EE 359: Wireless Communications

Advanced Topics in Wireless

3/12/2020
 Future Wireless Networks
Ubiquitous Communication Among People and Devices

Next-Gen Cellular/WiFi
Smart Homes/Spaces
Autonomous Cars
Smart Cities
Body-Area Networks
Internet of Things
All this and more …
 Challenges
 Network/Radio Challenges 5-6G AdHoc

 Gbps data rates with “no” errors Short-Range

 Energy efficiency
 Scarce/bifurcated spectrum
 Reliability and coverage
 Heterogeneous networks
 Seamless internetwork handoff
 Device/SoC Challenges BT
Radio
 Performance GPS
Cellular
 Complexity Cog
 Size, Power, Cost, Energy
Mem
 High frequencies/mmWave WiFi
 Multiple Antennas CPU mmW
 Multiradio Integration
 Software-Defined (SD) Radio:
Is this the solution to the device challenges?
BT
FM/XM A/D
Cellular GPS

DVB-H
A/D
Apps DSP
Processor WLAN A/D
Media
Wimax
Processor A/D

 Wideband antennas and A/Ds span BW of desired signals

 DSP programmed to process desired signal: no specialized HW

Today, this is not cost, size, or power efficient

SubNyquist sampling may help with the A/D and DSP requirements
“Sorry America, your
airwaves are full*”

Source: FCC
 On the Horizon,
the Internet of Things
What is the Internet of Things:

 Enabling every electronic device to be connected

to each other and the Internet

 Includes smartphones, consumer electronics,

cars, lights, clothes, sensors, medical devices,…

 Value in IoT is data processing in the cloud

Different requirements than smartphones: low rates/energy consumption

 Are we at the Shannon
limit of the Physical Layer?
We are at the Shannon Limit
“The wireless industry has reached the theoretical limit of
how fast networks can go” K. Fitcher, Connected Planet
“We’re 99% of the way” to the “barrier known as
Shannon’s limit,” D. Warren, GSM Association Sr. Dir. of
Tech.
Shannon was wrong, there is no limit
“There is no theoretical maximum to the amount of data
that can be carried by a radio channel” M. Gass, 802.11
Wireless Networks: The Definitive Guide
 “Effectively unlimited” capacity possible via personal cells
(pcells). S. Perlman, Artemis.
What would Shannon say?
We don’t know the Shannon
capacity of most wireless channels
 Time-varying channels.
Channels with interference or relays.
Cellular systems
Ad-hoc and sensor networks
Channels with delay/energy/$$$ constraints.

Shannon theory provides design insights

and system performance upper bounds
 Current/Next-Gen
Wireless Systems
 Current:
 4G Cellular Systems (LTE-Advanced)
 6G Wireless LANs/WiFi (802.11ax)
 mmWave massive MIMO systems
 Satellite Systems
 Bluetooth
 Zigbee
 WiGig
 Emerging
 5G Cellular and 7G WiFi Systems
 Ad/hoc and Cognitive Radio Networks Much room
 Energy-Harvesting Systems For innovation
 Chemical/Molecular
Future Cell Phones
Everything
Burden for wireless in one device
this performance is on the backbone network
San Francisco

BS
BS

LTE backbone is the Internet

Internet
Boston
Nth-Gen Phone Nth-Gen
Cellular System Cellular

BS

Much better performance and reliability than today

- Gbps rates, low latency/energy , 99.999% coverage
The Licensed Airwaves are “Full”

Also have Wifi

And mmWave
10s of GHz of Spectrum

Source: FCC
What is the Internet of Things:
Enabling every electronic device to be
connected to each other and the Internet

Includes smartphones, consumer electronics,

cars, lights, clothes, sensors, medical devices,…

Value in IoT is data processing in the cloud

Different requirements than smartphones: low rates/energy consumption

 Are we at the Shannon
capacity of wireless systems?
We don’t know the Shannon
capacity of most wireless channels
Channels without models: molecular, mmW, THz
Time-varying channels.
Channels with interference or relays.
Cellular systems
Ad-hoc and sensor networks
Channels with delay/energy/$$$ constraints.

