Nice!
Two consecutive 0dB points would also be a good indication but of course there is also the inter-sample clipping thing, so maybe some more complicated method involving interpolation could be used, but would eat up DSP power...)
This is important because even a with a correction that never exceeds 0dB you can have saturation on 0dBfs real world signals when phase is modified, be it with IIR (normal filtering or all pass filters) or FIR (phase linearization).
The reason is quite simple: phase modification will alter temporal positions of peaks at different frequencies, and result in a different summed level. It will be lower than original most of the time, but can also be higher to an extend that is difficult (or impossible) to predict.
In this situation a clipping led is a good indication that the level has to be lowered inside the DSP (or in front of it) for that particular material.
Having volume control at the source or inside the DSP solves this problem, because you almost never approach 0dBfs (and if you do and your system is well adjusted chances are that you also reach territories where you amplifier and/or loudspeakers and/or ears will not be at their best), but having the volume control inside the DAC with integer (or clipped fractionals) data in front of it imposes some more care.
Two consecutive 0dB points would also be a good indication but of course there is also the inter-sample clipping thing, so maybe some more complicated method involving interpolation could be used, but would eat up DSP power...)
This is important because even a with a correction that never exceeds 0dB you can have saturation on 0dBfs real world signals when phase is modified, be it with IIR (normal filtering or all pass filters) or FIR (phase linearization).
The reason is quite simple: phase modification will alter temporal positions of peaks at different frequencies, and result in a different summed level. It will be lower than original most of the time, but can also be higher to an extend that is difficult (or impossible) to predict.
In this situation a clipping led is a good indication that the level has to be lowered inside the DSP (or in front of it) for that particular material.
Having volume control at the source or inside the DSP solves this problem, because you almost never approach 0dBfs (and if you do and your system is well adjusted chances are that you also reach territories where you amplifier and/or loudspeakers and/or ears will not be at their best), but having the volume control inside the DAC with integer (or clipped fractionals) data in front of it imposes some more care.
A possible solution would be to automatically lower the level inside the DSP when clipping is detected and rise the level in the DAC accordingly...
Some sort of compensated limiter with no consequence on dynamics.
You should easily be able to do this with some of the processing blocks offered by Audioweaver.
cheers
My understanding of the Haas effect is that differnt sounds with wavefronts that comes less than 1 ms appart will be heard as one sound with the same spectral content as the two signals. So preringing 0.1 ms before the square wave will be heard as belonging to the square wave sinuses. (Then I don't think humans can hear the frequencies of the preringing in question)
So, sorry, but I don't think the normal human ear can hear risetimes higher than that of a 20k sinewave.
Torgeirs
Last edited:
There is a serial EEPROM on board used to store all of the settings so you don't lose them
So that makes it impossible for the end user to implement that "compensated limiter" thing.
Could there be a choice to assign the volume control button to either the ESS one or one inside the DSP ?
Think the ringing frequency is outside the human hearing, but now I found tests that show time delays much lower than 1 ms can be detected by the human ear for certain waveforms and fir filters:
http://lib.tkk.fi/Dipl/2008/urn011933.pdf#page61
So longer fir filters are not nesessary better of axis.
http://lib.tkk.fi/Dipl/2008/urn011933.pdf#page61
So longer fir filters are not nesessary better of axis.