Stereo @48kbps: very-high-band content missing

Git SHA: 7ddadf2c

Encoding stereo at 48kbps and decoding to stereo can result in missing high band content. This can not be seen with the testvectors, but can be e.g. reproduced using track number 27 (Castanets) from the SQAM material, available here: https://tech.ebu.ch/publications/sqamcd.

File encoded with float reference codec:

File encoded with fixed point codec:

One can see that significant HF content is missing in the fixed point version. In debugging, I found that some frequency bins in that HF region are decoded as explicitly coded lines in the function decoder_tcx_invQ, while this does not happen in the fixed point implementation where only the bandwidth extension is used to fill this region later in the decoding process. The difference is audible for signals with very prominent high frequency content (e.g. electronic music), but not very prominent as it only happens very high in the spectrum.

Code item with:

./IVAS_cod -stereo 48000 48 27.wav bit
./IVAS_dec stereo 48 bit out.wav