That won't fix the group delay.
-
WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
What is group delay?
- Thread starter KTN46
- Start date
Raindog123
Major Contributor
It might, but it would not be a linear filter/system anymore. And I am not sure how practical it is for consumer audio applications.
Hmm? Any sequence of linear systems is a linear system.
Raindog123
Major Contributor
No, I was saying that a [pretty random] group delay [in theory] can be equalized with a non-linear filter. If we have the ability to properly characterize the channel, and the channel does not change.
Last edited:
You would need for the phase of your filter to be discontinuous to have a point at zero phase and another point wrap around 360 degrees without traversing the in between.
FYI, thank you everyone for explaining, now I get why group delay is not the time delay. I believe now that time delay causes group delay and the two would be equal in a system with flat zero phase response. Would it be fair to say that group delay measures the steepness of the change in phase and the measure of group delay (units are in seconds) is what time delay would produce an equivalent steepness of change in phase?
Last edited:
Raindog123
Major Contributor
BTW, what if you have a simple "system" X that passes a signal of two tones - f1 + f2 - through. Let's say, this system delays f2 by a certain fixed delay τ (which is X's group delay)... If you now construct a filter F that does nothing to f2 but delays f1 by the same τ, which is still a linear filter. The net effect of X+F would be a constant delay across f1 and f2, ie some delay but _zero_ group-delay.
Sure, but if X is a minimum phase filter the correction filter won't be, contrary to what someone was claiming above.BTW, what if you have a simple "system" X that you put a signal of two tones - f1 + f2 - through. Let's say, shis system delays f2 by a certain fixed delay τ (which is X's group delay)... If you now construct a filter F that does nothing to f2 but delays f1 by the same τ, which is still a linear filter. The net effect of X+F would be a constant delay across f1 and f2, ie some delay but _zero_ group-delay.
Raindog123
Major Contributor
Oops, while rushing to respond I missed the fact of you referring to a minimum-phase (casual) system. We/I am good now!
Raindog, can you please bring up something to do with "Quantum" in this thread. I just feel like it is missing in this rather large and ongoing engineer catfight! I'm all about abusing the word Quantum!Oops, while rushing to respond I missed the fact of you referring to a minimum-phase (casual) system. We/I am good now!
Raindog123
Major Contributor
Funny you've asked...
(Those without IEEE Xplore can get it here.) I do not think the idea went anywhere... maybe in the waiting for cheap quantum computers.For much-higher-than-audio frequencies, the Rydberg atom excitation sensors (receivers, apertures) -- promising out-of-this-world sensitivity -- are a hot topic these days... But again, this is RF... but maybe someone will think of a cool consumer audio application?
dc655321
Major Contributor
- Joined
- Mar 4, 2018
- Messages
- 1,599
- Likes
- 2,255
Here's an example of phase response of a 2-way speaker with LR 24db/octave XO at 1800Hz. Blue response is with phase left uncorrected and orange is with phase response corrected for XO deviation:
Although you can see quite a difference in those 2 phase responses
Unwrap the phase (blue) and much what little drama there is will disappear.
Never even bothered to investigate gd/phase for higher frequencies.
Have never seen any credible documentation for audibility, and when looking at phase when listening to different systems, there does not seem to be any correlation at all - such as, a speaker with flat phase does not seem to have any advantage over one where phase shifts considerably.
Testing for audibility is easy, just create a phase shift on one sample and compare abx. But even if it is possible to get a positive for audible difference, the question is, does it have a negative impact on sound quality.
Now if the relative time delay gets large enough, there will be a difference. But when talking about phase, we usually mean a delay less than one period, and for freqs >1K that delay is very small.
It's difficult to give a complete explanation of group delay in the time domain, but it's possible to demonstrate the difference between phase delay and group delay for individual examples. Here, I consider an all-pass filter with the following frequency response:
Using the phase response of the filter, we can calculate the group delay and phase delay:
At 10 kHz, the difference between group delay and phase delay is rather big. What happens if we apply a short burst of signal at that frequency?
