Slashdot Mirror
The Journey of Radios From Hardware to Software
An anonymous reader writes "The New York Times is carrying a story all about the process of replacing radios with software. The article tells the tale of Vanu Bose, son of the man who started the Bose company, and his quest to bring software to what was previously a hardware-only enterprise. He met a lot of resistance in the 90s to his ideas, because processor technology was not up to the task. Now that technology has caught up with Vanu, his software (and other products like it) are increasingly replacing now-outdated hardware components. 'Well-established companies like Motorola and Ericsson now use elements of software-defined radio for their base stations. But Mr. Bose was the first to come to market with software that could handle multiple networks with the same equipment. Software radio appears to offer an elegant solution to what has been a vexing problem: how to have a single handset, like a cellphone, communicate across multiple networks. For instance, the G.S.M. standard, for global system for mobile communications, is used broadly in Europe, and most notably in the United States by AT&T.'"
It's amazing what can sometimes be done in software. You can make a simple AM-band transmitter using a microcontroller and two resistors -- with everything done in software. MCUs are fun!
Paleotechnologist and connoisseur of pretty shiny things.
It's not that a single software-defined radio is all that important. It's that you can do the transforms on the incoming waveform and then extract N different channels with one signal processing system. That's what's been making cellular base stations go for almost two decades. (All the hard work is on the receive side; transmission is easy.)
First generation cellular base stations (i.e. AMPS) had one big analog card per channel, each heavily shielded from its neighbors. The amount of hardware required was huge, and cell sites tended not to be fully populated with channel cards, so they were easy to overload.
Then things started to go digital, with combinations of analog and DSP components processing the signal. Both GSM and CDMA inherently assume digital processing, and in early systems, hard-wired special purpose components were used. As CPUs get faster, there's a steady trend toward using general purpose CPUs.
It's still rare to actually process RF directly in software. Usually, there's a local oscillator and mixer to down-convert the desired band to a working IF frequency, which is then digitized and processed. So it's only necessary to digitize at maybe 10-100MHz, not up in the gigahertz range.
For lower bands, though, a true software RF receivers are available. These just suck up everything from 0 to 30MHz and digitize it. An attached PC does all the hard work.
It's not quite a "duh" nor is it clear without apply a lot of brain-power. In an analog radio there is no "hard-wired logic", its just the physical properties of the components engineered to work with electromagnetic waves. Software radios are only possible because digital processors are cheap/small/fast enough to approximate what the analog components are physically doing by crunching some very clever math (FFTs) in real-time.
GNU Radio.
This is about radios in a variety of communication devices. Like cell phones. And cell phone towers. Especially cell phone towers. Not so much your Sony Walkman et al.
The World Wide Web is dying. Soon, we shall have only the Internet.
Something many people don't understand is that the Nyquist criterion applies to the bandwidth of the recovered signal, not to its carrier frequency. So if you want to recover a 10-kHz wide signal at 800 MHz, you don't need to sample at 1600 MHz... you just need to sample at 20 kHz, using an ADC with lots of front-end bandwidth.
That's an oversimplification, but it may be what you were thinking of.
Got no highs? Got no lows?
Only midrange! Must be Bose.
I may make you feel, but I can't make you think.
I think the title of this posting could be construed as a bit misleading, in that it says 'radio' but when you read the attributed article, they're talking about cellphones, not things like broadcast radio or other areas where RF transmission and reception are necessary. It may or may not be obvious to anyone, but there's no way that tuned RF circuits can be completely eliminated, at least if you're talking about over-the-air transmissions; you still need to at least provide amplification (which must be a tuned circuit) and impedance-match to your antenna (which again must be a tuned circuit).
The Joint Tactical Radio System (JTRS or 'jitters') is currently being tested by the US Army... it aims to be "everything in one box," and it was "originally planned to span a frequency range of 2 megahertz to 2 gigahertz. JTRS has been expanded to frequencies above 2 GHz to satisfy space communications requirements"
:)
That's a direct quote from the Wikipedia article (which looks like it's pretty accurate), located here: http://en.wikipedia.org/wiki/JTRS
I'm in the Army, and buddies of mine have played with it and can attest that "it's pretty cool"
Here's to the crazy ones
In theory, software (solid-state digital circuits) has huge advantages over hardware. Software offers extreme flexibility, no wear-and-tear, etc. If it worked as well as it should, in theory, there wouldn't be mechanical linkages just about anywhere, anymore. All the controls in your car would be electric, and a significant portion of your car's engine would be gone (no more cam shaft). The same would be true of most everything... If not replacing significant numbers of mechanical components, at least using software to precisely control it, and getting much better efficiency as a result. Yes, your refrigerator, microwave, etc. could all greatly benefit from software control.
There are just two big problems that have made software control a non-starter.
First is customization. Put a spring in the mechanics of an engine, and I can replace it with a shorter/longer/stronger/weaker spring. I can heat it up to weaken it, grind it down, etc... With software, you are given a black box, binary-only, with no documentation on how it works, and definitely no common interface to access and modify it. So every time car companies add another function to their cars' onboard computers, and take away mechanical systems, there's extreme resistance, as buyers know they're out of the loop, and if they want to adjust anything, or if something should go wrong, they can only take it to the select few company-blessed shops, which have paid the necessary bribes to get enough info to do just a few basic things with the onboard computer. And you're entirely screwed if you want more changes than that, because the company doesn't WANT you to, and without man millions of dollars on the line, you're not even a blip on their radar.
Despite what many believe, cost is almost never a problem. For low cost products, low-end micro-controllers can be found for pennies, and even cheaper are the basic I/O elements like thermistors, power meters, transistors, relays, etc. Yet even the dirt cheap processors sold today can do many millions of calculations per second, far faster than could be needed for damn near any products.
Second, and perhaps more important, is reliability. Computer hardware is EXTREMELY reliable. You can go buy a dirt cheap commodity CPU, RAM and MOBO, and be pretty damn sure it will run for 20 years without a SINGLE error. The only big exception to this is power supplies... a marginal one, not supplying enough power will cause a crash, but that generallyonly happens in the case of the cheapest no-name junk. What's more, go up a small step to a high quality MOBO, ECC RAM, redundant PSUs, UPS, etc., and you'll never ever see a hardware-induced glitch.
The reliability problem comes ENTIRELY from poor software, and mostly commonly available kernels, at that. People don't want to believe that, but the facts are that computers are 100% pure math machines, and math is 100% accurate. A computer will do exactly what you tell it to do, but most people are trying to program their computer through several million lines of indirection... If you write, in hex, a simple loop with a bit of processing, a computer will run it error-free, from here until doomsday, but programing a complex system in hex is much too hard, and human programmers aren't perfect enough to do so.
The only real possibility to ensure reliability with reasonable development time is something very much like a micro-kernel. You need a tiny bit (a few hundred KBs) of EXTREMELY-thoroughly audited code, that can very strictly manage memory, do strict input and bounds checking, carefully manage communications between independent modules of code, instantly tear-down and restart any bit of code which shows the slightest signs of an error, and also strictly ensuring real-time operation.
I'm not endorsing any product here. The fact is nothing like this exists. That is why we aren't seeing mechanical systems having components being replaced by software as quickly as they can be redesigned. Open source operating systems
Slashdot gets worse every day... Pipedot: News for nerds, without the corporate slant
If a job's not worth doing, it's not worth doing right.