Recording digital data to analog tape... revisited

[H. Peter Anvin]

Followup to: <[email protected]>

Problem is, he doesn't have access to the raw video.

But I been thinkin' - a freaking submarine has got to be a pretty
high-budget project, wouldn't you think? I think it'd be worth it to hack
into the stupid camera, and probe it, and figure out where to inject the
danged telemetry signal on top of the video. It will be only one point,
after all. You might want to pick up sync, so two. Not counting ground.
Well, you get the idea.

The nice thing with using the audio tracks presumably is that the
telemetry and picture get synchronized. Otherwise I'd say use
solid-state storage... much more reliable.

These days, if I'd do something similar I'd use a (good) webcam and
record everything including video to hard disk.

-hpa

 

[Rich Grise]

I agree wholeheartedly with your desire to keep the camcorder intact.
This is as a result of a few attempts to repair VCRs----which are
mechanically much larger. The idea of loosening a screw on the

I think a similar approach would work on an audio recorder---but
probably at a lower baud rate. I used an SX chip from UBICOM as a
modulator/demodulator. For lower baud rates, a PIC would probably
do the job with less power dissipation.

OK, since you put it that way - we're back to "how to put telemetry
on an audio tape." Remember, it's a camcorder, so it might have pretty
good audio response - probably better than 10 KHz. How much telemetry
is there in a sub, for heaven's sakes? How fast are you expecting
stuff to change? Just make an ascii stream and put it through a 9600
BPS modem chip in half-duplex. Do you really need more than 960
characters per second of data?

Thanks,
Rich

 

[Tom]

OK, since you put it that way - we're back to "how to put telemetry
on an audio tape." Remember, it's a camcorder, so it might have pretty
good audio response - probably better than 10 KHz. How much telemetry
is there in a sub, for heaven's sakes? How fast are you expecting
stuff to change? Just make an ascii stream and put it through a 9600
BPS modem chip in half-duplex. Do you really need more than 960
characters per second of data?

Back in the 80s I did a small project that would record data onto a regular
stereo tape deck. Put the data on the left channel and the clock on the right
channel. If I remember correctly all I did was use a resistor divider and
capacitor to go directly from TTL to audio line level for recording. For
playback, I had a pair of op-amps for converting from line level back to TTL.
I had to massage the data a little bit before recording in order to keep the
DC balance near 0 but otherwise it worked fine for low bitrates, it was pretty
immune to tape stretch and dropouts.

Obviously if your camcorder isn't stereo then this won't work.

--Tom.

 

[Lewin A.R.W. Edwards]

OK, since you put it that way - we're back to "how to put telemetry

on an audio tape." Remember, it's a camcorder, so it might have pretty
good audio response - probably better than 10 KHz. How much telemetry

Guys! Gals! Disguised eel-monkeys! This is a dead issue!!

1) The project has two forks. Fork A is to record a telemetry stream
on the audio track of the camcorder. Fork B is to use the same analog
output to record the same stream on a regular cassette recorder for
applications where the user is using this same module to control a DSC
or electrically-operated automatic film camera. Fork A already worked
nicely when I started this discussion. My problem with Fork B turned
out to be a stupid interconnect issue (albeit with really weird
symptoms).

2) I have a hardware and firmware solution that works well, thanks for
all the responses. I am doing the decode on a PC at the moment but
will eventually build a little box that feeds the playback signal
through a slicer and passes it to a digital input on a micro driving
an LCD.

stuff to change? Just make an ascii stream and put it through a 9600
BPS modem chip in half-duplex.

You missed a few steps, such as "buy a 9600 BPS modem chip that is
readily available to everybody who might read about this project, and
will remain so for some time" and "build the support circuitry for a
9600 BPS modem chip" (and then "work out how to decode the result
easily"). Everything I need to do can be done in an 8-bit micro with
plenty of horsepower to spare, and the resulting on-tape format is
really, really easy to decode. Plus it uses off-the-shelf
general-purpose parts.

Do you really need more than 960
characters per second of data?

No. This is not a throughput issue and never was.

 

[CBFalconer]

.... snip ...

I've managed to get 230KB async from a UART through a kilometer
of intercom wire (simulating an oceanographic cable) by putting
out full cycles of pseudo sine waves for 1 bits and nothing at all
for zero bits. The signals were transformer coupled for impedance
matching and DC isolated so the cable could be used for HV power.

You should be able to go a lot further and faster with balanced
pairs, provided you avoid DC effects. The basic driver is a
differential pair, with a current source 2I in the emitter
circuit. The normal current mirroring methods can be used to
generate currents I from the positive rail into the collector
circuits of that pair, so that each output line now switches a net
+-I. There is only one switch, avoiding nasty crossover
distortions, which can make the receiver get slowly varying signals
just at its most sensitive point. The DC component of injected
signal is zero over the differential line pair. Other methods can
ensure the average on each half of the line is also at zero net DC.

The drive is then high impedance, so the line conditions depend
solely on the line termination. In particular any biases set for
receivers are set for the overall line, in just one place, with
current injection. Most cables have well defined impedances, in
particular almost all twisted pair telephone cable looks like 100
ohms or so at 1 Mhz up. With the above configuration the single
line is a buss, and you can fairly freely add transmitters and
receivers, provided you can compensate for the varying point to
point propagation times. There are ways to handle this. One is by
defining a transmission direction and adding a master clock line.
Now all you have to do is put all transmitters to the left of all
receivers, assuming direction is left to right. No tuning.

