Icera, Inc.

Lucent Technologies Inc.

Graphcore Limited

Nvidia Technology UK Limited

Alcatel Italia S.p.a.

Nvidia Corporation

Alcatel

Agere Systems Guardian Corp.

Carlo Luschi

Steve Allpress

Tarik Tabet

Edward Andrews

Simon Nicholas Walker

Abdelkader Medles

Philip Jones

Jonathan Wallington

Simon Huckett

Ran-Hong Yan

Arnaldo Spalvieri

Stephen Alan Allpress

Gang Wang

Paul Edward Strauch

Syed Aon Mujtaba

Godfrey Da Costa

Franco Guglielmi

Alexandr Kuzminskiy

Magnus Sandell

Dominic Alexander Masters

Phil Jones

Stephen Felix

Simon Christian Knowles

Alan Graham Alexander

Stephan Ten Brink

Israel Bar-David

Mrudula Gore

Nallepilli S Ramesh

Konstantinos Samaras

Steve Felix

Colman Hegarty

Andrea Sandri

Badreddine Noune

Donal Price

Antoine Labatie

Laolu Lijofi

Jonathan Peter Lucas

Helen Byrne

Luke Benjamin Hudlass-Galley

Zachary Frank Eaton-Rosen

Vishwambhar Rathi

Simon Walker

Nallepilli Ramesh

A computer-implemented method of training a deep neural network, comprising, for each of one or more batches of training examples: processing the data in a forward pass through the layers of the network, by: applying a set of network weights to the input data to obtain a set of weighted inputs, normalising the weighted inputs based on statistics computed for each training example, transforming the normalised inputs by affine transformation parameters, applying an activation function to the transformed normalised inputs to obtain post-activation values, and normalizing the post-activation values based on one or more proxy variables sampled from a distribution defined by proxy distribution parameters, the normalization applied independently of training example; processing the data in a backward pass through the network to determine updates to learnable parameters comprising network weights, affine transformation parameters, and proxy distribution parameters, and updating the learnable parameters to optimise a predefined loss function.

Processing data batches in a multi-layer network

A processing device comprising: a control register configured to store a scaling factor; at least one execution unit configured to execute instructions to perform arithmetic operations on input floating-point numbers provided according to a first floating-point format, wherein each of the input floating-point numbers provided according to the first floating-point format comprises a predetermined number of bits, wherein the at least one execution unit is configured to, in response to execution of an instance of a first of the instructions: perform processing of a first set of the input floating-point numbers to generate a result value, the result value provided in a further format and comprising more the predetermined number of bits, enabling representation of a greater range of values than is representable in the first floating-point format; and apply the scaling factor specified in the control register to increase or decrease an exponent of the result value.

Processing device for intermediate value scaling

A method of processing batches of data in a computer comprising a plurality of pipelined stages each providing one or more layers of a machine learning model. The method comprises: processing a first batch of data in the pipeline processing stages, each layer of the model using an activation function and weights for that layer to generate an output activation, wherein an output layer generates an output of the model. The method further comprises, for each layer: computing an estimated gradient of a loss function; generating updated weights by processing the estimated gradient with respect to the weights for the first batch using a learning rate for the model; and storing the updated weights for processing on the next batch of data. Updated weights are generated using a modulation factor based on the number of processing stages between that layer and the output layer.

Batch processing in a machine learning computer

A data processing system for training a neural network, the data processing system comprising: a first set of one or more processing units running one model of the neural network, a second set of one or more processing units running another model of the neural network, a data storage, and an interconnect between the first set of one or more processing units, the second set of processing units and the data storage, wherein the data storage is configured to provide over the interconnect, training data to the first set of one or more processing units and the second set of one more processing units, wherein each of the first and second set of processing units is configured to, when performing the training, evaluate loss for the respective training iteration including a measure of the dissimilarity between the output values calculated based on the different modes running on the first and second set of processing units, wherein the dissimilarity measure is weighted in the evaluation of the loss in accordance with a parameter that is updated between different training iterations.

System having multiple processing unit sets for training neural networks

One aspect provides a method of processing a signal transmitted over a channel in a wireless communication system. In one embodiment, the method comprises receiving at a receiver the signal transmitted over the channel, estimating at intervals at least one parameter of a time varying communication environment of the system, monitoring at least one processor-related criterion of a processor at the receiver, and selecting a signal processing function from a plurality of signal processing functions implementable by the processor. The selecting of the signal processing function is based on both the at least one parameter and the at least one processor-related criterion. Each signal processing function comprises a plurality of code blocks which process the received signal. Each code block of the plurality of code blocks comprises a sequence of instructions for execution by on a processor platform of the processor.

Processing signals in a wireless communications environment

Wireless receivers are described for receiving signals from a transmitter. A receiver can include a plurality of antennas each for receiving a version of a signal via a different propagation channel and providing that version to a respective input. Signal processing means can be included and configured to operate diversity processing of a supplied number of said inputs for use in performing detection of said signal. The receiver can include channel parameter estimation means, configured to estimate one or more channel parameters on the propagation channels. The receiver can also include selection means configured to select only a subset of said inputs to implement a specific dimensionality of the receiver diversity processing, in dependence on the one or more channel parameters indicative of channel conditions on said propagation channels. The diversity processing can be linear or non-linear. Related methods and software implementations and computer program products are also described.

