Fundamentals of Radio Interferometry

Rick Perley, NRAO/Socorro

ATNF Radio Astronomy School

Narrabri, NSW
29 Sept. – 03 Oct. 2014
Topics
• Introduction: Sensors, Antennas, Brightness, Power
• Quasi-Monochromatic Approximation
• The Basic Interferometer
– Response to a Point Source
– Response to an Extended Source
– The Complex Correlator
– The Visibility and its relation to the Intensity
– Picturing the Visibility
Essentials of Sensors (Antennas)
• Coherent interferometry is based on the ability to correlate the
electric fields at spatially separated locations.
• Doing this requires transport of the electric field E(s,n,t), or a
surrogate, at various locations r to a central location for analysis.
• In the radio regime, the normal practice is to convert the E-field to a
voltage V(n,t) which can be conveyed to a central location for
processing. Note that information on direction s is lost.
• The ideal sensor is a device which responds to the electric field at
some place and converts this to a voltage which faithfully retains the
amplitudes and phases of the electric fields.

EM waves in A real sensor will modify the

amplitudes and phases, but in a
simple, slowly varying way.

Voltage out (preserving relative amplitude

and phases of all input fields)

2014 Narrabri Radio Astronomy School 3

Brightness and Power.
• Imagine a distant source of emission,
described by brightness I(n,s) where
s is a unit direction vector. dW Solid Angle
• Power from this emission is s
intercepted by a collector (`sensor’)
of area A(n,s).
• The power, dP (watts) from a small
solid angle dW, within a small Sensor Area
frequency window dn, is A
dP  I( , s)A( , s) dν dΩ
• The total power received is an
dn Filter width
integral over frequency and angle,
accounting for variations in the
responses. P Power collected
P   I ( , s) A( , s)d dW

2014 Narrabri Radio Astronomy School 4

Quasi-Monochromatic Radiation
• Analysis is simplest if the fields are monochromatic.
• A perfectly monochromatic electric field (dn = 0), cannot exist in
nature – it would both no power and would last forever.
• So we consider instead ‘quasi-monochromatic’ radiation, where
the bandwidth dn is very small, but not zero.
• Then, for a time dt ~1/dn, the electric fields will be sinusoidal, with
unchanging amplitude and phase.
• Consider then the electric field from a small solid angle dW about
some direction s, within some small bandwidth dn, at frequency n.
• We can write the temporal dependence of this field as:

• The amplitude and phase remains unchanged to a time duration

of order dt ~1/dn, after which new values of A and f are needed.

2014 Narrabri Radio Astronomy School 5

Simplifying Assumptions
• We now consider the most basic interferometer, and seek a
relation between the characteristics of the product of the voltages
from two separated antennas and the distribution of the brightness
of the originating source emission.
• To establish the basic relations, the following simplifications are
introduced:
– Fixed in space – no rotation or motion
– Quasi-monochromatic (signals are sinusoidal)
– No frequency conversions (an ‘RF interferometer’)
– Single polarization
– No propagation distortions (no ionosphere, atmosphere …)
– Idealized electronics (perfectly linear, no amplitude or phase
perturbations, perfectly identical for both elements, no added
noise, …)
 Basic Concepts of Interferometry
s s

Sensor 2 b Sensor 1

• There are two sensors, separated by vector baseline b

• Radiation arrives from direction s – assumed the same for
both (far-field).
• The extra propagation path is L = b.s
• The time taken for this extra path is
• For radiation of wavelength l, we have a phase given by:
(radians)

2014 Narrabri Radio Astronomy School 7

 The Stationary, Quasi-Monochromatic
Radio-Frequency Interferometer

Geometric
s s
Time Delay

b
The path lengths
from sensors
X to multiplier are
assumed equal!
multiply

average
Unchanging Rapidly varying,
with zero mean

Note: Rc is not a function of time or location!