 

[Rich Grise]

You missed a few steps, such as "buy a 9600 BPS modem chip that is
readily available to everybody who might read about this project, and
will remain so for some time" and "build the support circuitry for a
9600 BPS modem chip" (and then "work out how to decode the result
easily"). Everything I need to do can be done in an 8-bit micro with
plenty of horsepower to spare, and the resulting on-tape format is
really, really easy to decode. Plus it uses off-the-shelf
general-purpose parts.

Well, all I was really trying to say was "Make a data stream that you can
record", and it seems you've already done that.

Thanks for the clarification.

Cheers!
RIch

 

[Mark Borgerson]

You should be able to go a lot further and faster with balanced
pairs, provided you avoid DC effects. The basic driver is a
differential pair, with a current source 2I in the emitter
circuit. The normal current mirroring methods can be used to
generate currents I from the positive rail into the collector
circuits of that pair, so that each output line now switches a net
+-I. There is only one switch, avoiding nasty crossover
distortions, which can make the receiver get slowly varying signals
just at its most sensitive point. The DC component of injected
signal is zero over the differential line pair. Other methods can
ensure the average on each half of the line is also at zero net DC.

I'll look into that idea if I work with long cables again and have
actual twisted pairs to work with. I was looking at some of
the techniques used for DSL communications. I would love to
find a pair of DSL modems that accepted serial from the microcontroller
and produced DSL signals for coupling to the cable. Didn't seem to
be anything workable when I was looking about three years ago.

The drive is then high impedance, so the line conditions depend
solely on the line termination. In particular any biases set for
receivers are set for the overall line, in just one place, with
current injection. Most cables have well defined impedances, in
particular almost all twisted pair telephone cable looks like 100
ohms or so at 1 Mhz up. With the above configuration the single
line is a buss, and you can fairly freely add transmitters and
receivers, provided you can compensate for the varying point to
point propagation times. There are ways to handle this. One is by
defining a transmission direction and adding a master clock line.
Now all you have to do is put all transmitters to the left of all
receivers, assuming direction is left to right. No tuning.

My application was for an oceanographic cable, which can be quite
different from a telephone cable. In particular, I had to cope
with the fact that one of the wires might have about 300V DC
to ground (or the other wire). I've also heard, but not been
able to verify, that the cable characteristics change as the cable
is reeled out into the ocean. I do know that we had be be able
to cope with slip ring noise also--which was handled with ACK/NAK
and packet repeats.

Mark Borgerson

 

[CBFalconer]

.... snip ...

My application was for an oceanographic cable, which can be quite
different from a telephone cable. In particular, I had to cope
with the fact that one of the wires might have about 300V DC
to ground (or the other wire). I've also heard, but not been
able to verify, that the cable characteristics change as the cable
is reeled out into the ocean. I do know that we had be be able
to cope with slip ring noise also--which was handled with ACK/NAK
and packet repeats.

That sounds like moving power on the same copper. My approach
might be to use the common mode of a twisted pair to supply one of
the power potentials (including ground). Thus, if a 100 ohm line
is terminated by two 50 ohm resistors at each end, they look like
25 ohms per end to the common mode supply. The remote end can use
DC/DC converters to trade volts for amps. Redundant signal paths
mean redundant power paths with means of detecting partial
failures.

Changes in line characteristics look like taps on the line. The
system should be fairly immune to these when both ends are
terminated with a reasonable match, provided the taps are
reasonably high impedance. Obviously a short circuit doesn't let
anything past it.

You can even apply TDR (time domain reflectrometry) to the
installed line, and decide exactly what you have. You can then
compensate for its deficiencies or tell the maintenance people
where to look.

 

[Mark Borgerson]

That sounds like moving power on the same copper. My approach
might be to use the common mode of a twisted pair to supply one of
the power potentials (including ground). Thus, if a 100 ohm line
is terminated by two 50 ohm resistors at each end, they look like
25 ohms per end to the common mode supply. The remote end can use
DC/DC converters to trade volts for amps. Redundant signal paths
mean redundant power paths with means of detecting partial
failures.

Yes, power is often moved along the same copper. When towing
multiple instrument packages at the end of a kilometer of
cable, you can quickly run out of conductors. The package I
worked up was designed to allow 4 or 5 instruments with modest
data rates to mux their data together over the same pair of
conductors.

Changes in line characteristics look like taps on the line. The
system should be fairly immune to these when both ends are
terminated with a reasonable match, provided the taps are
reasonably high impedance. Obviously a short circuit doesn't let
anything past it.

You can even apply TDR (time domain reflectrometry) to the
installed line, and decide exactly what you have. You can then
compensate for its deficiencies or tell the maintenance people
where to look.

"installed line" is probably not the proper term for an armored cable
that gets winched in and out several times a day. They do use TDR
to locate breaks and pinches on the line, though. However repairing
a problem is extremely difficult---particularly 400 miles off the
coast of nowhere! Oceanographic cables often grow shorter over their
lifetime as new terminations are made or you get kinks over the
sheaves---particularly as you bring the package aboard. The worst---
and very expensive case---is a problem in the middle of a 1 or 2KM
cable. You will often see large spools of cable sitting in the
shore facilities. I sometimes wonder whether they are too short,
too few conductors, or simply too expensive to repair.


Mark Borgerson