Information Theoretic Concepts of 5G (slides)

Information Theoretic Concepts of 5G

Ivana MaricEricsson Research

Joint work withSong-Nam Hong, Dennis Hui and Giuseppe Caire (TU Berlin)

IEEE 5G Silicon Valley SummitNovember 16, 2015

Outline

I What is new in 5G

I Multihop Communications for 5G

I Channel coding for 5G

Outline

I What is new in 5G

I Multihop Communications for 5G

I Channel coding for 5G

Outline

I What is new in 5G

I Multihop Communications for 5G

I Channel coding for 5G

5G - What is New?

I Applications

5G - What is New?

I Applications

I RequirementsI 1000x mobile data, 100x user data rates, 100x connected

devices, 10x battery life, 5x lower latencyI Sustainable, secure

5G - What is New?

I Applications

I Requirements

I Architecture - Common network platform

5G and Spectrum

Design

I Low frequencies: widecoverage

I mmW band: short range,low complexity

Ultra-dense Networks in mmW Bands

Dense deployments

I Due to limited range

I For higher throughput

Ultra-dense Networks in mmW Bands

Backhaul for thousands ofaccess points?

I Backhaul today:P2P, line-of-sight

I Tomorrow:Wireless multihop backhaul

I Access points relay eachother’s data

Ultra-dense Networks in mmW Bands

Backhaul for thousands ofaccess points?

I Backhaul today:P2P, line-of-sight

I Tomorrow:Wireless multihop backhaul

I Access points relay eachother’s data

Remove Houses: Mesh Network

source 2

source 3

source 1

User 1

User 3

User 2

User 4

Efficient multihop scheme? What should relays do?

Information Theory: Relay Channel is 44 Years Old

Multihop Schemes in Practice

I Large body of IT results

I Efficient multihop schemes developed; capacity bounds, scalinglaws and capacity in some cases determined

I Not much practical impact

I Too complex?

I There was no need?

I 5G will deploy multihop communications

Multihop Schemes in Practice

I Large body of IT results

I Efficient multihop schemes developed; capacity bounds, scalinglaws and capacity in some cases determined

I Not much practical impact

I Too complex?

I There was no need?

I 5G will deploy multihop communications

Multihop Schemes in Practice

I Large body of IT results

I Efficient multihop schemes developed; capacity bounds, scalinglaws and capacity in some cases determined

I Not much practical impact

I Too complex?

I There was no need?

I 5G will deploy multihop communications

Multihop Communications for 5G

Multihop Backhaul for Ultra-dense Networks

Multihop MTC?

70000 tracking devices 9 Gbyte/user/hour

480 Gbps/km2

Multihop Backhaul

Current Proposal for 5G

Interference-avoidance routing

I Each relay performsstore-and-forward

I Establish routes iteratively

I Works well in lowinterference

Does not work in high interference

Current Proposal for 5G

Interference-avoidance routing

I Each relay performsstore-and-forward

I Establish routes iteratively

I Works well in lowinterference

Does not work in high interference

Current Proposal for 5G

Interference-avoidance routing

I Each relay performsstore-and-forward

I Establish routes iteratively

I Works well in lowinterference

Does not work in high interference

Current Proposal for 5G

Interference-avoidance routing

I Each relay performsstore-and-forward

I Establish routes iteratively

I Works well in lowinterference

Does not work in high interference

Current Proposal for 5G

Interference-avoidance routing

I Each relay performsstore-and-forward

I Establish routes iteratively

I Works well in lowinterference

Does not work in high interference

Current Proposal for 5G

Interference-avoidance routing

I Each relay performsstore-and-forward

I Establish routes iteratively

I Works well in lowinterference

Does not work in high interference

Current Proposal for 5G

Interference-avoidance routing

I Each relay performsstore-and-forward

I Establish routes iteratively

I Works well in lowinterference

Does not work in high interference

Decode vs. Quantize

Routing

I Each relay has to decodemessages

I Worst relay is a bottleneck

Quantize

I Any relay can quantize sourcesignal

I Noisy network coding (NNC)[Avestimehr et.al, 2009], [Lim

et.al, 2011],[Hou & Kramer, 2013]

Decode vs. Quantize

Routing

I Each relay has to decodemessages

I Worst relay is a bottleneck

Quantize

I Any relay can quantize sourcesignal

I Noisy network coding (NNC)[Avestimehr et.al, 2009], [Lim

et.al, 2011],[Hou & Kramer, 2013]

