Providing Unlimited Wireless Capacity

Bob Brodersen

Berkeley Wireless Research CenterAdaptrum, Inc and SiBEAM, Inc.

Univ. of California, Berkeley

More Capacity – Do we really need it ?? Major bandwidth driver is moving from voice to

video and data – a 1000-10,000 fold requirement increase

User base is moving to be a significant fraction of the 6 Billion world population

Devices (without an attached human) will communicate wirelessly » Source to HD display with uncompressed video => 4

Gbit/sec» A home will have a 1000 radios

This is not going to be addressed by improving any radio system we have now

Looks like we need it ….

Claim: There are technological solutions to providing this capacity which is based on abandoning the property rights model of “owning” frequency spectrum

Assertions: » The concept of fixed frequency spectrum allocation has

become fundamentally flawed» We need to exploit wireless communication strategies

that exploit the time, space and frequency degrees of freedom

» Exploiting these new approaches could allow essentially “unlimited capacity”

Why has Frequency Allocation become a Flawed Concept?

The applications are continually changing and allocation doesn’t mean use

Frequency is only one of the 3 degrees of freedom to use to avoid interference - time and space are actually more effective

UWB, 60 GHz and Cognitive radios individually exploit the 3 DOF

Frequency » Separate users by using different frequency bands

– traditional method using analog filtering» Exploit wide bandwidths and DSP» Exploit higher frequencies

– present CMOS technology allows use up to 100GHz Space and Angle

» Reduce transmit power – decreases radius of omni-directional cells

» Exploit the angular nature of the spatial channel– Multiple antennas

Time» Impulse filtering» Sense interference and avoid

UWB, 60 GHz and Cognitive radios individually exploit the 3 DOF

Frequency » Separate users by using different frequency bands

– traditional method using analog filtering» Exploit wide bandwidths and DSP (UWB) » Exploit higher frequencies

– present CMOS technology allows use up to 100GHz (60 GHz) Space and Angle

» Reduce transmit power – decreases radius of omni-directional cells (UWB)

» Exploit the angular nature of the spatial channel– Multiple antennas (60 GHz)

Time» Impulse filtering (UWB)» Sense interference and avoid (Cognitive Radio)

Talk Organization

A discussion of the emerging techniques that are exploiting the three degrees of freedom in new ways» UWB» Cognitive Radios» 60 GHz

How these can be combined to achieve “unlimited capacity”

Lets start with UWB…

Breakthrough! For the first time the regulators allowed sharing of the frequency spectrum» Underlay – Allow sharing of the spectrum if the

interference is negligible » Ultrawide bandwidths were allowed

Fundamental choices remain on how to best to use the wide bandwidth» Stay with conventional frequency domain

thinking» Or exploit the time degree of freedom

UWB Frequency and Time domain strategies

Time Frequency

FrequencyTime

TimeUWB Impulse

FrequencyTime

FrequencyUWB OFDM(multiple sine waves)

Traditional user(narrowband)

Noise

Pow

erP

ower

Pow

er

Time domain filters (block the impulse in time) can be easily made adaptive (unlike frequency domain)

Time domain interference

Interference in frequency domain

UWB is allowed in over 11 GHz of spectrum

0 20 40

UWBUWB

UWB

60 80 100 GHz

Comm Vehicular

Limited power allows a high level of spatial reuse Chip sets are available for both OFDM and

Impulse approaches But ….. If this is such a good idea why

has it not been commercially successful

What happened …

Two competing approaches were attempted which resulted in a standards battle which was waged without good technical input

The (comfortable) frequency domain approach (OFDM) had too high a complexity and too low a performance

The time domain (impulse) approach is too different and needs much more R&D in theory and implementation

Cognitive Radios

Basic idea » Exploit the time degree of freedom by sensing if

a signal is present » Then take steps to assure there isn’t

interference This is quite restricted, others use a more

expanded definition

A Cognitive Radio using Time and Frequency

Sense the spectral environment over a wide bandwidth Transmit in “White Space” Detect if primary user appears Move to new white space Adapt bandwidth and power levels to meet QOS requirements

Pow

er

Frequency

PU1

PU2

PU3

PU4

Cognitive Radio system level view

Network coordination

Medium Access

Sensing Signal Processing

Sensing radio

Resource Allocation

Wideband signaling

Wideband radio

Physical Layer

NetworkLink Layers

Spectrum sensing is the key enabling functionality and

must be very sensitive to limit unwanted interference

Sensing Weak Signals

A new radio functionality – requires new algorithms and understanding

A/D N pt. FFT

x(t)Averageover T

Energydetect

Threshold

High SNR Low SNR

Spectrum density

Energy Detector

A/D N pt. FFTCorrelate

X(f+a)X*(f-a)Averageover T

x(t)Featuredetect

Spectral correlation

Cyclostationary Feature Detector

Sensing Performance (Danijela Cabric)

