The detector and the atmosphere are two important sources
of noise for interferometric observations in the infrared.
Noise from the former source is present
in all observations and intensity variations due to changes in
atmospheric transparency occur on many nights.
Fluctuations of the refractive index of the atmosphere along the light path
between the source and the detector, the so-called scintillation noise,
can dominate in bright objects and in the thermal regime.
While the ubiquitous photon noise is characterized by its standard
deviation being proportional to the square root of the total intensity,
scintillation noise has the interesting property of being linearly
proportional to the source intensity. This is the reason it becomes
increasingly important for bright stars and in wavelength regions where
the sky background becomes very large. It is generally not a factor
in traditional grating spectroscopy because each detection element
`sees'
radiation originating from only a narrow wavelength region, and as a
result the intensities at the detector are quite low and photon noise
continues to dominate. However, scintillation noise is
an important consideration in interferometry where the detector is
always
measuring radiation multiplexed from a wide range of wavelengths. In
fact,
scintillation can introduce a multiplex disadvantage when observing
bright
stars.
Detector and scintillation noise both have a 
frequency dependence.
This suggests that they might both be suppressed by modulating the
signal so that it is sampled at high
frequencies. In infrared photometry this is usually accomplished by use
of a
chopping secondary mirror, which is an example of amplitude modulation.
Although chopping can also be used for interferometric observations, the same result can be achieved more efficiently
with high frequency, low amplitude modulation of the path difference, a
technique referred to as phase or internal modulation.
The CFHT FTS uses internal modulation and as a consequence the nature of the recorded interferogram is changed. In the case of a monochromatic source, the measured
signal now represents the change of the signal given by 1.2 occurring over a small increment of retardation. As a result, the constant
component of the interferogram, 
, and low frequency
variations in this term are not modulated and can be totally eliminated
providing the frequency of the modulation is chosen to be high enough.
For the CFHT FTS, modulation frequencies of 20, 150, and 300 Hz are available at the turn of a switch (see Error Signal/Servo Control Panel).
Generalizing to the broadband case,
if the path difference is modulated with an amplitude 
about
some mean path difference 
, the signal recorded by a synchronous
detector will be (from 1.8)
Trigonometric identities can be used to reduce this to the form
It is apparent from this expression that the spectral distribution can
be recovered when modulation is applied, although a sine transform must
be
used. Moreover, the result should be divided by 
(
).
Besides suppressing noise, modulation also makes the
location of ZPD easy to identify because
it now occurs as a zero crossing, instead of a
maximum in signal intensity, which can be more difficult to locate
accurately.
The optimum amplitude for modulation occurs when
(
) 
() is maximized. This
occurs when
where 
is the wavenumber at the middle of the spectral region being
studied and 
is the optimum amplitude of the modulation. This
condition is satisfied when
where 
is an integer. In other words, modulation is most efficient when
its
amplitude is an odd multiple of the central wavelength of the region
being
observed divided by 4.