2014 Narrabri Radio Astronomy School 8

 Pictorial Example: Signals In Phase
If the voltages arrive in phase:
b.s = nl, or tg  n/n (n is an integer)

•Antenna 1
Voltage

•Antenna 2
Voltage

•Product
Voltage

•Average

2014 Narrabri Radio Astronomy School 9

 Pictorial Example: Signals in Quad Phase
If the voltages arrive in quadrature phase:
b.s=(n +/- ¼)l, tg = (4n +/- 1)/4n

•Antenna 1
Voltage

•Antenna 2
Voltage

•Product
Voltage

•Average

2014 Narrabri Radio Astronomy School 10

 Pictorial Example: Signals out of Phase
If the signals arrive with voltages out of phase:
b.s=(n + ½)l tg = (2n + 1)/2n

•Antenna 1
Voltage

•Antenna 2
Voltage

•Product
Voltage

•Average

2014 Narrabri Radio Astronomy School 11

 Some General Comments
• In all cases, the output is a steady voltage, with the amplitude
dependent upon the signal strength, and the phase relationship.

• The averaged product RC is dependent on the received power, P =

A2/2 and geometric delay, tg, and hence on the baseline orientation
and source direction:

• Note that RC is not a a function of:

– The time of the observation -- provided the source itself is not variable.
– The location of the baseline -- provided the emission is in the far-field.
– The actual phase of the incoming signal – the distance of the source
does not matter, provided it is in the far-field.

• The strength of the product is dependent on the antenna collecting

areas and electronic gains – but these factors can be calibrated for.

2014 Narrabri Radio Astronomy School 12

Pictorial Illustrations
• To illustrate the response, expand the dot product in one dimension:

– Where u = b/l is the baseline length in wavelengths,

– a is the angle w.r.t. the baseline vector
– l  cos a  sin q is the direction cosine
s

q
a
b

• Consider the response Rc, as a function of angle, for two different

baselines with u = 10, and u = 25 wavelengths:

2014 Narrabri Radio Astronomy School 13

Whole-Sky Response for u = 10
• When u = 10 (i.e., the baseline is 10 wavelengths long), the response is
RC  cos(20 l )
• There are 21 fringe maxima over the whole hemisphere, with maxima at l =
n/10 radians.
• Minimum fringe separation 1/10 radians
-10 -8 -5 -3 0 2 5 7 9 10
Whole-Sky Response for u = 25
For u = 25 (i.e., a 25-wavelength baseline), the response is
RC  cos(50 l )
• There are 51 whole fringes over the hemisphere.
• Minimum fringe separation 1/25 radians.

-25 25
From an Angular Perspective q
0
3
Top Panel: 5

The absolute value of the 7

response for u = 10, as a +

function of angle. -
+ 9
The ‘lobes’ of the response
-
pattern alternate in sign.
10
+
Bottom Panel:
The same, but for u = 25.
Angular separation between
lobes (of the same sign) is
dq ~ 1/u = l/b radians.
Hemispheric Pattern
• The preceding plot is a meridional cut
through the hemisphere, oriented along
the baseline vector.
• In the two-dimensional space, the fringe
pattern consists of a series of coaxial
cones, oriented along the baseline vector.
• The figure is a two-dimensional
representation when u = 4.
• As viewed along the baseline vector, the
fringes show a ‘bulls-eye’ pattern –
concentric circles.

2014 Narrabri Radio Astronomy School 17

The Effect of the Sensor

• The patterns shown presume the sensor (antenna) has

isotropic response.
• This is a convenient assumption, but doesn’t represent reality.
• Real sensors impose their own patterns, which modulate the
amplitude and phase, of the output.
• Large antennas have very high directivity -- very useful for
some applications.
• Small antennas have low directivity – nearly uniform response
for large angles – useful for other applications.
The Effect of Sensor Patterns
• Sensors (or antennas)
are not isotropic, and
have their own
responses.
• Top Panel: The
interferometer pattern
with a cos(q)-like
sensor response.
• Bottom Panel: A
multiple-wavelength
aperture antenna has a
narrow beam, but also
sidelobes.
• Note that the phase will
also be modified.
The Response from an Extended Source
• The response from an extended source is obtained by summing the
responses at each antenna to all the emission over the sky, multiplying
the two, and averaging:

• The averaging and integrals can be interchanged and, providing the

emission is spatially incoherent, we get

• This expression links what we want – the source brightness on the

sky, In(s), – to something we can measure - RC, the interferometer
response.
• Can we recover In(s) from observations of RC?
• NB I have assumed here isotropic sensors. If not, a directional
attenuation function must be added.

