The company has applied for six patents covering the technology and the first monitors using it will go on sale next spring.
I know what they're doing is commendable and all, but COME FREAKING ON! Their solution with capacitors and relays is totally obvious. And if they go ahead and get the patent, does that prevent other manufacturers from making similar improvements? How is this in anyone's interest other than Fujitsu Siemens?
So... how does this square with this
wireless "theft" thing reported earlier? Maybe it's using the prostitution analogy... illegal both to provide and to partake?
Shouldn't this be easy to circumvent in custom hardware? So my understanding is that the IP numbers must be stored in memory in the web server so that the server knows where to send request responses. Why not just have a custom network card do a little more decoding and just maintain hashes of the incoming IP addresses that get filled in with the real addresses when response packets get sent out? Then the system RAM never gets the IP addresses at all. Better yet, do the crypto hash in hardware (new key once in a while, of course) so that the IP addresses never even get stored inside the network card's memory.
Y'know, I've seen a lot of comments here about why it doesn't matter where you go, but not many posts actually *answering* the question of how to get into a prestigious school.
So. I got into Caltech about eleven years ago (ugh I feel old...). My vital stats were about the same: good GPA, one of the best public high schools in my state, club activity, high SATs, etc. But basically up until that point I was just another of the thousands of faceless young Asian males applying to college. Teacher rec letters were important, but in the final analysis I think it was the essays that got me in. See, I had *fun* with them. Caltech's app deadline (iirc) was a bit later than most, so by the time I got around to filling it in I was kinda tired of the whole thing and just stopped caring. So I sat back and let my sheer love of my field (biology, with a heavy dose of CS and physics) shine through. I kinda liked the "fill this space with something interesting" part of the app... I wonder if it's still there...
And I totally had a blast at Caltech. For the first time im my life the courses were actually challenging, and oh my god I learned so much from them. Actually the teaching, especially in the biology department, was very much hit-or-miss, depending on how much the prof cared or didn't. But most the CS and physics and applied math courses I took totally rocked. Plus I got to take organic from a future Nobel laureate!
The thing that I haven't seen anyone mention is the research. So the bio dept's policy was that courses were there to build up your knowledge base so that you could get into a research lab and learn stuff while doing research. The low student-to-faculty ratio (3 undergrads to each faculty) meant that it was pretty easy to get into a lab. Basically I got to do cutting-edge neuroscience research the summer after my sophomore year because of it; that's actually pretty common at Caltech, and from what I've heard is pretty rare elsewhere. In addition, the prestige of the institute meant that the research seminars were always top-notch, so I got to hear about lots of other cutting-edge research too.
That said, the lack of women was a problem. Made for a pretty messed-up social atmosphere, and lots of women there get spoiled. And from what I've heard the EE department isn't actually all that great on hands-on stuff like building circuits. They're a lot more theoretical research-oriented. Lots of good work on information theory. So depending on where your interests lie I think CMU might actually be a better fit for EE/CS types. Certainly CMU is pretty famous for having a good engineering program.
Handbook of Mathematical Functions, by Abramowitz and Stegun. Lots of pointless function tables that can be replaced by a good desk calculator, but it does have more formulas than you can shake a stick at, esp. for special functions. Cheap, too, since it's a Dover book.
Video Microscopy, by Inoue' and Spring. Great ref for microscopy and imaging.
It's true that gnuplot's outputs tend to be a bit rudimentary for print-quality graphs, but you can go back and edit the postscript output to prettify it. I've used Illustrator (which understands postscript natively) to post-process the plot contours, do plot insets, make pretty axis lables, and shade regions between contours.
To me it just makes sense to separate the actual contour generation from the assembly of contours into final graphs. Since gnuplot is scriptable, I can do some simple data analysis (i.e. curve fitting) and get a log of exactly what I've done -- I can't count the number of times I look at a graph and go, "what did I do here again?". I can then manipulate the raw plots to get exactly the graph I want with my vector art program of choice. Combining these two functions can be convenient at times, but in my experience just tends to short-shrift both.
