Talk:Group delay and phase delay

Untitled
How is group delay defined for a frequency translating device (a mixer)? If the input and output frequencies are different, their phases cannot be directly compared. It would be possible to compare the phase to that of an ideal frequency transliting device. Are there other, better definitions? How can it be measured? (A simple example of a mixer is an AM radio. Signals at several MHz are translated down to audible frequencies).--HelgeStenstrom 13:29, 2 May 2006 (UTC)

article move and revamping.
i'm a little busy at the moment, but i do want to continue with this rewrite including a couple of drawings that i haven't done yet. as i said in the move comments, i want to start with first principles where both group delay and phase delay are defined and then have the article go over the specific places where group or phase delay (mostly the former) are used (optics, transmission lines, audio, etc.). also, thanks to User:81.240.215.127 for correcting my convolution integral mistake. r b-j 17:41, 6 March 2007 (UTC)

Big Improvement Possible ...
The article below provides a **much** more intuitive explanation of what group delay is:

http://sep.stanford.edu/sep/prof/pvi/spec/paper_html/node19.html BB


 * I do believe the concepts in the Standford article are mostly at the heart of what this Wiki should get across. But not everyone can follow the equations very well. A much simpler explanation of the same concepts is possible, and at each stage should be accompanied by graphs, along with the equations that go with those graphs. Ohgddfp (talk) 02:11, 16 March 2020 (UTC)

In my mind group delay is the change in phase as a function of frequency (i.e. phase delay non-linearity). As it stands the current wikipedia entry addresses this only obliquely. Perhaps a useful explanation for signal processing theoreticians who see the world through convoluting FIR filters but not a good explication for others. I suggest that the above article form the basis of the introduction with the beat frequency as an example. In fact I'd extend the example to show the post filter graphic that increases the beat frequency. The existing Laplace description should be retained in a "mathematics" section. —Preceding unsigned comment added by 75.107.113.194 (talk) 14:44, 8 September 2009 (UTC)


 * About: "... group delay is the change in phase as a function of frequency ...": How about "... group delay is the rate of phase change as the functon of frequency? Ohgddfp (talk) 17:43, 18 March 2020 (UTC)

Ungeekification
For those of us that can't comprehend what the laplace transform of an integral is, I think this article (or at the very least, the introduction) needs to be simplified. The opening paragraph should give a broad overview of the subject and what it is generally used for. Then, there should probably be a "derivation" section where all of the meaningless formulas can be shown. From reading this article, I have no idea what group delay is, and I'm an engineer that took calculus and differential equations all through college. Snottywong 02:17, 4 May 2007 (UTC)

I have to agree. I have a technical background (more on the computer science side, granted) and had no more idea of what group delay was when I finished the article than when I'd started. I suggest a single non-technical paragraph at the very beginning explaining in layman's terms what group delay is. Ivan Denisovitch 18:43, 6 June 2007 (UTC)


 * The phase response issues are complicated by their very nature. So try this:  Group Delay of a system, measured in units of time, is how much the phase angle of a sinusoid test signal changes between the input and output of that system, due to a small reduction in frequency, divided by the amount of that reduction in frequency.
 * Nope. Totally untrue (unless it's phase linear).  What you're talking about is Phase Delay, not Group Delay. 71.169.190.28 (talk) 14:59, 22 April 2011 (UTC)
 * About "... talking about is Phase Delay, not Group Delay."  Where I said how much the phase angle of a sinusoid test signal changes between the input and output of a system, that's English for delta phase angle (in degrees or radians).   Where I said a small reduction in frequency, that is English for minus delta frequency.  Now I divided one by the other.  In math speak, that's delta phase angle divided by minus Hertz.  The minus can be brought to the numerator without changing the equation.  So it's minus delta phase angle divided by delta Hertz.  Now this is a close approximation to Group delay which is precisely a rate of change, which is a derivative.  -d[Phase Angle as a function of frequency]/df, where f is frequency.  This derivative is the official definition of "Group Delay".


 * Absent practical issues, such as system and measurement noise, the smaller the change in frequency, the more accurate the group delay measurement becomes near that frequency. In this way, the negative of the "slope" of the graph of the system sinusoid "phase angle versus frequency response" between input and output at a given sinusoid test frequency is the group delay measured in seconds at that sinusoid frequency.  The practical meaning of group delay is the time delay of the frequency components of a signal "envelope", where the envelope is a modulating signal that modulates a high frequency sinusoid to create a passband signal, such as that used in radio.
 * The practical meaning of group delay is how long the group or envelope of a burst of a modulated sinusoidal signal is delayed, *not* the frequency component inside the envelope.
 * No, the "group" term refers to the group of frequencies that are in the passband, not a group of signal variations. A burst of a modulated sinusoidal signal is a bandpass signal as utilized in a radio channel, for example.  So Group Delay is really the time delay of the frequency components making up the DEMODULATED baseband (not passband) waveform, which has the same shape as that envelope.  The definition of Group Delay is -d[A(f)]/df of the passband signal, as the slope of Phase Delay Angle A of the passband as the function of passband frequency.  The envelope as a whole will indeed be delayed if all the frequency components making up the DEMODULATED baseband waveform are delayed by the same amount of time.   That is constant Group Delay over the passband, which translates as a constant Phase Delay (in seconds) of the DEMODULATED baseband waveform.  However, if the frequency components of the DEMODULATED baseband waveform (same shape as the envelope) are not delayed by the same amount, the shape of the DEMODULATED baseband waveform is also DISTORTED in addition to being delayed, and the Group Delay over the passband is also not constant.  It's the DISORTION of the DEMODULATED baseband waveform shape that makes group delay such an important specification in LTI systems that utilize modulation such as in analog television.  To keep the envelope distortion free, there is no need for the carrier to have a particular phase.  Group Delay is useful because it does NOT include information about the phase angle change of that carrier through the system, even though the carrier is part of the passband.  This is convenient because the carrier phase has no effect on envelope timing or envelope distortion. Oobsoiud (talk) 04:17, 28 April 2011 (UTC)


