CDMA is super-cool tech, but it has some issues. The biggest one being that the cell tower needs to receive the exact same energy from every handset - if the power-control on the handsets malfunctions, the whole cell goes down. In fact, this is *the* limiting issue for CDMA - even with decent power control, its still not perfect. If it was perfect, a generic CDMA tower capable of supporting x Walsh Functions* could support x users all on the same band within an area limited only by output signal power, multipath-delay (though Qualcomm's CDMA implementation uses a RAKE Receiver which redues the effects of multipathing), and interfering noise (white or otherwise). And yes, x could go to infinity.
Anyway, despite the problems with power-control, China and Europe are adopting their own versions of CDMA for 3G services.
*walsh functions are orthogonal - by modulating the output signal by the assigned walsh function (different for every transmitter) and then scrambling the resultant signal by a pseudo-random sequence (to improve the average PSD), all of the signals are orthogonal to each other and do not interfere as long as the power is exactly the same for all of them (power differences screw up the orthogonality).
No, SNR is the correct way to measure signal strength. In fact, its the only way to measure signal strength unless you plan on proving that Claude Shannon was wrong about everything. Inside the phone, eveything (noise and signal) get boosted to some set energy level anyway, so signal energy only makes sense in the context of its relation to noise energy (SNR).
IANAWE (I am not a wireless engineer - though I am a wireline engineer and know the basics of CDMA and wireless tech), but AFAIK the problem is due to multipath delay (creating Rayleigh Fading/ Frequency-Selective Fading). The effect of Rayleigh fading is that while the signal may be fine right now and right here, if you move the phone (or something in the area moves) a half-wavelength to the left or right (at 900 MHz, about 6.5 inches. Less at higher frequencies), the signal may suck. Hard.
If you're moving quickly (say, in a car), this effect is ameliorated somewhat as you zip in and out of the fades and your average SNR isn't too bad. Standing still (slow-fading), however, the phone can't deal with being in a fade for that long and drops the call.
Results: for the same mix of calls (i.e. home and roaming), a European cellphone bill is 65-70% if not less than the equivalent US bill.
Talking with my European colleagues (French, Belgian, Dutch), they seem to think that Uservice in the US is cheaper if you know how to pick your plan.
For example, for $39.95/month, I get virtually unlimited calling (inside the US) during business hours (400 minutes/month) and unlimited minutes (again, domestic) at any other time. My home area is virtually the entire North East US (NJ, DE, NY, PA, all of New England and I think Maryland) and outside this area (roaming) its 10 cents/minutes. My cell-phone bill only ever exceeds the minimum (around $44 after taxes and such) when I travel outside my home area, despite having around 500-1000 minutes/month of calls.
My company's field engineer in the Netherlands, on the other hand, pays a reasonable rate in the N.L. (I forget exactly what, but it wasn't too bad), but when he travels to Germany or Belgium, its 2 Euro/minute. Anyway, he says that his phone bill is usually over 100 Euro/month.
don't see why everyone thinks we will have a war with China. It's pretty ridiculous. We are on very excellent terms right now, and our economies are too interdependant on each other to go to war. Taiwan isn't something we'd risk a nuclear war over. It is not that important to the US. Sure, our government has said that we'd defend Taiwan against and invasion, but that's just a bluff. In the event that China did invade Taiwan and the U.S. did threaten war, China would probably back down. They aren't braindead. The Chinese wouldn't risk a losing nuclear war with the U.S. Nuclear war is a bad thing that nobody wants. Notice that despite all of their problems and differences, the Cold War never went nuclear.
War over Taiwan is an absurd thought - epeciually for the PRC. The Chinese Navy couldn't cross a puddle. Taiwan, on the otherhand, has been buying fairly modern US cruisers and destroyers as well as maintaining one of the world's most elite airforces (yes, China has the largest airforce, but much of it is Mig 15's and 19's, which don't really count). The only reason the PRC would try to invade Taiwan is if they wanted a cheap way to scrap their fleet:)
Yeah, but you can't mass-deploy 100 Mbps ethernet like that - the crosstalk would kill you. Ethernet is spectacularly inefficient in the way that it uses bandwidth (though GigE is much better from what I can tell) and the spectral compatibility with repsect to self XT (not to mention other comm protocols) is terrible. But that's not what ethernet is for - ethernet was designed to be simple and cheap and fast and short-ranged.
