XE31CZS Exercise - Multi-Band Signal Processing
Tasks to do:
- Signals for processing
- Sum of harmonic components
Generate the signal as sum of sinusoidals of different
frequencies over full frequency band
- Speech signal
sm1.bin - fs = 16 kHz (utterance
"Mobilní hlasová shránka" - male - raining)
sf1.bin - fs = 16 kHz (utterance "Mobilní hlasová shránka" - female)
sm2.bin - fs = 16 kHz (utterance "1 0 6 4 7" - male)
sf2.bin - fs = 16 kHz (utterance "1 0 6 4 7" - female)
- Split the signal into two symetrical frequency bands:
- Design filter bank into following two frequency bands:
0 - fs/4 a fs/4 - fs/2.
- Filter analyzed signal.
- Observe spectral characteristics of the signals in particular
Attention ! For sinusoidal observe short-time DFT spectrum, for
Speech observe spectrgrams !
- Add separated signals and evaluate power of error signal
(i.e. difference between original and output)
- Realize decimation 1:2 for original signal and also for signals in
ATTENTION !!! It is not possible to use function decimate
from Signal Processing Toolbox because it uses always low-pass
filtering !!! You must realize it by direct removing of given signal samples.
- Observe spectral characteristics of all signals before and after
decimation and explain results.
- Insert zeros between samples of decimate signals and observe
spectral characteristics of these signals.
- Realize the interpolation for both bands using suitable
filtering. Observe again spectral characteristics of output signals.
- Add signals after interpolation and evaluate again the power
of error signal and compare with previously evaluated results for
the synthesis without decimation and interpolation.
- Try to realize splitting into 4 frequency bands, including
decimation and interpolation. Take care about following problems:
- decimation order,
- changes of spectral characteristics in particular bands,
- changes in interpolation procedure.