| Pages:
1
2
3 |
woelen
Super Administrator
Posts: 7976
Registered: 20-8-2005
Location: Netherlands
Member Is Offline
Mood: interested
|
|
Home made spectrometer
I ordered a set of LEDs of different colors (red, orange, yellow, green, blue, UV plus white) and I intend to use these for building my own crude
spectrometer. These LEDS (except the white one) emit light at a very narrow band around a certain wavelength.
With a broadband sensitive lightsensor (or array of light sensors for different wavelength areas) and some electronics I can make a spectrometer.
With this setup I can measure at certain points, but I would like to measure at many more points. Does anyone of you know a means to produce light of
any desired wave length at acceptable strength? Of course I can work with bright white light and all kinds of filters, but that seems like a very
expensive and difficult setup. I would like most if I could make a light source, with adjustable wavelength.
Any ideas are welcome. My aim is to build a spectrometer at a tight budget of at most $200.
|
|
|
chromium
Hazard to Others
Posts: 284
Registered: 27-6-2005
Member Is Offline
Mood: reactive
|
|
Device that separates one particular selectable wavelength from white light is called monochromator. You can google it there is lot of links.
http://en.wikipedia.org/wiki/Monochromator
|
|
|
Marvin
National Hazard
Posts: 995
Registered: 13-10-2002
Member Is Offline
Mood: No Mood
|
|
I think you might be better going the other direction. A prism, or a replica grating, some optics and a flatbed scanner. What you are doing
currently sounds better suited to specific absorbtion measurements.
|
|
|
Pommie
Hazard to Self
Posts: 70
Registered: 6-2-2005
Location: Australia
Member Is Offline
Mood: No Mood
|
|
How about a mixture of powders from Florescent lights combined with UV LEDs as a mixed source? Some of the tubes used in fish tanks seem to have a
broad spectrum.
Mike.
|
|
|
chemoleo
Biochemicus Energeticus
Posts: 3005
Registered: 23-7-2003
Location: England Germany
Member Is Offline
Mood: crystalline
|
|
Marvin, where does a flatbed scanner come in? Are you suggesting to use this as a means of detecting transmitted light? I'd think the sensitivity
is not high enough...
Anyway, the way I'd do it is this:
Get yourself a broadspectrum light source, which is very intense. It can also be some sort of flash light. I can check with my colleague what lamps a
professional spectrometer uses.
Then, focus that light into a tight linear beam, as far as this is possible. The more linear, the better, as this will affect the minimal wavelength
range.
Feed this into a quartz prism, which is remotely controlled by a high qual servo.
Dispersed light exiting the prism falls onto a fixed slit (i.e. immovable) of very narrow width (< 1 mm), at i.e. 50 cm away from the prism.
Behind this is the sample chamber, for the cuvette containing the sample. Again it has to be quartz if you want to work in UV.
On the exit side of the cuvette is a photodiode, possibly a set of them, with different absorption profiles.
If you calibrate this properly (with a computer program), I am sure you could make this work.
Of course, the position of the prism controls what wavelength of light falls onto the slit, so the quality and calibration of the servo, and the
linearity of the beam will be the most important factors.
It's hard but it can be done 
[Edited on 21-11-2005 by chemoleo]
Never Stop to Begin, and Never Begin to Stop...
Tolerance is good. But not with the intolerant! (Wilhelm Busch)
|
|
|
I am a fish
undersea enforcer
Posts: 600
Registered: 16-1-2003
Location: Bath, United Kingdom
Member Is Offline
Mood: Ichthyoidal
|
|
You could greatly increase the resolution by using a wider range of LEDs. They can be bought with far more exacting specifications than the standard
"Red, Yellow, Green..." of general-purpose electronics catalogues.
As a starting point, check out the LED Muesum.
1f `/0u (4|\\| |234d 7|-|15, `/0u |234||`/ |\\|33d 70 937 0u7 /\\/\\0|23.
|
|
|
Twospoons
International Hazard
Posts: 1281
Registered: 26-7-2004
Location: Middle Earth
Member Is Offline
Mood: A trace of hope...
|
|
Your best broadband light source will be a thermal one - tungsten halogen, xenon flash, or a metal halide HID lamp. All the phospors emit on discrete
wavelengths, even the so-called broadband fluoros.
A source based on an HID lamp will cover long IR up to UVB or UVC if you get one without a UV filter.