Decode vs. Quantize

Routing

I Each relay has to decodemessages

I Worst relay is a bottleneck

Quantize

I Any relay can quantize sourcesignal

I Noisy network coding (NNC)[Avestimehr et.al, 2009], [Lim

et.al, 2011],[Hou & Kramer, 2013]

Noisy Network Coding

I No interference at relays: every signal is useful

I A relay sends a mix of data flows

I Can outperform other schemes

I Achieves constant gap to the multicast capacity

Implementation: NNC Challenges

I Full-duplex assumption

I Channel state information

I Relay selection

I Decoder complexity

I Rate calculation

We developed a scheme that has a lower complexity and improvedperformance [Hong, Maric, Hui & Caire, ISIT 2015, ITW 2015]

Implementation: NNC Challenges

I Full-duplex assumption

I Channel state information

I Relay selection

I Decoder complexity

I Rate calculation

We developed a scheme that has a lower complexity and improvedperformance [Hong, Maric, Hui & Caire, ISIT 2015, ITW 2015]

Relay Selection

Group relaying

source

destination

Relay Selection

Group relaying

source

destination

Relay Selection

Group relaying

source

destination

Relay Selection

Group relaying

source

destination

Relay Selection

Layered network

destinationsource

To Improve Performance: Adaptive Scheme

A relay chooses a forwarding scheme based on SNR

I Relays with good channels decode-and-forward

I The rest of relays quantize

How much to quantize?

To Improve Performance: Adaptive Scheme

A relay chooses a forwarding scheme based on SNR

I Relays with good channels decode-and-forward

I The rest of relays quantize

How much to quantize?

To Improve Performance: Adaptive Scheme

A relay chooses a forwarding scheme based on SNR

I Relays with good channels decode-and-forward

I The rest of relays quantize

destinationsource

quantize

quantize

quantize

quantize

decode

decode

How much to quantize?

To Improve Performance: Adaptive Scheme

A relay chooses a forwarding scheme based on SNR

I Relays with good channels decode-and-forward

I The rest of relays quantize

destinationsource

quantize

quantize

quantize

quantize

decode

decode

How much to quantize?

To Improve Performance: Optimized Quantization

A relay chooses number of quantization levels based on SNR

I Optimal quantization decreases the gap to capacity fromlinear to logarithmic

I NNC with noise-level quantization [Avestimehr et. al., 2009]

R(K) = log(1 + SNR) − K

I Optimal quantization [Hong & Caire, 2013]

R(K) ≥ log(1 + SNR) − log(K + 1)

To Reduce Complexity: Successive Decoding

Destination successively decodes messages from different layers

I Does not decrease performance in the considered network[Hong & Caire, 2013]

destinationsource

quantize

quantize

quantize

quantize

decode

decode

Summary

destination

message 1

quantize

quantize

quantize

quantize

decode

decode

source

message 2

I Relay selection via interference-harnessing

I Adaptive scheme: each relay chooses to decode or quantize

I Quantization level is optimized

I Destination performs successive decoding

I Successive relaying [Razaei et.al., 2008]

I Rate splitting reduces interference at DF relays

Performance Gains

I Derived closed form solution for the rate, for any relayconfiguration [Hong, Maric , Hui & Caire, ISIT 2015, ITW 2015]

I Better performance with a simpler scheme!

Performance Gains

I Derived closed form solution for the rate, for any relayconfiguration [Hong, Maric , Hui & Caire, ISIT 2015, ITW 2015]

I Better performance with a simpler scheme!

Channel Coding for 5G

Informationsource

Sourceencoder

Channelencoder

Modulator

Channel

User Sourcedecoder

Channeldecoder

Demodulator

Choosing Channel Codes for 5G

I Main considerationsI Performance, complexity, rate-compatibility

I LTE deploys turbo codes [Berrou et. al., 1993]I Perform within a dB fraction from channel capacity

I Why Beyond Turbo Codes?

I LDPC codes

I New classes of codes that are capacity-achieving with lowcomplexity encoder and decoder

Polar & spatially-coupled LDPC codes

Choosing Channel Codes for 5G

I Main considerationsI Performance, complexity, rate-compatibility

I LTE deploys turbo codes [Berrou et. al., 1993]I Perform within a dB fraction from channel capacity

I Why Beyond Turbo Codes?

I LDPC codes

I New classes of codes that are capacity-achieving with lowcomplexity encoder and decoder

Polar & spatially-coupled LDPC codes

Choosing Channel Codes for 5G

I Main considerationsI Performance, complexity, rate-compatibility

I LTE deploys turbo codes [Berrou et. al., 1993]I Perform within a dB fraction from channel capacity

I Why Beyond Turbo Codes?

