Playing a Phonograph Record with a Camera
(mostly written 10/09/2010 11:49 AM, revised April 15, 2025)
Some years ago, I and the electrical engineer Ed Nisley, a
former IBM'er who was then a columnist at the late
Dr. Dobbs Journal, did what was almost a thought experiment, with
one piece of data, which I attach. A phonograph record is a
kind of irregular grating. If you photograph it the right way, you
can recover the sound track. The little light
blips in the picture are about three hundred microns long
(thirty pixels), corresponding to a frequency of 20 Khz, and you
will notice that they exhibit spectra from side to side. It
ought to be possible to translate those colors into a
wavelength-sized measurement. Here is a write-up I did
at the time, essentially summarizing Nisley's
experimental notes.
... we did some playing around with the question of whether it
might be possible to use a camera or scanner to rip old phonograph
records. Fifty-year-old books are in common use, but fifty-year-old
sound recordings are not. At this point, there are probably
more electronic cameras and scanners than there are
still-functioning turntables. At any rate, Nisley took the
attached photograph with a scanner. There were apparently both
reflections per se and diffraction effects. Here are his comments:
----------------------------------------------------------------
Ed Nisley:
I subscribe to AudioXpress and saw an article on the latest
iteration of a laser turntable. It's more subtle than I thought...
apparently the latest rev comes heartbreakingly close to actually
working.
They use two lasers directed at some (no doubt patented) angle
to the two sides, then peek at the various reflected beams to derive
tracking plus audio output. It's not done from the top down, it's a
sideways thing. They're quite proud that the entire audio chain is
pure analog.
Back of the envelope: radius = 70 cm near the lead-out, so
circ = 200 mm. At 33.3 rpm, speed is 200 * 33.3 = 6600 mm/s.
Wavelength at 20 kHz is 6600 / 20000 = 0.333 mm. But the
amplitude is pretty small: the stylus tip is something on the order
of 20 microns... call it a mil. That'd be one honkin' scan: 1000 dpi
over a 14-inch square!
There being nothing like a good new problem to take your mind
off all your old problems, I slapped an LP on the scanner, ate it at
2400 dpi (10 micron), and did some serious enhancing. Looks like
hell, frankly. Obviously you'd need better lighting and, I think, a
-much- better scanner: it claims to have 2400 dpi optical
resolution, but I'm not sure I believe them.
the first scan came from the top center of the "page" where the
tracks were horizontal... which turned out to be light black on dark
black with minimal contrast. Couldn't get any image out of it at
all, no matter how much contrast enhancement I applied. The
histogram was a Gaussian-oid peak down around 0.1 intensity (scale 0
-> 1.0) and putting the black & white points at the peak's
edges didn't produce anything interesting.
The rest of the image was similar to that, except for two
radial bands of gold-colored reflections at 4 & 8 o'clock, where
the illumination was enough off vertical to create reflections.
Soooo, the picture you saw came from about 8 o'clock. I betcha
the "white" parts are the lands and the black parts are the grooves,
with reflections from the audio ripples in the edges. The histogram
was much broader and it didn't take much tweaking. I did apply a
mild sharpening filter.
I just stuck it under my stereo zoom microscope and, yup, that's how
it plays. The bright lines are the far side of the grove, the dark
lines are the lands, and the near side shows up as a modulation of
the brightness. I had to close one eye: the difference in angle was
enough to completely change the picture and I couldn't get visual
fusion.
The real problem is that an LP is wider and taller than a
letter-size page, so you can't scan the whole thing in one pass.
That means stitching four scans together, unless you decide to build
a new scanner, which sort of negates the overwhelming coolitude of
the original idea.
If you were going build a custom scanner, I'd go for rotating the
disk over a linear array with side illumination... but then you're
right back on the wheel of incarnation. You'd want to move a
magnified track image over a short array and have to re-invent disk
tracking. Eventually you'd be down to a single optical sensor
hovering over one track...
Mmmm, two TV images through a stereo microscope?
