The Light Bulb

I get some crazy ideas swirling around in my skull and I know I'm never going to make anything of them so I post them here. Maybe some corporate big-wig will see one and decide to implement it! Whatever. Anyways, if you have a bright idea you'd like to share just email me and I'd be happy to post it here with credits.

Jul 23, 2004

Phew, it's 30 degrees Celsius here in Vancouver (inside!) and I'm worrying about my computer overheating. I wasn't comfortable with her CPU temperature (74C under a prolonged full load) so I took her frontside bus (FSB) down to 100 MHz from 133 MHz, meaning that I slowed her down by 25%. The effect, you could guess, was to reduce her excess temperature (above ambient) by nearly 25% down to 66C.

That got me thinking. People overclock to try and get the best performance subject to the constraint that they don't want to harm the processor by overheating. Add that motive with the ability to adjust the FSB speed in real-time and you could have something pretty neat: a utility that dynamically changes the system speed to maintain a certain temperature. As long as your computer is running cool, it would increase the FSB (even overclocking the system) until it hit the temperature limit. If it went over that temperature, it would bring the clock speed back down (perhaps underclocking). That could ensure you always maximize your system performance without endangering the hardware.

Feb 9, 2003

Ok, unsolicited mail, or spam, is a known problem with internet e-mail. I usually get 5-10 spams per day in my account. It would be nice to have a filter that recognized spam and eliminated it automatically from my inbox. There are some programs that do just that but they work by looking for keywords that might indicate spam (such as “work from home” or “lose weight fast”). There are two problems with this kind of filter: (1) it may erroneously mark legitimate mail as spam and (2) it is easy for spammers to reword their messages to get around the filters.

Isn't there a better indicator of spam? Of course there is. It is that a single identical email is being sent to thousands or millions of different people. So a filtering system that took advantage of this fact should be able to easily discriminate spam from legitimate email.

My idea is a little program that creates a small, unique signature (such as a CRC) for each message waiting in the mailbox and submits the signatures to a repository. If the same signature had been submitted many times already then that message is probably spam and can be handled accordingly. The signature could be built from the message headers (from, subject, etc.) or the body. For security reasons it's important that the signatures are (fairly) unique to each message but cannot be decrypted to recover any of the original message.

Placing the repository on a centralized server would have costs which would need to be subsidized (eg. advertising) and would be susceptible to hacks and legal attacks by unethical (redundant?) spammers so a peer-to-peer network might be a better model.

Could spammers get around this mechanism? I think so, but at a cost. They would have to make each message they sent out unique. Instead of sending a single message to a million users they would be forced to send out a million slightly different messages. I have read that spamming is only barely profitable as it is and then, only because of the low cost of sending millions of copies of a single message, so this hurdle might be enough to cross the line so that it doesn't make economic sense.

Feb 7, 2001

I was watching a television program the other day about the construction of some monolithic structure and how they performed tests on a scale model to determine its tolerances and such. For this project they built a 1:40 (meaning 40 times smaller) scale model of the structure and subjected it to certain stress tests. But this only gives an approximation of how things will be handled on larger scales.

For instance, the strength of our muscles is roughly proportional to the square of our height whereas our weight is proportional to our height cubed (which explains why it's so much easier to lift your own body when you're a child). So you see, not everything scales in the same way, which could cause problems when you try to guess how dynamics at one size scale up to another.

So naturally, they try to make the scale model as close to the actual size as they can get away with (given cost constraints and such). But why do they do scale-testing on only one scale? I suspect you could get more accurate results and (maybe) save money by building multiple models on different scales. For example, instead of one 1:40 model build three models at 1:60, 1:120: and 1:240. Assuming the cost of building a model is proportional to its linear size this would be cheaper than the single model at 1:40.

But the real advantage would be that you could extrapolate your results to the size of the real project. If you determine each model can tolerate such-and-such stress at some scale then you can estimate more precisely what the tolerance at the maximum scale would be. (I'm guessing the proper extrapolation function would be a scaling power-law because–after all–we are talking about scaling.)

Jan 19, 2000

You've heard of that glass that can be switched between transparent and frosted, right? It works by having a layer of liquid crystals which normally scatter the light in all directions but when a current is applied they orient themselves so as to let the light pass through. Cool. I wonder if that idea could be extended to photosensitive liquid crystals (which I know nothing about) to automatically block out incoming bright light.

