04.02.2014 Views

Lecture Series in Mobile Telecommunications and Networks (1583KB)

Lecture Series in Mobile Telecommunications and Networks (1583KB)

Lecture Series in Mobile Telecommunications and Networks (1583KB)

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

From pla<strong>in</strong> old Telephony to flawless mobile audio communication<br />

have been converted, to <strong>in</strong>troduce wideb<strong>and</strong> services everywhere, where you have the GSM phone. However, that is<br />

not a st<strong>and</strong>ard but just an idea from research.<br />

<strong>Mobile</strong> audio – communication<br />

Now we have nice cod<strong>in</strong>g schemes, you might <strong>in</strong>troduce wideb<strong>and</strong>, super-wideb<strong>and</strong>, or artificial extension or whatever,<br />

but we have adverse conditions on the radio channel <strong>and</strong> we need error protection. There are very sophisticated<br />

techniques. You have heard about turbo-cod<strong>in</strong>g, turbo error protection, turbo error correction <strong>and</strong> error concealment.<br />

4 Turbo error protection<br />

The turbo concept can be extended, which was orig<strong>in</strong>ally def<strong>in</strong>ed for two channel codes which help each other <strong>in</strong> an<br />

iterative way at the receiver. It can be extended to a channel decoder <strong>and</strong> to a source decoder. At the receiv<strong>in</strong>g end, we<br />

have a channel decoder which produces soft <strong>in</strong>formation – <strong>and</strong> soft <strong>in</strong>formation means that the channel decoder does<br />

not produce 0 <strong>and</strong> 1, but it produces reliability <strong>in</strong>formation. It says, eventually it is 0 or 1, with the probability. Then we<br />

have a soft decision source decoder, where the decoder exploits some residual redundancy which is still <strong>in</strong> the<br />

parameters. If we model the speech tract, the vocal tract, then the parameters evolve more or less smoothly, because<br />

we do the analysis every 20 milliseconds, 50 times per second, <strong>and</strong> the vocal track doesn’t make any jumps.<br />

There is still correlation <strong>in</strong> the parameters – <strong>and</strong> redundancy <strong>in</strong> terms of the distribution. The parameters are not white<br />

noise. This can be used such that we do a prelim<strong>in</strong>ary parameter decod<strong>in</strong>g <strong>and</strong> we exploit the redundancy on the level<br />

of the parameters, <strong>and</strong> we feed that <strong>in</strong>formation back – it is called extr<strong>in</strong>sic <strong>in</strong>formation – through the channel decoder<br />

<strong>and</strong> the channel decoder tries once more. There is a second decod<strong>in</strong>g us<strong>in</strong>g the knowledge from the channel end, the<br />

additional auxiliary <strong>in</strong>formation from the source decoder.<br />

In this slide, the source decode has better <strong>in</strong>formation <strong>and</strong> can improve once more <strong>and</strong> so, by several iterations, you<br />

have iterative source channel decod<strong>in</strong>g <strong>and</strong> we can improve tremendously.<br />

Example 5: iterative source-channel decod<strong>in</strong>g<br />

Let’s listen to one example of how we can improve by<br />

spend<strong>in</strong>g at the receiv<strong>in</strong>g end on just additional signal<br />

process<strong>in</strong>g complexity. If the channel is clear, we do not need<br />

it but, if the channel is gett<strong>in</strong>g worse, we can improve it by<br />

additional process<strong>in</strong>g. [Example of speech record<strong>in</strong>g played]<br />

In this case, we used just pla<strong>in</strong> PCM to demonstrate the<br />

concept. [Music played] We can ma<strong>in</strong>ta<strong>in</strong> the quality if we<br />

are allowed to use this complexity.<br />

<strong>Mobile</strong> audio – communication<br />

Now I would like to share a vision of what more we can do<br />

than just <strong>in</strong>troduc<strong>in</strong>g wideb<strong>and</strong> <strong>and</strong> artificial wideb<strong>and</strong><br />

extensions.<br />

Iterative source-channel decod<strong>in</strong>g (ISCD)<br />

This is a comparison with state-of-the-art decod<strong>in</strong>g, where we<br />

have hard-decision decod<strong>in</strong>g that is table look-up decod<strong>in</strong>g<br />

<strong>in</strong> the source decoder, or soft decod<strong>in</strong>g <strong>in</strong> the source decoder<br />

without iteration. We then do some iterations – here, <strong>in</strong> this<br />

case, it is 10 iterations – <strong>and</strong> we see a significant<br />

improvement. This is the parameter signal-to-noise ratio, the<br />

quality of the filter coefficients of the speech model, for<br />

example. This is the channel quality here, <strong>and</strong> we have a<br />

good channel here, <strong>and</strong> a bad channel. If we have a dem<strong>and</strong><br />

for a certa<strong>in</strong> quality, we can ma<strong>in</strong>ta<strong>in</strong> it, even <strong>in</strong> very adverse<br />

channel conditions.<br />

The Royal Academy of Eng<strong>in</strong>eer<strong>in</strong>g 53

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!