is the Learn++ algorithm for supervised neural networks.
[0042] Figure 4 An example of a method 400 for blindly updating a detection model according to some examples of the present disclosure is shown. At operation 410, a receiver demultiplexes a first and a second stream of data. For example, by performing Figure 2 operation. As previously described, for each predetermined time period, a photon count is determined by the observed photons that strike the photon detector of the receiver. The photon count is submitted to one or more detection models. For example, for two streams generated by two light sources at two different power levels, a first detection model may provide a probability that the photon count is generated by the first light source activated alone at the first power level, a second detection model may provide a probability that the photon count is generated by the second light source activated alone at the second power level, and a third detection model may provide a probability that the photon count is generated by both the first and second light sources activated at the first and second power levels. In some examples, a fourth detection model may provide a probability that the photon count is not generated by the light source being activated. In other examples, if the photon count is below a minimum photon count, it may be determined that no light source may have been activated during the time period. In some examples, the model may be a Poisson probability.
[0043] As previously described, each detection model can be associated with a specific bit allocation for each data stream. In a simple modulation scheme, where a light source is activated to signal a '1' for the bit stream it transmits and a light source is turned off to signal a '0' for the bit stream it transmits: a first detection model can correspond to a bit allocation of 1 for a first stream corresponding to a first light source and 0 for a second stream corresponding to a second light source, a second detection model can correspond to a bit allocation of 1 for a second stream corresponding to a second light source and 0 for a first stream corresponding to the first light source, and a third detection model can correspond to a bit allocation of 1 for each stream. In some examples, the model with the highest probability given the observed photon counts is selected and bits are allocated to the bit streams using the bit allocation corresponding to that model.
[0044] In operation 415, an optimization metric may be identified, calculated, updated, or otherwise determined. For example, an error correction code (ECC) scheme may be applied to the first and second streams of data. Example ECC schemes include convolutional codes, block codes, Reed-Solomon codes, Hamming codes, Turbo codes, low-density parity-check codes (LDPC), etc. These ECC schemes may not only be able to detect errors, but also correct one or more errors in the data. In some examples, the optimization metric may be an error rate, such as the percentage of decoded data with correctable or uncorrectable ECC errors. Since ECC is applied to both streams, the metric may be a combined ECC error rate, or a separate error rate for each separate stream.
[0045] At operation 417, the receiver compares the optimization metric to the optimization condition to determine whether the optimization metric matches the optimization condition. For example, if the error rate exceeds an error threshold. If the optimization metric does not match the optimization condition (e.g., the error rate does not exceed the threshold), the process continues back to demultiplexing the data. If the optimization metric does match the optimization condition (e.g., the error rate exceeds the threshold), the process proceeds to operation 420. In some examples, the optimization condition can be a single error adjusted bit by bit.
[0046] At operation 420, one or more of the detection models may be updated. For example, the detection model may be updated based on an observed data set. The observed data set may include received photon counts and corresponding actual bit allocations derived from error correction operations performed on predicted bit allocations determined during demultiplexing. For example, when there are no ECC errors, the system may store received photon counts and associated bit allocations resulting from the photon counts. When there are ECC errors, the erroneous bits may be corrected, and the system may store received photon counts and associated corrected bit allocations.
[0047] As previously described, the model can be updated by calculating a new average photon count using the observed data of the Poisson distribution detection model. In other examples, the model can be updated by making one or more adjustments to the model - for example, moving the median or distribution of the model. In some examples, multiple random shifts can be made and then checked with the observation set to see if the error rate on the observation set will decrease. The random shift that produces the minimum error rate on the observation set can be used as the updated model. In examples where the detection model is a machine learning model, the model can be updated by retraining the model using the observation set and / or training data set or utilizing an incremental learning algorithm to update the model of a machine learning model (e.g., a neural network model and other supervised machine learning models).
[0048] In other examples, instead of storing the observed data, a running average photon count can be maintained for each corrected bit (determined by the ECC) assignment. Thus, the average photon count and bit assignment (determined by the ECC) are added to the running average photon count for that particular bit assignment, and the model is immediately updated with that particular average photon count. This allows the bit assignments to react quickly to changing conditions. In some examples, the change in average photon count can be applied to all detection models, but in other examples it can be applied only to the detection models corresponding to the ECC corrected bit assignments.
[0049] In examples where the model is a machine learning algorithm, each corrected bit combination and associated photon count may be fed to an incremental learning algorithm to update the model on a bit-by-bit basis.
[0050] In some examples, if the ECC errors are uncorrectable, some data may be lost and a retransmission sequence may be initiated. In some examples, uncorrectable ECC errors may not be used to optimize the model. In some examples, uncorrectable ECC errors may still be factored into the optimization metric and used as a factor in whether the model is optimized. In other examples, if the number or rate of uncorrectable ECC errors exceeds a threshold, the receiver may initiate a new training sequence.