Shannon theory provides design insights

and system performance upper bounds
Enablers for Increasing Wireless Data
Rates in next-generation networks
Wireless Network Design
Rethinking cellular system design
Hetnet optimization

PHY/MAC Techniques
Utilizing more spectrum (mmW/THz)
(Massive) MIMO
New modulation, coding, and detection
for rapidly varying channels 1971
1980s
New MAC strategies 2014
New PHY and MAC Techniques
New Waveforms
More bandwidth/energy efficient/robust to changing channels
More flexible and efficient subcarrier allocation

New Coding Techniques

New channels amd new waveforms
New requirements (low latency and low complexity)

New Detection Techniques

Lower complexity, robust to impairments, blind

New Multiple Antenna Techniques

New space-time modulation, coding, and detection methods
Massive MIMO

New Access Techniques

Efficient access for low-rate IoT Devices
Sparse code MAC, GFDM, OTFS, variants of OFDMA
Massive non-orthogonal MAC

Access/interference mitigation for unlicensed LTE

OFDM Alternatives: Motivation
In OFDM, Doppler spread causes
impulse responses to be rapidly time
varying.
Phase change must be small within
frame

Joint with T. Dean, M. Chowdhury, and N. Grimwood

Dual of OFDM: FDM-FDCP
Frequency Division Multiplexing with Frequency
Domain Cyclic Prefix (FDM-FDCP)
Design for high-Doppler, zero (or low) delay spread
Applies time-frequency duality to OFDM
FDM-FDCP

Differences from OFDM:

Cylic prefix is placed in frequency domain
Element-wise equalization in time domain
 T.
Performance in AWGN
mmWave Massive MIMO
10s of GHz of Spectrum Dozens of devices

Hundreds
of antennas

mmWaves have large attenuation and path loss

For asymptotically large arrays with channel state
information, no attenuation, fading, or interference
mmWave antennas are small: perfect for massive MIMO
Bottlenecks: channel estimation, complexity, propagation
Ideal beamforming disappears with shadowing
Blind MIMO Decoding
Geometric Interpretation

-1

Joint with T. Dean and M. Wooters

Problem Statement

Receiver only knows the constellation xi,j

Blind decoding: jointly estimate A and X
Previous approaches:
 Complexity |Χ|n
Assume sparsity in A or X
Blind Decoding by Fitting a Parallelepiped

Gaussian noise will not greatly

distort this shape

maximize   | (∗)
 
𝑼
 
subject to

is a non-convex optimization problem

 
Constrained gradient descent works but is slow.
The Vertex Hopping algorithm uses concepts from solving mixed-
integer linear programming to solve .
Gradient Descent Solution
Works but is slow Have shown for proper X
Slow: O(n12)
All optima are solutions
for n<6
Local optimal for n≥6

All optima lie

on vertices

Objective function is
monotonic on edges
Vertex Hopping: Key Idea
Search vertices of feasible region as if
solving a linear program
Find a suitable initial vertex
Hardest step: O(n4k)
Pick random feasible point, compute

gradient, move to boundary, project

gradient to boundary, move to boundary, …

Search vertices for global optimum

Only a couple hops in practice
AWGN and Fading Performance
Runtime Performance
Vertex Hopping* Gradient Descent
n k Pr[success] Time (s) Pr[success]
 
Time (s)
2 8 1.0 0.99
3 13 1.0 0.99
4 18 1.0 0.99
5 18 1.0 0.97
6 22 1.0 0.93
8 30 0.99 0.80
10 45 0.99 0 -
12
12 60
60 0.99
0.99 0
0 --

 Implemented in Rust
*

MATLAB’s fmincon
ML in PHY layer design
Analog Channel N( f )

Message Transmitter H( f ) A/D Receiver Message

x(t ) y (t )

• PHY transmitter and receiver design typically based

on a mathematical channel mode
• Accurate channel models may not be known
• Models may not enable computationally efficient PHY
algorithms (decoding, detection, message recovery)
• How does an ML-based approach solve this?
• No need for an underlying model or all its parameters
• Learn the design directly from data
Machine Learning: Some History
• First Artificial Intelligence Workshop in 1956
• Predicted machine intelligence on human scale within a
generation
• Significant funding pumped into AI from 60s-80s
• “AI Winter”: 1970s-1990s
• 2000s: AI focused on machine learning, cognitive systems,…

1950s 1960s 1970s 2000s

ML Today is a Bandwagon

Should we jump on?

