Slashdot Mirror
Embedding Data Signals In White Noise
Anemophilous Coward writes "ZDNet has the following article which describes a company that 'has devised a method for sending wireless signals over ordinary audio speakers so that humans can't hear them. With this same technology, radio stations can unobtrusively transmit ads, Web site URLs, or information about music and artists to in-car cell phones.'" Here is some further reading about the company, Intrasonics.
Blow your mind
Others have already observed that it's not a frequency thing, but let me expand on that.
Frequencies are already optimized for human hearing, and it's not usually possible to send, say, a 40,000Hz signal on most anything you can think of, analog or digital. Standard phones have a bandwidth of something like 3 K Hz. CDs of course top out around 20,000Hz, give or take a bit. (It's not a perfect cutoff at 22050, it's a curve, so there isn't quite a point you can say is "the limit".) I don't know for certain but I'd bet FM can't transmit those frequencies and be compliant with FCC regulations. (Of course the tech could do it in theory, but the radio station may have to leave their allocated frequencies to do it; I don't know for certain.) AM could do it in theory but based on the low quality of the signal I hypothesize that something is preventing high frequencies from getting through.
Finally, the coup de grace is that our speakers are optimized for human hearing, pretty much no matter what. Covering the bases from 20Hz - 20,000Hz is a hard enough problem without pushing the required range up another couple of octaves.
In fact, what the company is proposing seems to be in some sense the inverse of MP3 coding. MP3 coding strips the signal of things that you can't hear through by analysing what is psychoacoustically masked in the original signal. The MP3 encoding process can then focus on just the parts of the signal you do hear, which is obviously going to require less space, except in some pathological cases where the whole sound is perceivable (like a pure sine wave tone).
From what I understand of the marketing, the part of the signal that an MP3 encoder strips out is exactly where they would place their data. They can stick any data they want in there and we just plain won't hear it, but a computer+microphone doesn't have this problem.
Interesting corrolary: The time frame this will work in is limited, as digital transmission usually uses compressed audio, and the act of compressing the audio will preferentially eliminate this data. (Or does digital radio transmit an uncompressed stream?) They'd better get marketing this now, so that there's an installed base and they can try to later create receivers that will re-add their signal on the receiving side. Of course, if all anybody is using this for is advertising, I can't imagine we're going to go out of our way to buy "Advertising Enabled!" digital radio receivers.
In europe, there exists a similar technology called RDS for "Radio Data System". It's on the air for about 10 years now and allows for these cool features since then:
- Show the Station name in your radio display
- Show what's playing
- Certain stations are transmitted over several frequencies. RDS knows the alternative frequencies of your stations and automatically switches to the best frequency
Check out the Portable People Meter from Arbitron. It can recognize subaudible watermarks in music including over radio, Musak, and even some streaming audio compressions. Arbitron uses it for ratings purposes.
Of course, then there is IBOC from Ibiquity which is an on-channel digital enhacement for AM and FM signals, part of which could be used for datacasting, as part of most DTV signals will.
22050 is an absolute upper cutoff in the encoding system, and in theory you could encode a 22,049 Hz signal into Red Book audio. I said it's not a perfect cutoff because it's impossible to build an audio filter to cut off precisely at 22050 Hz, which is beyond the scope of a Slashdot comment; please consult a signal processing resource on how signal filtering works.
;-) I just like to be precise when possible, and prefer practical realities to theoretical ones, which is why I'm in Engineering college.
At the recorder, you must cutoff signals over 22050 or risk the horrible problem of aliasing (again, out of scope of a Slashdot post but pretty interesting). Since you can't have a perfect cutoff filter, you generally can't record 22049Hz signals except with extreme attenuation (in the specific case of 22049, it will well below the noise floor). Generally, when the CD players re-construct the sound, they will also do some filtering as a side-effect of how they do it. So you can't generally play back a 22049Hz signal either, even if you directly encode it onto a CD.
So while you can encode it, you can't record it directly and you can't play it back, so in a very real practical sense, 22049Hz is not usable with CDs. And so on and so forth for the other frequencies between 20000 and 22050. It's a smooth curve (and not necessarily the same one for two pieces of equipment, though my impression is that they have standardized somewhat because it's cheaper that way), so in a real-world CD recording and playback application, in a very real way there's no particular cutoff frequency you can directly point at, even though there's one in theory.
In general, it's a pretty pedantic point.
This, by the way, is part of the reason that CD's sample at 44100, instead of 40000. 40000 would be somewhat more efficient with the storage medium, but you'd have problems with the fact that you have no room to filter out the higher frequencies without hitting "good" ones as well. There are other concerns too, that's not the whole story, but it is a very significant part of it. In fact that goes for this whole post; I'm skimming over a lot because this is only a Slashdot post. (Like "20-20,000 is only a convenient fiction", the exact way filters behave, etc.)