Expanding the Scope of SETI
at the Allen Telescope Array
Gerry Harp, B. Wilcox, J. Arbunich, P. Backus, R.
Ackermann, J. Jordan, T. Kilsdonk, Samantha Blair,
J. C. Tarter and ATA Team
Search for Extraterrestrial Intelligence
Exo-technology, Exo-civilizations
G. R. Harp
Tasks for ET
• Get our attention (help us locate them).
– ET will send us a signal that is easily
recognized as artificial.
• Tell us something.
– Sending any signal costs money (power).
– They will encode info s.t. we can interpret it.
• They have a limited power budget.
– They have to spend some power getting our
attention and some more to send info.
G. R. Harp
How do we know signal is
artificial?
• Uncertainty principle argument?
G. R. Harp
Uncertainty Principle is a
Tautology For Sampled Signals
• Uncertainty principle for signals of unknown bandwidth
present in wave fields:
t f  1
Heisenburg
• As soon as we measure the signal (convert to power), the
wavefunction collapses and the uncertainty goes away:
t f  1
PERIOD!
Nyquist, Shannon
• We can’t classify signals as artificial or natural based
on uncertainty principle, all measured signals have
equal “uncertainty” = 0.
G. R. Harp
For Example
Power (10^-26 Watts)
Measured Frequency Power Spectrum
100
80
• Measurement
time = t = 1s.
• What is signal
bandwidth f?
60
t f 
1 s  1 Hz = 1!
40
20
0
230
240
250
Frequency (Hz, arb. offset)
G. R. Harp
Hey! Don’t use a degenerate
example!
Measured Frequency Power Spectrum
Power (10^-26 Watts)
100
80
60
40
20
0
230
240
250
Frequency (Hz, arb. offset)
• Measurement
time = 1s
• Signal f = 2Hz
• 2 Hz in 1 Hz
resolution ->
Signal has short
coherence time,
t = 0.5 s.
t f  1!
You can’t beat
Nyquist!
G. R. Harp
Natural Law of the Universe
Well known but never identified.
Measured Voltage
• Natural signals can be broad or peaky
• But they always look like noise.
6
5
Peaky
Broad
4
5
2.5
0
-2.5
3
-5
1
Measured Voltage
2
125
135
Time (ns)
145
125
135
Time (ns)
145
5
2.5
0
0
2
4
6
8
10
Measure the time
variation of small
bandwidth here.
0
-2.5
-5
G. R. Harp
An Unfortunate Natural Law of
the Universe
• Natural signals always look like noise.
• This is a natural outcome of the central
limit theorem (Rice).
• Simply put: If there are N independent
sources contributing to your measurement,
then the signal converges to Gaussian
noise as N  
• It doesn’t matter what the spectrum of a
single source is!
G. R. Harp
An Unfortunate Natural Law of
the Universe
• Natural signals always look like noise.
• Counter example – pulsars.
Vela Pulsar
• Why? Because
N = 1. Central
limit theorem
does not apply.
G. R. Harp
An Unfortunate Natural Law
of the Universe
Simulaton
NASA
Counter example – blazars. Why? Because
N ~1.
G. R. Harp
How do we know signal is
artificial?
• Uncertainy principle argument?
• Shannon’s theorem – Allows us to quantify the
amount of information signals contain.
– A signal that transmits maximal informaion on channel
always looks like Gaussian White Noise.
– Just like natural sources!
– To recognized signal as artificial it must be redundant.
– Corrolary: Signals containing no redundant information
are impossible to decode (w/o key).
– Signals must contain redundancy or we have no hope
of understanding the message.
G. R. Harp
Required properties for ID of
ETI Signal
• Must be noticeably artificial
– Must contain less than the maximum information
content that could be conveyed Redundant!
• Must be persistent
– Not their problem, but ours. We can’t be sure of
direction of signal arrival unless we can make many,
many measurements.