Or run screaming?

• Communication theorists have an important role to play in

• Formulating problems where ML can “beat” theory in system design
• Developing upper and lower bounds on ML performance
• Developing mathematical foundations of ML (ITSoc JSAIT)
Uplink Detection with SBRNN
Single cell mmw uplink (sparse)
The BS has multiple antennas
Each device has a single antenna
Can be extended to multi-antenna case
Use Sliding BRNN detector to detect the
received symbols at the BS without any
knowledge of CSI
Performance: SBRNN vs Viterbi
 Parameter  Value
SBRNN is close to the optimal Viterbi, Carrier Frequency 28 GHz

and outperforms Viterbi-cut Bandwidth 800 MHz

Transmit Power 11 dBm
SBRNN trained on a single SNR range
Tx-Rx Separation distance 60 m
generalizes well to other SNRs. Tx-Rx Antenna Gains 24.5 dBi
Modulation Scheme BPSK

4 RX Antennas 128 RX Antennas

Performance: SBRNN w/ CSI imperfection
Training Overhead and Run Time Efficiency

Training: # of samples needed for 90% detection accuracy

Testing: run time of detecting one 200-bit message1
Receive Antennas Sliding BRNN Viterbi

4 0.244 s 12.461 s
128 0.264 s 52.681 s
ViterbiNet: learn just p(y|x)
 Equalization of Nonlinear Channels
Volterra equalizer: uses Volterra series approximation
Tradeoff of distortion reduction and noise enhancement
We derive optimal training SNR
We propose an alternate NN approach
1.7 dB gain over no equalization
.6 dB gain over optimized Volterra
1.7db

.6 db

38
 Rethinking Cellular System Design
How should cellular
CoMP Small
Cell systems be designed?
Relay

Will gains be big or

DAS incremental; in capacity,
coverage or energy?
Cellular systems reuse channels/timeslots in different cells
 Traditional design assumes system is “interference-limited”
 Capacity unknown; upper bound based on BC/MAC with pooled antennas
 No longer the case with recent technology advances:
 MIMO, multiuser detection, cooperating BSs (CoMP) and relays
 Raises interesting questions such as “what is a cell?”
 Dynamic self-organization (SoN) needed for deployment and
optimization
 Small cells are the solution to
increasing cellular system capacity
In theory, provide exponential capacity gain
Cloud Cellular networks are
Optimization increasingly hierarchical
Large cells for coverage
IP Network Small cells for capacity
and power efficiency
Cell resource optimization
Small cell BS is best done in the cloud
Macrocell BS
HetNet Optimization Challenge
Algorithmic complexity
Frequency allocation alone is NP hard
Also have MIMO, power control, CST, hierarchical
networks: NP-really-hard
Advanced optimization tools needed, including a
combination of centralized (cloud) distributed, and locally
centralized (fog) control
Cloud Optimization
ML can also play a role
Fog
Optimization
Next challenge:
optimizing caching
Small cell BS
and edge computing
Macrocell BS
 Software-Defined Network Architecture
(generalization of NFV, SDN, cloud-RAN, and distributed cloud)
Video
Video Security
Security
Vehicular
Cloud Computing
Vehicular
Networks
M2M
M2M App layer Health
Health
Networks

Freq.
Freq. Power Self QoS
QoS CS
CS
Allocation
Allocation Control Healing ICIC Opt.
Opt. Threshold
Threshold

