modulated data over GSM voice (was Re: Data over normal GSM call)

[Ian Stirling]

Harald Welte wrote:
> Just to give you a "summary judgement":
> 
> Running any kind of voice-encoded data over a regular voice channel of a
> GSM phone is _extremely_ unlikely to work.
> 
> There are a number of different codecs in use.  Which codec is
> determined by the network.  There is echo cancellation at potentially
> multiple locations during the call.  There might be one or multiple
> transcoders of the voice codec along the road.
> 
> If you can manage to design a modulation and coding scheme that survives
> all (or even most) of the stages above, I think you have achieved
> something great.  I doubt you will get more than 300bps though :)


It has been done - for the minimal case of going through two actual real 
phones.
1300bps.
Basically a voice synthesizer driven by data, and a voice recognition 
system on the other end.
It uses the very properties that GSM/... codecs for speech are intended 
to preserve in the face of small amounts of bit errors - synthesizing an 
artificial throat, and exciting this model with data, so you don't get 
into parts of the codec that are chaotic or horribly lossy.

Unfortunately, I can't find the original paper I found.