[0051] Now turn to Figure 5 , a method 500 for updating a detection model according to some examples of the present disclosure is shown. Method 500 is applicable to a Poisson detection model. At operation 510, an average photon count of a first detection model is determined based on observed data and / or training data. For example, if the first detection model corresponds to a bit assignment of 1 for the first stream and 0 for the second stream, the photon counts corresponding to any ECC-corrected 1,0 bit assignments from any observed data (and / or training data) can be added and divided by the number of samples to determine the average photon count. In some examples, photon counts (or average photon counts) from one or more previous training stages may also be used to calculate an updated detection model. At operation 515, the new average photon count is used to determine an updated first detection model. For example, a Poisson detection model may be given by the following equation:
[0052] Equation 1:
[0053] At operation 520, an average photon count for a second detection model is determined based on the observed data and / or training data. For example, if the second detection model corresponds to a bit assignment of 1 for the second stream and 0 for the first stream, the photon counts corresponding to the 0,1 bit assignments corrected by any ECC from any observed data can be added and divided by the number of samples to determine the average photon count. In some examples, photon counts (or average photon counts) from one or more previous training phases may also be used to calculate the updated detection model. At operation 525, the new average photon count is used to determine the updated second detection model. For example, Equation 1 shown above.
[0054] At operation 530, an average photon count for a third detection model is determined based on the observed data and / or training data. For example, if the third detection model corresponds to a bit assignment of 1 for the second stream and 1 for the first stream, the photon counts corresponding to the 1,1 bit assignments corrected by any ECC from any observed data can be added and divided by the number of samples to determine the average photon count. In some examples, photon counts (or average photon counts) from one or more previous training phases may also be used to calculate the updated detection model. At operation 535, the new average photon count is used to determine an updated second detection model. For example, Equation 1 shown above.
[0055] In some examples, the receiver may use additional detection models in those examples (eg, for 0,0) and may also update those detection models. Figure 5 An example of method 500 is shown in which all detection models are updated, but in other examples, not all detection models may be updated. For example, the optimization metric and condition may be specific to a particular bit combination. In these examples, only models having a bit combination in which the optimization metric matches the optimization criteria may be updated.
[0056] As previously described, the update to the model can be calculated using the corrected bit allocation values. The corrected bit allocation can be determined by the ECC applied to each stream. The ECC of each stream can be independent of another stream. That is, the ECC applied to the first stream can be different from the ECC applied to the second stream, and so on.
[0057] The observed data sets may be retained for a specific period of time. For example, a predetermined number of recently received data sets may be stored and used to adjust the detection model. In other examples, the receiver may store the observed data for each model separately. Thus, for example, a first detection model may have a buffer for storing observed data corresponding to bit combinations associated with the first detection model, a second detection model may have a buffer for storing observed data corresponding to bit combinations associated with the second detection model, and so on. Once the buffer is full, the oldest observed data may be deleted and replaced with new observed data, and so on. This ensures that the observed data is not dominated by data corresponding to a single detection model and provides a broad basis for calculating all detection models.
[0058] Example Transmitter and Receiver
[0059] Now turn to Figure 6 , a schematic diagram of a system 600 for increasing optical fiber bandwidth according to some examples of the present disclosure is shown. A first transmitter 605 may include a processing circuit 610 to convert a data stream to prepare it for transmission on an optical fiber. Example operations include applying ECC, encryption, modulation operations, etc. The converted bits are used as signals for a controller 620 to instruct a light source 615 to selectively turn on or off to represent the converted bit stream according to a modulation scheme. For example, the light source 615 is turned on in response to a '1' in the bit stream and turned off in response to a '0' in the bit stream. The controller 620 can set the power of the light source 615 based on the power level indicated in the allocated power level allocation scheme and based on the current stage of the power level allocation scheme. A power level allocation scheme is any formula or plan for coordinating different power levels across two or more transmitters. For example, a transmitter can alternate which transmitter transmits at a specific power level. The power level allocation scheme can be divided into one or more stages. A stage specifies a unit of a power level allocation scheme, in which each transmitter served by the scheme is assigned a power level for a defined duration or until a defined event occurs. The duration may be time-based, data-length-based (e.g., a defined number of time slots), etc. In some examples, the detection model used by the receiver may be specific to the current phase of the power level allocation scheme. The power level allocation scheme may be described by one or more data structures. For example, a formula, table, chart, or other indicator. In the case of a modulation scheme that changes power, the power level may be the average power level over a particular time slot. An indication of which power level allocation scheme is active and which phase is active may be stored in the power level allocation scheme storage 665.
[0060] Light source 615 transmits light on an optical communication path, which may be through a medium such as an optical fiber to a receiver. Example light sources may include LED or LASER light sources. Controller 620 and processing circuit 610 may be general purpose processors or may be specially designed circuits configured to implement the techniques described herein. Power level allocation scheme storage 665 may be flash memory, read-only memory (ROM), or other temporary or non-temporary storage.
[0061] The first transmitter 605 and the second transmitter 650 may be transceivers, as they may have associated receivers, such as receivers 625, 658. The power level allocation scheme may be allocated by the receiver (which may also be a transceiver 660) by reaching an agreement with the second transmitter 650, etc. The allocated power level allocation scheme may be one of a library of predetermined allocation schemes stored in a power level allocation scheme storage 665. In some examples, the allocated power level allocation scheme may be based on a scheme in the library of allocation schemes, but modified for one or more of the specific transmitters and receivers involved in the communication session. In other examples, the allocated power level allocation scheme may be customized for a specific communication session. The power level allocation scheme storage 665 may store a specific allocation scheme, a selection of a specific allocation scheme, any customization in use, a current stage, etc.