G. R. Harp
2 Correlators + MIRIAD
ATA - 42 Dishes
100 kHz to 3 kHz resolution, 100
1’ pointing (at night, 2.5’ day)
MHz BW
Wide FOV = 53’ @ 4 GHz
2 (3) Beamformers
Real time, time series voltages, 70
0.7mm RMS Surfaces (at night,
MHz BW
~3 mm day)
Points synthetic beam anywhere,
0.5 – 11.2 GHz
supports nulls
4 x 100 MHz bands
SETI spectrometer
Typically 0.7 Hz resolution, .25 – 1
Hz / second drift rates
• Conventional SETI strategies:
– Focus attention on a small region of
sky, look for ~1Hz signals
– Extraordinary efficiency
• Unconventional SETI strategies
– a “sky survey” covering large areas
of the sky with modest sensitivity
– a “targeted search” looking for
signals with large autocorrelation
(time slip spectroscopy).
Shostak
G. R. Harp
ATA is a Unique Telescope
New Search Spaces
• ATA has multiple beamformers, nulling
technology to discriminate RFI.
G. R. Harp
Multiple BF + Nulling
• Simultaneously measure signal at one
point (beam 1) while excluding signal from
that direction in beam 2, beam 3, etc.
• Anticoincidence rejects signals that appear
in both beams (must be RFI)
FOV
FOV
x Null
Beam
Beam
x Null
G. R. Harp
Ongoing Searches at ATA
Shostak
Kassim, et al.
Galactic Center Survey: Sky survey of 20
square degrees near the galactic center.
Gal. Center = 26000 LY. Survey detects
transmitters with 20,000 x EIRP
(transmitted power) of Arecibo
planetary radar at the galactic center.
Give or take 10 dB.
G. R. Harp
Ongoing Searches at ATA
Cygnus X3 Region
Pike and Drake, 1964
Shostak
Galactic Center Survey: Sky survey of 20
square degrees near the galactic center.
Cygnus X-3 Survey: Sky survey of 4 sq.
deg. near Cygnus. Includes the x-ray binary
star Cygnus X-3.
Distance to Cyg X3 is 1.5x distance
to GC.
Detects transmitters with 50% of the
sensitivity of galactic survey (GC).
G. R. Harp
Ongoing Searches at ATA
Galactic Center Survey: Sky survey of 20
square degrees near the galactic center.
Cygnus X-3 Survey: Sky survey of 4 sq.
deg. near Cygnus. Includes the x-ray binary
star Cygnus X-3.
Exoplanet Search: Targeted survey, have
observed 146 stars with planets in the
Waterhole band, ongoing.
Most are between 10-1000 LY.
1000 LY ~ 700x sensitivty of GC.
10 LY = 104x more improvement.
Kalas, et al.
Shostak
G. R. Harp
Ongoing Searches at ATA
Habitable Zone, where astrophysics
might permit life.
HabCat Catalog: Turnbull and Tarter
Shostak
Galactic Center Survey: Sky survey of 20
square degrees near the galactic center.
Cygnus X-3 Survey: Sky survey of 4 sq.
deg. near Cygnus. Includes the x-ray binary
star Cygnus X-3.
Exoplanet Search: Targeted survey, have
observed 146 stars with planets in the
Waterhole band, ongoing.
HabCat Search: Targeted survey, Turnbull
and Tarter compiled list of stars that could be
suitable hosts for habitable planets.
17,000 stars, 4-55000 LY
4 LY – 20,000x better than GC search
55000 LY = 25% sensitvity of GC
G. R. Harp
Ongoing Searches at ATA
30 Polar
Cap Sources
always up
Galactic Center Survey: Sky survey of 20
square degrees near the galactic center.
Cygnus X-3 Survey: Sky survey of 4 sq.
deg. near Cygnus. Includes the x-ray binary
star Cygnus X-3.
Exoplanet Search: Targeted survey, have
observed 146 stars with planets in the
Waterhole band, ongoing.
HabCat Search: Targeted survey, Turnbull
and Tarter compiled list of stars that could be
suitable hosts for habitable planets.