And to tell the truth, this particular VDSL implementation isn't going anywhere. QAM is going to get clobbered when it comes to standardization by DMT. QAM VDSL is made up of 3 or 4 massive downstream channels and a similar number of upstream channels interleaved in a frequency-domain-duplex fashion, with each channel having a different symbol rate (somebody correct me if I'm wrong, but I just can't see any other way to do this with QAM unless the channels all have the same bandwidth, which I don't believe they do). This means that you need separate analog receivers for each of the channels - yuck. And QAM is also disadvantaged by the fact that these massive channels have to deal with interference by notching out the offending frequencies in their equalizer, and given limitations on the number of taps you can have in an equalizer at these kind of symbol rates, the notches are going to either be inneffective or way too wide.
DMT VDSL, on the otherhand, is effectively 4096 (or 8192) separate QAM channels, each 4.3125 KHz wide. This means that a) Everything runs at the same symbol rate, greatly simplifying frequency-domain-duplex arrangements and b) If there is interference at a particular frequency, you just load less bits there (or shut it off completely), and this can be done with a 4.3125 KHz resolution.
A) means DMT modems will be cheaper and B) means they will deal more gracefully with noise which in turn means it can be more widely deployed. Both those things mean more money for the telco, and telcos like money:)
Similarly, but from the other end, I work at a communications DSP company and I'd estimate average productivity to be around one line of production code per day per worker. Once the main development effor is complete on a project, problems tend to be quite obscure and can, in extreme but suprisingly common cases, take thousands of tests to duplicate, examine, and debug.
This means a lot of time spent doing absolutely jack shit. Fortunately, we have understanding managers who understand this. However, their managers and the marketing folks they interact with do not. "Why is [whoever] sleeping on the couch?!? We have to solve this problem for [big customer]!" is a common phrase heard around here.
Some people don't sleep on the couch waiting for the test to trap the problem - they look busy, and they get the respect from marketing guys for working all out to solve the problem. Some of us consider automated test-time to be nappy time - we get the disdain of the marketing guys. Fortunately, the end result, when it comes to solving the problem is similar and the engineering managers recognize this and couldn't give a shit what the marketing guys say.
With a different management, though, I imagine this could be quite a problem - what the fuck am I supposed to be kept busy with while waiting for obscure problem #129 to show up so I can debug it? They sure as hell don't know, but to some people, if I'm not working *really* hard 40 hours/week, well, why bother employing the guy.
Well, that's... interesting from my perspective. Can't say that I'd like it much, but interesting none-the-less.
I'm a geek, but I really dug metal-shop and auto-shop. In fact, if I have anything good to say about the US education system, it would be that it allows you to do damn near anything in high-school and still go to college and do damn near anything in college and still get the job of your choice.
In high-school, I did take a computer course, and I loved it, but I also took auto-shop and metal shop and really liked that as well. In college, I studied physics and took a four year lab practical in alcoholism, and somehow managed to get my current job as a DSP guy. The ducations were nice, if meaningless - anything and everything can be learned on the job, so I see little point in over-specializing in school.
It's not like those demonisations were unfounded. At Iwo Jima, the Japanese fought to the last man - a successfully invaded, heavily defended island yielded double-digit numbers of POWs. Even on the Eastern Front in Europe, this was unheard of. At Okinawa, Japanese citizens threw themselves off cliffs in the face of advancing American forces. The message was clear - this people, at this time, will not ever surrender unless faced with certain destruction.
An invasion of mainland Japan would not only have caused the deaths of at least 250K US soldiers, it would have decimated the population of Japan. The atomic bombings were a blessing in disguise - they gave the Japanese a way out without facing anihilation - even the emperor and his advisors realized they faced certain defeat in the face of such weapons, regardless of how hard they fought and resisted.
And when you consider the atrocities that were taking place in China, Korea, and Indochina, the numbers killed in the atom bombings would've been more than surpassed in a few months time by the Imperial Japanese Army.
While the MIPS thing is much less a problem than it used to be, storage space will always be scarce in high-performance embedded applications (like comm. DSPs).
