Qualcomm Incorporated

Other

Jilei Hou

Chao Wei

Changlong Xu

Jian Li

Danlu Zhang

Sharad Deepak Sambhwani

John Edward Smee

Peter S Gaal

Hao Xu

Joseph Binamira Soriaga

Jing Jiang

Wanshi Chen

Bibhu P Mohanty

Pavan Kumar Vitthaladevuni

Yu Zhang

Peng Cheng

Aziz Gholmieh

Rohit Kapoor

Ozcan Ozturk

Ravi Agarwal

Siddhartha Mallik

Arjun Bharadwaj

Juan Montojo

Haitong Sun

Naga Bhushan

Alexei Yurievitch Gorokhov

Gabi Sarkis

Tingfang Ji

Ravi Palanki

Yang Yang

Roberto Padovani

Hari Sankar

Amir Farajidana

Durga Prasad Malladi

Xipeng Zhu

Liangming Wu

Paul E Bender

Chong Li

Aamod Dinkar Khandekar

Heechoon Lee

Thomas J Richardson

Peter J Black

Jibing Wang

Lin Yang

Vinay Chande

Neng Wang

Minghai Feng

Yibo Jiang

Henry David Pfister

Jiming Guo

Navid Hassanpour Ghady

Jason Frank Hunzinger

Jinghu Chen

Stefano Tomasin

Xiaohui Liu

Weiyan Ge

Arnold Jason Gum

Brian K Butler

Jamie Menjay Lin

Jeffrey Alexander Levin

Thomas Wilborn

Gaojin Wu

Alessandro Vanelli Coralli

Rashid Ahmed Akbar Attar

Mostafa El-Khamy

Christopher G Lott

Daniel Fred Filipovic

Ying Wang

Rashid Ahmed Attar

Qingxin Chen

Bo Chen

Alan Barbieri

Mingxi Fan

Jin Won Lee

Kenneth David Easton

Chengjin Zhang

Eugene Baik

Tijmen Pieter Frederik Blankevoort

Haijun Zhao

Shengshan Cui

Kiran Kiran

Liang Zhang

Guodong Li

Peter John Black

Christos Louizos

Jianqiang Zhang

Matthias Reisser

Yadong Lu

Wei Chao

Bibhu Prasad Mohanty

Methods and apparatus for constructing polar codes are provided. A transmitter determines at least one set of parameters corresponding to data to be transmitted, and a set of sorting indices corresponding to bits of the data to be transmitted based on the set of parameters, the set of sorting indices indicating a position set of the bits to be transmitted. The transmitter polar encodes the data based at least on the set of parameters and the set of sorting indices to generate a coded block of the data, and transmits the coded block of the data.

Methods and apparatus for constructing polar codes

A method for compressing a deep neural network includes determining a pruning ratio for a channel and a mixed-precision quantization bit-width based on an operational budget of a device implementing the deep neural network. The method further includes quantizing a weight parameter of the deep neural network and/or an activation parameter of the deep neural network based on the quantization bit-width. The method also includes pruning the channel of the deep neural network based on the pruning ratio.

Joint pruning and quantization scheme for deep neural networks

Aspects of the disclosure relate to wireless communication systems configured to provide techniques for polar coding control information together with combined cyclic redundancy check (CRC) information. The combined CRC information may include a number of CRC bits selected to jointly decode and verify the control information to reduce the CRC overhead.

CRC bits for joint decoding and verification of control information using polar codes

The disclosure relates in some aspects to information encoding. Information encoding may involve puncturing bits of a codeword or repeating bits of a codeword. The disclosure relates in some aspects to selecting a puncturing or repetition pattern. In some aspects, a puncture pattern for data encoding is selected based on a criterion that the output and the repetition input of an XOR are not erased. In some aspects, a repetition pattern for data encoding is selected based on a criterion that repetition not be applied for the output and the repetition input of an XOR.

Puncturing and repetition for information encoding

Aspects of the disclosure relate to wireless communication devices configured to encode information blocks to produce code blocks and interleave the code blocks utilizing an interleaver including a plurality of rows and a plurality of columns, where the number of columns of the interleaver varies between the rows. In some examples, the interleaver includes a right isosceles triangle-shaped matrix of rows and columns. In other examples, the interleaver includes a trapezoid-shaped matrix of rows and columns.