Network Optimization

UNIFIED CONTROL PLANE

HW layer
Distributed Antennas
WiFi Cellular mmWave … Ad-Hoc
Networks
SDWN Challenges
Algorithmic complexity
Frequency allocation alone is NP hard
Also have MIMO, power control, CST, hierarchical
networks: NP-really-hard
Advanced optimization tools needed, including a
combination of centralized (cloud) distributed, and locally
centralized (fog) control
Cloud Optimization

Fog
Hardware Interfaces X2 X2
X2
Optimization
X2

Seamless handoff
Small cell BS
Resource pooling
Macrocell BS
Fog-Based Resource Allocation
Form virtual cells of multiple BSs

Joint with M. Yemini

Hierarchical Clustering
Avoids reclustering due to local changes
Minimizes max interference between clusters
Fog-Optimization vs. Centralized
Use clustering technique to cluster BSs, then optimize
power allocation to maximize uplink sum rate
Consider multiple clustering techniques (will also look at ML)
Nonconvex approximation for optimization
Single-User Decoding per BS Joint Decoding in Virtual Cell

10x loss

55% loss
 Ad-Hoc Networks

 Peer-to-peer communications
 No backbone infrastructure or centralized control
 Routing can be multihop.
 Topology is dynamic.
 Fully connected with different link SINRs
 Open questions
 Fundamental capacity region
 Resource allocation (power, rate, spectrum, etc.)
 Routing
 Cooperation in
Wireless Networks

 Many possible cooperation strategies:

 Virtual MIMO, relaying (DF, CF, AF), one-
shot/iterative conferencing, and network coding
 Nodes can use orthogonal or non-orthogonal channels.
 Many practice and theoretical challenges
 New full duplex relays can be exploited
 General Relay Strategies
TX1 RX1
X1
Y4=X1+X2+X3+Z4
relay
Y3=X1+X2+Z3 X3= f(Y3)

Y5=X1+X2+X3+Z5
X2
TX2 RX2

 Can forward message and/or interference

 Relay can forward all or part of the messages
 Much room for innovation
 Relay can forward interference
 To help subtract it out
 Beneficial to forward both
interference and message

• For large powers, this strategy approaches capacity

 Spectrum innovations beyond
licensed/unlicensed paradigms
 Cognitive Radios
CRTx CRRx
IP
NCR
NCR CR CR NCRRx
NCRTx

MIMO Cognitive Underlay Cognitive Overlay

 Cognitive radios support new users in existing
crowded spectrum without degrading licensed users
 Utilize advanced communication and DSP techniques
 Coupled with novel spectrum allocation policies

 Multiple paradigms
 (MIMO) Underlay (interference below a threshold)
 Interweave finds/uses unused time/freq/space slots
 Overlay (overhears/relays primary message while
“Green” Wireless Networks
Pico/Femto
How should wireless
Coop
MIMO systems be redesigned
Relay
for minimum energy?

DAS Research indicates that

significant savings is possible

Drastic energy reduction needed (especially for IoT)

New Infrastuctures: Cell Size, BS/AP placement,
Distributed Antennas (DAS), Massive MIMO, Relays
New Protocols: Coop MIMO, RRM, Sleeping, Relaying
Low-Power (Green) Radios: Radio Architectures,
Modulation, Coding, Massive MIMO
DAS to minimize energy
Optimize distributed BS antenna location
Primal/dual optimization framework
Convex; standard solutions apply
For 4+ ports, one moves to the center
Up to 23 dB power gain in downlink
Gain higher when CSIT not available

6 Ports
3 Ports
 Energy-Constrained Radios
Transmit energy minimized by sending bits very slowly
Leads to increased circuit energy consumption

Short-range networks must consider both transmit and

processing/circuit energy.
Sophisticated encoding/decoding not always energy-efficient.
MIMO techniques not necessarily energy-efficient
Long transmission times not necessarily optimal
Multihop routing not necessarily optimal

Sub-Nyquist Sampling
Sub-Nyquist Sampled Channels
Analog Channel N( f )

Message Encoder H( f ) Decode Message

x(t ) y (t ) r
C. Shannon

Wideband systems may preclude Nyquist-rate sampling!

