New quantum-dot spectrometer is small enough to function within a smartphone | Hacker News Reader

New quantum-dot spectrometer is small enough to function within a smartphone

newsoffice.mit.edu

64 pointsuser_23571134 comments

It's a way to build a multispectral imager out of a CCD imager, with lots of different narrow-band filters over different pixels. Humans have 3 color sensors; some birds have 22, which helps in recognizing specific plants. This is more useful if it extends outside the human visual range, into IR and UV.

Applications are somewhat limited. It's useful for forensics [1], farming (weed and sick plant identification)[2], and some kinds of industrial inspection. For farmers, it's more useful on a drone than on a smartphone; they can fly over the fields and look for trouble spots.

[1] http://www.beneaththewaves.net/Photography/Uses_of_Multispec... [2] http://www.marcusuav.com/using-multispectral-cameras-with-sm...

To those who are more familiar with spectrometers:

What's the most interesting "consumerish" application that you can think of for a spectrometer? I imagine health related applications are an easy target because spectrometers are already widely used in biology. Are there other, more off the wall applications that don't make sense now, given the cost and size of a spectrometer?

I don't know much about spectrometers, but I thought [these guys](https://www.kickstarter.com/projects/903107259/scio-your-six...) had some good suggestions when I saw their campaign last year.

Depending on how good the specm is, just about as many uses as a camera. What's this leaf? Whats this red stuff? What kinda soil and I standing on? What is the air pollution like today? Is this bleach? Is that really gold? What is this IC made of? What is my baby's heart rate, does she have the flu? What grade of steel is this wrench? How old is this picture of gram? How new is this gasket? Is this chicken what I have, or fish (from the lips of Mrs. Lachey (nee Simpson) herself)?

It's mostly what questions as to materials, depending on how good the specm is. Combine this with your phone's camera for other optical measurements (IR, size, shape, position, etc), some gps data, maybe some tracking of your amazon history, and you have a device that might be able to tell you what anything ever seen is to reasonable certainty. There is without any doubt, a large market for that.

I could see this being a huge boon for Augmented Reality applications. It's frankly quite exciting. I'm kind of jittering right now. Imagine how difficult it would be for an application to determine the difference between a real person and a suitably detailed mannequin. The human might be able to tell, but the application wouldn't. But test the "skin" of the form and it will know, and then know different overlays to apply.

Oh man oh man oh man.

Human -> facial recognition -> overlay Twitter feed.

Mannequin -> check for NFC tags -> show product details + highlight sales.

Banana -> ask the user if they're going to eat it -> add to daily calorie intake.

Generic plate of food -> estimate density of individual ingredients based on image coverage -> save the world from heart disease!

I'm dead serious, how soon can we get this in a device? I have work to do.

I think you'll still need to carefully prepare the samples.

Yeah, I started trying to research how this could be done and realized this :(

Still feels like there is something there. I mean, can't we consider a photo a type of spectrograph, just for only a few, very broad bandgaps of light? Perhaps we can spend some of a modern, high-resolution CCD's spatial resolution on wavelength resolution. With high-magnification lenses, you'd then be able to spot-check an area of a certain material from across a room. You'd not be able to do the entire room in one shot, but you'd be able to scan a room. Combine it with photogrammetry methods to process a roving needle as a user walks into a room, looks around, etc.

Depending on how good the specm is...

The article says that besides being small it's also high bandwidth and high resolution. Not sure how many distinct things you can identify based on spectral analysis alone, but I have high hopes for one day owning a Tricorder like device. Being able to fingerprint and identify real world stuff we might ingest seems huge, am I drinking Coke or Diet Coke? Vodka or Gin?

> am I drinking Coke or Diet Coke? Vodka or Gin?

Society didn't take MyVessyl.com really well.

Your friends might roll their eyes at you putting Soylent in your Vessyl and monitoring your daily caloric intake on your Apple Watch either, but nerds want what they want.

Could it be used to test if drugs are the genuine stuff?

That would be a great use case. And really, not even for the illegal stuff either. Is this an M&M or advil? How much and what isomer of caffeine is in this drink? Is there shellfish in this? Is that poison ivy?

I mean, really simple stuff for anyone to get hurt with chemically, things like this can help with. Maybe not the full functionality like we see on star trek. But, I'm sure people with allergies, ravers with molly, moms that want to know what is safe for their babies, etc can all have something that can help mitigate risk in some small way.

The right way to do that would be with FTIR or Raman spectroscopy, and this sensor is capable of neither.

At best, this is an ersatz replacement for the tunable filter on a hyperspectral imaging system. There are some advantages (off hand, acquisition rate is a big one) to a more diverse mosaic filter, but other than in using QDs as the actual filter medium, this idea is not at all new.

