Even though audio tape is referred to as an "analog" medium, the individual magnetic domains are almost all in one of two states, just as most of the points in a newspaper photograph will either be black or white, and for basically the same reason: while it is in theory possible to produce grays by printing a 'medium' amount of ink on an area, it's very difficult to produce a consistent gray that way; printing a black and white pattern of dots costs some resolution, but makes it much easier to maintain consistency. Similar issues exist when recording signals on a magnetic tape.

Normal analog playback devices work by measuring the average polarization of the magnetic domains. The behavior is roughly analagous to copying a newspaper photograph by making a somewhat 'blurry' scan of it. This approach will yield decent, albeit not stellar, results if the photograph is in perfect condition. If the paper is distorted or discolored anywhere, however, or if the ink has faded in some areas, such defects will show up in the copied image.<p>

What I would suggest doing with audio tapes would be more akin to taking a high-resolution scan of the newspaper photograph and using the known regularity of the halftone pattern as a spacial and density reference. Any defects caused by deformation, discoloration, or fading could be corrected provided such distortions were not so severe as to prevent the scanner from seeing where the dots were. Even when the distortion was so severe as to prevent reading the dots, this would often be detectable, and this could be used to generate a mask for areas to be "reconstructed".<p>

What I would propose doing would be to have a player for analog tapes operate on this same principle. Measure the domains indidually and use such measurements to adjust playback speed and volume in real-time.

To begin playback, the player measures the bias frequency and amplitude when running the tape at 1.875"/sec; it also starts putting measured signals into a buffer. Once the player has enough data captured (probably about a second), it could then start playing the buffer, normalizing it for a fixed bias frequency and amplitude. If, in part of the tape, the bias frequency were to drop 1% and then increase back to normal, the player would output the samples thus recorded slightly faster than it had read them off the tape (since the reason for the change would most likely have been that the tape stretched slightly there). Likewise, if there was a spot where the bias signal dropped in amplitude 10%, the player would boost the signal on playback so as to keep a uniform bias signal (since that would most likely indicate a bad spot on the tape).

Using this approach, it should be possible to get high quality playback even of tapes which have been damaged or, in some cases, which were recorded on marginal equipment or tapes. Indeed, it may be possible to make 20-year-old tape sound better than it did when it was new.
-- supercat, Jul 06 2004

supercat: The bias frequency cannot be detected on playback. It is a above-hearing (supersonic) signal added to linearise the recorded signal.
-- neelandan, Jul 07 2004


Is it still detectable (electronically) on the tape though?
-- half, Jul 07 2004


I'm not sure but isn't this what happens when Hi-8 video is played on Digital-8 machines? (I know that the audio on Hi- and standard- 8 video is digital, but the video signal is analog.)
-- angel, Jul 07 2004


half: No.
-- neelandan, Jul 07 2004


[half]: To expand on [neelandand]'s somewhat terse reply, the bias signal simply 'resets' the magnetic domains to their default state.
-- angel, Jul 07 2004


I wish I had gone to the same finishing school as [angel]
-- po, Jul 07 2004


So do I.
-- angel, Jul 07 2004


The bias frequency is on the order of 80Khz to 170Khz. Although a 'normal' playback head would have to wide a gap to read such signals very well, I would not think it should be particularly difficult to construct a head with a gap narrow enough to read such a signal.
-- supercat, Jul 07 2004


I like it, but think it may be a bit ambitious to go for the magnetic domains themselves as they are really tiny. I've no idea quite how tiny, but a little googling for a number give me 60-100nm. Assuming we can read something that small without Heisenberg, Shrödinger or limits of current technology getting in the way, we still have to store the enormous amount of data that comes from this.

Each domain has an orientation so we need more than a bit to describe it (plus, they're nor laid out in a nice grid which may complicate things). Say we can describe each domain with a byte, and we assume the size of a domain to be a square 100nm on a side.

Area density of domains = 10^14 m^-2
A strip 1cm by 2mm (say a cm of tape) would contain 2 x 10^9 domains.
This would take 2 x 10^9 bytes or just under 2 terabytes to store.

I still like the idea, though. (btw, can someone check my maths?)
-- st3f, Jul 07 2004


It would not be necessary to detect the individual domains, any more than it would be necessary to detect individual molecules of ink to scan and enhance a newspaper photograph. What is needed is sufficient resolution to resolve the biasing wave; since that's only an order of magnitude higher in frequency than the signals of interest, I would think that should be quite feasible.
-- supercat, Jul 07 2004



random, halfbakery