Wireless receiver with receive diversity

Method, receiver and computer program product for processing a signal transmitted over a wireless network from a plurality of spatially separated transmit antennas of a transmitter using a Multiple-Input Multiple-Output transmission. The signal is received at a plurality of receive antennas, the signal comprising a plurality of data streams. The channel quality for each of the data streams in the received signal is determined and based on the determined channel quality of the data streams, the number of independent data streams that can be supported in the Multiple-Input Multiple-Output transmission of the signal is determined. An indication of the determined number is transmitted to the transmitter.

Rank signaling in spatial multiplexing

A method of processing digital samples of a signal received at a receiver of a wireless communication system includes monitoring channel conditions and generating a channel indicator including at least one channel parameter by performing at least one of: estimating a channel mobility parameter and comparing it with a threshold; estimating a channel parameter of the energy of the channel outside a predefined temporal window, and comparing it with a threshold; estimating a channel temporal duration parameter and establishing if it meets predetermined criteria; estimating a channel-zero location parameter and establishing if it meets predetermined criteria; estimating a received-signal signal-to-disturbance power ratio, and comparing it to a threshold; estimating an estimated-channel-response signal-to-disturbance power ratio; estimating the degree of non-stationarity of the disturbance at the receiver input; and selecting one of a plurality of processing routines for processing the digital samples based on said channel indicator. Related receivers are also described.

Processing digital samples in a wireless receiver

Techniques are described for optimizing the processing facilities of a receiver in a wireless communication environment, taking into account processing performance set against the computing resources and/or power consumption required to obtain that processing performance. A radio receiver for a wireless communications system can include channel equalization means arranged to receive samples of an incoming signal and to generate an equalized output, said channel equalization means including means for processing said digital samples in accordance with an equalizer algorithm. The receiver includes means for estimating parameters of a channel over which the incoming signal has been transmitted. The receiver can also include means for selecting from a plurality of equalizer algorithms an equalizer algorithm for execution by the processing means based on at least one said estimated channel parameter. Related methods and algorithms are described. Related computer program products are also described.

Radio receiver in a wireless communications system

An apparatus, method and program for processing a signal received over a wireless channel. The apparatus comprises: a channel searcher configured to determine a profile of the channel in the form of energy values for a plurality of rays at respective channel positions, and an equalizer for equalizing the signal based on a variable window of the equalizer. The channel searcher is configured to estimate dispersion of the channel by determining a measure of variance of ray distance from a reference position within the channel profile weighted by ray energy. The equalizer is coupled to the channel searcher and arranged to adapt the window in dependence on the measure of energy-weighted ray distance variance.

Equalisation of a signal received over a wireless channel

A method of processing digital samples of a signal received at a receiver of a wireless communication system includes monitoring channel conditions and generating a channel indicator including at least one channel parameter by performing at least one of: estimating a channel mobility parameter and comparing it with a threshold; estimating a channel parameter of the energy of the channel outside a predefined temporal window and comparing it with a threshold; estimating a channel temporal duration parameter and establishing if it meets predetermined criteria; estimating a channel-zero location parameter and establishing if it meets predetermined criteria; estimating a received-signal signal-to-disturbance power ratio and comparing it to a threshold; estimating an estimated-channel-response signal-to-disturbance power ratio; estimating the degree of non-stationarity of the disturbance at the receiver input; and selecting one of a plurality of processing routines for processing the digital samples based on said channel indicator. Related receivers are also described.

A method, receiver and program for processing a signal received using a wireless communication channel by a receiver in a wireless cellular network. The method comprises: receiving signal samples of a signal to be processed from a serving cell; identifying a set of dominant interfering cells generating an interfering signal above a level; using the number of cells in the set to select an interference scenario; and using the selected interference scenario and at least one parameter related to the serving cell and the interfering cells to select a processing function for processing the signal.

Cellular communications system

A method, receiver and program for equalising a radio signal comprising a sequence of data samples multiplexed with a sequence of pilot samples. The method comprises; calculating equaliser coefficients by computing cross-correlations of the received signal and known pilot samples available at the receiver and auto-correlations of the received signal; and equalising the received signal using the calculated coefficients.

Equaliser for wireless receivers with normalised coefficients

Method, receiver and computer program product for processing a signal, wherein the signal is received over a wireless network from a transmitter, the signal comprising a plurality of data streams. A selected one of the data streams is preferentially treated by reporting an adjusted value of a characteristic of the selected data stream to the transmitter and the received signal is decoded using a technique of successively decoding data streams in which the selected data stream is the first data stream to be decoded, a signal corresponding to the selected data stream being removed from the received signal prior to decoding at least one of the unselected data streams in the received signal.

Deciding the order of SIC detection by adjusting MCS

A method of transmitting an RF signal over a wireless communication network is provided. The method comprises determining a respective weighting factor for each of a plurality of digital signals each corresponding to a respective channel, the weighting factors weighting the digital signals to produce a composite signal intended for transmission as an RF signal via a power amplifier. The method further comprises executing instructions on a processor to dynamically calculate a metric related to the non-linearity of the power amplifier's transfer characteristics for the composite signal using the determined weighting factors. The method further comprises supplying to the power amplifier a signal for transmission as an RF signal and amplifying the signal for transmission at the power amplifier to transmit an RF signal over the wireless communication network via at least one antenna. The method further comprises controlling the transmission based on the metric related to the amplifier non-linearity.

Estimating signal characteristics

Growing community of inventors

Oxford, United Kingdom

Carlo Luschi