Help needed in extracting signal from noise

[a***@gmail.com]

I'm trying to form a gray-scale image of a scene by scanning it with a laser in daylight.

The setup is, I stare at the entire scene with a stationary lens and a single photodetector (and this does not scan, but takes in the entire scene), and then I scan a laser spot across the scene to build up a rastered image.
The light contribution from the entire background is 8000X brighter than the light from the laser spot.
I have heard that it is possible to extract faint signals from a noisy background by modulating the signal (the laser spot) and extracting the matching frequency components from the noise, which sounds crazy to me because, well, the noise is 8000X greater than the signal, and any detector is going to have a hard time seeing, say, 1000 photons on top of 8,000,000.

What am I missing? I'm not a signal-processing guy, if that wasn't obvious already.

Can modulating the laser spot enable me to build up a grayscale image of the scene, and if so, how, or should I take a different approach?

[Phil Hobbs]

Post by a***@gmail.com
I'm trying to form a gray-scale image of a scene by scanning it with a laser in daylight.
The setup is, I stare at the entire scene with a stationary lens and
a single photodetector (and this does not scan, but takes in the
entire scene), and then I scan a laser spot across the scene to build
up a rastered image. The light contribution from the entire
background is 8000X brighter than the light from the laser spot. I
have heard that it is possible to extract faint signals from a noisy
background by modulating the signal (the laser spot) and extracting
the matching frequency components from the noise, which sounds crazy
to me because, well, the noise is 8000X greater than the signal, and
any detector is going to have a hard time seeing, say, 1000 photons
on top of 8,000,000.
What am I missing? I'm not a signal-processing guy, if that wasn't obvious already.
Can modulating the laser spot enable me to build up a grayscale image
of the scene, and if so, how, or should I take a different approach?

That's a hard problem. Not only is the sun very bright, but the
resulting photocurrent has full shot noise. Executive summary: by
going sufficiently slowly you can probably do it, but you won't like the
frame times you'll have to use.

To get a vaguely decent image, you need at least 20 dB signal-to-noise
ratio in each pixel. That is, if the photocurrent due to the total
scene brightness is I_s, then your received photocurrent I_L from the
laser has to be at least

I_L >= 10*sqrt(2 e I_s B)

where e is the electron charge (1.602E-19 C), B is the measurement
bandwidth. B ~= 1/(2t), where t is the pixel rate. Thus the minimum
integration time per pixel is

t_min = 1/(2 Bmin) = 100 * 2 e I_s / (I_L ^ 2).

You can plug in your actual measured photocurrent values and see what
you're looking at.

It'll require modulating the laser at a few kilohertz and using lock-in
detection to suppress fluctuations such as wind-driven motion. On-off
modulation loses you half your integration time, so the scan will need
twice as long as calculated above.

Since you have to have a scanner anyway, why not descan the detector?
That way you're always looking at the illuminated spot with a narrow
field of view. You potentially win by a factor of

K ~= (lens FOV)**2
-------------
(beam divergence)**2

That gets to be a very big number pretty fast. The down side is that
you have to keep the system properly aligned, which introduces some
mechanical complications.

Cheers

Phil Hobbs

[a***@gmail.com]

Post by Phil Hobbs

Post by a***@gmail.com
I'm trying to form a gray-scale image of a scene by scanning it with
a laser in daylight.
The setup is, I stare at the entire scene with a stationary lens and
a single photodetector (and this does not scan, but takes in the
entire scene), and then I scan a laser spot across the scene to build
up a rastered image. The light contribution from the entire
background is 8000X brighter than the light from the laser spot. I
have heard that it is possible to extract faint signals from a noisy
background by modulating the signal (the laser spot) and extracting
the matching frequency components from the noise, which sounds crazy
to me because, well, the noise is 8000X greater than the signal, and
any detector is going to have a hard time seeing, say, 1000 photons
on top of 8,000,000.
What am I missing? I'm not a signal-processing guy, if that wasn't obvious already.
Can modulating the laser spot enable me to build up a grayscale image
of the scene, and if so, how, or should I take a different approach?