Their approach is pretty cool, where the activity of each gene corresponds to one bit (actually one analog "voltage", but I digress) that can be independently controlled. Unfortunately each cell in their "computer" is expected to behave similarly, so the approach won't scale.
The problem is that each gene is gonna be at least 1000 base pairs, roughly. Compare that to a typical bacterial genome (~5,000,000 base pairs) or the human genome (~3,000,000,000 base pairs), keeping in mind that large portions of those genomes are there to, well, keep the organism alive. Right now they're not even talking about taking over whole entire genomes here, just plasmids and viruses. That'll get you in the ballpark of 100,000 base pairs, or 100 bits, at most.
Oh yeah, for each transition in a circuit here you'll have to make a new batch of proteins. That'll take minutes to hours. Not exactly stellar clock speeds.
Which of course begs the question of how nature gets anything done at all. It's still pretty mysterious actually, but part of it comes down to the fact that your cells use feedback in a much more nuanced way than just "on" and "off". There's also lots of parts re-use, but probably the most important thing of all is communication and coordination *between* your cells. Like the fact that each neuron in your head does something different, and all of them put together make up something interesting and useful (hopefully). Rudimentary cell-cell communication circuits are already being constructed, and I'd like to see these scientists incorporate some of that into their work.
Actually TFA doesn't say anything about eliminating "crank" articles; they just excluded non peer-reviewed ones. It's unfortunate that the submitter used that sort of language.
And as someone who works in science and has had papers rejected by peer review, I basically have faith that the system works. Of course you'll have occasional abuses, but my experience has been that most review comments are helpful and give rise to a better finished product and more appropriate publication. Just because a paper goes against the stream doesn't mean that it's reject outright. The data and methods are probably scrutinized more closely, but they will eventually get published if they hold up. There exists for every paper that gets written a journal willing to publish it.
I looked into this over this past summer. The thing about scopes is that they don't need a high *sustained* transfer rate, just a high burst rate. If you've ever looked at one of the new-fangled Tektronix LCD multi-GHz digital scopes it'll be apparent that it updates the displayed data only a few times a second. Besides which, the LCD only updates at like 60 to 100 Hz. So if you're acquiring at 100 MHz for a 1000 pixel wide screen it works out to 10 microseconds per screenful. If you update the screen at 10 Hz, i.e. about as fast as your eye can keep up, 99.99% of the incoming data never gets displayed.
So it seems like you'd have much better luck doing a parallel port scope if you could buffer the high-speed (e.g. 100 MHz) data stream, say with a fast 2Kx8 FIFO or an n-bit synchronous counter hooked up to a dual-port SRAM. The computer can then read the data out at its leisure through a parallel port interface. Doing simple I/O programming on the parallel port I think you get much better than the 10 kHz needed for 10 Hz refresh @ 1000 samples per screen. And you'd never worry about stupid Windoze timing making you lose samples.
IMNSHO the FIFO you'd need is pretty deep as FIFOs go; such a deep FIFO will be a few dozen dollars, and I think DPSRAMs can be had cheaper. 8-bit ADC chips are can easily go up to like 100 MHz for not too much money. Keep in mind though that this is *per channel*. You can always get wider memory for multiple channels, which should be cheaper than getting multiple IC's. And you'd need glue logic. IIRC the external glue logic in my design came out to like 20 high-speed 74xx family IC's. Or you can just put them all on a PAL. (proximity to a good EE lab with PAL burning hardware and expertise would be a plus) With a decent investment of like $100 and lots of time you should get a parallel-port scope with max speeds of several 10s of MHz at least. Try doing that with a POS ISA card and software-generated timing!
BTW get the high-bandwidth breadboard and zero-propagation-time components at your nearest physics warehouse, where you'd normally go to get massless string and frictionless pulleys:)
Slashdot Mirror
From TFA:
The company has applied for six patents covering the technology and the first monitors using it will go on sale next spring.
