Abstract: This document relates to training of machine learning models. One example method involves providing a machine learning model having a first classification layer, a second classification layer, and an encoder that feeds into the first classification layer and the second classification layer. The example method also involves obtaining first training examples having explicit labels and second training examples having inferred labels. The inferred labels are based at least on actions associated with the second training examples. The example method also involves training the machine learning model using the first training examples and the second training examples using a training objective that considers first training loss of the first classification layer for the explicit labels and second training loss of the second classification layer for the inferred labels. The method also involves outputting a trained machine learning model having the encoder and the first classification layer.

Abstract: This document relates to training of machine learning models. One example method involves providing a machine learning model having a first classification layer, a second classification layer, and an encoder that feeds into the first classification layer and the second classification layer. The example method also involves obtaining first training examples having explicit labels and second training examples having inferred labels. The inferred labels are based at least on actions associated with the second training examples. The example method also involves training the machine learning model using the first training examples and the second training examples using a training objective that considers first training loss of the first classification layer for the explicit labels and second training loss of the second classification layer for the inferred labels. The method also involves outputting a trained machine learning model having the encoder and the first classification layer.

Abstract: Consistent with the present disclosure, the above-described subcarrier noise, which may be characterized as a linear filtering effect, may be reduced or eliminated by providing a first multiple-input multiple output (MIMO) circuits at the transmit end of an optical link and providing a second MIMO circuit at the receive end of the optical link. The first MIMO may include a first plurality of filters, each of which may include a finite-impulse response (FIR) filter having variable coefficients or tap weights that may be changed or adapted to minimize subcarrier noise associated with the modulator, as well as D/A and analog circuitry, at the transmit end of the optical link. In addition, the second MIMO may include a second plurality of filters, each of which may also include an FIR filter having variable coefficients or tap weights that may be changed or adapted to minimized subcarrier noise associated with the optical hybrids, as well as A/D and analog circuitry, at the receive end of the optical link.

Abstract: Consistent with the present disclosure, multiple forward error correction (FEC) encoders are provided for encoding a respective one of a plurality of data streams. A mechanism is provided to mix or interleave portions of the encoded data such that each subcarrier carries information associated with each data stream, as opposed to each subcarrier carrying information associated with only a corresponding one of the data streams. As a result, both higher SNR and low SNR optical subcarriers carry such information, such that errors occurring during transmission are distributed and not concentrated or limited to information associated with a single data stream. Accordingly, at the receive end, each FEC decoder decodes information having a similar overall error rate. By balancing the error rates across each FEC encoder/decoder pair, the overall ability to correct errors improves compared to a system in which mixing or interleaving is not carried out.

Abstract: This document relates to training of machine learning models. One example method involves providing a machine learning model having a first classification layer, a second classification layer, and an encoder that feeds into the first classification layer and the second classification layer. The example method also involves obtaining first training examples having explicit labels and second training examples having inferred labels. The inferred labels are based at least on actions associated with the second training examples. The example method also involves training the machine learning model using the first training examples and the second training examples using a training objective that considers first training loss of the first classification layer for the explicit labels and second training loss of the second classification layer for the inferred labels. The method also involves outputting a trained machine learning model having the encoder and the first classification layer.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for transmitting signals with a high data rate. In some implementations, an apparatus includes a first digital signal processor outputting first data at a first data rate. A second digital signal processor outputting second data at a second data rate. A filter circuitry receiving and up-sampling the first and second data. Additionally, the apparatus includes a combiner circuit that receives the first up-sampled data and the second up-sampled data, the combiner circuit combining the first and second up-sampled data to provide a multiplexed output, the multiplexed output having a third data rate that is greater than the first data rate or the second data rate.