 * As long as the group delay is constant over the range of frequencies contained in the passband (radio frequency) signal, all the frequency components of the recovered (demodulated) intelligence, which is the waveform whose shape is the envelope shape, are delayed by the same amount of time, and therefore the recovered intelligence (envelope) suffers no waveform distortion due to phase response irregularities.
 * Now you're getting it right.


 * However, the radio frequency carrier sinusoid does not necessarily have the same time delay compared to the time delay of the individual frequency components of the envelope. But since the carrier alone contains no intelligence, the wrong carrier time delay does not contribute to phase distortion of the signal envelope (aka modulating signal or intelligence).  Therefore in a modulated system such as radio, constant group delay over the range of frequencies of the radio channel is all that is needed to prevent waveform distortion of the demodulated (recovered) signal that is due to phase response irregularities.


 * Another kind of linear time invarient (LTI) system works with "baseband" signals and does not involve modulation. An audio amplifier is an example of such an LTI system.  And so to insure that the entire waveform (not just an envelope) suffers no waveform distortion due to phase response irregularities, the graph of system "sinusoid phase angle versus frequency response" in case of "baseband" signals such as plain audio, must not only be a straight line over the frequencies contained by the signal, but also the extension of that line going straight down to lower frequencies must pass through the origin at zero Hz, even if there is no signal energy at all at such low frequencies.  This is the only way to guarantee that all sinusoidal components of the signal have the same time delay through the system in order to prevent waveform distortion due to phase reponse irregularities.  This linear (straight line) phase response through the origin translates to a group delay graph that is constant all the way down to zero Hz.  A group delay that is constant only over the frequencies contained in the signal, but not constant down to zero Hz, still imparts unequal time delays to the sinusoidal components of the signal that lie in the frequency range where the group delay is constant, which in turn causes waveform distortion due to phase response irregularities.  In other words, in an LTI system processing baseband signal, constant group delay over the frequencies contained in the signal DO NOT guarantee freedom from waveform distortions caused by phase response irregularities.
 * ???? How do you get a constant group delay over the baseband frequencies while not getting a constant phase delay?  Consider, for the moment what happens at DC: Group delay = Phase delay. (assuming polarity is preserved).  71.169.190.28 (talk) 14:59, 22 April 2011 (UTC)
 * Both Group Delay and Phase Delay are measured in seconds. A constant "Phase Delay" (measured in seconds) can only happen when the graph of the Phase Response versus Frequency (in degrees or radians) is a straight line that passes through the origin.  Under this condition, the Group Delay is indeed constant with frequency.  However, if the entire graph of the Phase Response is shifted vertically, the Phase Delay (which is Phase Angle Delay divided by frequency equals time delay in seconds) is no a constant time delay at all frequencies.  This results in a disorted waveshape (if not a sinusoid), but at the same time, the Group Delay, which DOES NOT RETAIN the information about where the Phase Response (angle measurement) graph intersects the vertical axis at zero Hz, remains exactly the same.  Since Group Delay is -dA/df  (A is the angle and f is frequency), that represents the slope of the graph for Phase Response versus Frequency, and moving this graph vertically does not change the slope, and therefore the Group Delay also doesn't change.  Yet, with the Phase Response versus frequency graph NOT going though the origin, but is still a straight line, the Phase Response is no longer proportional to frequency, and therefore the "Phase Delay" (Phase_Angle / f) in seconds is also no longer constant with frequency, which in turn distorts the waveform because the different frequency components of the waveform now have different time delays.  At the same time, the Group Delay hasn't changed at all because the slope of the Phase Angle as a function of frequency does not change when its entire graph is moved vertically. Oobsoiud (talk) 04:17, 28 April 2011 (UTC)


 * Fortunately, the ear is relatively insensitive to such distortion in audio, but for video, such distortion is very visible, and for digital pulses, too much distortion causes a higher bit error rate. A more direct measure of performance that measures freedom from phase distortion for baseband signals (such as what an audio amplifier works with) is "Phase Delay", also measured in units of time.  The vertical scale of such a graph is system sinusoid phase angle change between input and output, divided by the frequency, which translates to delay in seconds.  The horizontal scale of that graph is simply the sinusoid frequency corresponding to that time delay of the test sinusoid.  So the "Phase Delay" is a graph of time delay of a sinusoid frequency component on the vertical scale, versus the frequency of that sinusoid component on the horizontal scale.  If the time delay of all the  frequency components of a baseband signal (audio or video, for example) are the same as each other, then there is no waveform distortion that is due to phase response irregularities. Am358 (talk) 13:43, 22 April 2011 (UTC)