[0062] The receiver 625 may be a fiber optic receiver, but may also be an out-of-band receiver such as a WiFi receiver, a Bluetooth receiver, an Ethernet receiver, etc. The receiver 625 may receive instructions from the transceiver 660 that are passed to the controller to turn the light source 615 on or off during model training of the receiver.
[0063] The second transmitter 650 may include similar components as the first transmitter 605. For example, a controller 654, a light source 656, a processing circuit 652, a receiver 658, a power level allocation scheme storage 670, etc. In some examples, if the first transmitter 605 and the second transmitter 650 are in the same device, one or more components may be shared between the first transmitter 605 and the second transmitter 650. In addition, the first transmitter 605 and the second transmitter 650 may send multiple data streams to the transceiver 660 on the optical cable at multiple different wavelengths. Therefore, the first transmitter 605 and the second transmitter 650 can use the two techniques of the present invention to send multiple data streams simultaneously on the same optical fiber by changing the power level, and also send multiple data streams using different wavelengths.
[0064] Figure 7A schematic diagram of a receiver 700 according to some examples of the present disclosure is shown. For example, the receiver 700 can be an example receiver as part of the transceiver 660. The receiver 700 may include a photodetector 705 that detects and / or counts photons received through an optical communication path such as an optical fiber within a predetermined time period (e.g., a time slot). The photon counts are passed to a controller 710. The controller 710 may determine the individual bits in the bit stream using one or more detection models stored in a model storage 735. For example, the model may include one or more Poisson distributions that may return the probability that the photon count corresponds to one or more specific bit combinations for each stream. The specific detection model to be used may be selected based on the current stage of the current power level allocation scheme. The current stage and / or the selected power level allocation scheme may be stored in the power level allocation scheme storage 740.
[0065] For example, consider a simple power level allocation scheme where two light sources are transmitting simultaneously across the same communication path (e.g., optical fiber) at the same wavelength. The power level allocation scheme alternates which of the two light sources corresponding to two different data streams is activated at a high power level on a bit-by-bit basis. In the first bit, stream 1 is a high power light source and stream 2 is a low power light source. The photon counts received during the time period when the first bit is transmitted are submitted to a first set of detection models, which includes models trained to detect the first light source activated at high power (the second light source is off), the second light source activated at low power (the first light source is off), and both activated at their respective specified powers. The detection model that returns the highest score (e.g., detection probability) is used to assign values to the bit streams. For example, if the detection model trained to detect the first light source activated at high power (the second light source is off) returns the highest probability, then a '1' is assigned to the bit stream corresponding to the first light source and a zero is assigned to the bit stream corresponding to the second light source (e.g., based on a modulation scheme where a '1' is indicated by the activation of a light source and a '0' is indicated by the light source being off).
[0066] In the second phase, stream 1 is a low power light source and stream 2 is a high power light source. The photon counts received during the time period in which the second bit is to be transmitted are submitted to a second set of detection models, which includes models trained to detect the first light source activated at low power (without activating the second light source), the second light source activated at high power (without activating the first light source), and both emitting '1's at their respective designated powers. The detection model that returns the highest score (e.g., detection probability) is used to assign values to the bit streams. For example, if the detection model trained to detect the first light source activated at low power (without activating the second device) returns the highest probability, then '1' is assigned to the bit stream corresponding to the first light source, and zero is assigned to the bit stream corresponding to the second light source.
[0067] Each bit stream determined by the controller is then passed to processing circuits 715 and 720, respectively, which decode the bit stream and perform various operations (e.g. Figure 6 The model updater 755 may be configured to perform an inverse operation of the operations performed by the processing circuits 610 and 652 of the transmitter in the model storage 735 (e.g., a physical layer, a transport layer, or other network layer) and output a bit stream to a higher level layer (e.g., a physical layer, a transport layer, or other network layer). Example inverse operations may include detecting and correcting errors using ECC. When the processing circuits 715, 720 decode the ECC, the correct bit allocation and photon counts are stored in the observed data storage area 745. As the demultiplexing operation proceeds, the model updater 755 monitors one or more optimization metrics and compares these optimization metrics with one or more optimization conditions stored in the optimization condition data storage area 760. The optimization conditions may be predetermined by an administrator (and input into the optimization conditions through one or more user interfaces), input by a system designer, and the like. If the optimization metrics satisfy the optimization conditions, the model updater may update the detection model in the model storage 735 using the observed data in the observed data storage area 745. In an example where the model is optimized in a bit-by-bit manner, the model updater 755 continuously updates the model 735.
[0068] The calibration component 725 can initially train the model and can include a model training component 730 that can instruct the emitter (via the emitter 750) to emit various test data sequences. The model can be constructed using the photon counts observed by the photodetector 705. For example, by creating a model using average photon counts, or by submitting photon counts to a training module, such as Figure 8 Training component 810.