I know what they're doing is commendable and all, but COME FREAKING ON! Their solution with capacitors and relays is totally obvious. And if they go ahead and get the patent, does that prevent other manufacturers from making similar improvements? How is this in anyone's interest other than Fujitsu Siemens?
So... how does this square with this wireless "theft" thing reported earlier? Maybe it's using the prostitution analogy... illegal both to provide and to partake?
Shouldn't this be easy to circumvent in custom hardware? So my understanding is that the IP numbers must be stored in memory in the web server so that the server knows where to send request responses. Why not just have a custom network card do a little more decoding and just maintain hashes of the incoming IP addresses that get filled in with the real addresses when response packets get sent out? Then the system RAM never gets the IP addresses at all. Better yet, do the crypto hash in hardware (new key once in a while, of course) so that the IP addresses never even get stored inside the network card's memory.
Y'know, I've seen a lot of comments here about why it doesn't matter where you go, but not many posts actually *answering* the question of how to get into a prestigious school.
So. I got into Caltech about eleven years ago (ugh I feel old...). My vital stats were about the same: good GPA, one of the best public high schools in my state, club activity, high SATs, etc. But basically up until that point I was just another of the thousands of faceless young Asian males applying to college. Teacher rec letters were important, but in the final analysis I think it was the essays that got me in. See, I had *fun* with them. Caltech's app deadline (iirc) was a bit later than most, so by the time I got around to filling it in I was kinda tired of the whole thing and just stopped caring. So I sat back and let my sheer love of my field (biology, with a heavy dose of CS and physics) shine through. I kinda liked the "fill this space with something interesting" part of the app... I wonder if it's still there...
And I totally had a blast at Caltech. For the first time im my life the courses were actually challenging, and oh my god I learned so much from them. Actually the teaching, especially in the biology department, was very much hit-or-miss, depending on how much the prof cared or didn't. But most the CS and physics and applied math courses I took totally rocked. Plus I got to take organic from a future Nobel laureate!
The thing that I haven't seen anyone mention is the research. So the bio dept's policy was that courses were there to build up your knowledge base so that you could get into a research lab and learn stuff while doing research. The low student-to-faculty ratio (3 undergrads to each faculty) meant that it was pretty easy to get into a lab. Basically I got to do cutting-edge neuroscience research the summer after my sophomore year because of it; that's actually pretty common at Caltech, and from what I've heard is pretty rare elsewhere. In addition, the prestige of the institute meant that the research seminars were always top-notch, so I got to hear about lots of other cutting-edge research too.
That said, the lack of women was a problem. Made for a pretty messed-up social atmosphere, and lots of women there get spoiled. And from what I've heard the EE department isn't actually all that great on hands-on stuff like building circuits. They're a lot more theoretical research-oriented. Lots of good work on information theory. So depending on where your interests lie I think CMU might actually be a better fit for EE/CS types. Certainly CMU is pretty famous for having a good engineering program.
Good luck in college! Remember to have fun!
Handbook of Mathematical Functions, by Abramowitz and Stegun. Lots of pointless function tables that can be replaced by a good desk calculator, but it does have more formulas than you can shake a stick at, esp. for special functions. Cheap, too, since it's a Dover book.
Video Microscopy, by Inoue' and Spring. Great ref for microscopy and imaging.
It's true that gnuplot's outputs tend to be a bit rudimentary for print-quality graphs, but you can go back and edit the postscript output to prettify it. I've used Illustrator (which understands postscript natively) to post-process the plot contours, do plot insets, make pretty axis lables, and shade regions between contours.
To me it just makes sense to separate the actual contour generation from the assembly of contours into final graphs. Since gnuplot is scriptable, I can do some simple data analysis (i.e. curve fitting) and get a log of exactly what I've done -- I can't count the number of times I look at a graph and go, "what did I do here again?". I can then manipulate the raw plots to get exactly the graph I want with my vector art program of choice. Combining these two functions can be convenient at times, but in my experience just tends to short-shrift both.
p.s. no, I haven't used Grace. Looks tempting...
