This is truly incredible. I've spent the last 4 years neck deep in live video broadcasting solutions for consumer, commercial, and government use cases, and this just blew my mind.
My immediate thought is, even easier to install reversing cameras for older cars. Now my second thought is, spying devices, a wholly more troubling prospect.
Maybe I'm just too old-fashioned, but I generally dislike the idea of reversing cameras. I only reverse if I'm completely sure of my surroundings, based on turning back and using my eyes as well as the mirrors. Looking anywhere else tends to feel like a distraction.
480p at 10fps at 16ft is really incredible. That's enough distance to put cameras all over cars to sense environments around it. My first immediate thought is to retrofit old cars with lane-changing assist camera with this in traffic.
Perhaps it is, as cameras require looming and looming requires removing trim and finicking around which is expensive in man hours. Especially if literally every car that comes in is different.
Also fat looms the length of a car can still cost hundreds of dollars said and done.
Cost of installation is greatly reduced if all you need to do is line it up and rivet it on and there is no need to customise the power system to different cars (ie 12V, 24V) and loom layouts (ie old looms are crappy, hard to get apart for splicing, wires are old and might break during retrofit, some cars have +ive chassis)
It seems transmitting the raw sensor data would make it tricky to do this securely. Would it be possible to encrypt the signal with some sort of hardware cipher?
How about through a wall? That would make security applications in an apartment a bit less attractive, and I’m sure there are others. I wonder if encryption would be possible without much extra power overhead.
Analog encryption might be possible if the sender and receiver measure some common optical signal (assuming optically local security is what you're after).
There is a power source: the radio signal, but there's only microwatts available. Nevertheless, it's not impossible that there might be some clever solution to the problem, however, since the signal is analog that makes things a bit more challenging.
They address the security possibilities in the paper:
> Our current implementation does not account
for security. However, to secure the wireless link between
the camera and reader, we can leverage the fact that our
digital core processes the PWM signal. Each wireless
camera can be assigned a unique pseudo random security
key. Based on this key, the camera’s digital core can
modulate the width of the PWM-encoded pixel value
using an XOR gate. The reader, which knows the security
key, can map the received data to the desired pixel values
by performing the analogous operation.
If the static image could be somehow "bankswitched", "shifted" or updated... I am thinking a huge one time pad. The problem is doing it power effeciently of course.
Thinking about, the main problem will be the penguin problem [1]. Which for a video is even worse as if a shape is moving, from seeing the pixels that are changing you can tell most of what is going on.
what is there to attack? if someone is in range to get the backscattered signal they can just see what the camera is looking at...its 16ft LOS, for privacy concerns at such low powers any walls would attenuate the backscatter.
The new batteryless wireless sensors coming out of uw are incredible. Dr. Smith is leading some great work. My interest started when I heard of the WISP [0]. The RF power harvesting design is so simple and elegant. It is amazing what can be done with uW of power!
Although I don't recognise myself in the style of the comment you're responding to, I would like to point out a (watch) battery will not die in 10 minutes when drawing microwats only. I think we want to separate low energy usage, which is a cool feat by itself, and RF harvesting (which goes well with low energy usage, but so might a small, cheap, battery).
I'm not an RF engineer, but the video mentioned that the 'traditional' WiFi component would draw 1W. I assumed that, if you add a battery, it's to power this part.
No, backscatter is the only legit part of this writeup, everything else is made up.
"we simulate an ASIC, which achieves 60 fps 720p and 1080p HD video streaming for 321 µW and 806 µW, respectively" simulate, how convenient!
"Our inter and intra-frame compression algorithms reduces
total bandwidth requirements by up to two orders of
magnitude compared to raw video" so they also invented mpeg4 level of compression 1mW codec.
You might wonder how such an amazing compression codec works? its middle-out. They transmit "fullhd" by transmitting ~100x50 resolution intra frames ":). Everything is Simulated, Estimated, Planned, Theorized, Calculated and Faked. That cool 1080 YT clip of them walking around the corridor just after showing you face mounted camera models? Never happened, prototype is BW 112×112, "simulated" again. Their setup can do 1080@60 of a static picture, there is only enough bandwidth to sustain ~2fps without dropping data. They even conveniently kept all the calculations ignoring color (3x the data).