[0069] In some examples, the controller 710 may also select and control a power level allocation scheme. For example, a scheme may be selected and / or customized by communicating with the transmitter. This may occur periodically before and / or during a communication session with the transmitter. In other examples, where the transmitters agree on a power level allocation scheme, the controller 710 receives a message indicating which power level allocation scheme is active. The controller may determine the current phase by messaging to and / or from one or more transmitters (e.g., for QoS-based methods or modifications), based on an elapsed time from the last phase, and the like.
[0070] Figure 8An example machine learning component 800 according to some examples of the present disclosure is shown. The machine learning component 800 can be implemented in whole or in part by the model training component 730 and / or the model updater 755. The machine learning component 800 can include a training component 810 and a prediction component 820. In some examples, the training component 810 can be implemented by a different device than the prediction component 820. In these examples, a model 880 (e.g., a detection model) can be created on a first machine and then sent to a second machine.
[0071] The machine learning component 800 utilizes a training component 810 and a prediction component 820. The training component 810 inputs feature data 830 into a feature determination component 850. The feature data 830 can be photon counts, phases, etc. In some examples, the feature data can be explicitly labeled with the bit allocation for each stream, the light source currently emitting, the power level at which the light source currently emitting is emitting, etc. For example, the feature data can be or include data from an observed data set.
[0072] The feature determination component 850 determines one or more features of the feature vector 860 from the feature data 830. The features of the feature vector 860 are a set of information inputs and are information determined to predict the bit allocation of each stream. The features selected to be included in the feature vector 860 can be all feature data 830, or in some examples, can be a subset of all feature data 830. In examples where the features selected for the feature vector 860 are subsets of the feature data 830, a predetermined list of feature data 830 included in the feature vector can be utilized. The machine learning algorithm 870 can utilize the feature vector 860 (along with any applicable labels) to generate one or more detection models or other models 880.
[0073] In the prediction component 820, the current feature data 890 (e.g., photon counts) can be input to the feature determination component 895. The feature determination component 895 can determine the same feature set or a different feature set as the feature determination component 850. In some examples, the feature determination components 850 and 895 are the same component or different instances of the same component. The feature determination component 895 produces a feature vector 897, which is input into the model 880 to determine results 899 such as bit allocation, phase, power level allocation scheme, etc.
[0074] The training component 810 can operate in an offline manner to train the model 880. However, the prediction component 820 can be designed to operate in an online manner. It should be noted that the model 880 can be regularly updated through additional training and / or user feedback. For example, an observed data set can be used to retrain or update the model 880.
[0075] The machine learning algorithm 870 can be selected from many different potential supervised or unsupervised machine learning algorithms. Examples of supervised learning algorithms include artificial neural networks, convolutional neural networks, Bayesian networks, instance-based learning, support vector machines, decision trees (e.g., iterative bisection 3, C4.5, classification and regression trees (CART), chi-square automatic interaction detector (CHAID), etc.), random forests, linear classifiers, quadratic classifiers, k-nearest neighbors, linear regression, logistic regression, support vector machines, perceptrons, and hidden Markov models. Examples of unsupervised learning algorithms include expectation maximization algorithms, vector quantization, and information bottleneck methods. Unsupervised models may not have training components 810. In some examples, the model 880 can determine the bits of each stream based on the detected photons. In other examples, the model 880 can generate scores or probabilities for each stream that sends a specific bit.
[0076] As described above, the machine learning model can also be used to select a power level allocation scheme. In these examples, the feature data 830, 890 can be information that predicts an appropriate power level allocation scheme. The features discussed above can be used as the feature data 830, 890 - such as power budget, transmitter characteristics, receiver characteristics, etc. The result 899 can be a ranking and / or selection of a power level allocation scheme.
[0077] The modulation scheme used here has been relatively simple (on or off to represent '1' or '0'). In other examples, different modulation schemes may be used. For example, if the light source and receiver are capable, WDM, phase shift modulation, amplitude modulation, and other advanced modulation forms may be utilized in addition to the techniques described herein. For example, multiple bit streams may be split into multiple wavelengths - each of which may have multiple data streams sent using the methods disclosed herein. Similarly, for power modulation, the power allocation scheme of the present invention may allocate multiple power levels to each transmitter - each of which is a specific bit combination. Thus, a first transmitter may be assigned power levels 1, 2, and 3 (to indicate '01', '10', and '11' bits, respectively), while a second transmitter may be assigned power levels 4, 5, and 6 (to indicate '01', '10', and '11' bits). In this example, the system may allocate power levels so that the average photon counts for each power level combination are different enough so that the probability distributions are far enough apart so that the error rate is low.
[0078] Fig. 9A block diagram of an example machine 900 is shown, on which any one or more techniques (e.g., methods) discussed herein can be executed. In alternative embodiments, the machine 900 can operate as a standalone device or can be connected (e.g., networked) to other machines. In a networked deployment, the machine 900 can operate in a server-client network environment with the capabilities of a server machine, a client machine, or both. In one example, the machine 900 can act as a peer machine in a peer-to-peer (P2P) (or other distributed) network environment. The machine 900 can be a personal computer (PC), a tablet computer, a set-top box (STB), a personal digital assistant (PDA), a mobile phone, a smart phone, a web tool, a network router, a switch or a bridge, or any machine capable of executing instructions (sequential or other) specifying the actions to be taken by the machine. The machine 900 can implement a transmitter and / or receiver disclosed herein. In addition, the machine 900 can include a transmitter and / or receiver disclosed herein. The machine 900 can implement any method disclosed herein. Further, while only one machine is illustrated, the term "machine" shall also be taken to include any collection of machines that individually or jointly execute a set (or multiple sets) of instructions to perform any one or more of the methodologies discussed herein, such as cloud computing, Software as a Service (SaaS), and other computer cluster configurations.