Sub-Nyquist sampling well explored in signal processing

 Landau-rate sampling, compressed sensing, etc.
 Performance metric: MSE

H. Nyquist

We ask: what is the capacity-achieving sub-Nyquist

sampler and communication design
Capacity and Sub-Nyquist Sampling
 Consider linear time-invariant sub-sampled channels
Preprocessor

 Theorem: Capacity-achieving sampler t  n(mTs )

q(t) s1 (t ) y1[n]

zzzz
p(t )
zzzz
zz
 zzzzz
s (t )
zzzzz
y[n] or
t  n(mTs )

yi [n]
si (t )

Optimal filters suppress aliasing t  n(mTs )

sm (t ) ym [n]

Sub-Nyquist sampling is optimal for some channels!

Example: Multiband Channel
 Consider a “sparse” channel, and an optimally
designed 4-branch filter bank sampler

- Outperforms single-
branch sampling.

- Achieves full-capacity
above Landau Rate

Landau Rate: sum of total bandwidths

 Wireless Sensor Networks
Data Collection and Distributed Control
• Smart homes/buildings
• Smart structures
• Search and rescue
• Homeland security
• Event detection
• Battlefield surveillance

 Energy (transmit and processing) is the driving constraint

 Data flows to centralized location (joint compression)
 Low per-node rates but tens to thousands of nodes
 Intelligence is in the network rather than in the devices
 Where should energy come from?

• Batteries and traditional charging mechanisms

• Well-understood devices and systems

• Wireless-power transfer
• Poorly understood, especially at large distances and with
high efficiency

• Communication with Energy Harvesting Radios

• Intermittent and random energy arrivals
• Communication becomes energy-dependent
• Can combine information and energy transmission
• New principles for radio and network design needed.
Distributed Control over Wireless
Automated Vehicles
- Cars
- Airplanes/UAVs
- Insect flyers

Interdisciplinary design approach

• Control requires fast, accurate, and reliable
feedback.
• Wireless networks introduce delay and loss
• Need reliable networks and robust controllers
: Many design challenges
•
 Chemical Communications

 Can be developed for both macro (>cm) and

micro (<mm) scale communications
 Greenfield area of research:
 Need new modulation schemes, channel impairment
mitigation, multiple acces, etc.
Applications

Data rate: .5 bps

“fan-enhanced” channel
Current Work
Slow dissipation of chemicals
leads to ISI
Can use acid/base
transmission to decrease ISI
Similar ideas can be applied
for multilevel modulation and
multiuser techniques
Currently testing in our lab
New equalization based on
Sending text messages with windex and vinegar
machine learning
Stanford Report:
Increased data rate 10x November 15, 2016
 Applications in Health,
Biomedicine and Neuroscience
Neuroscience
-Nerve network
Body-Area (re)configuration
Networks -EEG/ECoG signal
processing
- Signal processing/control
for deep brain stimulation
- SP/Comm applied to
bioscience
Recovery from Nerve Damage

ECoG Epileptic Seizure Localization

EEG

ECoG
Epileptic Seizure Focal Points
Seizure caused by an oscillating signal moving across neurons
 When enough neurons oscillate, a seizure occurs
 Treatment “cuts out” signal origin: errors have serious implications
Directed mutual information spanning tree algorithm applied
to ECoG measurements estimates the focal point of the seizure
Application of our algorithm to existing data sets on 3 patients
matched well with their medical records

ECoG
Data
Summary
The next wave in wireless technology is upon us
This technology will enable new applications that will change
people’s lives worldwide

Future wireless networks must support high rates for

some users and extreme energy efficiency for others
Small cells, mmWave massive MIMO, Software-Defined
Wireless Networks, and energy-efficient design key enablers.

Communication tools and modeling techniques may

provide breakthroughs in other areas of science
 The End
 Thanks!!!
 Good luck on the final and final project
 Have a great spring break, stay safe and healthy