Upping the clock-rate doesn't cost much, but for high performance, all the memory needs to be on die, and more memory means larger die-size, which means higher cost, which is bad. The chips I work with can do well over a couple BIPs (all engines totalled), but has only 32K of addressable on die memory (of varying word-sizes) and that memory is still around 85% of the total die-area. You could say there's some pressure to reduce the amount of memory we need, or at least keep it under control:)
Yow, you're gonna have a tough time if you ever work on very small systems.
I work on chips with often less than 1K instructions of code-space in some of the processing units and three-deep call stacks as a rule. A new function call is a major design decision and you had better be damn sure that's what you want to do - the overhead alone can make it untenable, not to mention running out of stack-depth.
That's embedded programming. There is no OS, there are no components. There's a chip and your code, and you had better know exactly what each of those things is gonna do under any circumstance. A black-box is the bane of your existance in that circumstance.
Not all the world is a PC with enormous amounts of memory and MIPS, and knowing how to deal with small systems should be part of any decent CS/SE education.
"Sometimes I wonder what would happen it I and my lab partner decided to drop out of college and start our own company in featuring natural language information systems. We probably would have failed anyway, but I was too young and stupid NOT to have tried."
I know how you feel:)
In my freshman year in college (1995), a good friend of mine suggested that we say goodbye to this bullshit and startup an online auction company (AFAIK, this pre-dated e-bay). He had written a script that converted the Usenet auction listings into HTML and had done some work on categorizing those listings. I was helping out for shits and giggles, but when he proposed it as a serious business opportunity, I explained how it would a) Most likely be illegal and b) Result in such bad karma we'd be reborn as houseflies.
Of course, the plan, as it stood, wasn't to just rip usenet listings but also to get a ative subscriber base and secure the transactions for a fee. I don't know how well it would've worked out, but three years later I felt like kicking my own ass...
In eastern China, anybody who wants to watch Red TV probably watches their own TV (or 2).
Once you get away from the coast, though, its pretty desperate rural poverty. I'd wager that most of the farming communities in central and western China don't even have a single TV in the entire village.
And Sunday is when I decided to stop being a physicist:)
And I totally agree about our hunter-gatherer brains not being able to grasp the nature of universe at fundamental levels. I studied physics, especially quantum stuff (two specialized classes in QED), and while I kinda-sorta udnerstand it, its mostly just mathemtical constructs that happen to be correct and consistent.
In contrast, I know work as a DSP programmer, a pretty complex field that uses a lot of classical physics. I feel that I can explain every signal-processing/digital-communications concept I'm familiar with to anybody without using anything more than (very) basic math and a whiteboard. With quantum physics, complex math is *everything* - there is no there, there - only math.
Richard Feynman made a valiant effort in his layman's version of QED, but it still misses some very fundamental points because he was unwilling (for the purposes of the book, and rightly so) to delve into the constructs that actually explain what's happening.
Of course, because all that gold served Spain so well.
What's that? Over 30 state bankruptcies in 200 years? Having half the world granted to them by papal decree and then having to surrender it due to lack of funds? Hey, if we can't pay our bills, the only thing to do is get more gold.../sarcasm
The gold standard suffers from the exact same inflationary problems as fiat money. If the government needs more gold, they'll just dig it up, like Spain did. And remember, gold-extraction technology is much more advanced now. Besides which, gold is completely unregulatable - if gold is money people are gonna dig it up on their own accord (unless you plan on shooting prospectors).
The US, and most of the world, was officially or not, off the gold standard by the 1930's. In the US, the myth of a gold standard applied until 1972, at which point there wasn't enough available gold in the world to properly represent the US economy, so the fiction was dropped.
Since 1935 or so (when the gold standard was mostly dropped), the world economy has grown more than it did in the preceding 1000 years. Having the money supply being restricted artificially by the availability of a particualr metal is like trying to swim wearing cement shoes.
The manager who hired me at my company did so only because I sent in a Space Invaders game written in straight 80x86 assembly. I had no degree (Junior drop-out) and little experience writing drivers (which was what the job was for), but my demonstration of technical proficiency was enough to get me the interview (whereupon he grilled me about the screen-refresh and keyboard handling algos I used).
Not my manager anymore - I've moved into the DSP side of things (very similarly at that - I was pissed when one of the DSP guys wouldn't fix a problem so I fixed it myself and then he brought me over) and people there are just shocked that I don't have a degree - everyone else has at least a masters and then there's dropout Tim who never took a signal processing course:)
I've been trying to get our hiring procedure revised - I think hiring smart people over those with degrees/experience would be a major boon, but I've encountered resistance. But I keep trying...