[0079] As described herein, examples may include logic or multiple components, components or mechanisms, or may operate on them. A component is a tangible entity (e.g., hardware) that is capable of performing a specified operation and may be configured or arranged in a certain manner. In one example, a circuit may be arranged in a specified manner (e.g., internally or relative to an external entity such as other circuits) as a component. In one example, all or part of one or more computer systems (e.g., independent, client or server computer systems) or one or more hardware processors may be configured by firmware or software (e.g., instructions, application parts or applications) as a component that performs a specified operation. In an example, the software may reside on a machine-readable medium. In one example, the software causes the hardware to perform a specified operation when executed by the underlying hardware of the component.
[0080] Thus, the term "component" is understood to encompass a tangible entity that is physically constructed, specifically configured (e.g., hardwired), or temporarily (e.g., temporarily) configured (e.g., programmed) to operate in a specified manner or to perform part or all of any of the operations described herein. Considering the example of temporarily configured components, it is not necessary to instantiate each component at any time. For example, where the components include a general-purpose hardware processor configured using software, the general-purpose hardware processor can be configured as different components at different times. The software can configure the hardware processor accordingly, for example, to constitute a particular component at one instance in time and to constitute a different component at a different instance in time.
[0081] The machine (e.g., computer system) 900 may include a hardware processor 902 (e.g., a central processing unit (CPU), a graphics processing unit (GPU), a hardware processor core, or any combination thereof), a main memory 904, and a static memory 906, some or all of which may communicate with each other via an interconnect (e.g., a bus) 908. The machine 900 may also include a display unit 910, an alphanumeric input device 912 (e.g., a keyboard), and a user interface (UI) navigation device 914 (e.g., a mouse). In an example, the display unit 910, the input device 912, and the UI navigation device 914 may be a touch screen display. The machine 900 may additionally include a storage device (e.g., a drive unit) 916, a signal generating device 918 (e.g., a speaker), a network interface device 920, and one or more sensors 921, such as a global positioning system (GPS) sensor, a compass, an accelerometer, or other sensors. The machine 900 may include an output controller 928, such as a serial (e.g., universal serial bus (USB), parallel or other wired or wireless (e.g., infrared (IR), near field communication (NFC), etc.) connection to communicate or control one or more peripheral devices (e.g., printer, card reader, etc.).
[0082] The storage device 916 may include a machine-readable medium 922 on which is stored one or more sets of data structures or instructions 924 (e.g., software) that embody or are used by any one or more of the techniques or functionality described herein. The instructions 924 may also reside, completely or at least partially, within the main memory 904, within the static memory 906, or within the hardware processor 902 during execution by the machine 900. In an example, one or any combination of the hardware processor 902, the main memory 904, the static memory 906, or the storage device 916 may constitute a machine-readable medium.
[0083] Although machine-readable medium 922 is illustrated as a single medium, the term “machine-readable medium” may include a single medium or multiple media (eg, a centralized or distributed database, and / or associated caches and servers) configured to store one or more instructions 924 .
[0084] The term "machine-readable medium" may include any medium capable of storing, encoding, or carrying instructions executed by the machine 1800 and causing the machine 900 to perform any one or more of the techniques of the present disclosure or capable of storing, encoding, or carrying data structures used by or associated with such instructions. Non-limiting machine-readable medium examples may include solid-state memory and optical and magnetic media. Specific examples of machine-readable media may include: non-volatile memory, such as semiconductor memory devices (e.g., electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)) and flash memory devices; disks, such as built-in hard disks and removable disks; magneto-optical disks; random access memory (RAM); solid-state drives (SSD); and CD-ROM and DVD-ROM disks. In some examples, the machine-readable medium may include non-temporary machine-readable media. In some examples, the machine-readable medium may include machine-readable media that is not a transient propagation signal.
[0085] The instructions 924 may further be sent or received over a communication network 926 via the network interface device 920 using a transmission medium. The machine 900 may communicate with one or more other machines using any of a variety of transmission protocols (e.g., frame relay, Internet Protocol (IP), transmission control protocol (TCP), user datagram protocol (UDP), hypertext transfer protocol (HTTP), etc.). Example communication networks may include a local area network (LAN), a wide area network (WAN), a packet data network (e.g., the Internet), a mobile telephone network (e.g., a cellular network), a plain old telephone (POTS) network, and a wireless data network (e.g., the Institute of Electrical and Electronics Engineers (IEEE) 802.11 family of standards, known as The IEEE 802.16 family of standards, known as IEEE 802.15.4 family of standards, Long Term Evolution (LTE) family of standards, Universal Mobile Telecommunications System (UMTS) family of standards, peer-to-peer (P2P) networks, etc. In one example, the network interface device 920 may include one or more physical jacks (e.g., Ethernet, coaxial, or telephone jacks) or one or more antennas to connect to the communication network 926. In one example, the network interface device 920 may include multiple antennas to use at least one of single-input multiple-output (SIMO), multiple-input multiple-output (MIMO), or multiple-input single-output (MISO) technology for wireless communication. In some examples, the network interface device 920 may use multi-user MIMO technology for wireless communication.
