D-link DCS-2000 audio problems

    I am looking for a web cam (IP camera) to use as an aid for
    our son in studying music.

    The basic requirements are:
    - IP camera (room where piano is has no PC)
    - decent audio (more important than video)
    - decent video quality (not too fussy here, low FPS is OK)
    - can record both the video *and* the audio streams
    - ideally, complete with microphone
    - wireless not important (but possible option)

Device information: H/W version is A3, firmware version: 1.07 (was 1.05 initially, updated from .ca site)

The video is OK. I realize at any rate that high quality video means a lot of disk space (we'd want to record a lesson, then listen to it later). Fine, no issue here: we're running in B&W, 3 FPS.

But the audio was simply horrible in quality. An old analog phone does much better. I tried various parameter settings (both in camera and in DVI export), but to no avail. The bad quality was equally audible in IE and thru IP surveillance (monitor and playback).

Some research revealed a few things about this camera that aren't on D-link's web site (but nonetheless interesting):

• The audio stream (24kbps) is G.722.1.
• This camera is very similar to the Vivotek IP3132 camera.
• The protocol over the wire is not RTP or related RFC protocol (unfortunately). I still haven't figured out what it is (the part that happens after HTTP, ports 5001-5003).

• There were glitches in the waveform (of a recorder -- quite close to sine waves, so easy to analyse) that appeared at points that were multiples of 20 msec (the G.722.1 packet size) plus 10 msec. Too regular not to be suspicious. Here's one example (hardly convincing by itself, but when it is seen many times at predictable points, it becomes suspicious):
• Not surprisingly, the FFT of the WAV showed a lot of energy at frequencies which were multiples of 50 Hz (the inverse of 20 msec). Following shows five notes on the recorder. Time runs vertically from the bottom, frequency horizontally:
  Note that the pattern of 50 Hz harmonics is independent of the frequency of the note played. Also, this pattern is more obvious starting at 1 KHz (the meaning of this escapes me totally at this point).
• A G.722.1 S/W codec confirmed that the problem was with the encoding on the camera (I used an ethernet trace tool to grab the audio stream and extract the G.722.1 stream from it), not the decoding on the PC. The results were very similar with this decoder as what IE or IP Surveillance produced.

At some point, out of curiosity, I tried rolling back to firmware V1.05. Lo and behold, that fixed the problem! So if you have audio problems, try another firmware version, eg. 1.05.

Here are some short audio samples that show the problem (WAV, mono, 16 bit, 8 KHz):

• bad audio (the slow wow-wow-wow is an artefact of the downsampling -- web space is limited)
• good audio (after rolling back to 1.05 firmware): an order of magnitude better!

At this point, I wonder if it's not a bad bit in the memory which happened to be used to store audio samples in V1.07, but not in V1.05 (things having been shifted around). But I don't know, and probably never will.

Other useful things to know

• Interactions with XP's System Restore function. This could be a significant issue in some contexts.

  While using IP surveillance under Windows XP, I discovered that my system restore folder (F:\System Volume Information\_restore{...}" was filling up at an alarming rate. It turns out that some of the file extensions used by IP surveillance, in particular the *.hgd extension, are backed up by System Restore. So as IP surveillance would delete the older raw files (I have cycle recordings turned on), these files would be saved by System Restore, resulting in no actual freeing of space. And I ended up in a situation where IP surveillance would generate files for 1 second duration, and I could only see the last one (so total of one second of saved recordings!).

  The problem in some cases is that the above behavior will consume all space in the system restore folder, effectively flushing all other restores, and thus rendering the restore useless.

  A list of extensions monitored by System Restore is available at: http://msdn.microsoft.com/library/en-us/sr/sr/monitored_file_extensions.asp

  There are a couple of workarounds:

  • Disable System Restore completely.
  • Disable System Restore on the disk where the recordings are being made; the system seems to forget this setting tho when you remove the USB drive (which I do), so it's not practical for me.
  • The following registry key is my current solution: hklm->system->controlset001->control->backuprestore->filesnottobackup

       [HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\BackupRestore\FilesNotToBackup]
       "USB drive"=hex(7):46,00,3a,00,5c,00,2a,00,20,00,2f,00,73,00,00,00,00,00
    

• I record to a USB stick with 512 MB (good for about 8 hours of music). The minimum free space can't be reduced below 200 MB in the GUI, a bit high in this case. But, luckily, I found the following, which solved my problem:

     [HKEY_LOCAL_MACHINE\SOFTWARE\D-LINK\Matrix\Monitor Settings]
     "Cycle Recording"=dword:00000001
     "Reserved Diskspace"=dword:00000018
  

  Got to love the Registry!

Comments from D-link