 * it needs a drawing. it needs a drawing of an LTI system (a box that has an input and output that we say is an LTI system), with an amplitude modulated sinusoid going it and an amplitude modulated sinusoid coming out, where the frequencies of the sinusoids are the same, where the output sinusoid is delayed by the phase delay relative to the input sinusoid and where the output envelope is a replica of the input envelope but scaled by some factor (that is |H(j&omega;[sub]0[/sub])| ) and is delayed by the group delay.  that says the whole thing in a picture. (i don't have the time nor the tools to do it.) r b-j 03:09, 4 May 2007 (UTC)
 * Whilst I agree to a certain degree that this article may be too technical for some people, I think it'll be pretty hard to make this article accessible to people without a solid technical/scientific background (not computer science). --Taraborn (talk) 14:53, 20 September 2008 (UTC)
 * I think pictures with graphs relating phase response graphs to test sinusoids will be a big help. Oobsoiud (talk) 04:17, 28 April 2011 (UTC)

Authors in section "Group delay in the audio field"
I think in this section one of the author names is Blauert (with a 't' at the end). I tried to change it but it doesn't seem to be effective. Maybe it has something to do with the referencing thing. Sorry I don't have time to look into this now, so if someone knows what's wrong... — Preceding unsigned comment added by 82.66.50.169 (talk) 09:23, 25 May 2011 (UTC)

True time delay
Is constant group delay the same as True time delay? If so, the later article might be replaced with a redirect to this one. Sohaibafzal (talk) 07:14, 14 July 2011 (UTC)


 * I think you are right. Transmission lines are not my specialty but it appears that true time delay is simply the ideal case where there is no frequency smearing of group delay and phase delay. Binksternet (talk) 14:28, 14 July 2011 (UTC)
 * This is the first I had known of that other True time delay article. There is no reason for that article to exist.  As far as I'm concerned, feel free to redirect it, since it's hard to get an article deleted.  That's such a dumb article.  Why not have another one titled False time delay? (Boy, if that link ain't red, I'll be astonished.) 70.109.184.37 (talk) 22:45, 14 July 2011 (UTC)
 * Sorry, Sohaib, I couldn't wait. I put in the redirect and also a deletion request. 70.109.184.37 (talk) 22:53, 14 July 2011 (UTC)


 * Um, I have restored the true time delay article for the time being. "It's dumb" is not a legitimate argument for deletion/redirecting and "it's hard" is not a good reason to skip real deletion processes.
 * A Google search does turn up many sources, including papers cited by others, using this term prominently, so it is a legitimate term and I believe it warrants to be covered on Wikipedia. Merging is fine IMO. Here's a source talking about both group delay and true time delay, but I'm too ignorant to make sense of it. -- intgr [talk] 22:11, 27 September 2014 (UTC)
 * My reading of the article and True time delay is that true time delay is not the same as a group delay. Rather, it is used to describe a property of a device: the property of uniformly delaying all relevant frequencies equally, and so the signal is not distorted. So, we can say that if a transmission line has true time delay, then it will not distort. So, looking back, I agree with the interpretation Binksternet and recommend merge to a section on Group delay and phase delay, as the smaller page consists of nothing but a definition and benefits from the context of the broader article. Klbrain (talk) 14:18, 8 July 2017 (UTC)
 * ✅ Klbrain (talk) 15:50, 21 July 2017 (UTC)

Thresholds for group delay need more information
The threshold in the presented table [500 Hz	3.2 ms, 1 kHz	2 ms, 2 kHz	1 ms, 4 kHz	1.5 ms, 8 kHz	2 ms] needs additional information from the cited paper:
 * 1) Values are for diotic presentation (monotic presentation show similar thresholds; also, the monotic ones are presented only for 1,2 and 4 kHz)
 * 2) Values are for bandlimited sound impulses under unechoic conditions (25 micro seconds-wide rectangular impulses) - the thresholds for 'speech, noise, music, harmonic series' were found to be higher but not presented in the study
 * 3) Values are not presented for trained subjects - from their experiment, the threshold halved at 4 kHz for one trained subject: from 0.86 to 0.4 ms. So, for 'worst case' the thresholds are actually lower.
 * 4) Tested method was paired forced-choice [compared to the final test they did - triadic method, forced choice]

The table might need the word threshold for 'group delay', since it is not clear if the values are for group delay or phase delay.

I would mention these facts so that people would not take those values as they are. At least mention that more popular types of signals (music for instance) have higher thresholds. Also that these are for anechoic conditions.

For the worst-case scenarios (trained subjects), the thresholds are actually lower and are not reported for all frequencies. — Preceding unsigned comment added by 5.13.214.42 (talk) 16:31, 23 December 2012 (UTC)

Incorrect statements
I must say that the whole article needs serious improvement. It contains several innacuracies. For example, already in the definition:

I strongly suggest a complete rewrite of the article. — Preceding unsigned comment added by 138.4.36.49 (talk) 10:34, 17 January 2013 (UTC)
 * Group delay is not a definition for "signal processing". It is an efect tha happens to any signal that travels through any linear system. Signal processing makes use of it, as for example also does microwave engineering, analog filter designers, or communications engineering.
 * In the article it is stated that group delay affects to "signal amplitude" and phase delay affects to "signal phase". This is deeply untrue.
 * It links group delay with audibility. That is nonsense, because the audibility of any signal is independent of the precise instant it arrives. I strongly guess that he refers to "group delay dispersion".
 * etc


 * Well, feel free to be wp:bold. The worst that can happen is that your edits get reverted.  As best as I can tell, the math in the Introduction section is completely correct.  The group delay applies to the "group" or envelope of a modulated sinusoid.  The phase delay applies to the phase of the sinusoid.  The optics and audio sections I haven't paid much attention to. 71.169.181.144 (talk) 18:40, 17 January 2013 (UTC)

Add some diagrams
Jidanni (talk) 23:28, 11 December 2013 (UTC)

External links modified
Hello fellow Wikipedians,

I have just modified one external link on Group delay and phase delay. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:
 * Added archive https://web.archive.org/web/20150930080912/https://community.klipsch.com/forums/storage/3/1027021/7805blauert.pdf to https://community.klipsch.com/forums/storage/3/1027021/7805blauert.pdf

When you have finished reviewing my changes, you may follow the instructions on the template below to fix any issues with the URLs.

