In algorithm designs especially for real-time critical systems, no error is permitted at all.. even when the software is running real-time, 24 hours a day, 7 days a week, 365 days a year non stop..

I wonder how extensively are the available audio coders being tested.. for example encoding an audio clip which is 3 - 4 hours long without any errors? In my experimental audio coder, one or two errors can appear after about 2 hours of continous music clip..

You should try analysing your code with tools like BoundsChecker (NuMega) - i.e. there could be some memory leaks somewhere or usage of global variables.

Also, bad hardware could cause some round-issues etc...

If hardware is OK and code written without leaks or memory corruptions it should be able to operate as long as there is electricity

I read about testing procedures for Applications software designs.. but I hardly think that those testing procedures can apply to "critical systems" softwares...

For example, the designs of control software for a computerised car braking systems.. You can't have any bugs at all.. How do you justified that for every 1 000 000 car drivers who stepped on the brake pedals, one would fail due to software bug and die in a horrible road accident ?

It's somewhat tedious but you can do better than that.

http://www.prover.com/
http://www.safelogic.se/home/default.asp

And of course if the models or properties you provide for the system are buggy or incomplete, you're back at square 1. Garbage in, garbage out...

How important is it for you to have a 100% bug free AAC implementation?

I am not really too sure about this.. In some critical areas such as the sf coding, which is predictively coded.. One single error can cause the entire frame to be decoded wrongly.. It can cause sudden spikes at the decoder which is very noticeable.

Of course 100% perfection is desireable.. I am wondering if AAC were to be used for radio broadcasts, I am sure that there are going to be some data corruption due to the transmission channels! How well can AAC handle this sort of data errors?

For radio broadcasts the ER (Error Resilient) tools in the MPEG-4 standard will be used to minimise sound problems due to bit errors. In the DRM standard the AAC data is also reordered so that statistically the most errors will occur in the least important places.

Menno