Incoherent sensing time goes as 1/(SNR2)

Coherent sensing time goes as 1/SNR

Log

Tim

e

(Cyclostationary)

Incoherent processing

Coherent sensing

Cyclostationary

Coherent sensing - ATSC signal

Correlation of fixed header is used by Adaptrum in FCC trials being held now – highest performing results

Adaptrum TV-band CR prototype

CR Transceiver

CR Baseband/MAC FPGA (Altera)

Bowtie wideband UHF antenna

ATSC sensitivity measurement result

-122 -121.5 -121 -120.5 -120 -119.5 -119 -118.5 -118 -117.50

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Input Signal Power (dBm/6 MHz)

Pro

babi

lity

of S

ucce

ssfu

l Det

ectio

n

Channel 21 (515 MHz)

Channel 59 (743 MHz)

Planned Bay Area Cognitive Radio - 400 MHz experimental testbed

Using the Space Degree of Freedom to improve sensing

Spatially separated sensing radios can make independent measurements

Single radio sensitivity can be improved by the use of multiple antennas using beamforming

Spatially separated sensing radios

Expected result for independent measurements:

Primary System

Tx

Rx

Decoding SNR

Exploit spatial diversity in Sensing SNRs

Pd, network=1-(1-Pd, radio)N

Experimental SetupLocation (11,9)

Location (16,3)

Central combining and processing

Fiber provides 1/3 mile separation between radios and platform

controlledPHY and MAC

integration

Sensing radio

Sensing radio

Sensing PHY/MAC processors

Network Spectrum Sensing Results

Prob. of false alarm

Prob

. of d

etec

tion

1 radio

5 radios

If spacing >> λ/2 a few cooperative radios give big improvements

Danijela Cabric, Mubaraq Mishra and Anant Sahai

Dynamic range problem in wideband sensing P

SD

Freq.

A/DLNA AGC

Fixed LOBand of interest

Wideband sensing is required to quickly sense the open bands» Small signals need to be sensed in the presence of strong

interference and then processed digitally» Places difficult requirements on RF front end and A/D

Multi-antenna spatial processing provides two solutions

Multi-antenna spatial processing to improve sensing

Primary user f1 Phased antenna array

Improvements with N antennas » allow suppression of up to N-1 large

signals» provide up to an N times increase in

sensitivity We’ll find more uses for these arrays

Primary user f2

Time DomainWideband Signal

Interference Estimation

A/DDigital

Processing

Attenuate Strong Interferenceand Reduce DR to ADC

+-

RX

One possible implementation

Time Domain Interference Cancellationto address the dynamic range problem (Jing Yang)

Adaptive Filter

+-

r

ADC

DAC

AGC( )I S N

ADCAGCProgrammable Analog Delay LineProgrammable Analog Delay Line

ADC

Cancellation ADCN Bits

ADC

Residue ADCM Bits

DAC

Yields N+M equivalent bits of dynamic range

Simulated interference attenuation

Attenuate the strong interference and reduce the dynamic range to the Residue Attenuate the strong interference and reduce the dynamic range to the Residue ADC by 35dBADC by 35dB

Extending the effective number of bits for this system by nearly 6 bitsExtending the effective number of bits for this system by nearly 6 bits

10

0

-10

-20

-30

-40

-50

-60

Mag

nit

ud

e (d

B)

190 192 194 196 198 200 202 204 206 208 210

Frequency (MHz)

20

10

0

-10

-20

-30

-40

Mag

nit

ud

e (d

B)

190 192 194 196 198 200 202 204 206 208 210

Frequency (MHz)

Strong FSK modulated interfering signal

Moderate sinusoidal interfering signal

Before After

Attenuated strong interfering signal

CR signal

The opportunity of higher frequencies

0 20 40 60 80 100 GHz

Effectively no use above 5 GHz Antenna arrays require only a small area Absolutely necessary to get to gigabit/sec rates

0-3 GHz >> 99% of all wireless transmission

7 GHz of unused and unlicensed spectra

Use of Higher Frequencies (e.g. 60 GHz)

Conventional wisdom is that lower frequencies are better » Only line of sight operation is possible and can’t

penetrate materials» The technology to process signals is expensive» As you go up in frequency there is an inherent

“path loss” that reduces range

Not True!!!

Material Penetration – actually not so bad

60 GHz 2.5 GHz

Pine board – ¾ ” 8 db 1.5dB

Clay Brick 9 dB 2 dB

Glass with wire mesh 10.2 dB 7.7 dB

Asphalt Shingle 1.7 dB 1.5 dB

Plywood – ¾ ” 7 dB 1.5 dB

Clear Glass 6.4 dB 3.6 dB

Atmosphere per 100m

1.5 dB 0 dB

What about Oxygen absorption?