Helicopter: "helico" -> spiral, "pter" -> with wings
|
|
|
IrC
International Hazard
Posts: 2710
Registered: 7-3-2005
Location: Eureka
Member Is Offline
Mood: Discovering
|
|
If only one frequency at a time is needed I wonder why a tunable dye laser is not used? Also, Marvin is right the charge coupled imaging device is
more than sensitive enough to do the job.
|
|
|
Tacho
National Hazard
Posts: 582
Registered: 5-12-2003
Member Is Offline
Mood: No Mood
|
|
Please, criticize my idea for a homebuilt spectrometer:
(see attached scheme)
1) two cuvettes are put in the device, one with sample and the other with the solvent alone (reference). Identical photosensors are placed behind the
cuvettes ;
2) photosensor 2, behind the reference, keeps servo light shutter with an aperture that keeps it's output (photosensor's output that is)
constant no matter how bright the halogen lamp is;
3) a cheap and dirty interface with the computer increases the brightness of the lamp by Pulse Width Modulation, slowly, all the way from low infrared
to everything+uv. Of course, the shutter closes as the lamp gets brighter. Reference photosensor output is constant and the sample photosensor output
floats (hopefully).
4) another cheap and dirty interface reads the output of the photosensor 1 as the brightness of the lamp is increased.
5) the computer does some math. I think the change in values not the value itself will bring the information. Something in the lines of dX/dY along
the brightness path.
|
|
|
Pommie
Hazard to Self
Posts: 70
Registered: 6-2-2005
Location: Australia
Member Is Offline
Mood: No Mood
|
|
| Quote: | Originally posted by Tacho
Please, criticize my idea for a homebuilt spectrometer:
|
Doesn't a dim light still have a broad but shifted spectrum and so to actually work out which wavelengths have been absorbed would require very
accurate photosensors and a lot of number crunching.
If you used a slit and a rotating prism in place of your shutter such that the line of light was horizontal and passed through both samples and onto
the 2 sensors, I think you would get a better result.
You could still use something like a PIC to control a model servo. I would think that 2048 positions would be possible. A pic could also read both
sensors and send the data to the PC. If your familiar with microcontrollers, have a look at the 16F88 at microchip.com.
Mike.
|
|
|
unionised
International Hazard
Posts: 5102
Registered: 1-11-2003
Location: UK
Member Is Offline
Mood: No Mood
|
|
"If only one frequency at a time is needed I wonder why a tunable dye laser is not used?"
IRC, have you seen the prices of a tuned laser and a light bulb?
Laser sources do get used in spectrometry- but not often. The commonest lamps in spectrophotometers are tungsten/halogen or deuterium lamps.
|
|
|
Marvin
National Hazard
Posts: 995
Registered: 13-10-2002
Member Is Offline
Mood: No Mood
|
|
Using a scanner povides a number of very helpful features. No moving parts. Well, the motor will be moving but this does not have to be connected to
the bar or even working, you only need to fake the end stop to trick the software into thinking its scanning a page. No writing software, its all
written for you, you only need to use a very basic program to add all the lines together (for less noise).
The optics will be hard particually if the sensor is used directoly (only about 1 inch long), but very sensitive and you only need to disperse a small
distance.
Tacho,
I had the same idea for an infra red spectrometer. The math would be very hard, black body emission curves and sensor curves and integrating to get
something soluable. Scanning the spectrum would take a long time. Even at the bright end you need extremely low noise, as the noise and the
difference in the signal are in the same place. Worse still at the dull red end, your light is neerly just red (with an IR cut off), but its also
very very faint, so you need a huge dynamic range for the sensor, very very long capture times for the low noise requirement. Resolution would suck.
I can't prove this but I'm certain it would. In the case of an optical spectrometer this would mean thin absorbsion lines would just
vanish. With a crude dispersive spectrometer you can use atomic emission, and while they may blur they can't just dissapear.
Tunable dye lasers have the best resolution of any method, but they contain a spectrometer cavity themselves, so building one would be much harder.
|
|
|
Quibbler
Hazard to Self
Posts: 65
Registered: 15-7-2005
Location: Trinidad and Tobago
Member Is Offline
Mood: Deflagrated
|
|
Search for LED color chart. It would seem you can cover the whole range with reasonable resolution.
Another interseting development is the tunable LED - not available as yet and will probably be expensive.
I seem to remember that LEDs used to change color if you pushed them too hard - green ones would turn slightly orange. Doesn't seem to happen
with new ones - or is it just another sign I'm losing my mind.
|
|
|
Twospoons
International Hazard
Posts: 1281
Registered: 26-7-2004
Location: Middle Earth
Member Is Offline
Mood: A trace of hope...
|
|
| Quote: | Originally posted by Tacho
...Identical photosensors ... |
And there you have a rather fundamental problem! It is much better to use one sensor and swap cuvettes - that way all of the measuring equipment is
identical for both the test and reference samples, and all the matching, calibration and drift issues just go away.
I like Marvin's idea of using a scanner - you get a nice sensor typically 4000 pixels or bigger, plus some handy optics. The killer is going to
be dark current in the sensor if you need long exposure times due to a weak source. This can be overcome to some extent by summing successive
exposures - a process called oversampling. This also reduces the noise in the signal by the square root of the number of samples.
