Great explanation. I never assumed that the phase graph would have built-in feature to keep the line continuous but assumed that the zero would be reference time for all frequencies. I understand that both graphs have use cases so you should be aware about what the phase graph you're using actually displaying!
Interresting ! You gave an good explaination of the phenomenon called "Group Delay" which is a very complicated topic... but put simply is the "delay" of the enveloppe of a signal ... so it delays a group of frequency, that actually is "phase distortion" , your 120Hz single cycle actually stop being an single frequency but a group of frequencies centered around 120Hz (check the RTA ;) ) we can re name it a "burst" and so the phase shift caused by a filter, "delays the group of those frequencies ... yeah definitly mind fucking ... i don't have the level for those math but the group delay express the slope of the phase function, so its derivative stuff... thanks our 2ch analyser can display it for us... actually I show to student some kind of equivalent experiment exaclty like your to express delay , phase shift, group delay, and phase delay!
Thanks, I have been battling with interpreting phase from speakers, unfortunately your answers then open up many more questions! come to the conclusion you are best off not messing with phase at all. Now, that also brings up the question, would ironing out kinks in the group delay and making them near zero achieve better sound? I thought FIR was a safe option because pre-ringing is beyond our dynamic hearing range, but there you showed pre-ringing equal in amplitude to the main signal, so guess that burst my bubble.
ive never had the luxury of seeing phase graphs, and this has been VERY helpful. especially that tip about how it shows phase differences for IIR filters and how it could be mistaken as time-misalignment. just one of the hidden tips they dont teach you. also i have a question. does going up on the phase graph mean going forward in time?
“Does going up on the phase graph mean going forward in time” is the MILLION DOLLAR QUESTION that I might tackle in another video, but first I have to do my research to make sure I don’t end up looking like an idiot lol
@@devinlsheets_alphasound I would say that phase delay or a filter group delay is a real delay, but it is frequency dependent. For a lowpass filter the lower frequencies are delayed more than the upper ones. It is not too noticeable for a single filter, but chain 100 in a row and it is easy to hear the lows being delayed much longer than the highs. As far as forward in time, it is all relative to where zero is. By the way, the wiggle you get from an impulse response (without the 120 Hz cycle) can be used as a set of FIR coefs. And if you take the fft of it, it will show the frequency response of the filter. They are all different ways of looking at the same thing. My channel offers free loopback test software that you may find interesting, it does impulse and a bunch of other tests I use to characterize guitar pedals.
almost a year later. ive learnt a lot since. i found out that its the _steepness_ of the phase graph that determines how it shifts in time. more specifically, its the group delay. the steeper down the phase graph gets (going up in frequency), the further back in time that part of the frequency response will be, OR how much more the group delay will be. that means a flat part of the graph is, as we had seen, not shifted in time. whenever it goes up, it is earlier in time. whenever it goes down, it is later in time. there is a wonderful Wikipedia article on this titled "phase delay and group delay"
What it means, is that it's the SLOPE of the line that represents the delay. The problem therefore is that the first plot mode is the wrong way to show delay, because the Y axis of the plot doesn't correlate with the slope.
hi, not related to the subject , but i had a question and you might know much better , would a digital to analog conversion be better before , or after amplification ?
Ah, yes! Investigating the difference between phase and time! I have seen you struggling on different places on the internetz.. Do understand that phase shift as it is associated with causal filtering (as in minimum phase) usually is not linear. Meaning it's derivative = groupdelay is not a constant. So you can NOT repeat NOT compensate that phase shift with time in 'time-alignment' (man, it is signal alignment , but hey) This is a pledge for our industry to stop using the word phase for the overall 'time' difference between different parts of signals: First strip of the simple linear time difference, that's easy , everyone understands. Then we can start talking about phase difference. For an even better understanding of what phase does to your signals use 'wavelets' instead of sine waves.. nice one! .
So we should always bypass all EQ processing before aligning phase graphs for multiple speakers ? Even bypass the lopass filters for subs ? Is that correct ?
I would say no, as EQ adds delay depending on the type of filter, so for example Linkwitz Riley have constant delay, but others may not, tbh this is a minefield!, not sure many really understand it fully.