2014 Narrabri Radio Astronomy School 20

A Schematic Illustration in 2-D
• The correlator can be thought of multiplying the actual sky brightness by a
cosinusoidal coherence pattern, of angular scale ~l/b radians.

• The correlator then integrates (adds) the modified brightness pattern over
the whole sky (as weighted by the antenna response.
• Pattern orientation set by baseline
geometry.
• Fringe separation set by
(projected) baseline length and
wavelength. l/b rad.
• Long baseline gives close-packed
fringes Source
• Short baseline gives widely- brightness
separated fringes
• Physical location of baseline
unimportant, provided source is in
the far field.
- + - + - + -
Fringe Sign
 A Short Mathematics Digression –
Odd and Even Functions
• Any real function, I(x,y), can be expressed as the sum of two real
functions which have specific symmetries:

An even part:

An odd part:

I IEIE IIO
O
I
= +

2014 Narrabri Radio Astronomy School 22

Why One Correlator is Not Enough
• The correlator response, Rc:

is not enough to recover the correct brightness. Why?

• Express the brightness as the sum of its even and odd parts:
I  I E  IO
• Then form the correlation:

• Since the cosine fringe pattern is even, the response of our

interferometer to the odd brightness distribution is 0.
• The Odd symmetric component, IO is invisible, and is lost.
• Hence, we need more information if we are to completely
recover the source brightness.

2014 Narrabri Radio Astronomy School 23

 Why Two Correlations are Needed

• To recover the ‘odd’ part of the brightness, IO, we need an ‘odd’

fringe pattern. Let us replace the ‘cos’ with ‘sin’ in the integral, to
find:

since the integral of an even times an odd function is zero.

• Thus, to provide full information on both the even and odd parts
of the brightness, we require two separate correlators.
– An ‘even’ (COS) and an ‘odd’ (SIN) correlator.
• Note that this requirement is a consequence of our assumption
of no motion – the fringe pattern and the source intensity are
both fixed.
• One can build a correlator which ‘sweeps’ its fringes across the
sources – providing both fringe types.
Making a SIN Correlator
• We generate the ‘sine’ pattern by inserting a 90 degree phase shift
in one of the signal paths.
s s

b
A Sensor

X 90o
multiply

average
Define the Complex Visibility
• We now DEFINE a complex function, the complex visibility, V, from the
two independent (real) correlator outputs RC and RS:

where

• This gives us a beautiful and useful relationship between the source

brightness, and the response of an interferometer:

• This is a Fourier transform – but with a quirk: The visibility distribution is

in general 3-dimensional, while the brightness distribution is only 2-
dimensional. More on this, later.
 The Complex Correlator and Complex
Notation
• A correlator which produces both ‘Real’ and ‘Imaginary’ parts – or the
Cosine and Sine fringes, is called a ‘Complex Correlator’
– For a complex correlator, think of two independent sets of projected
sinusoids, 90 degrees apart on the sky.
– In our scenario, both components are necessary, because we have assumed
there is no motion – the ‘fringes’ are fixed on the source emission, which is
itself stationary.
• The complex output of the complex correlator also means we can use
complex analysis throughout: Let:

V  A cos(t )  Re Ae
1
- it

V  A cos[ (t - b s / c)]  Re Ae
2

-i ( t -b s / c )

• Then:
P  VV  A e
corr 1 2
* 2 - i bs / c
Wideband Phase Shifters – Hilbert Transform

• For a quasi-monochromatic signal, forming a the 90 degree

phase shift to the signal path is easy --- add a piece of cable
l/4 wavelengths long.
• For a wideband system, this obviously won’t work.
• In general, a wideband device which phase shifts each spectral
component by 90 degrees, while leaving the amplitude intact,
is a Hilbert Transform.
• For real interferometers, such an operation can be performed
by analog devices.
• Far more commonly, this is done using digital techniques.
• The complex function formed by a real function and its
Hilbert transform is termed the ‘analytic signal’.