Now things start to get interesting. We see here that each individual peak is delayed a little bit, just enough to make the output the negative of the input, kind of. But the overall envelope of the signal is delayed quite a bit more! These two different delays correspond to the values at 10 kHz that can be extracted from the previous graph (around 0.3ms group delay, and 0.05ms phase delay).
For some further nice visualizations on how this can look in the time domain I recommend the Wikipedia page on Group velocity. But note that the x-axis there is space, and not time, as they're looking at how waves propagate in media where velocity depends on frequency. Luckily that is not the case in air, so this is somewhat irrelevant for Hi-Fi, but it might help with understanding group delay. And the animations are very pretty!
What is Group-Delay is a justified question and may even be a concern for audio-purists.
In the topic of analog audio (and imo) it is not a great big issue, as Group-Delay was born of RF-dictates (hardware, medium, etc) used to xmt/rcv coherent RF signals bearing intelligence (modulated).
It does NOT mean that a digital (impulse/step/Tr/Tf) equivalent of an audio signal is immune to such Group-Delay issues ; since these signals can be defined/resolved in the frequency domain as a complex RF signal (both using measurements + theory (objectively).
MY EXAMPLE: I have a AVPre/Pro, which is used to route input RCA audio, thru a power amp, en-route to L/R speakers.
I also take this RCA input audio to stream it other places in the house; using xmt/rcv AoIP dongles. Since rigging these together, there has been a very pronounced delay between main speaker outputs and other speakers in the house. It should be noted that the Dante dongle' outputs, also get 'processed' thru XLR inputs of another power amp. First weeks of this annoying delay, I was blaming the AES-67 accrued delays in the chain. It turned out that my main AVPre/Pro (a Rotel) is the more delayed output... than same signal going thru 50feet of RG-6a, and multiple PoE network switches (etc.). I was confused at first that the AES route had significantly LESS delay than the Rotel's internal delays.
This is NOT Group-Delay!
NOTE: I was able to lessen this 'thru-put' delay by introducing additional [?] delay, via DanteController GUI.
In the topic of analog audio (and imo) it is not a great big issue, as Group-Delay was born of RF-dictates (hardware, medium, etc) used to xmt/rcv coherent RF signals bearing intelligence (modulated).
It does NOT mean that a digital (impulse/step/Tr/Tf) equivalent of an audio signal is immune to such Group-Delay issues ; since these signals can be defined/resolved in the frequency domain as a complex RF signal (both using measurements + theory (objectively).
MY EXAMPLE: I have a AVPre/Pro, which is used to route input RCA audio, thru a power amp, en-route to L/R speakers.
I also take this RCA input audio to stream it other places in the house; using xmt/rcv AoIP dongles. Since rigging these together, there has been a very pronounced delay between main speaker outputs and other speakers in the house. It should be noted that the Dante dongle' outputs, also get 'processed' thru XLR inputs of another power amp. First weeks of this annoying delay, I was blaming the AES-67 accrued delays in the chain. It turned out that my main AVPre/Pro (a Rotel) is the more delayed output... than same signal going thru 50feet of RG-6a, and multiple PoE network switches (etc.). I was confused at first that the AES route had significantly LESS delay than the Rotel's internal delays.
This is NOT Group-Delay!
NOTE: I was able to lessen this 'thru-put' delay by introducing additional [?] delay, via DanteController GUI.
There is no divide between F-domain and T-domain and they are fully transformable in both theory + practice, like a grand-unification theory, which has stood the test of time..
I believed you w/o you having to add this part to your reply.
They sure do, and for full effect add another graph (or combined into one) with a burst of e.g. a 5kHz sine would be even more interesting as it will show the difference on a different frequency which is sort of the point here (I think, so many points made already
).
...now I'm even more confused. Does Dirac Live attempt to correct this type of issue?
Holmz
Major Contributor
Here are a few graphs showing this in the context of minimum phase filters: https://troll-audio.com/articles/linear-and-minimum-phase/
Wrapped or unwrapped, what you see there is Linkwitz-Riley 24db/octave phase distorion. My point here was that even when left uncorrected group delay looks pretty much the same as after correction.
Similar threads