[0086] Additional Notes and Examples
[0087] Example 1 is a method for receiving data via an optical communication path, the method comprising: demultiplexing first and second streams of data received via the optical communication path using multiple detection models; determining whether an optimization metric corresponding to an output of the demultiplexing operation matches an optimization condition; and in response to determining that the optimization metric corresponding to the output of the demultiplexing operation matches the optimization condition, updating at least one of the multiple detection models based on an observed data set, the observed data set comprising received photon counts and corresponding actual bit allocations determined by an error correction operation performed on the bit allocations determined during the demultiplexing operation.
[0088] In Example 2, the subject matter of Example 1 includes wherein the optimization metric is an error rate determined based on an error correction code (ECC), and the optimization condition is an error rate threshold.
[0089] In Example 3, the subject matter of Example 2 includes, wherein the error rate threshold is a single error.
[0090] In Example 4, the subject matter of Examples 1-3 includes wherein the first stream is transmitted by a first light source at a first power level and the second stream is transmitted by a second light source at a second power level, the first stream and the second stream being transmitted simultaneously at a same frequency.
[0091] In Example 5, the subject matter of Examples 1-4 includes, wherein the detection model is a Poisson probability model.
[0092] In Example 6, the subject matter of Examples 1-5 includes, wherein updating at least one of the plurality of detection models comprises calculating a Poisson distribution based on an average photon count of data in an observed data set that matches an actual bit allocation corresponding to the at least one of the plurality of detection models.
[0093] In Example 7, the subject matter of Examples 1-6 includes, wherein demultiplexing a first stream of data and a second stream of data received via an optical communication path using multiple detection models includes: submitting a count of photons detected using a photon detector during a time slot to the multiple detection models; and selecting bit allocations for the first and second streams corresponding to a detection model that returns a highest probability among the multiple detection models.
[0094] In Example 8, the subject matter of Examples 1-7 includes determining errors in a first bit allocation generated by demultiplexing using an error correction code (ECC); and saving a photon count of the generated first bit allocation and an ECC-based correction bit allocation as part of an observed data set.
[0095] In Example 9, the subject matter of Examples 1-8 includes, wherein updating at least one of the plurality of detection models comprises updating at least one of the plurality of detection models without using a training sequence.
[0096] In Example 10, the subject matter of Examples 1-9 includes, wherein updating at least one of the plurality of detection models comprises retraining at least one of the plurality of detection models using a training dataset of the observed dataset and training data collected during a training operation.
[0097] In Example 11, the subject matter of Examples 1-10 includes, wherein updating at least one of the plurality of detection models comprises applying the observed data set in an incremental learning algorithm to produce an updated at least one detection model of the plurality of detection models.
[0098] In Example 12, the subject matter of Examples 1-11 includes, wherein the error correction operation uses ECC.
[0099] Example 13 is a device for receiving data via an optical communication path, the device comprising: a processor; a memory comprising instructions, which, when executed by the processor, cause the device to perform operations including: demultiplexing first and second streams of data received via the optical communication path using multiple detection models; determining whether an optimization metric corresponding to an output of the demultiplexing operation matches an optimization condition; and in response to determining that the optimization metric corresponding to the output of the demultiplexing operation matches the optimization condition, updating at least one of the multiple detection models based on an observed data set, the observed data set comprising received photon counts and corresponding actual bit allocations determined by an error correction operation performed on the bit allocations determined during the demultiplexing operation.
[0100] In Example 14, the subject matter of Example 13 includes, wherein the optimization metric is an error rate determined based on an error correction code (ECC) and the optimization condition is an error rate threshold.
[0101] In Example 15, the subject matter of Example 14 includes, wherein the error rate threshold is a single error.
[0102] In Example 16, the subject matter of Examples 13-15 includes, wherein the first stream is transmitted by a first light source at a first power level and the second stream is transmitted by a second light source at a second power level, the first stream and the second stream being transmitted simultaneously at a same frequency.
[0103] In Example 17, the subject matter of Examples 13-16 includes, wherein the detection model is a Poisson probability model.
[0104] In Example 18, the subject matter of Examples 13-17 includes, wherein the operation of updating the at least one of the plurality of detection models includes calculating a Poisson distribution based on an average photon count of data in the observed dataset that matches an actual bit allocation corresponding to the at least one of the plurality of detection models.
[0105] In Example 19, the subject matter of Examples 13-18 includes, wherein the operation of demultiplexing the first and second streams of data received through the optical communication path using multiple detection models includes: submitting the photon counts detected using the photon detector during the time slot to the multiple detection models; and selecting bit allocations for the first and second streams corresponding to the detection model among the multiple detection models that returns the highest probability.
[0106] In Example 20, the subject matter of Examples 13-19 includes, wherein the operations further include: determining errors in a first bit allocation produced by the demultiplexing using an error correction code (ECC); and saving the photon count that produced the first bit allocation and the ECC-based correction bit allocation as part of an observed data set.