What I envision is a liquid crystal which defaults to no particular direction but when a bright light (eg. the sun) is shone on it, it orients itself to face the light face on, effectively reflecting it. The stronger the light, the more accurate the orientation. You could maybe have a low-level current running through it too, so that in the absence of a strong light source, the glass would be transparent.

Hopefully the view from behind this kind of glass on a sunny day would be really cool: the sun and the area around the sun would be darkened out but you could still see out at angles far away from the sun. Then you could still enjoy the beautiful view from your home on bright, sunny days.

May 29, 1998

Why don't chip (integrated circuits, not snacks) manufacturers build a true random number generator into chips? Current techniques of generating random numbers actually apply deterministic chaos to initial values (seeds) to produce pseudo-random numbers. Start with the same seed and you'll get the same sequence of “random” numbers every time. In fact, trying to create generators which “you can't tell aren't random” is a serious avenue of research (Press et al. 1992). Truly random numbers would be invaluable in simulating natural phenomena on computers.

How to generate a random bit with exactly equal odds of 0 or 1.

It seems simple enough to produce truly random numbers. First of all we need to produce a random bit (binary digit–a 0 or 1). To do that we rely on Quantum Mechanics (which, to our best knowledge, is truly random). We build a little circuit with a barrier across it, so that an electron shouldn't be able to cross it. But Quantum Mechanics tells us that if the barrier is small enough there is a chance that the electron could tunnel its way across. We try to build the barrier so that an electron has a 50% chance of tunneling, but I'm guessing that might be difficult to do so we set up a little logic circuit as follows (see the figure, A, B and C store binary digits):

  1. testing sub-circuit: iff not C and not A=B then return A and reset A=B=C=0 (Done!), else
  2. toggle C; if C then close switch A and open switch B and vice versa
  3. release an electron from the tunneling sub-circuit with some probability p. If electron tunneled then set bit A or B (depending on switches) to 1, else 0.
  4. back to step 2.

That's not very clear but if you stare at it long enough it should make sense (you may need to unfocus your eyes ;-) ). Basically, the idea is to generate two random bits with some probability p and store them in A and B. Then if A and B are different then each possibility (A=0,B=1 or A=1,B=0) is equally likely so we can just set our final output to A–a random bit with exactly equal odds of being 0 or 1.

We have to use this convoluted technique because we can't produce a chip with p=50% exactly. What if the barrier was too thick and p was 10%? This circuit would still produce an output bit that was random with a 50% chance of being in either state (it would just take a little longer). Likewise, that is why we can't use two tunneling sub-circuits: they might not have exactly the same tunneling probabilities. It is critical that this circuit be able to produce perfect random bits or it will be useless.

Notice I've been deliberately vague about how to build the “tunneling sub-circuit”. That's because I don't know how. But I'm sure any tech. at a chip manufacturing plant would have a good idea. I think this circuit is feasible with current technologies because I have heard that we are approaching circuit sizes where spontaneous tunneling is something that needs to be worried about. Let's put it to good use!

One last comment: now we've got a random-bit generator but how do we generate random numbers? Well, if you want to generate a random 16-bit integer then just fill each of the 16 bits with a random bit from the generator. Easy, no?

(Boy, would I love to see a circuit like this implemented–it would make my life so much easier!)

Visit John Walker's HotBits page if you are seriously interested in perfectly random numbers. He claims (I haven't read it closely) to have constructed a little device which generates perfect random numbers and can send you an automatically generated list of up to 2048 random bytes (0 to 255) over the net.

It looks like Intel will be adding a random number generator to its chips as part of a cryptography effort.

Apr 19, 1998

Why don't jets get there tires up to speed before they land? I don't know if you've looked closely at a runway lately but there is a lot of rubber being wasted. All it would take is a wind turbine being mounted to the side of the tire to let air speed spin it prior to landing. It could be quite thin considering the speed they land at. I'm positive this would decrease the amount of tire wear.

— Kevan Kobasiuk

Nov 18, 1997

(People are going to start thinking I'm obsessed with clean showers…)

We get a lot of rain out here on the Wet Coast and, combined with the mild weather, this provides prime conditions for mildew growth. The floor of our deck outside is discoloured by a pinkish mildew. (Hey! It's a rental…so what do I care?) This Fall I noticed some strange grey stuff around some old decaying leaves on the deck. Imagine my surprise when I realized that it was actually the clean floor of the deck shining through!