Millimeter wave radios

Misconception: Implementing millimeter wave radios requires exotic materials» Conventional state-of-the-art digital CMOS can

be used to implement integrated radios up to 100 GHz

» Future technology scaling will allow even higher frequency operation (research is beginning into Terahertz operation)

60-GHz CMOS operation (130 nm)

1 mm

1.3 mm

Use of transmission line interconnect allows control of electrical and magnetic fields

Better control than at lower frequencies!

11-dB Gain@ 60 GHz

60-GHz CMOS Receiver front-end

CMOS integration means even a 60 GHz receiver will eventually cost about the same as a WiFi

Millimeter wave radios

Misconception: As you go up in frequency there is an inherent “path loss” that reduces range » This comes from only considering omnidirectional

antennas which have a size that is inverse with carrier frequency

» Solution is to keep area constant using directional antennas - then the received signal increases with frequency

Tx

Rx

Low Gain Tx Antenna

Area of sphere = 4r2

Captured energy

Antenna fundamentals: Receive Antenna Energy

2r4

rA

spherea of Area

antennareceive ofArea

Ar = Area of receiveantenna

Fraction of power received from an omnidirectional transmission

=

Effective transmit antenna power

Maximum increase in power in direction of beam

= (wavelength of carrier)

Effective maximum power as if it came from an omnidirectinal antenna = Pt Gt

Rx

Pt Gt

Tx

2

tt

A4G

At = Area of transmitter

The link budget with directional antennas

2t

t

A4G

2tr

2t2r

ttr r

AA1P

r4

AGPP

Receive power improveswith frequency!!

Rx

Captured energyIn Area Ar

Wasted energyTx

22 dB more gain at 60 GHz over 5 GHz if antenna size is kept constant

(Compare to Friis “path loss” formula ) 2tr2

trr4

GGPP

Millimeter wave radios

Misconception: Only line of sight transmission is possible» millimeter waves reflect like lower frequency

waves, so adaptive directional antenna arrays can choose strongest signal

» millimeter waves have higher material penetration loss, but this can be compensated for with the higher power and antenna gain

Non line of sight transmission

Transmitter in Source

Phased Array Antennas» Power used more efficiently for better reception, longer distance,

higher bandwidth» Dynamically steers beam to specific receiving station

SiBEAM Module

Phased array circuitry

N antennas allow N power amps to transmit in parallel Phase accuracy requirement is very low as the number

of antennas go up (2 bits for 16 antennas)

1 2 3 4 5

x 104

0

2

4

6

8

10

12

Number of Symbols

Eig

en

valu

es

Algorithms for adaptive beamforming

1

2

3

4

0 0 0

0 0 0' '

0 0 0

0 0 0

y x

Separate the multipath into separate channels through an SVD

Choose the strongest one and put all the energy into it

σ1

σ2

σ3

σ4

Blind tracking

)(ty

ArrayProcessing

)(tx

ArrayProcessing

1st path, σ1=10

2nd path, σ2=6

As beams get narrow the capacity increases (Ada Poon)

Capacity 2 A -2W T log SNR

Carrier frequency

Cumulative scattering solid angle

Transmit Array area

Polarization

The old (Shannon) Formula:

Capacity W T log SNR

With spatial processing:

Transmission interval

Bandwidth

New “spatial” degreesof freedom provide multiplicative increases

The time frequency degrees of freedom

How do we get to “unlimited capacity”?

Two ways – probably more» Angular isolation of beamformed signals

requires not only co-location but the same angle

– This essentially eliminates interference – Add in “UWB-like” spectrum sharing and cognitive

techniques to achieve essentially “unlimited”

» Increasing the number of users with multiple antennas provides an unlimited increase if frequency or antenna area is increased

Angular isolation

A transmits towards B with phased array B only receives in direction of A Transmitter C doesn’t interfere with B

C

A B

Angular isolation

Interference between beamformed signals has to not only be in the same space, but also have the same angle

C

A B

Another solution to aligned receivers

Start with link AB Add aligned link CD What do we do now….

DC A B

Use a reflection

Usually at least 2-3 reflections This is requiring more resolution in the

phase shifters to control sidelobes

Capacity increases with number of RX&TX=N

Scheme No.ofAntennas

CapacityIncrease

Frequency/Time Subdivision N 1

Packet Multihop N N1/2

Cooperative Diversity (Tse) N N

MIMO at each user (Chung) N2 N2

MIMO with reflections N A/2 (N A/2)2

Back to our Frequency Allocation Chart

If we use all 3 degrees of freedom then a chart like this really is meaningless

A future allocation chart…

Now how do we get to this!!

Shared Spectrum

Shared Spectrum

Shared Spectrum

Shared Spectrum

Shared Spectrum

Shared Spectrum

Shared Spectrum