PiHI Search 1: Targeted survey, 100 HabCat
stars near magic frequency of 4.462 GHz
( times the HI line frequency of 1.421 GHz).
Almost virgin frequency territory. After Sagan.
Choose ~100 Habcat stars within 200 LY of
earth and in polar cap.
100 LY = 70,000 x more sensitivity than GC
This is a little unfair – GC measures millions
of stars at once, PiHI measures 1.
Harp et al.
Shostak
G. R. Harp
PiHI Targeted Search
•
•
•
•
•
•
Last summer, Bethany Wilcox – dynamite!
No SI surveys have ever gone this high.
“Contact” Frequency =  HI = 4462 MHz
94 target stars chosen from HabCat.
F9-G7 (sun-like) stars within ~62 pc.
Stage 1: 30 MHz BW, 0.7 Hz resolution, 200 sec.
observations, ½ Hz / sec drift rate
• 30 MHz BW is large enough to encompass any
proper motion (statistical analysis of catalog)
• Stage 2: “Pi Hole” ( times waterhole) 2010
G. R. Harp
PiHI Targeted Search
• All targets were declination > 60° so that
observations could be done at any time, day
or night.
• Easy to schedule.
• Daytime obs have
sun in sidelobes.
• Choice of spherical
cap reduced solar
interference.
G. R. Harp
PiHI Targeted Results
(^_^);
(We have a very effective mechanism for identifying RFI)
G. R. Harp
PiHI Targeted Results
• Every 200 second observation, ~2000 signals would
pass the power threshold.
– Of those signals, ~800 would be immediately
dismissed as recent RFI or zero-drift.
– Almost all of the remainder failed the null
beam test.
– ~1 would be classified as 'confirm‘
– Prelude would re-observe to see if the signal
persisted. If so, proceed to off test.
• Ultimately 64 signals got to “first off” test. None of
these survived.
G. R. Harp
One that Got Away
Large Pulse Observed
Power (arb.)
.
350
300
250
Exactly the right place.
Requirement of
persistence is a
pain in the neck.
200
150
100
50
0
4463
Not observed in
follow-up
observations.
Dang It!
4463.5
4464
4464.5
Futher study
showed this
signal appeared
in both beams.
4465 (;_;)
Frequency (MHz)
G. R. Harp
Nothing for a dynamic young
scientist to do.
• Conventional SETI observing system is
~100% automated.
• Push into a different search space.
G. R. Harp
What new directions can we take?
To length scales describe interferometer.
Dish Size ~ 1 / FOV
Aperture Size ~ 1 / Resolution
42 ATA Dishes <> One ~40 m Single Dish
Aperture Size = Dish Size
Resolution = FOV
G. R. Harp
Differences between Interferometer and
Filled Aperture Telescopes
Radio Image of
Moon (ALMA)
Equivalent
Single-Dish
FOV and
Resolution
Element
Resolution
Element
Point Source
Transmitter
ATA FOV
At 7 GHz
Single pixel of ATA gets data only
from black dot
ATA can get signal from anywhere in
blue circle
For single dish, get one data point
All the signal in the green circle
Transmitter signal corrupted by
100 x more noise from the Moon.
ATA better distinguishes point source emitters from background.
ATA can image many 1000’s of points at once.
G. R. Harp
ATA is a Unique Telescope
New Search Spaces
• ATA has multiple beamformers, nulling
technology to discriminate RFI.
• Interferometer can see multiple points
on sky at once.
– Opens the door to
signals with Optical
Angular Momentum
– Nothing to do with
polarization (spin)
of signal!
G. R. Harp
Optical Angular Momentum
and its generalizations
June 2005
Two Signals,
images to send.
Imaged with ATA350
interferometer
Same data with 10 ħ
removed.
+=
ET applies
10 ħ OAM
before
summing.
Extra info
looks like
structure.
Repeats.
G. R. Harp
Problem with OAM
Point Source
OAM = 0 ħ
Point Source
OAM = 1 ħ
 = |0,0>
 = |1,1> + |1,-1>
• First, OAM is not so special, just one of  number of
bases for expressing E field.
