[digi-voice] Digital voice over high-speed packet

[Nicolai Heilemann]

Hi Ken,

If you are plannung to use "broadband" packet radio (I know that in the
century of DSL and WLAN we are a bit behind, but I am still happily
using my 9k6 FSK modem :) ), I guess, your demands for extremely high
compression are not too high, and I would recommend you'd rather trade
some bandwidth to allow a bit higher audio quality during coding (it
will get worse during transmission anyway ;)).

For that high (channel) bitrate you also shouldn't be too concerned
about the packet header overhead. It might be a bit inefficient to have
a header twice or three times as large as your real content, but as long
as you remain below the transmission bit rate and you have the channel
on your own, that should be fine.
I haven't had a too close look at the codec so far, so I cannot say how
many bits you will actually need per audio frame amd can't estimate how
many audio frames could actually be packed into the 256 bytes payload
(minus the space needed for the RTP header, if you want to use it), but
what I have found was that the codec uses 20ms frames.  Imagine you are
collecting 4 audio frames per packet frame (as I did last time, to get
an overall bitrate below 9k6), you will already have a codec delay of
80ms, plus the computing time of your AX25 network driver/TNC), and you
are approching perceptual delays quickly.  Furthermore the more audio
frames you stuff into one AX.25-packet the more severe packet losses
will occur, as the error concealment of the decoder needs surrounding
data to reconstuct missing frames, thus will become less effective with
increasing size of the gap.
I have just looked back at my data (LPC), and found that with one audio
frame (22.5ms) per packet frame I would have had an overall bit rate of
21.3 kBit/s(AX.25 header, RTP header,one extra byte which I used to
distinguish between RTP and RTPC and a payload of 14 Bytes). Too much
for 9k6 bit still sufficient for your 38k transmission.
So don't be put off too much by the header overhead ;).

Much luck and fun with your experiments.

vy 73 es 55 de

Nicolai, DL9TS


> Thanks all for the info...
>
> I have been experimenting with a codex called speex (www.speex.org) 
> and can get decent results down to about 5kbps, just after the first 
> night.  I am sure I can get it down less than that.
>
> There is a LOT of header overhead with AX.25 and that is something 
> that will have to be death with for my little project.  I was hoping 
> to overcome this problem by going to very high speeds.  I just 
> purchased a Kantronics TNC, supposedly capable of up to 32kbps and 
> even 56kbps with a little circuit change.
>
> I have been looking around for some alternate protocol that gives a 
> larger data segment.  Any suggestions???
>
> 73s
> Ken  KC7RAD