That's a hard problem. Not only is the sun very bright, but the
resulting photocurrent has full shot noise. Executive summary: by
going sufficiently slowly you can probably do it, but you won't like the
frame times you'll have to use.
To get a vaguely decent image, you need at least 20 dB signal-to-noise
ratio in each pixel. That is, if the photocurrent due to the total
scene brightness is I_s, then your received photocurrent I_L from the
laser has to be at least
I_L >= 10*sqrt(2 e I_s B)
where e is the electron charge (1.602E-19 C), B is the measurement
bandwidth. B ~= 1/(2t), where t is the pixel rate. Thus the minimum
integration time per pixel is
t_min = 1/(2 Bmin) = 100 * 2 e I_s / (I_L ^ 2).
You can plug in your actual measured photocurrent values and see what
you're looking at.
It'll require modulating the laser at a few kilohertz and using lock-in
detection to suppress fluctuations such as wind-driven motion. On-off
modulation loses you half your integration time, so the scan will need
twice as long as calculated above.
Since you have to have a scanner anyway, why not descan the detector?
That way you're always looking at the illuminated spot with a narrow
field of view. You potentially win by a factor of
K ~= (lens FOV)**2
-------------
(beam divergence)**2
That gets to be a very big number pretty fast. The down side is that
you have to keep the system properly aligned, which introduces some
mechanical complications.
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510
http://electrooptical.net
http://hobbs-eo.com

Well, this is absolutely brilliant and is exactly what I needed.

Many thanks, Phil. I appreciate your help.

[Mikko OH2HVJ]

Post by a***@gmail.com

Post by Phil Hobbs
That gets to be a very big number pretty fast. The down side is that
you have to keep the system properly aligned, which introduces some
mechanical complications.
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs

Well, this is absolutely brilliant and is exactly what I needed.
Many thanks, Phil. I appreciate your help.

When building something like that, you should absolutely have and read
a book called 'Building Electro-Optical Systems: Making It all Work'.

The author of the book is known to be very helpful and modest person.

--
mikko

[g***@gmail.com]

Post by a***@gmail.com
I'm trying to form a gray-scale image of a scene by scanning it with a laser in daylight.
The setup is, I stare at the entire scene with a stationary lens and a single photodetector (and this does not scan, but takes in the entire scene), and then I scan a laser spot across the scene to build up a rastered image.
The light contribution from the entire background is 8000X brighter than the light from the laser spot.
I have heard that it is possible to extract faint signals from a noisy background by modulating the signal (the laser spot) and extracting the matching frequency components from the noise, which sounds crazy to me because, well, the noise is 8000X greater than the signal, and any detector is going to have a hard time seeing, say, 1000 photons on top of 8,000,000.
What am I missing? I'm not a signal-processing guy, if that wasn't obvious already.
Can modulating the laser spot enable me to build up a grayscale image of the scene, and if so, how, or should I take a different approach?

A narrow band pass filter (Interference filter) centered on the laser
wavelength would help a lot. (maybe you already have one?)
Seems like there will alos be a distance effect in the reflected laser
light. Objects twice as far away will only scatter 1/4 the photons
into your detector.... maybe that's what you want?

George H.

[Phil Hobbs]

Post by g***@gmail.com

Post by a***@gmail.com
I'm trying to form a gray-scale image of a scene by scanning it with a laser in daylight.
The setup is, I stare at the entire scene with a stationary lens and a single photodetector (and this does not scan, but takes in the entire scene), and then I scan a laser spot across the scene to build up a rastered image.
The light contribution from the entire background is 8000X brighter than the light from the laser spot.
I have heard that it is possible to extract faint signals from a noisy background by modulating the signal (the laser spot) and extracting the matching frequency components from the noise, which sounds crazy to me because, well, the noise is 8000X greater than the signal, and any detector is going to have a hard time seeing, say, 1000 photons on top of 8,000,000.
What am I missing? I'm not a signal-processing guy, if that wasn't obvious already.
Can modulating the laser spot enable me to build up a grayscale image of the scene, and if so, how, or should I take a different approach?

A narrow band pass filter (Interference filter) centered on the laser
wavelength would help a lot. (maybe you already have one?)
Seems like there will alos be a distance effect in the reflected laser
light. Objects twice as far away will only scatter 1/4 the photons
into your detector.... maybe that's what you want?
George H.

If the filter is on a flat surface, angle tuning will limit the FOV, though.

Cheers

Phil Hobbs

[Helpful person]

Post by a***@gmail.com
I'm trying to form a gray-scale image of a scene by scanning it with a laser in daylight.
The setup is, I stare at the entire scene with a stationary lens and a single photodetector (and this does not scan, but takes in the entire scene), and then I scan a laser spot across the scene to build up a rastered image.
The light contribution from the entire background is 8000X brighter than the light from the laser spot.
I have heard that it is possible to extract faint signals from a noisy background by modulating the signal (the laser spot) and extracting the matching frequency components from the noise, which sounds crazy to me because, well, the noise is 8000X greater than the signal, and any detector is going to have a hard time seeing, say, 1000 photons on top of 8,000,000.
What am I missing? I'm not a signal-processing guy, if that wasn't obvious already.
Can modulating the laser spot enable me to build up a grayscale image of the scene, and if so, how, or should I take a different approach?

A much better solution is to scan the laser and detector together. Combine them with a beam splitter, add a lens and then scan.