Qualcomm Incorporated

Ren Li

Kwokleung Chan

Hyun Jin Park

Samir Kumar Gupta

Hui-Ya Liao Nelson

Sameer Wadhwa

Ankit Srivastava

Seyed Arash Mirhaj

Darrell L Krulce

Erik Visser

Kostadin Dimitrov Djordjev

Pei Xiang

Murray Robert Jarvis

Te-Won Lee

Prajakt Kulkarni

Brian Lee Momeyer

Dinesh Ramakrishnan

Jeremy Patrick Toman

Jongwon Shin

Phuong Lam Ton

Aaron Douglass Lamb

Joseph Robert Fitzgerald

Michael Leviant

Ian Ernan Liu

Andre Gustavo P Schevciw

Shaun William Van Dyken

Noam Dagan

Deepak Kumar Challa

Ian Clarkson

Htet Naing

Suren Mohan

Andre Gustavo Pucci Schevciw

Xiaofei Chen

Qilin Lu

Samuel Jon Eubanks

Andre Gustavo Schevciw

Joseph R Fitzgerald

Ian Charles Clarkson

Darell L Krulce

Luis D Oliveira

Certain aspects provide an apparatus for signal processing in a neural network. The apparatus generally includes first computation in memory (CIM) cells configured as a first kernel for a neural network computation, the first set of CIM cells comprising one or more first columns and a first plurality of rows of a CIM array. The apparatus also include a second set of CIM cells configured as a second kernel for the neural network computation, the second set of CIM cells comprising the one or more first columns and a second plurality of rows of the CIM array. The first plurality of rows may be different than the second plurality of rows.

Computation in memory architecture for phased depth-wise convolutional

Devices and systems for controlling pressure during sensing are disclosed. Such sensors and/or systems may be embodied in a wearable user device. The wearable user device may include a cuff configured to apply a pressure to a portion of a user at a plurality of discrete pressure levels over a plurality of time periods; a biometric sensor configured to obtain a plurality of sensor measurements associated with a blood vessel of the user at corresponding ones of the plurality of discrete pressure levels during corresponding ones of the plurality of time periods; and a wearable structure including the cuff and the biometric sensor.

Devices and systems for controlling pressure during sensing

Certain aspects of the present disclosure provide techniques for performing machine learning computations in a compute in memory (CIM) array comprising a plurality of bit cells, including: determining that a sparsity of input data to a machine learning model exceeds an input data sparsity threshold; disabling one or more bit cells in the CIM array based on the sparsity of the input data prior to processing the input data; processing the input data with bit cells not disabled in the CIM array to generate an output value; applying a compensation to the output value based on the sparsity to generate a compensated output value; and outputting the compensated output value.

Sparsity-aware compute-in-memory

Certain aspects of the present disclosure provide a method, including: storing a depthwise convolution kernel in a first one or more columns of a CIM array; storing a fused convolution kernel in a second one or more columns of the CIM array; storing pre-activations in one or more input data buffers associated with a plurality of rows of the CIM array; processing the pre-activations with the depthwise convolution kernel in order to generate depthwise output; modifying one or more of the pre-activations based on the depthwise output to generate modified pre-activations; and processing the modified pre-activations with the fused convolution kernel to generate fused output.

Compute in memory architecture and dataflows for depth-wise separable convolution

Certain aspects of the present disclosure provide a method for performing voice activity detection, including: receiving audio data from an audio source of an electronic device; generating a plurality of model input features using a hardware-based feature generator based on the received audio data; providing the plurality of model input features to a hardware-based voice activity detection model; receiving an output value from the hardware-based voice activity detection model; and determining a presence of voice activity in the audio data based on the output value.

Context-aware hardware-based voice activity detection

A compute-in-memory bitcell is provided that includes a pair of cross-coupled inverters for storing a stored bit. The compute-in-memory bitcell includes a logic gate formed by a pair of switches for multiplying the stored bit with an input vector bit. A controller controls the pair of switches responsive to a sign bit during a computation phase of operation and controls the pair of switches responsive to a magnitude bit during an execution phase of operation.

Compute-in-memory with ternary activation

An adaptive active noise cancellation apparatus performs a filtering operation in a first digital domain and performs adaptation of the filtering operation in a second digital domain.

Systems, methods, apparatus, and computer-readable media for adaptive active noise cancellation

This disclosure describes circuit configurations that may be used for active noise cancellation in the digital domain. In particular, this disclosure proposes the use a down sample unit and an up sample unit, rather than memory-based delay circuits, to achieve one or more desired delays in digital adaptive noise cancellation circuits or other circuits that use delay for signal processing. The delay achieved by the down sample unit and the up sample unit may be tunable so as to allow flexibility in producing the necessary delay for different active noise cancellation circuit configurations. Many different adaptive noise cancellation circuit configurations are discussed, and the techniques may also be useful for other types of circuits, such as low-latency equalization circuits.

Delay techniques in active noise cancellation circuits or other circuits that perform filtering of decimated coefficients

Techniques are described for assisting near distance communications. A first device comprising a receiver, a sensor and a processor may be configured to perform the assisted near distance communication techniques. The receiver may receive, from a second device located within a conversational distance to the first device, monophonic audio data representative of the near distance communication. The sensor may generate a sensor signal representative of spatial information of the near distance communication. The processor may render, based on the spatial information and the monophonic audio data, multi-dimensional audio data in which the near distance communication originates in a soundfield from a location of the second device relative to the first device. The processor may next output the multi-dimensional audio data to a transducer so as to reproduce the near distance communication in multiple dimensions.

Assisted near-distance communication using binaural cues

Methods and apparatuses are presented for controlling an application on a device. In some embodiments, a method may include detecting that a user is maintaining an object or gesture at a position hovering near the device for a threshold length of time. The method may also include anchoring an initial position of the object or gesture to the device based on the detection of the maintained position, and controlling the application using the anchored initial position. In some embodiments, controlling the application using the anchored initial position may include manipulating the application based on detecting within a stable zone associated with the anchored initial position a change in height of the gesture or object relative to the device, and not manipulating the application whenever the object or gesture is detected to move along a plane above and parallel to the device and within the stable zone.

Methods for system engagement via 3D object detection

A method includes detecting a signal at a first set of receivers of a plurality of receivers of a device. The plurality of receivers includes the first set of receivers and a second set of receivers. The first set of receivers corresponds to selected receivers and the second set of receivers corresponds to non-selected receivers. The method includes predicting, based on the signal, an expected blockage of a signal path between a source of the signal and a first selected receiver of the first set of receivers, and selecting a particular receiver of the second set of receivers as a newly selected receiver in response to predicting the expected blockage.

Systems and methods for predicting an expected blockage of a signal path of an ultrasound signal

A method includes receiving a first output from a first sensor of an electronic device and receiving a second output from a second sensor of the electronic device. The first sensor has a first sensor type and the second sensor has a second sensor type that is different from the first sensor type. The method also includes detecting a gesture based on the first output and the second output according to a complementary voting scheme that is at least partially based on gesture complexity.

Gesture detection based on information from multiple types of sensors

A method includes receiving an ultrasound signal at an ultrasound receiver from an ultrasound transmitter. The method also includes detecting movement of at least one object based on the received ultrasound signal and at least one previously received ultrasound signal. The method further includes modifying a parameter of an acoustic echo canceller in response to the detected movement.

Acoustic echo cancellation based on ultrasound motion detection

Growing community of inventors

San Diego, CA, United States of America

Ren Li