Well, it is an absolute cutoff*, but due to timing imperfection when playing the sample, the frequency response will tend to roll off a bit before 22050 Hz, which is what I think he was getting at. The better the timing, the less the roll-off before the Nyquist.
* Sorta. Since the interpolation (done implicitly by the output hardware) between two samples will unlikely be perfect, if you were to resample the output signal at a higher frequency, you would most likely see an alias of the signal at higher frequencies (side-lobes). A higher sampling rate would help with this problem (by doing the interpolation explicitly), but only marginally - not really worth it.
In Scenario 1, don't forget that the dummy is taking up the spot of a warhead on the launch vehicle, so if instead of sending 10 warheads, it might be 1 warhead and 9 dummies.
And besides, you can always just make more kill vehicles and get the dummies as well. If this causes the other side to build more dummies, which are cheaper than kill vehicles, they also have to build more delivery rockets, which are vastly more expensive than the kill vehicle, so the unwinnable arms-race is quite winnable.
Hehe, I grew up in PA and when I moved to Jersey, I constantly got introuble by trying to pump my own gas. One attendant threatened to call the cops, and then I figured it all out, there was no self-serve in this state (but incredibly, gas prices are pretty much the same).
Being a Western Pennsylvania farmboy, I found it to be quite wimpy:)
A lot of VLSI shops aren't big enough to buy their own supercomputer for simulation, so they make do by simplifying their sims sot that they'll finish before the heat-death of the universe on whatever hardware is available and hoping for the best when they tape-out.
This isn't going to change the little sims, everyone will still do that (it catches most of the problems) but if it were possible to spend $50K on some supercomputer time for doing a fully detailed sim before blowing a $1M on mask and fabrication costs, I bet a lot of companies would jump at the chance.
In otherwords, you wouldn't replace what you currently do with this, but rather add an occassional thing off your wishlist.
Yuppers, I did the same thing with some manufacturing-test software.
I like clean switch/case statements, even if only n options are even plausible, I still stick in a default so I can catch anything wrong. So I put in for several defaults that could never, ever, happen, a message along the lines of "What the fuck? How did I get here?"
Well, one of our customers got that message. I have absolutely no idea how, I looked for weeks and everything seemed clean (must have been a memory problem or something), but I got yelled at for it.
Note to self, remove stuff like that, you dumbass:)
But Microsoft has a massive cash reserve that keeps growing (and zero debt). ~$40 Billion last I checked, enough to cover all operating expenses for years with zero (!) revenue.
Cash, you can't lie about (and get away with). Either you have that much money in your account or you don't, period.
Debt is a little more slippery, and its easy to fudge the numbers (as Enron did) when you have lots of it, but again, MS has zero debt, meaning no outstanding loans or bonds or other weird debt financing scheme. Lying about having zero debt is impossible (again, to get away with).
They may play some games with revenue numbers, but it hardly matters when compared to their warchest. MS isn't going chapter 11 anytime soon.
But, as a DSP engineer, I do find myself reliant on Matlab, for two reasons:
1) Its damn convenient for plotting and analysing data. If, for instances I think my equalizer might be screwed up, I can dump the coefficients into a file, and plot them in matlab in both the time and frequency domains with great ease.
Compare that with the equivalent C code, and I think you'll find one of them is a much, much simpler way to check filter response:)
2) Our chips are programmed entirely in assembly and all the code is heavily optimized, but we still need to use compromised algorithms due to cycle and memory constraints, as well as dealing with the limitations fixed-point processing (loss of precision is a killer). In most DSPs, getting sqare-root and phase-offset results is a pretty dicey process, and you gotta hope the algorithm you're using is close enough because there is no 'good' algorithm that can reasonably be implemented.
Using Matlab, its easy to knock out an idealized version of an algorithm and compare the results with the compromised version that will actually run on the chip.
Is that a sick joke? According to their website, you too can have a car that can travel 100 miles bedfore needing a 6 hour charge-up, all for "a price somehwere between a Porsche and a Ferrari".
This is the problem with electric cars. Yes, you can make them fast, yes you give them a long range, yes you can make them cheap, but I have yet to see an electric car that does even two of those things (cheap, fast, long-range).