Cheers.— InternetArchiveBot  (Report bug) 16:54, 25 March 2017 (UTC)

More Ungeekification
This article is a perfect example of the need for an "Einsteinian" explanation, where Einstein said, "Everything should be made as simple as possible, but no simpler." Ohgddfp (talk) 17:03, 14 March 2020 (UTC)

So, about a particular paragraph under the previous headline, "talk:Ungeekification", where that paragraph begins with: "Another kind of linear time invarient (LTI) system works with "baseband" signals and does not involve modulation". That paragraph I believe, is absolutely corrrect. But furthermore, it contrasts related ideas that are at the heart of the confusion people have when trying to understand "Group Delay". (Although 'group delay' and 'phase delay' are intimately related to each other, the article should be entitled only 'Group Delay".)  Ohgddfp (talk) 17:03, 14 March 2020 (UTC)

Here is what I believe needs to be clarified. I think amplitude modulation and demodulation in a radio channel is simple enough to provide a very good example of an LTI system having a particular 'group delay'. In the example, the radio signal (passband) consists of a high frequency sine wave functioning as the 'carrier', and the intelligence in the form of a baseband signal modifying the carrier's amplitude, causing the envelope of the passband signal to be identical in shape with the baseband waveform. Upon demodulation of the radio signal in the receiver, the baseband function is recovered from the envelope of the passband signal. Ohgddfp (talk) 18:31, 18 March 2020 (UTC)

If the magnitude response and the 'Group Delay' of the above example LTI system are both flat for the aperture (frequency range) responsible for creating a particular envelope, then the time delay of frequency components making up the baseband waveform will not be altered, and therefore the waveshape of that baseband waveform will also not be altered. Besides being flat, the amount of 'group delay' give the time delay of the baseband (and envelope) waveform. Ohgddfp (talk) 18:31, 18 March 2020 (UTC)

Additional explanation is needed regarding a baseband signal not involved with a modulating (RF) signal: A linear phase LTI filter has a flat group delay, but the inverse is not necessarily true. (A filter with flat group delay is not necessarily a linear phase filter.) The explanation in the aforementioned Ungeekification paragraph is also very revealing on this issue; example graphs matching the words should be included. Ohgddfp (talk) 18:31, 18 March 2020 (UTC)

For people who work in 'signal processing' a critical function in industry, this article is of high importance. Ohgddfp (talk) 17:10, 14 March 2020 (UTC)

About the wiki section: Group Delay in Audio
This section should be removed entirely. There is no group delay in audio. That's because analog continuous time audio signals are directly used in sound reproduction, and are generally baseband signals and not modulated signals. Therefore the LTI system frequency range of interest can have very non-linear phase, giving very bad waveform distortion, and yet still have a perfectly flat group delay in that same frequency range of interest, making the measurement of group delay useless. See the paragraph in "Talk:Ungeekification" that starts with "Another kind of linear time invarient (LTI) system works with "baseband" signals and does not involve modulation." This Wiki section, "Group Delay in Audio", minus the word "Group" could be part of a wiki sound reproduction article, and 'phase' could have a link to a separate wikipedia article that is simply entitled "Signal Phase" or some such thing. Ohgddfp (talk) 00:44, 16 March 2020 (UTC)


 * Audio engineer Glen Ballou has a section about group delay in his widely used textbook Handbook for Sound Engineers, the New Audio Cyclopedia. The section says, "A filter can exhibit a group delay over a group of frequencies covering a section of the audio spectrum if those frequencies are all subject to the same time delay. The group delay is given as the first derivative of the phase with respect to frequency." He says people can hear this kind of group delay if it is 3 milliseconds or more, in a range of 500 to 4,000 Hz.
 * I argue that if group delay in audio is treated in a textbook, then we should keep the section about it. Binksternet (talk) 05:20, 16 March 2020 (UTC)
 * In general, I think that is a mistake for such a misunderstood subject. Textbooks are known to have mistakes. A textbook that concentrates on the math of group delay is more reliable than textbooks that use group delay incidental to the main subject matter. A non-math book with mistakes relating to one math subject should be determined to be unreliable for that math subject.  And a known unreliable source, not for the subject as a whole, but for the specific math topic at hand, should never be used as a source. Ohgddfp (talk) 01:21, 17 March 2020 (UTC)


