Abstract: Technology is disclosed herein to execute an inference model by a processor which includes a reshape layer. In an implementation, the reshape layer of the inference model receives an output produced by a previous layer of the inference model and inserts padding into the output, then supplies the padded output as an input to a next layer of the inference model. In an implementation, the inference model includes a stitching layer at the beginning of the inference model and an un-stitch layer at the end of the model. The stitching layer of the inference model stitches together multiple input images into an image batch and supplies the image batch as an input to a subsequent layer. The un-stitch layer receives output from a penultimate layer of the inference model and unstitches the output to produce multiple output images corresponding to the multiple input images.

Abstract: In examples, a device comprises control logic configured to detect an idle cycle, an operand generator configured to provide a synthetic operand responsive to the detection of the idle cycle, and a computational circuit. The computational circuit is configured to, during the idle cycle, perform a first computation on the synthetic operand. The computational circuit is configured to, during an active cycle, perform a second computation on an architectural operand.

Abstract: A method of optimally distributing signal power for multiple users over a xDSL or wireless channel considers uses computationally efficient tools to achieve improved crosstalk avoidance. The method chooses between EQPSD and FDS signaling in a fashion that maximizes overall data rate. Rather than choosing EQPSD signaling in regions where there is low self-NEXT, but where the echo is high (relative to signal power), the method selects FDS signaling since the communication system acts in a manner similar to self-NEXT in such regions.

Abstract: A receiving modem (20) including functionality for adjusting the synchronization of its timing relative to a transmitting modem (10), by recovering timing from a plurality of pilot tones transmitted by the transmitting modem (10) in a multicarrier modulation signal, is disclosed. For a subset of the pilot tones, the receiving modem (20) sums phase error differences, for example according to a weighted sum, to derive a phase error estimate relative to reference signals, in which noncoherent random noise cancels out, to correct for timing differences between the modems (10, 20). For another subset of the pilot tones not used in the phase error estimate, the receiving modem (20) analyzes the demodulated signal for those tones to adjust reference signals for each of those pilot tones. The adjusting is performed either by deriving an adjusted complex reference signal, or by adjusting coefficients in a frequency domain equalizer (28).

Abstract: The present disclosure describes a method and system of tone ordering for discrete multi-tone (DMT) systems with tone grouping. Some illustrative embodiments may include a method, comprising, dividing a plurality of sub-channels within a discrete multi-tone symbol into a plurality of groups, selecting each sub-channel of the plurality of sub-channels in non-sequential sub-channel order from the plurality of groups, and assigning data bits in bit order to each sub-channel in the order in which each sub-channel is selected. Each group comprises an equal number of sub-channels, and each sub-channel comprises a frequency bin with a unique center frequency.

Abstract: The present invention provides, in one embodiment, a method of simplifying filtering with respect to a band plan. The method includes selecting a reflection point relative to the band plan and designating subchannels as unusable within a filterable separation of the reflection point that would otherwise introduce an asymmetry within the filterable separation. The present invention also provides a filtering simplifier for use with a DSL system having a band plan. The filtering simplifier includes a reflection point selector configured to select a reflection point relative to the band plan. The filtering simplifier also includes a subchannel designator coupled to the reflection point selector and configured to designate subchannels as unusable within a filterable separation of the reflection point that would otherwise introduce an asymmetry within the filterable separation.

Abstract: A digital subscriber line communication system 10 is provided. The digital subscriber line communication system 10 includes first and second transceivers 12, 14. The first transceiver 12 communicates discrete multitone symbols using frequency division duplexing. The first transceiver 12 is operable to estimate a propagation delay. The second transceiver 14 communicates with the first transceiver 12 and receives the estimated propagation delay from the first transceiver 12. The second transceiver 14 uses the estimated propagation delay to determine a round-trip propagation delay. The second transceiver 14 determines a transmission time advance based at least partially on the round-trip propagation delay.

Abstract: A device 8 is disclosed. The device 8 comprises a first component 10 operable to produce a plurality of discrete multitone symbols based in part on a subchannel spacing and a cyclic extension length and a second component 152 operable to determine the subchannel spacing and the cyclic extension length, the cyclic extension length selected based on the subchannel spacing to increase the number of discrete multitone symbols transmitted during a low noise interval of a cyclostationary noise environment.

Abstract: A communication system 10 initialized using an iterative training portion is disclosed. The communication system 10 comprises a first transceiver 12 operable to communicate according to a digital subscriber line standard and a second transceiver 14 operable to communicate according to the digital subscriber line standard with the first transceiver 12. The first and the second transceivers 12, 14 are operable to perform an initialization 50, including a handshake portion 52, a channel analysis portion 56, and a training portion 54, the first and second transceivers 12, 14 negotiating a plurality of iterations of the training portion 54.

Abstract: An integrated circuit 18 is provided that includes a memory 32 and a memory modification component 33. The memory 32 maintains a bits count, a gain, and a tone order for each of a plurality of discrete multi-tone sub-channels. The memory modification component 33 operable to control an in-service modification of at least some of the bits count, the gain, and the tone order using a single bits, gains and tone order table.

Abstract: Multicarrier modulated communications, involving a transmitting modem (30; 30?) and a receiving modem (40; 40?) that operate according to normal operating mode and a low-power, quiescent, mode, are disclosed. In the quiescent mode, transmitter power is saved by scaling down the amplitude of the transmitted symbols in each frame by a scaling factor (A), and transmitting multiple instances of the frames in sequence. On receipt, the repeated frames are summed, and the summed amplitudes scaled to account for the repetition factor (M) and the scaling factor (A), to recover the transmitted information. If the repetition factor (M) is greater than or equal to the square of the scaling factor (A), the signal-to-noise ratio in quiescent mode is at least as high as that in normal operation. The repeating, scaling, and summing may all be performed in the frequency domain or in the time domain, as desired.

Abstract: A number of basic equalization and demodulation structures have been shown to be appropriate for DMT systems depending on the channel, noise, and system parameters. These include single path, dual path, oversampled, and double rate structures. Using the fundamental computation units of two TEQs (FIR filters) and two FFTs, in conjunction with simple delays, downsampling and routing, single path, dual path, oversampled and double rate equalization structures can be realized from a common equalization structure.

Abstract: System and method for reducing interference to existing devices. A preferred embodiment comprises specifying a frequency range for a set of dummy signals, specifying a clipping function to ensure that the set of dummy signals do not exceed a maximum power constraint, incorporating a least squares solution for computing the set of dummy signals into the clipping function, and iterating the clipping function until a terminating condition is reached. The use of the clipping function limits the magnitude of the dummy signals, to ensure that dummy signals do not exceed a maximum power constraint.

Abstract: A method of optimally distributing signal power for multiple users over a xDSL or wireless channel considers uses computationally efficient tools to achieve improved crosstalk avoidance. The method chooses between EQPSD and FDS signaling in a fashion that maximizes overall data rate. Rather than choosing EQPSD signaling in regions where there is low self-NEXT, but where the echo is high (relative to signal power), the method selects FDS signaling since the communication system acts in a manner similar to self-NEXT in such regions.