DSP Robotics Support

I'm thinking of a MB processor which runs a different process for each spectral band. It is a generic question, but as an example, let's suppose it is a:
3 band multi-band processor
Low band (under say 100 Hz) receives no processing
Medium band (say 100-4 KHz) receives stereo expansion with delay
High band (say 4 -20 KHz) receives a compressor + gain boost + some FIR filter with known latency

How can I make sure that the three bands are in ideal time alignment and avoid 'smearing' and 'frequency cancelation' of specific frequencies within the bands ?

Should I test this by impulse response? Phase response?
Other ideas/suggestions ?
What are the correction measures, simple delays per specific 'early' bands ?

1. Start with a band splitter which, when the bands are combined, will give you a flat frequency response. My tutorial on FS Guru explains how to do it.
2. Make sure that the bands are processed all with the same delay. You can always add delay to a band that receives no processing.
3. Don't use IIR filters, they will mess up phase relations between bands. Use linear phase FIR filters (if you really need to filter within a band).
4. Check with an FFT spectrum analyzer.
(a) white preferrable, feed it with a single impulse. Might not work for nonlinear processing.
(b) blue if white does not work. Feed it with white noise and use averaging to get meaningful readings.
5. Come back if you need more help.

Hi,

I had started off with the tutorial as mentioned.
For the example I have a 3 band splitter (cross-over) which shows a flat frequency response, and a 'slow-climbing' phase response. (I named it MV for Martin Vicanek since this is Martin's example from the GURU, hope this is OK).



Can I ask why if this is a good or rather normal phase response curve?

After I better understand this, my plan is to add some example processing which will be different for each band.

The combined bands phase response is that of an allpass. Some people would prefer linear phase (= same delay of all frequency components), however that is not possible with IIR filters.