Spectral resolution is going to be all in the optics - and its going to be a tradeoff between resolution and light throughput.
Helicopter: "helico" -> spiral, "pter" -> with wings
|
|
|
vulture
Forum Gatekeeper
Posts: 3330
Registered: 25-5-2002
Location: France
Member Is Offline
Mood: No Mood
|
|
Dark Noise in the sensor can be greatly reduced by cooling it. It doesn't have to be LN2, 273K instead of 298K will already show improvement.
Besides, I don't think your scanner CCD would like LN2 temperatures.
One shouldn't accept or resort to the mutilation of science to appease the mentally impaired.
|
|
|
woelen
Super Administrator
Posts: 7976
Registered: 20-8-2005
Location: Netherlands
Member Is Offline
Mood: interested
|
|
I do not worry about the noise. The LEDs I ordered are very bright (12000 Cd) at their normal rated current. They also are well-specified (center
frequency specified in nm and specification of the 99% power bandwidth around that central frequency). In fact, the LEDs have a very narrow band of
just a nm or so of wavelength. My concern indeed is the number of sample points, or stated otherwise, the number of frequencies at which I can
measure. I'll see if I can find more different well-specified LEDs. I do not want to buy just some led with a vaguely specified color.
The monochromator or servo mechanism simply is not within my budget. I'm quite sure that these will be WAY beyond my budget of appr. $200.
If I have any results, then I'll let you know. I'm still waiting for the LEDs, but I expect them to arrive one of these days.
-----------------------------------------------
How would a digital camera be as spectrometer? If I take a standard light source, a standard test tube (quarz if UV is to measured as well) and I take
a picture of the test tube with solvent only and I take a picture of the test tube with the colored species how well could that be used for spectral
analysis. How good are the CCD chips of digitals camera's for that purpose. With some mathematics and analysis of the RGB data I should be
capable of deriving spectral information, isn't it?
Problem might be dynamic range. For my website I already noticed that many solution seem black at higher concentration. I need to dilute them for good
pictures, but of course that is not always acceptable in coordination complex analysis. Dilution may affect the complexes.
If anyone has an opinion on this, I would appreciate that.
[Edited on 22-11-2005 by woelen]
|
|
|
Twospoons
International Hazard
Posts: 1281
Registered: 26-7-2004
Location: Middle Earth
Member Is Offline
Mood: A trace of hope...
|
|
| Quote: | Originally posted by woelen
With some mathematics and analysis of the RGB data I should be capable of deriving spectral information, isn't it?
|
Uh, no. Too much information is lost. e.g. There's no way to determine if the red light falling on a red pixel is at 600nm or 620 nm or
650nm.
You can build a simple monochromator with a couple of slits and a piece of CD. Total budget of about $1. Of course, you get what you pay for ...
You should worry about the noise, as its likely to bury the signal you are trying to see.
Good point about the cooling, Vulture! A peltier cooler would be a simple way to take the sensor down to -20C or so.
Helicopter: "helico" -> spiral, "pter" -> with wings
|
|
|
woelen
Super Administrator
Posts: 7976
Registered: 20-8-2005
Location: Netherlands
Member Is Offline
Mood: interested
|
|
| Quote: | Originally posted by Twospoons
You should worry about the noise, as its likely to bury the signal you are trying to see.
|
Sorry for me not being clear. I'm not going to use a scanner's CCD-array, but some light sensitive diodes, with integrated amplifier. I have
made a light-intensity meter, which I use in photography and the sensor I used there has a dynamic range of 1 : 10000 or even more. I can measure at
1/1000 of a second with that, but also at shutter times of a few tens of seconds. I have two of these sensors left and these I intend to use for my
meter.
|
|
|
Marvin
National Hazard
Posts: 995
Registered: 13-10-2002
Member Is Offline
Mood: No Mood
|
|
woelen,
Most of these objections arn't aimed towards your idea. Noise will not be a problem. What you are making is more like a colorometer, you'd
need 100's of LEDs to get anything like a proper spectrum. This may not be a disadvantage, most UV/Vis curves Ive seen are profoundly dull and
contain little information aside from the concentration of a few important coloured species, as opposed to IR which tells you things about the
molecule.
With a proper dispersive spectrometer either with a motorised system, or with a static CCD array type you'd have no problems working with things
like atomic emission lines as well as everything you're current plan does. And it would cost under $200. If not the scanner method then
servo's from an old printer would work fine, a glass prism and an old telecope would get you everything you need to adapt your existing light
meter.
Noise would not be a problem with either setup.
Twospoons, the process of adding together samples to reduce noise is called integration over time or sometimes 'frame integration'.