Reminds me of energy harvesting wristwatch crowdfunding HN was raving about few months back, conveniently ignoring particular design including battery able to power it for couple of years.
try 10-20 days if you believe in their simulated 800uW ASIC with magic middle out compression.
Assuming everything works according to their predictions and simulations (nothing was actually build to confirm that) you have to pump 1W from your cellphone to power 1mW wearable camera. Your cellphone will run out of battery after 4-6 hours.
Think about that for a second, they want you to recharge your phone every 6 hours to "save from inconvenience of recharging your wearable". This is they main proposed design, right there in YT presentation.
It really is. The bottom line is these sensors can scale to small sizes unlike watch batteries enabling remote sensing and integrated manufacturing. Think temperature sensors inside that watch battery. Strain sensors in existing and future infrastructure to monitor uninspectable areas for >50 years. There are a lot of novel applications.
Is this thing operating on the same principle as wireless charging? It can convert RF to power? And power consumption is so low that something like a wifi router can produce enough RF to satisfy it?
Same principle in that they both use electromagnetism, yes. But Qi and similar wireless charging using inductively couples coils. WISP power is generated from radio waves absorbed by the antenna generating a small current then boosted to a usable voltage for a low power microcontroller. The microcontroller can control the impedence of the antenna, changing the backscatter of the antenna, measure the change in backscatter you can get your 1s and 0s out of the sensor!
edit: wifi router could power these sensors, just a limited range.
I'm really impressed with the bandwidth. Being analog and only having back-scatter for the modulation, I thought it'd be way worse than current digital or old analog schemes, but from the paper, they're saying 2.8Mhz worst case for 720p@10fps, 0.98Mhz on average.
Somewhat related, a few days ago there was (is) a Kickstarter for a wireless 1080p surveillance camera that last a whole year with a single charge. Eufy
combine it with this and you get a 480p 10fps feed while no one’s moving, and a 1080p 60fps feed whenever motion is detected – the perfect combination.
This is a very cool first effort. I don't think it's suitable as a security system yet, because it seems like it would be so easy to overwhelm the relevant spectrum. Eventually maybe.
Depends on the kind of security system. I manage over 150 IP cameras in our internal facility CCTV system and it is expensive to pull Ethernet cables to 10 different spots near the ceiling in a single production room just 20x30ft. I would buy these for $300/each if it meant I only need a couple of $500 hubs at desk level per room.
I'm not worried about sabotage/jamming etc. We are trying to monitor OSHA-violations, accidents, and quality control. If I didn't have to worry about power, I would put multiple cameras in every warehouse aisle, loading dock, and doorway. Local storage is embarrassingly cheap but dragging CAT5/6 all over, supporting ton of PoE switches, and coordinating with facility staff to maintain cameras 20ft up the wall is very expensive and time consuming.
I would also buy like 20 of these for my house and live-stream my mini zoo of prairie dogs, goats, tortoises, and birds.
Seems like this could be useful for AR. I'm sure adding two front-facing cameras to the Vive Pro cut into both power and bandwidth budgets. I'd love to hear from an AR expert on this!
This is quite an interesting concept in general, with cool results. Could there be extensions to other real-time imaging modalities (ultrasound for example)? Anyone have a hypothesis?
The recived signal at 16 feet looks suprisingly good, you could probably fix it up with a bit of image processing and neural networks on the receiving end (eg. Smartphones)
I saw the presentation for their paper at NSDI last week. I was (inwardly) giddy the whole time. They had a live demo. It doesn't even resemble something that could be turned into a real product in the near term. It is, despite that, comically amazingly cool tech.
In the video and paper, they don't actually explain what backscatter is in this context. The wifi signal is what's being "backscattered". Here's some of the leadup papers about it:
Here's a video explaining WiFi backscatter. It really is an amazing concept even without doing HD video with it. All kinds of devices - I'm imaging home automation sensors - could become batteryless.
Thanks, that second paper is helpful. The OP is really the proof that passive Wifi is actually useful. In my opinion, the real advance here is in that second paper you cite, the fact that they've built Wifi enabled devices that are powered by the Wifi signal itself.
This is pretty interesting and seems transferable to any application where you have a data source you don't mind pouring energy into as long as that data source doesn't need dedicated power.