Slashdot Mirror
Anyway, despite the problems with power-control, China and Europe are adopting their own versions of CDMA for 3G services.
*walsh functions are orthogonal - by modulating the output signal by the assigned walsh function (different for every transmitter) and then scrambling the resultant signal by a pseudo-random sequence (to improve the average PSD), all of the signals are orthogonal to each other and do not interfere as long as the power is exactly the same for all of them (power differences screw up the orthogonality).
IANAWE (I am not a wireless engineer - though I am a wireline engineer and know the basics of CDMA and wireless tech), but AFAIK the problem is due to multipath delay (creating Rayleigh Fading/ Frequency-Selective Fading). The effect of Rayleigh fading is that while the signal may be fine right now and right here, if you move the phone (or something in the area moves) a half-wavelength to the left or right (at 900 MHz, about 6.5 inches. Less at higher frequencies), the signal may suck. Hard.
If you're moving quickly (say, in a car), this effect is ameliorated somewhat as you zip in and out of the fades and your average SNR isn't too bad. Standing still (slow-fading), however, the phone can't deal with being in a fade for that long and drops the call.
Results: for the same mix of calls (i.e. home and roaming), a European cellphone bill is 65-70% if not less than the equivalent US bill.
Talking with my European colleagues (French, Belgian, Dutch), they seem to think that Uservice in the US is cheaper if you know how to pick your plan.
For example, for $39.95/month, I get virtually unlimited calling (inside the US) during business hours (400 minutes/month) and unlimited minutes (again, domestic) at any other time. My home area is virtually the entire North East US (NJ, DE, NY, PA, all of New England and I think Maryland) and outside this area (roaming) its 10 cents/minutes. My cell-phone bill only ever exceeds the minimum (around $44 after taxes and such) when I travel outside my home area, despite having around 500-1000 minutes/month of calls.
My company's field engineer in the Netherlands, on the other hand, pays a reasonable rate in the N.L. (I forget exactly what, but it wasn't too bad), but when he travels to Germany or Belgium, its 2 Euro/minute. Anyway, he says that his phone bill is usually over 100 Euro/month.
Tim
don't see why everyone thinks we will have a war with China. It's pretty ridiculous. We are on very excellent terms right now, and our economies are too interdependant on each other to go to war. Taiwan isn't something we'd risk a nuclear war over. It is not that important to the US. Sure, our government has said that we'd defend Taiwan against and invasion, but that's just a bluff. In the event that China did invade Taiwan and the U.S. did threaten war, China would probably back down. They aren't braindead. The Chinese wouldn't risk a losing nuclear war with the U.S. Nuclear war is a bad thing that nobody wants. Notice that despite all of their problems and differences, the Cold War never went nuclear.
:)
War over Taiwan is an absurd thought - epeciually for the PRC. The Chinese Navy couldn't cross a puddle. Taiwan, on the otherhand, has been buying fairly modern US cruisers and destroyers as well as maintaining one of the world's most elite airforces (yes, China has the largest airforce, but much of it is Mig 15's and 19's, which don't really count). The only reason the PRC would try to invade Taiwan is if they wanted a cheap way to scrap their fleet
Tim
Yeah, but you can't mass-deploy 100 Mbps ethernet like that - the crosstalk would kill you. Ethernet is spectacularly inefficient in the way that it uses bandwidth (though GigE is much better from what I can tell) and the spectral compatibility with repsect to self XT (not to mention other comm protocols) is terrible. But that's not what ethernet is for - ethernet was designed to be simple and cheap and fast and short-ranged.
:)
And to tell the truth, this particular VDSL implementation isn't going anywhere. QAM is going to get clobbered when it comes to standardization by DMT. QAM VDSL is made up of 3 or 4 massive downstream channels and a similar number of upstream channels interleaved in a frequency-domain-duplex fashion, with each channel having a different symbol rate (somebody correct me if I'm wrong, but I just can't see any other way to do this with QAM unless the channels all have the same bandwidth, which I don't believe they do). This means that you need separate analog receivers for each of the channels - yuck. And QAM is also disadvantaged by the fact that these massive channels have to deal with interference by notching out the offending frequencies in their equalizer, and given limitations on the number of taps you can have in an equalizer at these kind of symbol rates, the notches are going to either be inneffective or way too wide.