 * Another scholarly treatment of the topic can be found at SynAudCon's online training page "Hand-in-Hand Phase Response and Group Delay " by Pat Brown, a respected expert. Binksternet (talk) 05:38, 16 March 2020 (UTC)
 * Thank you for the links. About the textbook sentence: " "A filter can exhibit a group delay over a group of frequencies covering a section of the audio spectrum if those frequencies are all subject to the same time delay." I am sure that everyone, including the author, knows that a filter exhibits a group delay over a group of frequencies ... regardless whether or not those frequencies are all subject to the same time delay.  That's because a filter always has a phase response that is a continuous function, and of course there is always a Group Delay that can be calculated from it by taking its derivative, which means that a filter always exhibits a group delay.  Therefore on the face of it, that textbook sentence makes no sense.  It could be corrected as: 'A filter can exhibit a flat group delay over a group of frequencies covering a section of the audio spectrum if those frequencies are all subject to the same time delay.'  The words can and flat mean that sometimes a filter will exhibit a flat group delay even if the phase response is extremely non-linear, causing severe waveform distortion of musical transients.  (Reasons for this are given by another contributor on this talk page.)  As a result, people can be deceived by this kind of misuse of group delay.  The proper use of group delay is to show the time delay of the envelope frequency components.  The author only mentions group delay in context with a study on perceptions. But I believe that is a flaw in the study for reasons I just gave.  Another study in perception: "However, interpretation of these data in terms of group delay perceptibility seems to be difficult."  Yeh it is.  Another case of misuse.  Some studies are known to have flaws. Ohgddfp (talk) 00:49, 17 March 2020 (UTC) Ohgddfp (talk) 00:52, 17 March 2020 (UTC)
 * The prosoundtraining segment is interesting. The author claims that "The group delay plot can identify the amount of pure delay ...". Well, only if the phase delay graph goes through the origin at zero Hz, something not mentioned by the author.  And whether or not it goes through the origin cannot be determined from the group delay because that is the information the group delay loses due to its being the minus derivative of the phase delay.  Delay, pure or otherwise, of a given single frequency (not a group of frequencies) can only be determined from the phase delay plot using the formula, τD=φ/(360f) Ohgddfp (talk) 00:52, 17 March 2020 (UTC)
 * My recommendation would be to expand the section with a summary of Pat Brown's analysis. Binksternet (talk) 00:05, 17 March 2020 (UTC)

My previous comments need review. My previous objections to Group Delay in audio stem from authors not mentioning "minimum phase". I am now looking for how to better find reliable sources and talk about them more completely. Now that I have reviewed the helpful links from Binksternet, reading those article pushed me into doing much more research so I can finally make helpful suggestions. Ohgddfp (talk) 04:23, 17 March 2020 (UTC)
 * Please don't modify or erase your comments after a reply has been posted. Rather, add new comments below. It's also allowed to strike out your comments using the HTML tag of followed by . Binksternet (talk) 04:45, 17 March 2020 (UTC)


 * Now it must be remembered that the group delay of an audio signal often shows the time delay of a group of frequencies that are not included in the audio signal itself. (In talk:ungeekification: "Group Delay is useful because it does NOT include information about the phase angle change of that carrier through the system, even though the carrier is part of the passband." Note that passband is the original audio signal in this context.)  So instead, demodulating the audio creates a new signal whose shape is identical to the envelope of the original audio signal. Group delay always shows the time delay of the frequency components of the new signal, without having to explicitly obtain that new signal, thus avoiding demodulation of the original audio signal in order to obtain that new signal.  So I understand that this is what makes the group delay concept complicated, and therefore so many audio books, including pro-audio text books, do not show enough detail to demonstrate that this is what is happening.  That doesn't diminish the utility of what those textbooks teach, but I believe that every use case in this wiki article should give an example that shows Group Delay being used in the context of how group delay actually works.  If the reader can gleen this context from the cited references, that would be adequate. Ohgddfp (talk) 18:10, 26 March 2020 (UTC)


 * I have seen in Google books some of Pat Brown's work, and looking for more of Pat Brown (where to find in Google Books? or free on-line?). I am looking for audio references that I can cite that include a specific example of interpreting an example group delay for loudspeaker driver timing, room corrections, speech analysis, etc., with a graph of the corresponding time domain audio signal example. Such an interpretation should show what the Group Delay reveals that simpler views do not show.  Now SynAudCon's online training page "Hand-in-Hand Phase Response and Group Delay " by Pat Brown is not a good example in the context of this wiki article because the example does not give an interpretation beyond what the time domain impulse plot already shows. Ohgddfp (talk) 22:33, 26 March 2020 (UTC)

The Lead Section
I am changing the lede, part of a series of improvements to this article. The subject matter is difficult to explain and understand, even for engineers. The characterization of some changes and other changes to come are:


 * 1) Accessible. Must be as simple as possible while still being entirely correct.  The emphasis must be on conceptualization.  To that end, no unneeded jargon.
 * 2) The topic is abut the mathematically quantifiable properties of signal processing devices such as amplifiers or filters. Anything other than how they are quantified is not real, and is off topic.
 * 3) The current lede (before my changes) mentioned "envelope". This is actually confusing for two reasons: One, in amplitude modulation, although the period of the intelligence (baseband modulating signal) is identical to the period and shape of the envelope, the frequency that is likely to be inferred regarding the passband signal does not exist in the passband signal at all, but exists only in the baseband signal. This is a point of very common confusion. And two, with other kinds of modulation more common these days, FM broadcast, complex modulating schemes like QAM, FSK and other digital signaling, SSB, often does not show a discernable envelope when looking at the passband (RF) signal.  So in terms of gaining a conceptual understanding, "envelope" is more likely to mislead rather than inform at an early point in the article (like the lede).
 * 4) Group delay, as mentioned by others on this talk page is quite complicated in concept. For example, it cannot give any difinitive or quantifiable information outside the context of modulation and demodulation.  And group delay cannot even give the time delay of the DUT frequency components.  Instead, in that context (mod and demod), group delay only gives the phase delay that the passband DUT contributes to the system as a whole, where both system input and system output are both baseband.  Correcting this point of common misconception is so important that is should be in the lede.
 * 5) "function of frequency for each component" ? ? ? The frequency is the component.  Yes, group delay is indeed a function of frequency.
 * 6) "time delay of the phase" ? ? ? ? Well, time delay predicted (given) by the phase delay property.  Yes, both group delay and phase delay can be calculated correctly from the phase response of the unit under test.
 * 7) Lnear phase? The "offical" term means that the phase response of the DUT goes through the origin.  But for people just learning or brushing up on this stuff, "linear" also can mean "straight line" that does not go through the origin.  But that would not really be linear phase in the technical sense, and such a phase response would indeed cause waveform distortion due to phase irregularities of the DUT.  All this, plus the needed graphs and lots of explanation cannot fit into a reasonable lede.  At this point, early in the article, the phrase only confuses without helping much.  — Preceding unsigned comment added by Ohgddfp (talk • contribs) 16:54, 20 October 2020 (UTC)  Ohgddfp (talk) 22:32, 20 October 2020 (UTC) Ohgddfp (talk) 18:25, 21 October 2020 (UTC)