I LDPC codes

I New classes of codes that are capacity-achieving with lowcomplexity encoder and decoder

Polar & spatially-coupled LDPC codes

Choosing Channel Codes for 5G

I Main considerationsI Performance, complexity, rate-compatibility

I LTE deploys turbo codes [Berrou et. al., 1993]I Perform within a dB fraction from channel capacity

I Why Beyond Turbo Codes?

I LDPC codes

I New classes of codes that are capacity-achieving with lowcomplexity encoder and decoder

Polar & spatially-coupled LDPC codes

Polar Codes [Arikan, 2009]

I First provably capacity-achieving codes with lowencoding/decoding complexity

I Outperform turbo codes for large block length n

I Best performance for short block length n

I Complexity O(nlogn)

I Better energy-efficiency for large n than other codes

I Code construction is deterministic

I No error floor

Polar Codes [Arikan, 2009]

I First provably capacity-achieving codes with lowencoding/decoding complexity

I Outperform turbo codes for large block length n

I Best performance for short block length n

I Complexity O(nlogn)

I Better energy-efficiency for large n than other codes

I Code construction is deterministic

I No error floor

Channel Polarization

I n instances of a channel are transformed into a set of channelsthat are either noiseless or pure-noise channels

I Polar code: send information bits over good channels

I Fraction of good channels approaches the capacity of theoriginal channel

Channel Polarization

I n instances of a channel are transformed into a set of channelsthat are either noiseless or pure-noise channels

I Polar code: send information bits over good channels

I Fraction of good channels approaches the capacity of theoriginal channel

Channel Polarization

I n instances of a channel are transformed into a set of channelsthat are either noiseless or pure-noise channels

I Polar code: send information bits over good channels

I Fraction of good channels approaches the capacity of theoriginal channel

Channel Polarization

I n instances of a channel are transformed into a set of channelsthat are either noiseless or pure-noise channels

I Polar code: send information bits over good channels

I Fraction of good channels approaches the capacity of theoriginal channel

Channel Polarization

I n instances of a channel are transformed into a set of channelsthat are either noiseless or pure-noise channels

I Polar code: send information bits over good channels

I Fraction of good channels approaches the capacity of theoriginal channel

Channel Polarization

I n instances of a channel are transformed into a set of channelsthat are either noiseless or pure-noise channels

I Polar code: send information bits over good channels

I Fraction of good channels approaches the capacity of theoriginal channel

Channel Polarization

I n instances of a channel are transformed into a set of channelsthat are either noiseless or pure-noise channels

I Polar code: send information bits over good channels

I Fraction of good channels approaches the capacity of theoriginal channel

Wireless Channel is Time-Varying

I Hybrid ARQ with Incremental Redundancy (HARQ-IR)I Send additional coded bits until decoding is successful

NACK

NACK

ACK

TX RX

Wireless Channel is Time-Varying

I Hybrid ARQ with Incremental Redundancy (HARQ-IR)I Send additional coded bits until decoding is successful

NACK

NACK

ACK

TX RX

HARQ-IR

I Encode for degraded channels W1 � W2 � . . . � WK withcapacities C1 ≥ C2 ≥ . . . ≥ CK

n1

k information bits

1st transmission

Rate R1 = k/n1

n2

2nd transmission

R2 = k/(n1 + n2)

Problem: HARQ-IR with Polar Codes?

I HARQ-IR requires the same information set for all codesI Polar code designed for fixed length n

I Information sets are different for different lengths

Our Solution: Parallel-Concatenated Polar Codes

I Encoder R1 > R2 (ex. K = 2)

information bits

DividerPolar encoderC(n1,R1)

Polar encoderC(n2,R2)

D

I Decoder

ReceiverPolar decoderC(n2,R2)

Polar decoderC(n1,R1)

Our Solution: Parallel-Concatenated Polar Codes

I Encoder R1 > R2 (ex. K = 2)

information bits

DividerPolar encoderC(n1,R1)

Polar encoderC(n2,R2)

D

I Decoder

ReceiverPolar decoderC(n2,R2)

Polar decoderC(n1,R1)

frozenbits

R2

I To choose D, used nested property of polar codes[Korada, 2009]

Capacity Result

Theorem [Hong, Hui & Maric, 2015]

For any sequence of degraded channels W1 � W2 � . . . � WK

there exists a sequence of rate-compatible punctured polar codesthat is capacity-achieving

I Details: arxiv.org/pdf/1510.01776v1.pdf

Summary

Constructed family of rate-compatible polar codes

I Achieves capacity

I Can be used for HARQ-IR

Thank You!