DMT VDSL, on the otherhand, is effectively 4096 (or 8192) separate QAM channels, each 4.3125 KHz wide. This means that a) Everything runs at the same symbol rate, greatly simplifying frequency-domain-duplex arrangements and b) If there is interference at a particular frequency, you just load less bits there (or shut it off completely), and this can be done with a 4.3125 KHz resolution.
A) means DMT modems will be cheaper and B) means they will deal more gracefully with noise which in turn means it can be more widely deployed. Both those things mean more money for the telco, and telcos like money
Tim
Similarly, but from the other end, I work at a communications DSP company and I'd estimate average productivity to be around one line of production code per day per worker. Once the main development effor is complete on a project, problems tend to be quite obscure and can, in extreme but suprisingly common cases, take thousands of tests to duplicate, examine, and debug.
This means a lot of time spent doing absolutely jack shit. Fortunately, we have understanding managers who understand this. However, their managers and the marketing folks they interact with do not. "Why is [whoever] sleeping on the couch?!? We have to solve this problem for [big customer]!" is a common phrase heard around here.
Some people don't sleep on the couch waiting for the test to trap the problem - they look busy, and they get the respect from marketing guys for working all out to solve the problem. Some of us consider automated test-time to be nappy time - we get the disdain of the marketing guys. Fortunately, the end result, when it comes to solving the problem is similar and the engineering managers recognize this and couldn't give a shit what the marketing guys say.
With a different management, though, I imagine this could be quite a problem - what the fuck am I supposed to be kept busy with while waiting for obscure problem #129 to show up so I can debug it? They sure as hell don't know, but to some people, if I'm not working *really* hard 40 hours/week, well, why bother employing the guy.
Tim
Well, that's... interesting from my perspective. Can't say that I'd like it much, but interesting none-the-less.
I'm a geek, but I really dug metal-shop and auto-shop. In fact, if I have anything good to say about the US education system, it would be that it allows you to do damn near anything in high-school and still go to college and do damn near anything in college and still get the job of your choice.
In high-school, I did take a computer course, and I loved it, but I also took auto-shop and metal shop and really liked that as well. In college, I studied physics and took a four year lab practical in alcoholism, and somehow managed to get my current job as a DSP guy. The ducations were nice, if meaningless - anything and everything can be learned on the job, so I see little point in over-specializing in school.
Tim
It's not like those demonisations were unfounded. At Iwo Jima, the Japanese fought to the last man - a successfully invaded, heavily defended island yielded double-digit numbers of POWs. Even on the Eastern Front in Europe, this was unheard of. At Okinawa, Japanese citizens threw themselves off cliffs in the face of advancing American forces. The message was clear - this people, at this time, will not ever surrender unless faced with certain destruction.
An invasion of mainland Japan would not only have caused the deaths of at least 250K US soldiers, it would have decimated the population of Japan. The atomic bombings were a blessing in disguise - they gave the Japanese a way out without facing anihilation - even the emperor and his advisors realized they faced certain defeat in the face of such weapons, regardless of how hard they fought and resisted.
And when you consider the atrocities that were taking place in China, Korea, and Indochina, the numbers killed in the atom bombings would've been more than surpassed in a few months time by the Imperial Japanese Army.
Tim
While the MIPS thing is much less a problem than it used to be, storage space will always be scarce in high-performance embedded applications (like comm. DSPs).
:)
Upping the clock-rate doesn't cost much, but for high performance, all the memory needs to be on die, and more memory means larger die-size, which means higher cost, which is bad. The chips I work with can do well over a couple BIPs (all engines totalled), but has only 32K of addressable on die memory (of varying word-sizes) and that memory is still around 85% of the total die-area. You could say there's some pressure to reduce the amount of memory we need, or at least keep it under control
Tim
Yow, you're gonna have a tough time if you ever work on very small systems.
I work on chips with often less than 1K instructions of code-space in some of the processing units and three-deep call stacks as a rule. A new function call is a major design decision and you had better be damn sure that's what you want to do - the overhead alone can make it untenable, not to mention running out of stack-depth.
That's embedded programming. There is no OS, there are no components. There's a chip and your code, and you had better know exactly what each of those things is gonna do under any circumstance. A black-box is the bane of your existance in that circumstance.