Abstract: Consistent with the present disclosure, an optical communication system is provided in which data is carried over optical signals including subcarriers. The subcarriers may be modulated with the standard modulation formats noted above, but the modulation formats are selectively assigned to the subcarriers, such that some subcarriers are modulated with different standard modulation formats than others. As used herein, a “standard modulation format” is one of BPSK, and n-QAM, where n is an integer greater than one. Such n-QAM modulation formats include of 3-QAM, 4-QAM (QPSK), 8-QAM, 16-QAM, 64-QAM, 128-QAM, and 256-QAM. By selecting the number of subcarriers and the types of modulation formats employed, an optical signal with an effective SE that is between that of the standard modulation formats can be generated for transmission over a distances that more closely matches the link distance.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for transmitting signals with a high data rate. In some implementations, an apparatus includes a first digital signal processor outputting first data at a first data rate. A second digital signal processor outputting second data at a second data rate. A filter circuitry receiving and up-sampling the first and second data. Additionally, the apparatus includes a combiner circuit that receives the first up-sampled data and the second up-sampled data, the combiner circuit combining the first and second up-sampled data to provide a multiplexed output, the multiplexed output having a third data rate that is greater than the first data rate or the second data rate.

Abstract: Consistent with the present disclosure, the above-described subcarrier noise, which may be characterized as a linear filtering effect, may be reduced or eliminated by providing a first multiple-input multiple output (MIMO) circuits at the transmit end of an optical link and providing a second MIMO circuit at the receive end of the optical link. The first MIMO may include a first plurality of filters, each of which may include a finite-impulse response (FIR) filter having variable coefficients or tap weights that may be changed or adapted to minimize subcarrier noise associated with the modulator, as well as D/A and analog circuitry, at the transmit end of the optical link. In addition, the second MIMO may include a second plurality of filters, each of which may also include an FIR filter having variable coefficients or tap weights that may be changed or adapted to minimized subcarrier noise associated with the optical hybrids, as well as A/D and analog circuitry, at the receive end of the optical link.

Abstract: Consistent with the present disclosure, an optical communication system is provided in which data is carried over optical signals including subcarriers. The subcarriers may be modulated with the standard modulation formats noted above, but the modulation formats are selectively assigned to the subcarriers, such that some subcarriers are modulated with different standard modulation formats than others. As used herein, a “standard modulation format” is one of BPSK, and n-QAM, where n is an integer greater than one. Such n-QAM modulation formats include of 3-QAM, 4-QAM (QPSK), 8-QAM, 16-QAM, 64-QAM, 128-QAM, and 256-QAM. By selecting the number of subcarriers and the types of modulation formats employed, an optical signal with an effective SE that is between that of the standard modulation formats can be generated for transmission over a distances that more closely matches the link distance.

Abstract: Apparatus and methods may provide improved equalizer performance, e.g., for optical-fiber-based communication systems. A least-mean-square (LMS) equalizer may include a decision feedback path containing feedback carrier recovery (FBCR), which may have low latency, and which may thus enable high-speed tap updating in the equalizer. Feed-forward carrier recovery (FFCR) may be applied, in parallel with the FBCR, to provide equalizer output by compensating, e.g., for phase noise, with improved carrier recovery/compensation, versus using FBCR to generate the output.

Abstract: Apparatus and methods may provide improved equalizer performance, e.g., for optical-fiber-based communication systems. A least-mean-square (LMS) equalizer may include a decision feedback path containing feedback carrier recovery (FBCR), which may have low latency, and which may thus enable high-speed tap updating in the equalizer. Feed-forward carrier recovery (FFCR) may be applied, in parallel with the FBCR, to provide equalizer output by compensating, e.g., for phase noise, with improved carrier recovery/compensation, versus using FBCR to generate the output.

Abstract: A digital signal processor (DSP) may include a receiver configured to receive an input signal. The DSP may include a processor component to perform carrier recovery on a set of digital signals representing a set of symbols associated with the input signal. The DSP may include an output component to provide information included in the set of digital signals representing the set of symbols. The DSP may be configured to perform, for the input signal, phase estimation with a latency of less than approximately 880 nanoseconds and having a power consumption of less than approximately 400 milliwatts at an update rate greater than approximately 4 Gigahertz. The latency being a propagation delay of the input signal.