Efficient interleaver design for polar codes

Methods, systems, and devices for wireless communication are described. A transmitter may generate a first segmentation based on a first subset of control information and a second segmentation based on jointly encoding the first subset and a second subset of the control information. The transmitter may polar encode the first segmentation to generate a first codeword and the second segmentation to generate a second codeword, and transmit the first and second codewords. A receiver may determine a first bit sequence corresponding to the first subset based on decoding a first codeword and determine an error detection code (EDC) and a second bit sequence corresponding to the second subset based on decoding a second codeword. The receiver may perform error detection on the first and second bit sequences based on the determined EDC, and output the first bit sequence and the second bit sequence or a decoding error.

Polar codes for uplink control information

Methods, systems, and devices for wireless communication are described for dynamic frozen bits of polar codes for early termination and performance improvement. A wireless device may receive a signal comprising a codeword encoded using a polar code. The wireless device may perform decoding of the codeword including at least: parity check of a first subset of decoding paths for making a decision on early termination of decoding of the codeword based on dynamic frozen bits, and generating path metrics for a second subset of the decoding paths that each pass the parity check based on the dynamic frozen bits, and performing error detection on a bit sequence corresponding to one of the second subset of the decoding paths based at part on error detection bits and the generated path metrics. The wireless device may process the information bits based on a result of the decoding.

Dynamic frozen bits and error detection for polar codes

Various aspects of the disclosure relate to rate matching techniques for block encoding. In some aspects, a decision regarding whether to use repetition-based rate matching or puncture-based rate matching is made based on a block size of information being encoded. In some aspects, repetition-based rate matching uses a bit-reversal permutation technique.

Rate matching for block encoding

Aspects of the disclosure relate to rate-matching a stream of bits encoded using polar codes. An exemplary method generally includes determining a mother code size (N) for transmitting an encoded stream of bits based, at least in part, on a minimum supported code rate for transmitting the encoded stream of bits (R), a control information size of the encoded stream of bits (K), a number of coded bits for transmission (E), and a maximum mother code size (N), encoding a stream of bits using a polar code of size (N, K) and storing the encoded stream of bits in a circular buffer, and performing rate-matching on the stored encoded stream of bits based, at least in part, on a comparison among the mother code size (N), the control information size of the encoded stream of bits (K), and the number of coded bits for transmission (E).

Rate-matching scheme for polar codes

Aspects of the disclosure relate to wireless communication systems configured to provide techniques for multiplexing dedicated control information for a plurality of users in a single information block and polar coding the information block to produce a polar code block of dedicated control information for transmission over a wireless air interface. The information block may further include group cyclic redundancy check (CRC) information for the information block and individual CRC information for each dedicated control information.

Efficient control channel design using polar codes

Certain aspects of the present disclosure support demodulation reference signal (DMRS) enhancement for higher order multi-user multiple-input multiple-output (MU-MIMO) communications. An example method generally includes determining a plurality of ports of a multi-dimensional array of transmit antennas and a number of spatial multiplexed layers for transmission to a plurality of user equipments (UEs), configuring a demodulation reference signal (DMRS) pattern by multiplexing the layers or the ports in the DMRS pattern, using an orthogonal cover code (OCC) and one or more code division multiplexing (CDM) groups, and transmitting DMRS symbols based on the configured DMRS pattern using the multiplexed layers and the ports.

DMRS enhancement for higher order MU-MIMO

Embodiments described herein relate to a method, comprising: receiving input data at a convolutional neural network (CNN) model; generating a factorized computation network comprising a plurality of connections between a first layer of the CNN model and a second layer of the CNN model, wherein: the factorized computation network comprises N inputs, the factorized computation network comprises M outputs, and the factorized computation network comprises at least one path from every input of the N inputs to every output of the M outputs; setting a connection weight for a plurality of connections in the factorized computation network to 1 so that a weight density for the factorized computation network is &#x3c;100%; performing fast pointwise convolution using the factorized computation network to generate fast pointwise convolution output; and providing the fast pointwise convolution output to the second layer of the CNN model.

Efficient inferencing with fast pointwise convolution

Growing community of inventors

San Diego, CA, United States of America

Jilei Hou