I see many uses DSP on subs but not their mids and tweets, how does that effect phase. Obviously the subs have to lag behind.. Does the DSP need cycle of an analog sound signal to do its thing? Some insight on this would be helpful!
In order for you to begin to understand the group delay correctly, you should use two microphones and two analyzers, and not look at the average readings on the same graph. You have a basic misconception about this. The group delay time is precisely a delay with a decrease in frequency, of course, with a distortion of the waveform, but the signal body will be delayed the more the lower the frequency and the higher the steepness of the filter. In the end, you answered incorrectly to people who hear the delay: it's not some kind of sound effect. Redo the experience with two speakers, with and two microphones, with two analyzers. You can place the speakers in different rooms for the purity of the experiment.
Not totally sure I follow how to set up your version of the experiment and what we would be testing or looking for on the graphs. Can you explain a different way maybe?
A sound expert who doesn't normalize or compress the volume of his own RUclips videos? Very strange. *Audacity's* Compressor effect is your friend: Threshold: -60 dB Noise Floor: -80dB Ratio: 10:1 Attack Time: 5.00 secs Release Time: 1.0 secs "Make-up gain for 0 dB after compressing" *CHECKED* "Compress based on Peaks" *CHECKED*
@@shazam6274 I downloaded this video and applied the above Compression settings to the audio and it worked beautifully. Now his voice isn't wildly varying in volume. You obviously did something wrong.
![Group Delay vs Phase Delay: What's the Difference? [DSP #18]](/img/1.gif)
Yes please, a deeper explanation of the phase graph is something I really need!
This helped me greatly understand the phase issues that can arise with iir vs fir filters. Thank you.
Great explanation. I never assumed that the phase graph would have built-in feature to keep the line continuous but assumed that the zero would be reference time for all frequencies. I understand that both graphs have use cases so you should be aware about what the phase graph you're using actually displaying!
Interresting ! You gave an good explaination of the phenomenon called "Group Delay" which is a very complicated topic... but put simply is the "delay" of the enveloppe of a signal ... so it delays a group of frequency, that actually is "phase distortion" , your 120Hz single cycle actually stop being an single frequency but a group of frequencies centered around 120Hz (check the RTA ;) ) we can re name it a "burst" and so the phase shift caused by a filter, "delays the group of those frequencies ... yeah definitly mind fucking ... i don't have the level for those math but the group delay express the slope of the phase function, so its derivative stuff... thanks our 2ch analyser can display it for us... actually I show to student some kind of equivalent experiment exaclty like your to express delay , phase shift, group delay, and phase delay!
That's the best explanation on phase ever. I still don't get it. But it's clearly the best explanation, still.
I laughed hard when you said "I've got an audio-file" because I thought for a second that you gonna show some "audiophile" ^'^
A dive into group delay might help to clarify exactly what's going on 🙂
Thanks, I have been battling with interpreting phase from speakers, unfortunately your answers then open up many more questions! come to the conclusion you are best off not messing with phase at all.
Now, that also brings up the question, would ironing out kinks in the group delay and making them near zero achieve better sound?
I thought FIR was a safe option because pre-ringing is beyond our dynamic hearing range, but there you showed pre-ringing equal in amplitude to the main signal, so guess that burst my bubble.
Great explanation. 👍 Phase graphs show change, not absolute timing.
ive never had the luxury of seeing phase graphs, and this has been VERY helpful. especially that tip about how it shows phase differences for IIR filters and how it could be mistaken as time-misalignment. just one of the hidden tips they dont teach you.
also i have a question. does going up on the phase graph mean going forward in time?
“Does going up on the phase graph mean going forward in time” is the MILLION DOLLAR QUESTION that I might tackle in another video, but first I have to do my research to make sure I don’t end up looking like an idiot lol
@@devinlsheets_alphasound I would say that phase delay or a filter group delay is a real delay, but it is frequency dependent. For a lowpass filter the lower frequencies are delayed more than the upper ones. It is not too noticeable for a single filter, but chain 100 in a row and it is easy to hear the lows being delayed much longer than the highs. As far as forward in time, it is all relative to where zero is.