• In order to see OAM (or …), must be able to resolve
dark from light on RHS.
• For source at 100 pc, need multiple transmitters spaced
at 0.5 pc, received in perfect phase at receiver.
• Because of ISM, this is impossible.
G. R. Harp
18-May-2008
How could
I have been
so naїve?
G. R. Harp
ATA is a Unique Telescope
New Search Spaces
• ATA has multiple beamformers, nulling
technology to discriminate RFI.
• Interferometer can see multiple points
on sky at once.
– Use correlator in conventional mode.
– Look for narrowband point sources in
images that might be beacons
– Cover large area of sky at once
G. R. Harp
Ongoing Searches at ATA
Harp, Wilcox, et al.
4 detection of
100 kHz signal
at 4455.3 MHz
Shostak
Galactic Center Survey: Sky survey of 20
square degrees near the galactic center.
Cygnus X-3 Survey: Sky survey of 4 sq.
deg. near Cygnus. Includes the x-ray binary
star Cygnus X-3.
Exoplanet Search: Targeted survey, have
observed 146 stars with planets in the
Waterhole band, ongoing.
HabCat Search: Targeted survey, Turnbull
and Tarter compiled list of stars that could be
suitable hosts for habitable planets.
PiHI Search 1: Targeted survey, 100 HabCat
stars near magic frequency of 4.462 MHz (
times the HI line frequency of 1.421 GHz).
Almost virgin frequency territory. After Sagan.
PiHI Search 2: Sky survey using ATA
correlator. Novel scheme runs commensally
with targeted survey.
G. R. Harp
PiHI Sky Survey
• “Contact” Frequency =  HI = 4462 MHz
• 94 regions chosen centered on HabCats.
• Stage 1: 100 MHz BW, 100 or 3 kHz
resolution, 2 minute observations
• All targets were declination > 60° so that
observations could be done at any time,
day or night.
G. R. Harp
Frequency Space
Double Difference Method
• Each correlator dump has ~824 images arranged by
frequency (edge images are corrupted by aliasing)
• Each contains 100 kHz (or 3 kHz) of image data.
• From these we calculate 824 numerical second derivative
images
• – (image1) + 2*(image2) – (image3) = Double Diff
–
+ 2*
–
=
–
+ 2*
–
=
G. R. Harp
Frequency Space
Double Difference Method
• Typical candidate signal
• Double difference causes single feature to show up
in 3 adjacent channels
4555.1 MHz
4555.2 MHz
4555.3 MHz
4555.4 MHz
4555.5 MHz
Second derivative feature at 4
• Noise level determined empirically from DD image
background (resistant to RFI)
G. R. Harp
Frequency Search Results
Double Difference Method
• 14 targets had peak values with an SNR greater
than 4.
• One target had a σ > 4 detection in both the 3 MHz
and 100 MHz band.
• These 14 were re-observed with correlator and
with conventional SETI instrument
• None of them were persistent.
• [Dang that persistence clause!]
G. R. Harp
ATA is a Unique Telescope
New Search Spaces
• ATA has multiple beamformers, nulling
technology to discriminate RFI.
• Interferometer can see multiple points
on sky at once.
• ATA has various fast dump modes. Can
look for fast time-variable signals.
G. R. Harp
Ongoing Searches at ATA
Harp,
Ackermann,
Arbunich
Shostak
Galactic Center Survey: Sky survey of 20
square degrees near the galactic center.
Cygnus X-3 Survey: Sky survey of 4 sq.
deg. near Cygnus. Includes the x-ray binary
star Cygnus X-3.
Exoplanet Search: Targeted survey, have
observed 146 stars with planets in the
Waterhole band, ongoing.
HabCat Search: Targeted survey, Turnbull
and Tarter compiled list of stars that could be
suitable hosts for habitable planets.
PiHI Search 1: Targeted survey, 100 HabCat
stars near magic frequency of 4.462 MHz (
times the HI line frequency of 1.421 GHz).