You could apply this to other types of sensors, such as perhaps accelerometers or microphones. Device locations could be a smart wearable, a surgical implant, or some kind of hard to reach diagnostic sensor.
I suppose we already use something similar in rfids but I guess the novelty comes from updating sensor data instead of a ROM. Very cool stuff.
This is an incredible outcome and really shows how much work these UW grad students and Dr Smith have been working on low power passive tech. Great stuff, I wish more people in software were versed in physical tech so they could create hardware to enable their ideas.
102 comments
[ 2.6 ms ] story [ 93.0 ms ] threadActually, there is a lot of great stuff that gets submitted that never gets anywhere.
It would be cool if somehow there was a manually edited "great but never got anywhere" stream for HN.
Edit: Now I've gone and said that, watch this get a 1000 points ;)
Most people drive and reverse way too fast.
Also fat looms the length of a car can still cost hundreds of dollars said and done.
Cost of installation is greatly reduced if all you need to do is line it up and rivet it on and there is no need to customise the power system to different cars (ie 12V, 24V) and loom layouts (ie old looms are crappy, hard to get apart for splicing, wires are old and might break during retrofit, some cars have +ive chassis)
> Our current implementation does not account for security. However, to secure the wireless link between the camera and reader, we can leverage the fact that our digital core processes the PWM signal. Each wireless camera can be assigned a unique pseudo random security key. Based on this key, the camera’s digital core can modulate the width of the PWM-encoded pixel value using an XOR gate. The reader, which knows the security key, can map the received data to the desired pixel values by performing the analogous operation.
[1] https://en.wikipedia.org/wiki/Block_cipher_mode_of_operation...
However, analog scrambling techniques should still be effective in many cases.
The tricky bit might be deciding where to start the stream...
[0] https://sensor.cs.washington.edu/WISP.html
"we simulate an ASIC, which achieves 60 fps 720p and 1080p HD video streaming for 321 µW and 806 µW, respectively" simulate, how convenient!
"Our inter and intra-frame compression algorithms reduces total bandwidth requirements by up to two orders of magnitude compared to raw video" so they also invented mpeg4 level of compression 1mW codec.
You might wonder how such an amazing compression codec works? its middle-out. They transmit "fullhd" by transmitting ~100x50 resolution intra frames ":). Everything is Simulated, Estimated, Planned, Theorized, Calculated and Faked. That cool 1080 YT clip of them walking around the corridor just after showing you face mounted camera models? Never happened, prototype is BW 112×112, "simulated" again. Their setup can do 1080@60 of a static picture, there is only enough bandwidth to sustain ~2fps without dropping data. They even conveniently kept all the calculations ignoring color (3x the data).
Reminds me of energy harvesting wristwatch crowdfunding HN was raving about few months back, conveniently ignoring particular design including battery able to power it for couple of years.
try 10-20 days if you believe in their simulated 800uW ASIC with magic middle out compression. Assuming everything works according to their predictions and simulations (nothing was actually build to confirm that) you have to pump 1W from your cellphone to power 1mW wearable camera. Your cellphone will run out of battery after 4-6 hours.
Think about that for a second, they want you to recharge your phone every 6 hours to "save from inconvenience of recharging your wearable". This is they main proposed design, right there in YT presentation.
edit: wifi router could power these sensors, just a limited range.
Edit: Added name
I'm not worried about sabotage/jamming etc. We are trying to monitor OSHA-violations, accidents, and quality control. If I didn't have to worry about power, I would put multiple cameras in every warehouse aisle, loading dock, and doorway. Local storage is embarrassingly cheap but dragging CAT5/6 all over, supporting ton of PoE switches, and coordinating with facility staff to maintain cameras 20ft up the wall is very expensive and time consuming.
I would also buy like 20 of these for my house and live-stream my mini zoo of prairie dogs, goats, tortoises, and birds.
https://passivewifi.cs.washington.edu/files/passive_wifi.pdf https://homes.cs.washington.edu/~gshyam/Papers/wifibackscatt...
http://iotwifi.cs.washington.edu/
That's terrifying.
You could apply this to other types of sensors, such as perhaps accelerometers or microphones. Device locations could be a smart wearable, a surgical implant, or some kind of hard to reach diagnostic sensor.
I suppose we already use something similar in rfids but I guess the novelty comes from updating sensor data instead of a ROM. Very cool stuff.