Turns out I lived inside this problem for a few years, quite some
decades ago: I did the track following code for the doomed IBM Video
Disk project. The goal was to put a one-micron beam on a one-micron
track while it wobbulated about 150 microns at 60 Hz, based on
sparse error data from the previous revolution. Bottom line: it
worked! The project crashed for (many) other reasons, but I could
align to a spiral track from a cold start and lay the beam on the
sensor every time.
Put me off of video for many moons, as we had only one tape cleared
by IBM legal for re-recording. I bet you didn't know ABBA's blonde
lead singer had (has?) really, -really- bad dentition...
I remain in awe of DVD tracking.
Neither of us had time to pursue this
adequately, so we decided to make a present of it to anyone who is
in a position to develop it. Nisley said:
"I hereby sign over all right, title, & interest to you. Should
you get insanely rich, send me a check with one significant figure
and a non-negative exponent... which would be tough on a GPL
project, methinks."
And I likewise assign all rights, etc. to the public, subject
to the same proviso. Feel free to pass this on to
whoever you think might be interested.
=========================================================
I sent the proposal off to various people, including a very
arrogant young man at Lawrence Berkeley Laboratories,
who lectured me for not appreciating the moral virtues of
using a $40,000 piece of equipment (grant-funded). At any rate, it
turned out that some undergraduates in
Sweden had experimented with the idea, but with
monochrome images, effectively throwing away the
diffraction data.
http://www.phys.huji.ac.il/~springer/DigitalNeedle/
http://www.s3.kth.se/kurser/2E1366/students/03/lightblue/index.html,
now:
https://web.archive.org/web/20071026054709/http://www.s3.kth.se/kurser/2E1366/students/03/lightblue/index.html
The Prince of Berkeley knew about this work, and provided me
with the above references, but he had apparently not
appreciated the implications. His approach was to ignore how much
the Swedes had done, because their first attempt was not
perfect. Of course, like most undergraduate projects, this one ended
permanently when the term ended, and the team members went their
separate ways. What the Prince of Berkeley would think
about a spectrometer with a frame of construction
paper...
At any rate, illumination by a single LED might be a more
elegant approach to photographing phonograph records.
Assuming you could get a half-silvered mirror reasonably cheaply,
you could, effectively put both the light source and the
camera on a line running perpendicularly through the
phonograph record's center hole, so that every ray-trace would be
perpendicular to the record tracks.
I’d say you could probably build a set-up
for less than a hundred dollars, the most expensive item being a
half-silvered mirror, which seems to be about thirty bucks on
Amazon.
https://a.co/d/8jyAzKp
You can get quasi-disposable LED Flashlights, with zoom
lenses, in packages for about two dollas each:
KunHe 6 Pack Small Mini LED Flashlight Single Mode Zoomable
Flashlights AA Battery Powerful Flashlights with Pocket Clip
Portable Bulk EDC Pen Flash Light for Christmas Gift
https://a.co/d/bwmIjeJ
I had qyuite forgotten about this work,
when I was reminded of it by a thread on Twitter/X,
https://x.com/kristine_froeba/status/1911619651907240271?s=43&t=Q2ZgMkIJpRJLdXkxvFU-CA
I dug out the old papers, put them on the website, and posted a
link.
My comments:
If you go about it the right way, it’s possible to read a vinyl LP
with a camera. You can treat the LP as a kind of irregular
diffraction grating, and rig up an apparatus with a half-silvered
mirror.Then you have a program which analyzes results, and generates
a sound track. A friend did some experimental work with a flatbed
scanner some years ago, and was able to optically extract the record
pits and grooves, but the half-silvered mirror would be better
because it allows both the light source and the camera to be one the
centerline of the record, so you don’t have to splice multiple
images. Here is a bit of experimental evidence, taken by my friend
Ed Nisley ( late of Dr. Dobb’s Journal). The blips are at
approximately 20 KHz.
[see image above]
Andrew D. Todd
1249 Pineview Dr., Apt
1
Morgantown, WV 26505
a_d_todd@rowboats-sd-ca.com
andrew2david2todd@iCloud.com
http://rowboats-sd-ca.com/
Index Home