I just revamped the lead again, adding a mention of the article coverage. Any mention of that was taken out of the Introduction. Ohgddfp (talk) 18:25, 21 October 2020 (UTC)

changed talk section title to Lead section Ohgddfp (talk) 17:46, 23 November 2020 (UTC)

Another new lead today. Harmonizes better with the upcoming math. Ohgddfp (talk) 17:13, 8 December 2020 (UTC)

Referring to 6 above, "time delay of the phase", which I complained about 1.5 years ago, is still in the article. I see it was quoted by at least one forum, so I will correct it now. Ohgddfp (talk) 21:16, 14 May 2022 (UTC)

Moving the lede to be part of the introduction. Also writing a new lede.
Hi all. I noticed that the so-called lede that I put up only hours ago looks more like an introduction, so I'm adding it to the beginning of the existing introduction. I am also re-instating some of the original lede, but heavily re-written. Ohgddfp (talk) 22:37, 20 October 2020 (UTC)

I'm looking at the introduction about the 3rd paragraph. About: "Variations in group delay cause signal distortion, just as deviations from linear phase cause distortion." Huh? er What? This gets into it too quick. Many who look up "group delay" have only a weak understanding of these terms. The foundational aspects of group delay is the device's phase response. To understand phase, one must understand "frequency", and to understand "frequency", one must understand at least the fourier series, even though a continuous fourier transform applies more generally. The article should be a development from simple to complex, so that each concept builds on the previous. The fourier series is more understandble at the beginning, and the concepts can be made quite visceral in the reader's mind when frequency spectrums are not continuous. Graphs should be used to illustrate this, eventually leading to the phase response of a system. Both group delay and phase delay can be calculated exactly from the system phase response. Issues dealing with phase are covered first, with some more vivid demonstrations, and then showing how phase delay and group delay come about as tools to more directly analyze signal distortion. Too much math at the beginning is not an efficient route to conceptional understanding. And certainly does not belong in the "introduction", which should only give a definitive use case and how the development of understanding is to proceed. Ohgddfp (talk) 09:45, 21 October 2020 (UTC)

Time to add background - Intro Changed a bit - What the background section entails
In the introduction, about: "Variations in group delay cause signal distortion, just as deviations from linear phase cause distortion." Well, yes it does, but not the other way around. Severe distortion caused by phase irregularities can easily occur in the frequency band of interest where the group delay is perfectly flat. Better not to bring this up until the reader has the "why and where for" of this. I already altered some of the material in the introduction. Now is the time to add new sections to address development of group delay and phase delay theory, starting with a gentle but quantitatively correct treatment on the fourier series. Ohgddfp (talk) 10:08, 21 October 2020 (UTC) So now I added a Background section.

Added Basic Sinusoid - What's Next
I still need to add a figure for this. I know about a nice animation in Wiki Commons for this, but I want to emphasize the details in the degrees, and also stick only with the more familure sine. It's easier to examine if nothing is moving. The next subsection is "Phase". I also need to insert something about fourier, which will be in the form of a lab experiment simulation, do-able in real life for 100 years now. Anyone can try this at home to get a visceral understanding of fourier series. Also some cosines multiplied by cosines, cosines multiplied by sines, etc. Some math will accompany this to show how it plays out quantitatively. Ohgddfp (talk) 10:58, 21 October 2020 (UTC)

Introduction Section
I made changes to the introduction. The heart of the article (plus the background theory section) will include simpler mathematics that still correctly quantifies in a mathematical and practical sense, all of the issues related to the topic. I do have plans to, later in the article, to use the higher level math that is already there. Such math, as it is certainly higher level to most readers, certainly doesn't belong in the introduction. Although the info about "approximations" seem right, I am uneasy about any discussion of approximations until the math with no approximations is covered first. To aid conceptual understanding, it' much easier (and the math is much easier) to deal with a "group of frequencies" in a literal sense, where the frequency spectrum is only the dirac impulses that result from the periodicity of test signals. For a useful actual signal, where the intelligence is always new information rather than an endless repeat of the same old information, an aperiodic signal requires the continuous fourier transform, which will also be covered by building on the simpler math that comes before it. Ohgddfp (talk) 18:44, 21 October 2020 (UTC)

I am moving the existing math out of the Introduction Section, because it belongs in the heart of the theory. Ohgddfp (talk) 11:52, 23 October 2020 (UTC)

The introduction was missing a critical condition for group delay of a device passing a "passband" signal equaling the phase delay for "baseband". That critical requirement is that the device phase response in frequency range contained in the signal must be a straight line. I updated both the lead and the introduction accordingly. Ohgddfp (talk) 17:52, 23 November 2020 (UTC).