Oversampling is a process intended to produce higher resolution using a higher sampling rate than is actually required for the job.
A scanner CCD is pretty sensitive, when you consider the light reflected from the paper is only focused be a lens a few millimeters in diameter an
effective 6cm away or so, transmission spectroscopy should be a breeze, even with very thin slits.
|
|
|
Twospoons
International Hazard
Posts: 1281
Registered: 26-7-2004
Location: Middle Earth
Member Is Offline
Mood: A trace of hope...
|
|
Marvin, the term 'oversampling' is also used to describe digitising the same thing many times and averaging the result to improve
signal to noise ratio. At least in my line of work it is.
Helicopter: "helico" -> spiral, "pter" -> with wings
|
|
|
Tacho
National Hazard
Posts: 582
Registered: 5-12-2003
Member Is Offline
Mood: No Mood
|
|
Twospoons,
The reference photosensor just controls the shutter to make sure any change in the sample sensor output is due to the sample absorbtion, not light
change.
Marvin,
I'm glad you had the same idea. Shows the principle is sound. I agree that resolution would suck, specially in IR range, but it still could be a
nice tool for the amateur to compare samples with known standarts, including the visible light range. Since all materials are available in my attic
junk box, I intend to give it a try . Then again, hell is paved with this...
Pommie,
I'm a Z80 man, and it took me great effort to evolve to 8051 (in fact, 89S8252). I'll stop there. No PICs for me.
I recently found out that old Turbo Pascal running on a pentium machine in a DOS enviroment (booted, not a windows' shell) is incredibly fast. I
mean really really fast. And it deals with the parallel with simple commands. That's good, because I don't have a nice ansi C compiller for
DOS, C doesn't deal easily with the parallel port and I like the old Turbo anyway.
|
|
|
Quibbler
Hazard to Self
Posts: 65
Registered: 15-7-2005
Location: Trinidad and Tobago
Member Is Offline
Mood: Deflagrated
|
|
Marvin is quite right most UV/Vis spectra are amazingly dull. Most measurements are to determine the concentration and for that all you need is a
monochromatic source.
That said I think this is a great idea I am going to make one I am thinking of mounting the leds on a disc and driving it using a stepper motor
probably use a LDR for detection to ADC to a PC. One thing to watch out for cuvets unless matched (=expensive) will show a lot of variability so I
ALWAYS use the same cuvet for sample and blank.
|
|
|
Marvin
National Hazard
Posts: 995
Registered: 13-10-2002
Member Is Offline
Mood: No Mood
|
|
Twospoons,
I think I see where the confusion has happened. Oversampling is sampling the data at a higher rate than required by the highest frequency componant.
It does not have to be done in one pass though, each pass can sampled with a slightly different phase, in the gaps of the previous sampling sets, so
to speak. The oversampled dataset can be made to produce lower noise and higher channel resolution data by a process called decimation.
When a sample is aquired at the same point, or for a DC value, this is not oversampling. Simply adding together the signal and divided by the number
is commonly refered to as 'signal averaging' or 'integration' the latter, particulally with imaging systems and CCDs as
'charge integration' together with something to differentiate it from on chip charge integration using long exposures.
Both processes in ideal situations reduce noise by SQRT(n) but since the position of the pixels are fixed on the CCD you cannot perform oversampling.
|
|
|
Twospoons
International Hazard
Posts: 1281
Registered: 26-7-2004
Location: Middle Earth
Member Is Offline
Mood: A trace of hope...
|
|
I say "toh - mah - toh" , you say "toh - may - toh".
Helicopter: "helico" -> spiral, "pter" -> with wings
|
|
|
Quibbler
Hazard to Self
Posts: 65
Registered: 15-7-2005
Location: Trinidad and Tobago
Member Is Offline
Mood: Deflagrated
|
|
I've had a look at the spectrum of some LEDs. It is possible to get a bit of tuning by putting too much current through.
Green LED
20 mA 565 nm
100 mA 573 nm
180 mA 587 nm
Red LED
20 mA 672 nm
100 mA 687 nm
200 mA 695 nm
in all cases the half height width was about 13 nm. It probably shortens the life of the led doing this and the high currents should not be continuous
due to the heating. But it seems possible to get about 20 nm of tuning. A word of caution some cheap LEDs have a much higher band width I bought some
cheapos and they have a bandwidth of about 50 nm (as oppossed to the 13 nm).
I've just re-read the specs on your LEDs
12000cd you might want to consider making a death ray instead. And are you sure that the band width is 1 nm you would be lucky to get that from a
laser diode.
[Edited on 25-11-2005 by Quibbler]
|
|
|
| Pages:
1
2
3 |
![[*]](images/xpblue/default_icon.gif){kind=icon}