Their approach is pretty cool, where the activity of each gene corresponds to one bit (actually one analog "voltage", but I digress) that can be independently controlled. Unfortunately each cell in their "computer" is expected to behave similarly, so the approach won't scale. The problem is that each gene is gonna be at least 1000 base pairs, roughly. Compare that to a typical bacterial genome (~5,000,000 base pairs) or the human genome (~3,000,000,000 base pairs), keeping in mind that large portions of those genomes are there to, well, keep the organism alive. Right now they're not even talking about taking over whole entire genomes here, just plasmids and viruses. That'll get you in the ballpark of 100,000 base pairs, or 100 bits, at most. Oh yeah, for each transition in a circuit here you'll have to make a new batch of proteins. That'll take minutes to hours. Not exactly stellar clock speeds. Which of course begs the question of how nature gets anything done at all. It's still pretty mysterious actually, but part of it comes down to the fact that your cells use feedback in a much more nuanced way than just "on" and "off". There's also lots of parts re-use, but probably the most important thing of all is communication and coordination *between* your cells. Like the fact that each neuron in your head does something different, and all of them put together make up something interesting and useful (hopefully). Rudimentary cell-cell communication circuits are already being constructed, and I'd like to see these scientists incorporate some of that into their work.
Actually TFA doesn't say anything about eliminating "crank" articles; they just excluded non peer-reviewed ones. It's unfortunate that the submitter used that sort of language.
And as someone who works in science and has had papers rejected by peer review, I basically have faith that the system works. Of course you'll have occasional abuses, but my experience has been that most review comments are helpful and give rise to a better finished product and more appropriate publication. Just because a paper goes against the stream doesn't mean that it's reject outright. The data and methods are probably scrutinized more closely, but they will eventually get published if they hold up. There exists for every paper that gets written a journal willing to publish it.
I looked into this over this past summer. The thing about scopes is that they don't need a high *sustained* transfer rate, just a high burst rate. If you've ever looked at one of the new-fangled Tektronix LCD multi-GHz digital scopes it'll be apparent that it updates the displayed data only a few times a second. Besides which, the LCD only updates at like 60 to 100 Hz. So if you're acquiring at 100 MHz for a 1000 pixel wide screen it works out to 10 microseconds per screenful. If you update the screen at 10 Hz, i.e. about as fast as your eye can keep up, 99.99% of the incoming data never gets displayed.
:)
So it seems like you'd have much better luck doing a parallel port scope if you could buffer the high-speed (e.g. 100 MHz) data stream, say with a fast 2Kx8 FIFO or an n-bit synchronous counter hooked up to a dual-port SRAM. The computer can then read the data out at its leisure through a parallel port interface. Doing simple I/O programming on the parallel port I think you get much better than the 10 kHz needed for 10 Hz refresh @ 1000 samples per screen. And you'd never worry about stupid Windoze timing making you lose samples.
IMNSHO the FIFO you'd need is pretty deep as FIFOs go; such a deep FIFO will be a few dozen dollars, and I think DPSRAMs can be had cheaper. 8-bit ADC chips are can easily go up to like 100 MHz for not too much money. Keep in mind though that this is *per channel*. You can always get wider memory for multiple channels, which should be cheaper than getting multiple IC's. And you'd need glue logic. IIRC the external glue logic in my design came out to like 20 high-speed 74xx family IC's. Or you can just put them all on a PAL. (proximity to a good EE lab with PAL burning hardware and expertise would be a plus) With a decent investment of like $100 and lots of time you should get a parallel-port scope with max speeds of several 10s of MHz at least. Try doing that with a POS ISA card and software-generated timing!
BTW get the high-bandwidth breadboard and zero-propagation-time components at your nearest physics warehouse, where you'd normally go to get massless string and frictionless pulleys