I am thinking also that the actual use case for group delay needs to be included in the introduction, emphasizing the benefits of group delay as a measurement tool in the context of a modulated signal, such as relatively easy calculation, especially with an oscilloscope, that reveals some aspects of the baseband signal after demodulation. Already in the article is the special case of a straight-line portion of the device's phase response as it relates to group delay. Also should be mentioned is a specific property and more general property of group delay device property for a modulated signal, in that the group delay function over the frequencies contained in the passband (modulated) signal reveals limits on the maximum possible time delay irregularities of the demodulated baseband signal, but cannot give actual time delay for a given frequency component of the demodulated baseband signal. As I am now preparing the simple math to demonstrate this aspect of group delay, the math itself will inform me of the best wording. So some better wording and graphics that harmonize better with the wording are forthcoming. Ohgddfp (talk) 15:50, 25 November 2020 (UTC)

Made some clarifying changes in what is now the Introduction Section. Ohgddfp (talk) 23:33, 14 May 2022 (UTC)

Currently called the "overview", this section outlines the complications of group delay, which is much harder to describe. More focused on the essential idea. No need to talk about cascading as before. Ohgddfp (talk) 17:15, 8 December 2020 (UTC)

The new introduction, introduced yesterday, is very bad. The first sentence of an article is critical. It should always define the topic. Instead, the new intro's first sentence says that phase and group delay "can be calculated exactly from an LTI device's phase vs frequency property", but gives the reader no clue whatsoever as to what they are, and even introduces additional confusion by introducing the undefined acronym "LTI". I understand the desire to be more rigorous in the lede, but the lede needs to also be clear and easy to understand. --Srleffler (talk) 21:48, 15 May 2022 (UTC)
 * It looks like it was reverted, which may be for the best right now. The lede is way too long anyway. I agree with some of the critisism. LTI should read "Linear Time Invariant (LTI)", which should be a link to the definition because a system needs to be substantially linear in order for the topic of this article to have a consistent meaning. Also, since this article is based on the concepts of simply "Phase", this should be a link as well. So I will put in a new lede. If the reader does not understand the concept of a function, there is no hope for the article to be useful to such a (normal person) reader.  So group delay and phase delay are complicated concepts, even for engineers. Understanding of the meaning of the links is a pre-requisit for grasping the concepts of this article. Otherwise it will be gobbledygook for all readers.  For example, the link "topics in Ohgddfp (talk) 21:16, 23 May 2022 (UTC)


 * I see in the above comment I was confusing the lede with the introduction. As far as the lede is concerned, it covers the general idea and has links to the other wiki pages that are absolutely required for the reader to move onto the deeper and more specific understanding in the Introduction.  In other words, the lede links are a pre-requisite for further understanding of the topic. The lede was very long with many redundancies and some inaccuracies.  I was able to shorten the lede and still give the prerequisites as links.  If the prerequisites are not somewhat understood by the reader than IMHO, the reader has no hope for any deeper understanding. Ohgddfp (talk) 01:03, 24 May 2022 (UTC)


 * I would like to discuss the underlining for the italizided sentence. This sentence is the single most revealing difference between Group Delay and Phase Delay, and I think the formatting should reflect this.  Thoughts? Ohgddfp (talk) 20:31, 26 May 2022 (UTC)
 * We don't generally use underlining for emphasis on Wikipedia. Stick with italics for emphasis. This is probably covered somewhere in the Manual of Style.--Srleffler (talk) 21:34, 28 May 2022 (UTC)


 * About "A varying phase response as a function of frequency, which can be described in terms of group delay and phase delay": Yes, phase response indeed can be described in terms of phase delay.  But in general, phase response cannot be described in terms of group delay. That's because group delay, being a derivative of the phase response, looses the constant term of the phase.  If we try to integrate group delay to get back to phase response, without any priori knowledge of the phase response, we can't because the constant term in the integrand is unknown.  How to modify this paragraph to reflect this reality ?  (By the way, the other way around, group delay being described in terms of phase response does work.) Ohgddfp (talk) 03:01, 29 May 2022 (UTC)
 * I think it is correct as written. The group delay and phase delay together describe the phase response.--Srleffler (talk) 16:44, 30 May 2022 (UTC)
 * Well, my position is that they do not "...together describe the phase response." Phase delay alone can correctly describe the phase response.  Here's proof.  Phase delay--a function of frequency measured in radians per second--is denoted as "tau as a function of omega". So phase delay is defined as: $$ \tau_\phi(\omega) = - \frac{\phi(\omega)}{\omega} \ $$, where phase response is denoted as "phi as a function of omega".  This equation can be algebraically manipulated into $$ \phi(\omega) \ = -    \tau_\phi(\omega) \omega $$ .  In words, the phase response, "phi_as_a_function_of_omega", equals minus "tau_as_a_function_of_omega" multiplied by omega.  This demonstrates phase delay alone as being able to completely and correctly, all by itself, describe the phase response. Not only is group delay not needed to describe phase response, group delay cannot even help describe phase response, due to group delay missing the (possibly non-zero) constant term of the phase response. Ohgddfp (talk) 13:13, 31 May 2022 (UTC)