Not all the world is a PC with enormous amounts of memory and MIPS, and knowing how to deal with small systems should be part of any decent CS/SE education.
Tim
"Sometimes I wonder what would happen it I and my lab partner decided to drop out of college and start our own company in featuring natural language information systems. We probably would have failed anyway, but I was too young and stupid NOT to have tried."
:)
I know how you feel
In my freshman year in college (1995), a good friend of mine suggested that we say goodbye to this bullshit and startup an online auction company (AFAIK, this pre-dated e-bay). He had written a script that converted the Usenet auction listings into HTML and had done some work on categorizing those listings. I was helping out for shits and giggles, but when he proposed it as a serious business opportunity, I explained how it would a) Most likely be illegal and b) Result in such bad karma we'd be reborn as houseflies.
Of course, the plan, as it stood, wasn't to just rip usenet listings but also to get a ative subscriber base and secure the transactions for a fee. I don't know how well it would've worked out, but three years later I felt like kicking my own ass...
Tim
Not quite.
In eastern China, anybody who wants to watch Red TV probably watches their own TV (or 2).
Once you get away from the coast, though, its pretty desperate rural poverty. I'd wager that most of the farming communities in central and western China don't even have a single TV in the entire village.
Tim
And Sunday is when I decided to stop being a physicist :)
And I totally agree about our hunter-gatherer brains not being able to grasp the nature of universe at fundamental levels. I studied physics, especially quantum stuff (two specialized classes in QED), and while I kinda-sorta udnerstand it, its mostly just mathemtical constructs that happen to be correct and consistent.
In contrast, I know work as a DSP programmer, a pretty complex field that uses a lot of classical physics. I feel that I can explain every signal-processing/digital-communications concept I'm familiar with to anybody without using anything more than (very) basic math and a whiteboard. With quantum physics, complex math is *everything* - there is no there, there - only math.
Richard Feynman made a valiant effort in his layman's version of QED, but it still misses some very fundamental points because he was unwilling (for the purposes of the book, and rightly so) to delve into the constructs that actually explain what's happening.
Tim
Of course, because all that gold served Spain so well.
/sarcasm
What's that? Over 30 state bankruptcies in 200 years? Having half the world granted to them by papal decree and then having to surrender it due to lack of funds? Hey, if we can't pay our bills, the only thing to do is get more gold...
The gold standard suffers from the exact same inflationary problems as fiat money. If the government needs more gold, they'll just dig it up, like Spain did. And remember, gold-extraction technology is much more advanced now. Besides which, gold is completely unregulatable - if gold is money people are gonna dig it up on their own accord (unless you plan on shooting prospectors).
The US, and most of the world, was officially or not, off the gold standard by the 1930's. In the US, the myth of a gold standard applied until 1972, at which point there wasn't enough available gold in the world to properly represent the US economy, so the fiction was dropped.
Since 1935 or so (when the gold standard was mostly dropped), the world economy has grown more than it did in the preceding 1000 years. Having the money supply being restricted artificially by the availability of a particualr metal is like trying to swim wearing cement shoes.
Tim
Technical writing is a very different skill than being able to write ficiton
:)
Not at my company
Tim
Ahh, one of the good guys :)
:)
The manager who hired me at my company did so only because I sent in a Space Invaders game written in straight 80x86 assembly. I had no degree (Junior drop-out) and little experience writing drivers (which was what the job was for), but my demonstration of technical proficiency was enough to get me the interview (whereupon he grilled me about the screen-refresh and keyboard handling algos I used).
Not my manager anymore - I've moved into the DSP side of things (very similarly at that - I was pissed when one of the DSP guys wouldn't fix a problem so I fixed it myself and then he brought me over) and people there are just shocked that I don't have a degree - everyone else has at least a masters and then there's dropout Tim who never took a signal processing course
I've been trying to get our hiring procedure revised - I think hiring smart people over those with degrees/experience would be a major boon, but I've encountered resistance. But I keep trying...
Tim
Well, it is an absolute cutoff*, but due to timing imperfection when playing the sample, the frequency response will tend to roll off a bit before 22050 Hz, which is what I think he was getting at. The better the timing, the less the roll-off before the Nyquist.