Meh, it's a use case scenario. When the differences in the materials you are testing for are so large, the sensor can make a pretty good bet for you. Take the case of; Is this Tylenol or a Skittle. Acetaminophen and fructose are very different and you can test for that pretty simply. To get very accurate readings you need a much more powerful pulse laser as the source, ture, but mose people don't need that. You can get 80% of the use cases with 20% of the 'tech', I think that's a great reason to stick one in every phone you can.

I'm not a big FTIR person, but at least for Raman, laser power is far less important than the linewidth (and you definitely don't need a pulsed laser) and the rest of the optical engine. I'm actually pretty bearish across the board on these consumer spectroscopy products mostly due to the importance of sample prep. Your use case statement is correct, distinguishing paracetamol from candy should be easy, but is really that a common need or a novelty? It seems to be the latter to me.

Neural networks would certainly benefit from the additional dimension, no more confusing furniture with animals[0].

[0] https://news.ycombinator.com/item?id=9749660

Yeah, it kinda acts as a proxy for 'touching' something. OUr finger get a lot of info from the surface texture, resonances in the material, tensile strength, etc. that a computer just can't have yet. A good specm could help fill in a bit of that data set.

There are some practical examples of home experiments that can be conducted with a spectrometer on the PublicLab wikipage here: http://publiclab.org/wiki/spectrometry-activities

I particularly like the idea of being able to test for fraudulent extra virgin olive oil (http://publiclab.org/notes/ygzstc/04-30-2014/detection-of-ol...).

I don't condone the majority of it's movement, but the essential oils community would be head over heels for a device like this... their oils are in the tens to hundreds of dollars for a few milliliters, so they are naturally very concerned about the quality and makeup of their oils.

I'd imagine that a ~$50-$100 device+app would sell extremely well.

More broadly, I imagine there are plenty of people dealing in... chemical substances, let's say... who would be interested in an on the spot test of purity they can rely on before parting with their cash.

Equally, I suppose there would be interest from law enforcement in the same technology.

Drug quality tester - checking purity of LSD, Ecstasy, Meth.

1) Test what percentage of jewellery is gold 2) Test whether that pearl is real? 3) How about if my water contains mercury?

The list is endless. A Spectrometer acts as a basic sense that is not fully evolved in animals.

I'd like a better understanding of why this is better than an ordinary RGB camera sensor by itself, at least as far as visible light is concerned. What would be an example of a wavelength that couldn't be disambiguated from some other wavelength by looking at its relative amplitude through R, G, and B filters? What do the extra degrees of freedom afforded by the quantum dot filters buy us? Could adding a fourth color filter to some of the elements in a CCD sensor accomplish the same thing?

Is it just a matter of the line width of a typical color filter being too broad?

It's not so much a matter of being able to distinguish one wavelength from another as it is being able to distinguish one wavelength from a combination of other wavelengths, or one combination for another.

For example, monochromatic orange light falls between about 590 and 620nm. But if you get a beam of red at half[1] the intensity of the orange mixed with a beam of yellow/green at half the intensity of the orange, you won't be able to tell the difference between that combo and actual monochromatic orange given a simple RGB array.

[1] Substitute "half" with whatever the appropriate scaling function is for the sensors and wavelength ranges in question.

To be accurate sensors need to be carefully calibrated. I think with a quantum dot they may be more stable over long periods (due to their inherent qualities).

RGB camera sensors can't identify specific wavelengths, all they can tell you is that a given wavelength is in some specific range. And when the light is not monochromatic then it becomes impossible to determine the spectral pattern.

I think these sensors implement n color vision, working a lot like the ear does. Instead of just detecting three colors you detect hundreds of colors. With each dot tuned to a fairly narrow band of light you get good selectivity.

Right, I'm just wondering how many discrete sensors you need before you can say you have a "real" monochromator. This scheme appears to work a lot like an RGB sensor, but with more than three filters.

One step closer to a real life tricorder.

Here's the link[1]. We're actually looking for micro UAV compatible hyperspectral cameras that have a significant number of pixels in any one direction.

[1] http://www.nature.com/nature/journal/v523/n7558/full/nature1...

I'd like to see these mounted under a car to warn you when you're hitting ice.

A lot of times you can't see it. the roads can be wet and then certain areas start freezing. You slow way down once you start seeing ice patches.

I don't think a spectrometer is the most efficient or effective way to detect ice.

Does ice return a different spectrum than liquid water?

No, but couple it with thermal imaging and "water, 0C" is ice and "water, >0C" is rain.

Reply on news.ycombinator.com