Abstract: A processor circuit is provided in a coherent optical receiver module. The processor receives a series of electrical signals over a time period, representative of a series of optical signals received at instants of time within the time period. Each of the electrical signals is indicative of a respective one of a plurality of points on an IQ plane, each of the points being spaced from one of a plurality of predetermined points in the IQ plane by a corresponding one of a plurality of distortion values. In addition, the processor circuit calculates one or more perturbative coefficients based on one or more of the distortion values and determines data from the series of electrical signals based on the perturbative coefficient.

Abstract: An optical receiver may include a digital signal processor to receive an input sample that includes transmitted data, transmitted by an optical transmitter, and nonlinear distortion. The digital signal processor may process the input sample to generate an estimated data value. The estimated data value may be an estimate of the transmitted data. The digital signal processor may remove the estimated data value from the input sample to generate a noise sample. The digital signal processor may determine a nonlinear distortion value based on the input sample, the estimated data value, and the noise sample. The nonlinear distortion value may be an estimate of the nonlinear distortion included in the input sample. The digital signal processor may remove the nonlinear distortion value from the input sample to generate an output sample, and may output the output sample.

Abstract: Consistent with the present disclosure, an optical communication system is provided in which data is carried over optical signals including subcarriers. The subcarriers may be modulated with the standard modulation formats noted above, but the modulation formats are selectively assigned to the subcarriers, such that some subcarriers are modulated with different standard modulation formats than others. As used herein, a “standard modulation format” is one of BPSK, and n-QAM, where n is an integer greater than one. Such n-QAM modulation formats include of 3-QAM, 4-QAM (QPSK), 8-QAM, 16-QAM, 64-QAM, 128-QAM, and 256-QAM. By selecting the number of subcarriers and the types of modulation formats employed, an optical signal with an effective SE that is between that of the standard modulation formats can be generated for transmission over a distances that more closely matches the link distance.

Abstract: A digital signal processor (DSP) may include a receiver configured to receive an input signal. The DSP may include a processor component to perform carrier recovery on a set of digital signals representing a set of symbols associated with the input signal. The DSP may include an output component to provide information included in the set of digital signals representing the set of symbols. The DSP may be configured to perform, for the input signal, phase estimation with a latency of less than approximately 880 nanoseconds and having a power consumption of less than approximately 400 milliwatts at an update rate greater than approximately 4 Gigahertz. The latency being a propagation delay of the input signal.

Abstract: A filtration apparatus for removing solids and particulates from a liquid stream flowing through a filter media. A magnet is located adjacent to the filter media and the liquid steam to attract metal particles in the liquid stream to the filter media. The magnet may be a rare earth magnet. A variable speed drive motor is coupled to a carrier supporting the filter media to advance the filter media through the liquid stream.

Abstract: Methods and systems are disclosed including receiving, with a processor circuit in a coherent optical receiver module, a series of electrical signals over a time period, representative of a series of optical signals received at instants of time within the time period, each of the electrical signals being indicative of a respective one of a plurality of points on an IQ plane, each of the points being spaced from one of a plurality of predetermined points in the IQ plane by a corresponding one of a plurality of distortion values; calculating, with the processor circuit, one or more perturbative coefficients based on one or more of the distortion values; and determining, with the processor circuit, data from the series of electrical signals based on the perturbative coefficient.

Abstract: An apparatus and method for removing solids and particulates from fluid by magnetic separation. First and second spaced rotatable drums move a closed loop belt. A portion of the belt is disposed in contact with the fluid flowing through a fluid passage disposed adjacent to a circumferential portion of the first rotatable drum where solids and particulates in the fluid are attracted to the belt by magnetic attraction from a magnetic source within the first rotatable drum. A wiper scrapes the solids deposited on the belt.