* Sorta. Since the interpolation (done implicitly by the output hardware) between two samples will unlikely be perfect, if you were to resample the output signal at a higher frequency, you would most likely see an alias of the signal at higher frequencies (side-lobes). A higher sampling rate would help with this problem (by doing the interpolation explicitly), but only marginally - not really worth it.
Tim
In Scenario 1, don't forget that the dummy is taking up the spot of a warhead on the launch vehicle, so if instead of sending 10 warheads, it might be 1 warhead and 9 dummies.
And besides, you can always just make more kill vehicles and get the dummies as well. If this causes the other side to build more dummies, which are cheaper than kill vehicles, they also have to build more delivery rockets, which are vastly more expensive than the kill vehicle, so the unwinnable arms-race is quite winnable.
Tim
Hehe, I grew up in PA and when I moved to Jersey, I constantly got introuble by trying to pump my own gas. One attendant threatened to call the cops, and then I figured it all out, there was no self-serve in this state (but incredibly, gas prices are pretty much the same).
:)
Being a Western Pennsylvania farmboy, I found it to be quite wimpy
And you're not the only cool Tim.
Tim
The chip guys. In fact, especially the chip guys.
A lot of VLSI shops aren't big enough to buy their own supercomputer for simulation, so they make do by simplifying their sims sot that they'll finish before the heat-death of the universe on whatever hardware is available and hoping for the best when they tape-out.
This isn't going to change the little sims, everyone will still do that (it catches most of the problems) but if it were possible to spend $50K on some supercomputer time for doing a fully detailed sim before blowing a $1M on mask and fabrication costs, I bet a lot of companies would jump at the chance.
In otherwords, you wouldn't replace what you currently do with this, but rather add an occassional thing off your wishlist.
Tim
Yuppers, I did the same thing with some manufacturing-test software.
:)
I like clean switch/case statements, even if only n options are even plausible, I still stick in a default so I can catch anything wrong. So I put in for several defaults that could never, ever, happen, a message along the lines of "What the fuck? How did I get here?"
Well, one of our customers got that message. I have absolutely no idea how, I looked for weeks and everything seemed clean (must have been a memory problem or something), but I got yelled at for it.
Note to self, remove stuff like that, you dumbass
Tim
But Microsoft has a massive cash reserve that keeps growing (and zero debt). ~$40 Billion last I checked, enough to cover all operating expenses for years with zero (!) revenue.
Cash, you can't lie about (and get away with). Either you have that much money in your account or you don't, period.
Debt is a little more slippery, and its easy to fudge the numbers (as Enron did) when you have lots of it, but again, MS has zero debt, meaning no outstanding loans or bonds or other weird debt financing scheme. Lying about having zero debt is impossible (again, to get away with).
They may play some games with revenue numbers, but it hardly matters when compared to their warchest. MS isn't going chapter 11 anytime soon.
Tim
LabView sucks :)
:)
But, as a DSP engineer, I do find myself reliant on Matlab, for two reasons:
1) Its damn convenient for plotting and analysing data. If, for instances I think my equalizer might be screwed up, I can dump the coefficients into a file, and plot them in matlab in both the time and frequency domains with great ease.
My script to do this is along the lines of:
load teq.dat;
plot(teq, 1);
teqfft=fft(teq, 512);
plot(20*log10(abs(teqfft(1:256))), 2);
Compare that with the equivalent C code, and I think you'll find one of them is a much, much simpler way to check filter response
2) Our chips are programmed entirely in assembly and all the code is heavily optimized, but we still need to use compromised algorithms due to cycle and memory constraints, as well as dealing with the limitations fixed-point processing (loss of precision is a killer). In most DSPs, getting sqare-root and phase-offset results is a pretty dicey process, and you gotta hope the algorithm you're using is close enough because there is no 'good' algorithm that can reasonably be implemented.
Using Matlab, its easy to knock out an idealized version of an algorithm and compare the results with the compromised version that will actually run on the chip.
Tim
Do the Sears Diehard batteries still have a lifetime guarantee?
:)
If so, the quick-recharge possibilities are great
Tim
Is that a sick joke? According to their website, you too can have a car that can travel 100 miles bedfore needing a 6 hour charge-up, all for "a price somehwere between a Porsche and a Ferrari".
This is the problem with electric cars. Yes, you can make them fast, yes you give them a long range, yes you can make them cheap, but I have yet to see an electric car that does even two of those things (cheap, fast, long-range).
Tim