[0107] In Example 21, the subject matter of Examples 13-20 includes, wherein the operation of updating at least one of the plurality of detection models comprises updating the at least one of the plurality of detection models without using a training sequence.
[0108] In Example 22, the subject matter of Examples 13-21 includes, wherein the operation of updating at least one of the plurality of detection models comprises retraining at least one of the plurality of detection models using a training dataset of the observed dataset and training data collected during the training operation.
[0109] In Example 23, the subject matter of Examples 13-22 includes, wherein the operation of updating at least one of the plurality of detection models comprises applying the observed data set in an incremental learning algorithm to produce at least one of the updates of the plurality of detection models.
[0110] In Example 24, the subject matter of Examples 13-23 includes, wherein the error correction operation uses ECC.
[0111] Example 25 is a machine-readable medium comprising instructions for receiving data via an optical communication path, which instructions, when executed, cause the machine to perform operations including: demultiplexing a first stream and a second stream of data received via the optical communication path using multiple detection models; determining whether an optimization metric corresponding to an output of the demultiplexing operation matches an optimization condition; and in response to determining that the optimization metric corresponding to the output of the demultiplexing operation matches the optimization condition, updating at least one of the multiple detection models based on an observed data set, the observed data set comprising received photon counts and corresponding actual bit allocations determined by an error correction operation performed on the bit allocations determined during the demultiplexing operation.
[0112] In Example 26, the subject matter of Example 25 includes, wherein the optimization metric is an error rate determined based on an error correction code (ECC) and the optimization condition is an error rate threshold.
[0113] In Example 27, the subject matter of Example 26 includes, wherein the error rate threshold is a single error.
[0114] In Example 28, the subject matter of Examples 25-27 includes, wherein the first stream is transmitted by a first light source at a first power level, the second stream is transmitted by a second light source at a second power level, and the first stream and the second stream are transmitted simultaneously at a same frequency.
[0115] In Example 29, the subject matter of Examples 25-28 includes, wherein the detection model is a Poisson probability model.
[0116] In Example 30, the subject matter of Examples 25-29 includes, wherein the operation of updating at least one of the plurality of detection models includes calculating a Poisson distribution based on an average photon count of data in an observed data set that matches an actual bit allocation corresponding to the at least one of the plurality of detection models.
[0117] In Example 31, the subject matter of Examples 25-30 includes an operation in which demultiplexing a first stream of data and a second stream of data received via an optical communication path using multiple detection models includes: submitting photon counts detected using a photon detector during a time slot to the multiple detection models; and selecting bit allocations for the first stream and the second stream corresponding to a detection model among the multiple detection models that returns a highest probability.
[0118] In Example 32, the subject matter of Examples 25-31 includes, wherein the operations further include: determining errors in a first bit allocation produced by demultiplexing using an error correction code (ECC); and saving the photon count that produced the first bit allocation and the ECC-based correction bit allocation as part of the observed data set.
[0119] In Example 33, the subject matter of Examples 25-32 includes, wherein the operation of updating at least one of the plurality of detection models comprises updating at least one of the plurality of detection models without using a training sequence.
[0120] In Example 34, the subject matter of Examples 25-33 includes, wherein the operation of updating at least one of the plurality of detection models comprises retraining at least one of the plurality of detection models using a training dataset of the observed dataset and training data collected during the training operation.
[0121] In Example 35, the subject matter of Examples 25-34 includes, wherein the operation of updating at least one of the plurality of detection models comprises applying the observed data set in an incremental learning algorithm to produce an updated at least one detection model of the plurality of detection models.
[0122] In Example 36, the subject matter of Examples 25-35 includes, wherein the error correction operation uses ECC.
[0123] Example 37 is a device for receiving data through an optical communication path, the device comprising: a unit for demultiplexing a first stream of data and a second stream of data received through the optical communication path using multiple detection models; a unit for determining whether an optimization metric corresponding to the output of the demultiplexing operation matches an optimization condition; and in response to determining that the optimization metric corresponding to the output of the demultiplexing operation matches the optimization condition, a unit for updating at least one of the multiple detection models based on an observed data set, the observed data set including received photon counts and corresponding actual bit allocations determined by an error correction operation performed on the bit allocations determined during the demultiplexing operation.
[0124] In Example 38, the subject matter of Example 37 includes, wherein the optimization metric is an error rate determined based on an error correction code (ECC) and the optimization condition is an error rate threshold.
[0125] In Example 39, the subject matter of Example 38 includes, wherein the error rate threshold is a single error.
[0126] In Example 40, the subject matter of Examples 37-39 includes, wherein the first stream is transmitted by a first light source at a first power level, the second stream is transmitted by a second light source at a second power level, and the first stream and the second stream are transmitted simultaneously at a same frequency.
[0127] In Example 41, the subject matter of Examples 37-40 includes, wherein the detection model is a Poisson probability model.
[0128] In Example 42, the subject matter of Examples 37-41 includes, wherein the means for updating at least one of the plurality of detection models includes means for calculating a Poisson distribution based on an average photon count of data in an observed data set that matches an actual bit allocation corresponding to at least one of the plurality of detection models.