Almost virgin frequency territory. After Sagan.
PiHI Search 2: Novel Sky survey using ATA
correlator. Novel scheme runs commensally
with targeted survey.
Time slip spectroscopy: Targeted survey,
prototype. Look for signals with narrow timedomain features. Esp. Autocorrelation
spectra.
G. R. Harp
I was pretty proud for thinking of
examining autocorrelation spectra
• But then: I was scooped again!
Dang!
G. R. Harp
Maximizing Information in Time
Slip Spectra
• Transmitter sends arbitrary signal with arbitrary
bandwidth, contains mucho information, e.g.
Encyclopedia Galactica
• After a short delay (s – 10s) send second copy
(can be overlaid on first signal)
• Very simple, very efficient algorithm discovers
signal, same technology as conventional search
• Equal or near-equal sensitivity as conventional
search
• Better yet, there is no ambiguity about where the
information lies.
G. R. Harp
Time Slip Spectroscopy
We start with bright objects, masers, blazars.
Civilization pumps a
maser with time
dependent signal
Acts as “natural”
amplifier for
signal.
Idea of using a
maser: J.M.
Weisberg, et al.
G. R. Harp
Time Slip Spectroscopy
Assume time series contains information, sent
multiple times. Could be simply repeating signal.
Here we show a more interesting case.
.
4
Signal
Voltage (arb.)
3
Delayed Signal
2
1
0
-1
-2
Send signal once.
Send signal again.
-3
30
46
62
Receiver
sees a superposition
of the two
signals.
Time (arb.)
G. R. Harp
Time Slip Spectroscopy
To IDENTIFY the signal, simply compute
autocorrelation spectrum (trivial with FFT).
.
3500
Zero lag autocorrelation = Total Power
Power (arb.)
3000
2500
Identify delayed signal using threshold
2000
1500
1000
500
0
-500
0
16
32
48
64
80
96
112
128
Time Delayrepetition
(arb.)
Once identified, deconvolve
to get
good measure of information-containing signal.
G. R. Harp
Time Slip Spectroscopy
What about ISM? Doesn’t that mess up wide
bandwitdh signals? (Dispersion, incoherent
scattering.)
YES!
But both the original signal and time delayed
signal are affected identically by ISM (provided
time delay is not too long (e.g. hours).
Therefore they will still correlate.
To decode signal will require more work, but at
least we know where to start!
G. R. Harp
Time Slip Spectroscopy
Proof of principle using ATA + Beamformer + Time capture.
Power (arb.)
Autocorrelation spectrum of
phase modulated GPS signal.
Downward trend of peaks
is expected. Data are not
normalized to # of samples.
0
4883
9767 14648
19531 24414
Time Delay (s)
29296
G. R. Harp
Time Slip Spectroscopy
The real thing, W3OH maser.
Autocorrelation shows
nothing special. We have ~3
such measurements so far.
Future: Longer delays, faster.
G. R. Harp
Time Slip Spectroscopy
Implementation uses same components as current SETI detector.
Channelizer
(forward)
Channelizer
(forward)
Square
(Power)
Channelizer
(inverse)
• Deep FFT
• Square Result
• Inverse Deep FFT
• Feed results to threshold
detector
• Exactly as fast as
conventional SETI
• ~2x as compute power.
Channelizer
(inverse)
Same ol’ SETI
detector as usual
G. R. Harp
Who is the ATA?