By the way, the wiggle you get from an impulse response (without the 120 Hz cycle) can be used as a set of FIR coefs. And if you take the fft of it, it will show the frequency response of the filter. They are all different ways of looking at the same thing. My channel offers free loopback test software that you may find interesting, it does impulse and a bunch of other tests I use to characterize guitar pedals.
almost a year later. ive learnt a lot since. i found out that its the _steepness_ of the phase graph that determines how it shifts in time. more specifically, its the group delay. the steeper down the phase graph gets (going up in frequency), the further back in time that part of the frequency response will be, OR how much more the group delay will be.
that means a flat part of the graph is, as we had seen, not shifted in time. whenever it goes up, it is earlier in time. whenever it goes down, it is later in time.
there is a wonderful Wikipedia article on this titled "phase delay and group delay"
What it means, is that it's the SLOPE of the line that represents the delay. The problem therefore is that the first plot mode is the wrong way to show delay, because the Y axis of the plot doesn't correlate with the slope.
Wow! This is the best
what is the software on the right monitoring the phase graph?
ElectroAcoustics Toolbox!
when system tuning, how does attenuating compared to boosting affect the phase alteration?
hi, not related to the subject , but i had a question and you might know much better , would a digital to analog conversion be better before , or after amplification ?
Ah, yes!
Investigating the difference between phase and time!
I have seen you struggling on different places on the internetz..
Do understand that phase shift as it is associated with causal filtering (as in minimum phase) usually is not linear. Meaning it's derivative = groupdelay is not a constant.
So you can NOT repeat NOT compensate that phase shift with time in 'time-alignment' (man, it is signal alignment , but hey)
This is a pledge for our industry to stop using the word phase for the overall 'time' difference between different parts of signals:
First strip of the simple linear time difference, that's easy , everyone understands.
Then we can start talking about phase difference.
For an even better understanding of what phase does to your signals use 'wavelets' instead of sine waves..
nice one!
.
A delay is a special case of phase; linear phase. So yes this video is not really about phase in general, but about how time delays affect phase.
sir, what software you use, plese inform me, thanks
So we should always bypass all EQ processing before aligning phase graphs for multiple speakers ? Even bypass the lopass filters for subs ? Is that correct ?
I would say no, as EQ adds delay depending on the type of filter, so for example Linkwitz Riley have constant delay, but others may not, tbh this is a minefield!, not sure many really understand it fully.
I see many uses DSP on subs but not their mids and tweets, how does that effect phase. Obviously the subs have to lag behind.. Does the DSP need cycle of an analog sound signal to do its thing? Some insight on this would be helpful!
What Is The Best Linear Phase Speaker Manegement System
coooooooool
Time is a fucked up thing
In order for you to begin to understand the group delay correctly, you should use two microphones and two analyzers, and not look at the average readings on the same graph. You have a basic misconception about this. The group delay time is precisely a delay with a decrease in frequency, of course, with a distortion of the waveform, but the signal body will be delayed the more the lower the frequency and the higher the steepness of the filter. In the end, you answered incorrectly to people who hear the delay: it's not some kind of sound effect. Redo the experience with two speakers, with and two microphones, with two analyzers. You can place the speakers in different rooms for the purity of the experiment.
Not totally sure I follow how to set up your version of the experiment and what we would be testing or looking for on the graphs. Can you explain a different way maybe?
So, would ironing out kinks in the group delay make it sound better?
Off-topic could you do a comparison. noise cancelling headphones versus normal earmuffs class 5. With the microphone head mannequin
Thanks
A sound expert who doesn't normalize or compress the volume of his own RUclips videos? Very strange.
*Audacity's* Compressor effect is your friend:
Threshold: -60 dB
Noise Floor: -80dB
Ratio: 10:1
Attack Time: 5.00 secs
Release Time: 1.0 secs
"Make-up gain for 0 dB after compressing" *CHECKED*
"Compress based on Peaks" *CHECKED*
Hilarious! Those settings will yield barely audible sound with zero effective dynamics due to the ridiculous times.
something you should read here : en.wikipedia.org/wiki/Dunning%E2%80%93Kruger_effect
@@Petermaler66 Yeah; ya just can't cure stoopid.
@@shazam6274
I downloaded this video and applied the above Compression settings to the audio and it worked beautifully. Now his voice isn't wildly varying in volume.
You obviously did something wrong.
@@Petermaler66
Your ad hominem is a logical fallacy.