Carl,
from all I know about Audyssey (which is no first-hand experience, I have to say), it looks that they do it right. They do not just simply correct amplitude without changing phase (which would be a linear-phase correction, not to be confused with correction to linear phase which is just the opposite: correcting phase to linear/flat no matter what the amplitude is), and the also do not apply simple minimum-phase correction, like simple analog EQing would do. They implement full-blown correction, "compacting" the direct sound (which means correction to linear phase and flat amplitude) plus cancellation of discrete reflections. Idealized, that is (many practical constraints that limit the achievable or desired perfection).
Note, however, that the term "minimum phase" is somewhat ambiguous and there are slightly different practical definitions in the analog and digital domains.
Often we find the definition of mixed-phase to be the following:
An analog filter that has a minimum-phase response (natural corresponding phase to the amplitude response) plus an excess-phase (or allpass-phase) component that accounts for any additional phase shift to the desired response. A typical multiway speaker system's anechoic room response would be of that kind. But it also extends to systems that cannot be made by analog means, like allpass filters which invert the phase response of a natural filter, taking away the excess phase and parts of the minimum phase as well, to yield overall linear phase regardless of amplitude -- this is what PhaseArbitrator does).
When we have additional reflections and other room effects (modes) things tend to get confusing in their defintions. According to Robert E.Greene (www.regonaudio.com) as long as the reflections are lower in level than the direct sound the system still is minimum phase (which is contradicting the statement from DDE as well as statements from knowledgable sources like John Kreskovsky, Earl Geddes et al), according to the definition of a minimum phase system as a system which can be corrected (to a perfect dirac, reflection-free response) without knowing the future of the signal, only the past and and the present of the signal is needed to be fed into a correction filter. The mentioned standard multiway speaker also is not a minphase system in that sense, here both definitions hold.
These things are still confusing me, I mean I'm in no way an expert in these matters, rather I'm still in the steep part of the learning curve on the subtleties of signal theory...
- Klaus
from all I know about Audyssey (which is no first-hand experience, I have to say), it looks that they do it right. They do not just simply correct amplitude without changing phase (which would be a linear-phase correction, not to be confused with correction to linear phase which is just the opposite: correcting phase to linear/flat no matter what the amplitude is), and the also do not apply simple minimum-phase correction, like simple analog EQing would do. They implement full-blown correction, "compacting" the direct sound (which means correction to linear phase and flat amplitude) plus cancellation of discrete reflections. Idealized, that is (many practical constraints that limit the achievable or desired perfection).
Note, however, that the term "minimum phase" is somewhat ambiguous and there are slightly different practical definitions in the analog and digital domains.
Often we find the definition of mixed-phase to be the following:
An analog filter that has a minimum-phase response (natural corresponding phase to the amplitude response) plus an excess-phase (or allpass-phase) component that accounts for any additional phase shift to the desired response. A typical multiway speaker system's anechoic room response would be of that kind. But it also extends to systems that cannot be made by analog means, like allpass filters which invert the phase response of a natural filter, taking away the excess phase and parts of the minimum phase as well, to yield overall linear phase regardless of amplitude -- this is what PhaseArbitrator does).
When we have additional reflections and other room effects (modes) things tend to get confusing in their defintions. According to Robert E.Greene (www.regonaudio.com) as long as the reflections are lower in level than the direct sound the system still is minimum phase (which is contradicting the statement from DDE as well as statements from knowledgable sources like John Kreskovsky, Earl Geddes et al), according to the definition of a minimum phase system as a system which can be corrected (to a perfect dirac, reflection-free response) without knowing the future of the signal, only the past and and the present of the signal is needed to be fed into a correction filter. The mentioned standard multiway speaker also is not a minphase system in that sense, here both definitions hold.
These things are still confusing me, I mean I'm in no way an expert in these matters, rather I'm still in the steep part of the learning curve on the subtleties of signal theory...
- Klaus
Last edited:
Thanks for the heads-up DSP and BTW nice entrance!
As with all good engineering feats, it's an exercise in compromise and I think most of the issues have been mentioned already. I like FIR's because I can control the complex response of the resulting filter i.e. I can make independent, arbitrary magnitude and phase. I don't know of a way to synthesize these in IIR.
Regarding the pre-ringing, I find that's only a problem with linear phase responses. In most cases you don't actually want a linear phase filter and as luck would have it, phase compensating filters seems to always have an impulsive response for the coefficients.
Having said that, I'd love to be able to synthesize IIR filter coefficients for truly arbitrary filter responses, just because of the lower latency. Peter Eastty did a nice tutorial at the last AES but it's still just pushing poles & zeros around til it looks right,
As with all good engineering feats, it's an exercise in compromise and I think most of the issues have been mentioned already. I like FIR's because I can control the complex response of the resulting filter i.e. I can make independent, arbitrary magnitude and phase. I don't know of a way to synthesize these in IIR.
Regarding the pre-ringing, I find that's only a problem with linear phase responses. In most cases you don't actually want a linear phase filter and as luck would have it, phase compensating filters seems to always have an impulsive response for the coefficients.
Having said that, I'd love to be able to synthesize IIR filter coefficients for truly arbitrary filter responses, just because of the lower latency. Peter Eastty did a nice tutorial at the last AES but it's still just pushing poles & zeros around til it looks right,
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.