SETI Institute
Ackermann, Rob
Backus, Peter
Barott, Billy
Bradford, Tucker
Davis, Mike
DeBoer, Dave
Dreher, John
Harp, Gerry
Jordan, Jane
Kilsdonk, Tom
Pierson, Tom
Randall, Karen
Ross, John
Tarter, Jill
UCB and HCRO
Milgrome, Oren
Backer, Don
Thornton, Doug
Blitz, Leo
Urry, Lynn
Bock, Douglas
Van Leuven, Jori
Bower, Geoff
Vanourtryve,
Cheng, Calvin
Cassandra
Croft, Steve
Welch, Jack
Dexter, Matt
Werthimer, Dan
Engargiola, Greg
Williams, Peter
Fields, Ed
Wright, Mel
Forster, Rick
Gutierrez-Kraybill,
Colby
Heiles, Carl
Helfer, Tam
Jorgensen, Susie
Kaufman, Jeff
Keating, Garrett (Karto)
MacMahon, Dave
Minex
Cork, Chris
Fleming, Matt
Vitouchkine, Artyom
Student Interns
Imara, Nia
Chubb, Kelsey
Adair, Aaron
Nadler, Zachary
Pearson, Ruth
G. R. Harp
End
G. R. Harp
Generalization of OAM:
Dual-beam Cross Correlation.
• OAM is only one basis. Infinite number of bases
(representations) of light.
• Another one is to take two beams and correlate them
coherently (astronomical data should never correlate)
• Can be done (RIGHT NOW!) with ordinary correlator.
• Dang! I ought to publish this!
• BUT!
• All such techniques rely on ability to resolve multiple
transmitters.
• Transmitters must be light years apart and perfectly in
phase at the receiver.
• This is literally impossible because of ISM.
• We can expect never to receive phase-coherent multiangle signals unless we have super-long baseline
interferometry and nearby transmitters.
G. R. Harp
How do we know signal is
artificial?
• Uncertainy principle argument?
• Shannon’s theorem -- the amount of information
they contain.
– Define “channel” as bandwidth/time for signal.
– A signal that transmits maximal informaion on channel
always looks like Gaussian White Noise.
– Corrolary: Such signals contain no redundant
information  they are impossible to decode w/o key
– Signals must contain redundancy or we have no hope
of understanding the message.
G. R. Harp
How much information is
contained in a signal?
Shannon’s Theorem
Any signal that is easy to find
doesn’t contain much information.
Information carrying capacity
of signal or “channel”, C
S

C  B log 2  1  
N

S = Average signal power in one sample
N = Average noise power in one sample
If B = 0, then there is no information (excepting…)
G. R. Harp
Senario 0
• ET uses all their power budget to send a
sine wave “beacon” to get our attention.
Call a stranger on the phone and then not
say anything. It’s a prank call!
G. R. Harp
Senario 1
• ET puts all their power into 2 discontinuous
sine waves while alternating polarization.
LHCP
RHCP
Signals are shown in sync, but ISM will cause
time difference of arrival (e.g. delay the RCHP)
• Detection signal to noise ratio reduced by
1
BWeff
2
– BWeff ~ inverse of bit rate.
– Even small bit rate gives large degradation in SNR.
– To discriminate against natural sources, bit rate cannot
exceed 200 bits per second. (Tarter, et al.)
– Very little info may be exchanged.
G. R. Harp
Detection Algorithm For Sine Waves
can be discontinous
Noise Picked Up Along Way
Electric Field 
ET Sends Signal
+
Time 
Composite Picked Up At Receiver
=
= S(t)
Signal
| Fourier Transform [S(t)]
|2

Noise
0
50
G. R. Harp
Senarios 0 or 1
• Advantages:
– Super Easy!
– FFT implementation of Fourier Transform is
fastest algorithm known to man.
• Very small information channel
• Problem: How do humans find the
information-containing signal?
– No way to point us toward information
G. R. Harp
Senario 2
• ET puts all their power into a single signal
containing 2 copies of arbitrary information
– P = PAvg / 2
• Use other half to send us a modulated signal
with simple encoding.
– P = PAvg / 2
• Problem: How do humans find the informationcontaining signal?
– No way to point us toward information
– Information content must be small or we won’t find it
G. R. Harp
Electric Field 
Example: Sine Wave
Time 
• Once you know the frequency, you know
everything, can predict future values with perfect
accuracy.
• Information content = 0 (or one very important
bit)
• Payoff: Easily distinguished from random nature.
G. R. Harp