[0129] In Example 43, the subject matter of Examples 37-42 includes, wherein a unit for demultiplexing a first stream of data and a second stream of data received via an optical communication path using multiple detection models includes: a unit for submitting photon counts detected using a photon detector during a time slot to the multiple detection models; and a unit for selecting bit allocations for the first stream and the second stream corresponding to a detection model among the multiple detection models that returns a highest probability.
[0130] In Example 44, the subject matter of Examples 37-43 includes a unit for determining an error in a first bit allocation produced by demultiplexing using an error correction code (ECC); and a unit for saving the photon count that produced the first bit allocation and the ECC-based correction bit allocation as part of an observed data set.
[0131] In Example 45, the subject matter of Examples 37-44 includes, wherein the means for updating at least one of the plurality of detection models comprises means for updating at least one of the plurality of detection models without using a training sequence.
[0132] In Example 46, the subject matter of Examples 37-45 includes, wherein the means for updating at least one of the plurality of detection models includes means for retraining at least one of the plurality of detection models using a training dataset of the observed dataset and training data collected during a training operation.
[0133] In Example 47, the subject matter of Examples 37-46 includes, wherein the means for updating at least one of the plurality of detection models comprises means for applying the observed data set in an incremental learning algorithm to produce the updated at least one of the plurality of detection models.
[0134] In Example 48, the subject matter of Examples 37-47 includes, wherein the error correction operation uses ECC.
[0135] Example 49 is at least one machine-readable medium comprising instructions that, when executed by a processing circuit, cause the processing circuit to perform operations to implement any of Examples 1-48.
[0136] Example 50 is a kit comprising means for implementing any of Examples 1-48.
[0137] Example 51 is a system implementing any of Examples 1-48.
[0138] Example 52 is a method of implementing any of Examples 1-48.
Claims
1. A method for receiving data via an optical communication path, the method include: demultiplexing a first stream and a second stream of data received via the optical communication path using a plurality of detection models; determining whether an optimization metric corresponding to an output of the demultiplexing operation matches an optimization condition; as well as in response to determining that an optimization metric corresponding to an output of the demultiplexing operation matches the optimization condition, updating at least one of the plurality of detection models based on an observed data set including received photon counts and corresponding actual bit allocations determined by an error correction operation performed on the bit allocations determined during the demultiplexing operation, wherein the optimization metric is an error rate determined based on an error correction code (ECC), and the optimization condition is an error rate threshold, The method further comprises: determining errors in first bit allocations produced by said demultiplexing using said error correction code (ECC); and The photon counts that produced the first bit allocation and the corrected bit allocation based on the ECC are saved as part of the observed data set.
2. The method according to claim 1, in, The first stream is transmitted by a first light source at a first power level, and the second stream is transmitted by a second light source at a second power level, the first stream and the second stream being transmitted simultaneously at the same frequency.
3. A device for receiving data via an optical communication path, the device include: processor; a memory including instructions that, when executed by the processor, cause the apparatus to perform operations including: demultiplexing a first stream and a second stream of data received via the optical communication path using a plurality of detection models; determining whether an optimization metric corresponding to an output of the demultiplexing operation matches an optimization condition; as well as in response to determining that an optimization metric corresponding to an output of the demultiplexing operation matches the optimization condition, updating at least one of the plurality of detection models based on an observed data set including received photon counts and corresponding actual bit allocations determined by an error correction operation performed on the bit allocations determined during the demultiplexing operation, wherein the optimization metric is an error rate determined based on an error correction code (ECC), and the optimization condition is an error rate threshold, The operations also include: determining errors in first bit allocations produced by said demultiplexing using said error correction code (ECC); and The photon counts that produced the first bit allocation and the corrected bit allocation based on the ECC are saved as part of the observed data set.
4. The device as claimed in claim 3, in, The error rate threshold is a single error.
5. The device as claimed in claim 3, in, The first stream is transmitted by a first light source at a first power level, and the second stream is transmitted by a second light source at a second power level, the first stream and the second stream being transmitted simultaneously at the same frequency.
6. The device as claimed in claim 3, in, The detection model is a Poisson probability model.
7. The device as claimed in claim 3, in, The operation of updating at least one of the plurality of detection models includes calculating a Poisson distribution based on an average photon count of data in an observed data set that matches an actual bit allocation corresponding to the at least one of the plurality of detection models.
8. The device as claimed in claim 3, in, The operation of demultiplexing a first stream and a second stream of data received through the optical communication path using a plurality of detection models comprises: submitting photon counts detected using the photon detector during a time slot to the plurality of detection models; and A bit allocation corresponding to a detection model among the plurality of detection models that returns a highest probability is selected for the first stream and the second stream.
9. The device as claimed in claim 3, in, The operation of updating at least one of the plurality of detection models includes updating the at least one of the plurality of detection models without using a training sequence.
10. The device according to claim 3, in, The operation of updating the at least one of the plurality of detection models includes retraining the at least one of the plurality of detection models using a training dataset of the observed dataset and training data collected during a training operation.
11. The device according to claim 3, in, The operation of updating at least one of the plurality of detection models includes applying the observed data set in an incremental learning algorithm to update at least one of the plurality of detection models.
