[OSM-talk] Camera and Dictaphone clocks

[David Earl]

I mentioned in the thread about continuous audio mapping yesterday that 
the audio was appearing to drift away from the corresponding waymarks 
after a while. I have now determined that this is due to the inaccuracy 
of the clock in my dictaphone. I have implemented a solution for audio 
in JOSM - see below - but this has implications for using camera images 
as well.

I recorded three hours off the radio this morning to include the pips 
at 9am and noon (for non Britishers: Greenwich time signal, an accurate 
clock which is broadcast on the hour most hours on BBC Radio 4). A 
suitable alternative would be to track the (very accurate) GPS clock, 
but that would have meant going outside.

I then measured the length of the recording between the start of each of 
the long pips in Audacity, whose timer is derived from the audio 
sampling rate and therefore reflects the clock in the dictaphone. Sure 
enough it came out at a remarkably inaccurate 3h00m17s (0.157% out): 
that means that my dictaphone loses about 2 minutes day!

I would not be at all surprised to find that camera clocks are also 
similarly inaccurate - they probably use the same silicon! A similar 
test for a camera is obviously to photograph a GPS display several hours 
apart and compare the time difference the image time stamp says with the 
GPS time difference shown.

The consequence is that synchronising the time by saying "NOW" or 
photographing the GPS clock is _not_ sufficient. If you're using voice 
or pictures just to annotate waypoints, it's just a minor inconvenience 
- the picture or sound isn't quite where you thought it would be on the 
track, but the waypoint is from the GPS, so accurate.

But if you're using the synchronised timestamp or offset into a 
recording to imply a feature's location, you could be 50m or more out 
after two-hours of a bicycle mapping session (10s error @ 5m/s).

For the new audio facilities in JOSM, I've added a voice recorder 
calibration box into the Audio Preferences panel (will be in tomorrow's 
build). You measure the purported length of a long chunk of audio to the 
true time, and enter the ratio, so in my case this is 1.00157. You could 
do this specially as I did, or during a mapping session just by putting 
sync cues at either end of your recording. You can measure it either in 
Audacity or similar as I did, or in JOSM itself by inserting audio 
markers at the geographical location of the sync cues and noting the 
time time difference of labels (dG), sync at the first mark, and then 
time the difference between the second mark and the second audio cue in 
the commentary (dA). The calibration is then (dA+dG)/dG. Of course if 
your device clock runs fast, you might have to go back a few seconds, 
and your ration would be a bit less than 1.

Of course, this assumes that the clock runs at a constant, if incorrect, 
rate both during a session, and, if you don't recalibrate each time then 
between sessions as well.

David