2014 Narrabri Radio Astronomy School 28

 Picturing the Visibility
• The source brightness is Gaussian, shown in black.
• The interferometer ‘fringes’ are in red.
• The visibility is the integral of the product – the net dark green area.
RC RS

Long Baseline
Long
Baseline

Short Baseline
Short
Baseline

2014 Narrabri Radio Astronomy School 29

Examples of 1-dimensional Visibilities.
• Picturing the visibility-brightness relation is simplest in one
dimension.
• For this, the relation becomes

• Simplest example: A unit-flux point source: I (l )  d (l - l0 )

• The visibility is then: -2iul0
V (u )  e
• For a source at the origin (l0=0), V(u) = 1. (units of Jy).
• For a source off the origin, the visibility has unit amplitude, and a
phase slope with baseline, rotating 360 degrees every 1/l0
wavelengths.

2014 Narrabri Radio Astronomy School 30

Point Source Visibility

Amplitude
2

1/l0 u

• For a point source, the visibility amplitude equals the source

flux density
• The phase slope (=2l0) gives the source position.
A Symmetric Double Point Source

• Mathematically, this is I (l )  d (l - l0 )  d (l  l0 )
• The Visibility is: V (u )  2 cos(2 ul0 )
which is a cosinusoid of amplitude = 2, reaching its maxima at
multiples of 1/l0 . The phase = 0.
• Note the symmetry: The brightness is real and even, so the
visibility is real and even.
Image Brightness Visibility

Angle Offset Baseline

2014 Narrabri Radio Astronomy School 32

 Extended Symmetric Source
• A point source has a constant visibility amplitude for all
baselines. An extended source’s visibility declines with
baseline.
• Consider a ‘top-hat’ source of width l0. Insertion into the
relation shows: sin( ul0 )
V (u )   sinc(ul0 )
 ulo
• A ‘Triangle’ source of full width 2l0 has a visibility function
2
 sin( u l0 
V (u )     sinc 2 (u l0 )
  u l0 
• For both of these, the visibility equals 0 at all multiples of u
= l0.

2014 Narrabri Radio Astronomy School 33

Examples of 1-Dimensional Visibilities
• Simple pictures are easy to make illustrating 1-dimensional visibilities.
Brightness Distribution Visibility Function

• ‘Top-Hat’
Sources

• Triangle
Sources

For these examples, the visibility peaks are all the same (=1), reflecting
the integrated flux density of the sources is the same (= 1).

2014 Narrabri Radio Astronomy School 34

Extended Symmetric Doubles

• Suppose you have a source consisting of two ‘top-hat’

sources, each of width l0, separated by l1 radians.

l1
l0

• Analysis provides: V (u )  sinc(u l0 ) cos( ul1 )

Which is an oscillatory function of period u = 1/l1
attenuated by a dying oscillation of period u = 1/l0.

2014 Narrabri Radio Astronomy School 35

 More Examples
• Simple pictures illustrating 1-dimensional visibilities.
Brightness Distribution Visibility Function

• Resolved
Double

• Resolved
Double

• Central
Peaked
Double

2014 Narrabri Radio Astronomy School 36

Another Way to Conceptualize …
• Return to the generalized definition of the visibility:

• The interferometer casts a phase slope across the brightness

distribution.
– The phase slope becomes steeper for longer baselines, or higher
frequencies.
– The phase slope is zero for zero baseline. (V(0) = S)
– The phase is zero at the phase origin.
– The amplitude response is unity (ignoring the primary beam)
throughout.
• The Visibility is the complex integral (sum) of the brightness multiplied by
the phase ramp.

2014 Narrabri Radio Astronomy School 37

Basic Characteristics of the Visibility
• For a zero-spacing interferometer, we get the ‘single-dish’
(total-power) response.
• As the baseline gets longer, the visibility amplitude will in
general decline.
• When the visibility is close to zero, the source is said to be
‘resolved out’.
• Interchanging antennas in a baseline causes the phase to be
negated – the visibility of the ‘reversed baseline’ is the
complex conjugate of the original. (Why?)
• Mathematically, the visibility is Hermitian. (V(u) = V*(-u)).
Some Comments on Visibilities
• The Visibility is a unique function of the source brightness.
• The two functions are related through a Fourier
transform. V (u, v)  I (l , m)
• An interferometer, at any one time, makes one measure of
the visibility, at baseline coordinate (u,v).
• `Sufficient knowledge’ of the visibility function will provide
us a `reasonable estimate’ of the source brightness.
• How many is ‘sufficient’, and how good is ‘reasonable’?
• These simple questions do not have easy answers…