The decaying leaves must produce some sort of toxin or some such thing which kills mildew. Does anybody know what it is? If it hasn't been done yet, perhaps it could be manufactured and used to clean showers and stuff. It's probably just some basic chemical but who knows?

Sept 13, 1997

What if cars had little radar or sonar sensors on their bumpers which can detect how far they are away from an object in front, behind, or beside them? And whenever the car got too close to an object a warning would sound from speakers in the general direction of the object (relative to the driver). If someone was about to get rear-ended the driver would hear a chirp from behind and have time to take evasive action.

You would want to make it work on the principle of time to impact, not distance. Based on current distance, speed and acceleration (of object relative to car) you could estimate time to impact. If this dropped down into the range of reaction times (eg. 1 second) a warning would sound. The closer, the louder (maybe). Each car would have to have an unique signal so that they wouldn't set off each other's warnings by mistake.

Since I first thought of this some minivan actually implemented a simplified version, just for backing up, I think.

Sept 13, 1997

You know how various Internet-based voice phone software is becoming a popular alternative to long-distance phone calls when both parties have Internet accounts? Well, it is. Of course the quality is kind of crappy but the advantage is that you only have to pay the relatively low price of your Internet account (if you have the necessary hardware) instead of paying long-distance phone charges. What if you started a company that extended this service to people not online. Only the dialing party would have to have a computer, which s/he would use to connect to a site in a certain city, punch in a phone number and the site would automatically make a local call to that number. Charge a ridiculously low rate compared with long-distance charges and you're in business!

Actually I think I've heard of somebody doing something similar to this since I first thought of it, but I posted it anyways. (Later…) Yup. Here's a review of an internet-based phone package which offers this functionality.

(Later still…) How about extending this service so neither party requires a computer? You dial a local Internet Telephony Service Provider (ITSP) and punch in the area code and number you want to dial. The ITSP makes a net connection to another ITSP in the desired area code and forwards the phone number. That service dials the number and ta-da! You're connected. After the call you get two small bills for a pair of local calls.

Sept 2, 1997

My glass shower door keeps getting grimy from our hard water (lots of chemicals). I wonder if it would be possible to develop some kind of spray that helped prevent this build-up. The problem is that water droplets dry on the door, leaving a white chemical deposit. Some people wipe down their showers after each use but I'm too lazy for that. If the water didn't bead up the way it does but just ran down the door there would be less on the door and it would leave less chemical deposit when it dried. What if there was some kind of hydrophillic (water-attracting) spray you could spray on a clean, dry door occasionally so that the water was more prone to run freely? I imagine the door would dry slower but there would be less water around so, when it did dry, there would be less grime. Would this work?

Aug 15, 1997

I surf the net from home, tying up our only telephone line. My wife gets irritated if she's trying to phone home and just keeps getting a busy signal. My phone company has lots of “goodies” like Call-Answer, Call-Display, etc. but these won't tell me if somebody is trying to get through. We all know Call-Waiting doesn't work with modems. Currently, the only solutions are a second phone line or a pager.

I'd like to see the phone company add some kind of Call-Email service. Whenever the phone is busy and a call comes in, I'd like an automated email message delivered to my mailbox, telling me that someone tried to call. There could be various levels of service: just a brief message relaying the time of the incoming call, or the time and number/name (like Call-Display), or it could act like a pager service where the caller is able to leave a message.

The whole point is that I receive my email rather quickly (typically within a minute of when it was sent) so I would quickly know that someone is trying to get through and I could get offline for a little while, to call them or to let them try again.

Looks like my phone company has finally implemented something along these lines. Where's my royalties? Teehee.

Aug 14, 1997

You know how all those astronauts/cosmonauts up on the space station Mir have problems with muscle and bone degradation because of the lack of gravity? Well, why not just swing them around on a big sling for about an hour a day? The centripetal acceleration would simulate gravity and may help keep the body strong. You could put a counterweight on the opposite end of the sling so the rotation doesn't throw the station out of orbit or anything.


Press, W. H., S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery. 1992. Numerical Recipes in C: The Art of Scientific Computing. Second. Cambridge: Cambridge University Press.



  • science/the_light_bulb.txt
  • Last modified: 2018-06-18 22:05 (6 years ago)
  • by Rik Blok