Group delay and phase delay in audio
About a recently added sentence from a paper by Leach: "The waveform of an audio signal can be reproduced exactly by a system that has a flat frequency response over the bandwidth of the signal and a phase delay that is equal to the group delay [over the bandwidth]." (IMHO, the reader will interpret the quote as if the author included my bracketed expression.) Although the sentence is true, this might lead the reader to believe that exact reproduction can be achieved, even if group delay and phase delay are not flat over the signal's bandwidth, as long as they're equal. But that would be wrong. Ohgddfp (talk) 06:02, 28 May 2022 (UTC)Ohgddfp (talk) 05:59, 28 May 2022 (UTC)

And worse, that reader's hypothetical conclusion is wrong for an unexpected reason. It turns out that if phase delay and group delay are not equal to each other, then a requirement for exact reproducton—a flat phase delay—is not achieved. Ohgddfp (talk) 05:59, 28 May 2022 (UTC)
 * I think the wording is fine. "The waveform of an audio signal can be reproduced exactly by a system that has a flat frequency response over the bandwidth of the signal and a phase delay that is equal to the group delay," reads to me that both conditions must be met, not one or the other.--Srleffler (talk) 21:39, 28 May 2022 (UTC)
 * Do you mean "both" of these two conditions being met?  1) Phase delay is flat   2)Group delay is flat Ohgddfp (talk) 01:18, 29 May 2022 (UTC)
 * No. 1) frequency response is flat, and 2) phase delay is equal to group delay. Of course these conditions also imply that both phase and group delay are flat. --Srleffler (talk) 16:36, 30 May 2022 (UTC)
 * There are systems (parts of which have non-minimum phase) that have perfectly flat magnitude frequency response, perfectly flat group delay, but at the same time wildly unflat phase delay. If a baseband signal were to be applied to such a system, severe signal distortion will result from the wildly unflat phase delay. Ohgddfp (talk) 17:22, 30 May 2022 (UTC). Now back to the original sentence, the sentence wording logically has both of the two meanings, "as long as they're equal and both perfectly flat" (which is the correct meaning), as well as "as long as they're equal and not flat" (the generally incorrect and impossible meaning). Since in general one meaning is true and the other meaning if false, then, as opposed to what I originally thought, the sentence, standing on its own as a statement, is actually false.  Only if the false meaning has already been previously established in the article as false, can the sentence be true only in that context. But that context has not been previously established, and I believe that forward references to rectify this problem (to teach the reader that one meaning had to be true and the other meaning had to be false), are too tough on the reader, especially with a topic as tough as this one. Ohgddfp (talk) 18:16, 30 May 2022 (UTC)
 * "As long as they're equal and not flat" is not a reasonable interpretation of the sentence we're discussing. --Srleffler (talk) 04:01, 31 May 2022 (UTC)
 * I'm not sure if there is agreement on what I already said about the mathematical possiblity of phase delay being wildly non-flat while simultaneously both the magnitude frequency response and group delay are flat.
 * Moving on to the sentence itself we are discussing, which is the quote from Leach:
 * "The waveform of an audio signal can be reproduced exactly by a system that has a flat frequency response over the bandwidth of the signal and a phase delay that is equal to the group delay."
 * Now I believe this quote ought to be a correct statement standing on its own. Sure, every statement can build on previous content, but a flat out statement such as this quote should also be correct when read in isolation without the reader needing more knowledge. So, looking at the logical meaning of the words in this statement, there is no mention of whether or not phase delay and group delay are flat. Without additional reader knowledge there is no reason why the reader would know anything about flat or non-flat. So without that addtional knowledge, the reader has a perfect right to logically conclude, on the wording alone, that the logical meaning of the words is both:
 * 1) "... phase delay is equal to group delay (and both are flat)"
 * and 2) "... phase delay is equal to group delay (and both are non-flat)".
 * Therefore, without explicitly mentioning flat or non-flat, the statement is therefore actually proporting that both meanings are true.
 * Now if any part of a statement is false, then the statement as a whole is false. One of the meanings--both are non-flat--which the reader has a right to believe is true, is of course false. Well, that makes the statement as a whole false. Ohgddfp (talk) 14:35, 31 May 2022 (UTC)


 * Or does it? Well, the real problem, although the statement is literally true (see how I am vacillating back and forth about this), I think it is reasonable to conclude that a reader will typically and incorrectly infer that it is possible for group delay to be equal to phase delay while at the same time their functions (over the frequency range of interest) are non-flat. So I'll wait a couple days for comments, and then lighten the reader's burden by following the quote from Leach with "This means that both phase delay and group delay functions are flat, which fullfills the conditions to eliminate linear distortion (as opposed to non-linear distortion) because the combination of a flat magnitude frequency response with a flat phase delay is necessary and sufficient to prevent linear distortion." In addition, the terms "linear distortion" and "non-linear distortion" should be links. Ohgddfp (talk) 11:50, 1 June 2022 (UTC)

Differential time-delay distortion
Under the heading "Group delay in audio", about


 * "The waveform of an audio signal can be reproduced exactly by a system that has a flat frequency response over the bandwidth of the signal and a phase delay that is equal to the group delay. Leach[15] introduced the concept of differential time-delay distortion, defined as the difference between the phase delay and the group delay, which is given by:
 * $$ \Delta\tau = \tau_\phi - \tau_g $$.
 * An ideal system should exhibit zero or negligible differential time-delay distortion.[15]":

I cannot find a single actual specific real life example of a use case for this differential time-delay distortion measurement, other than in the founding document by Dr. Leach. So this cannot be an example of "Group delay in audio" where no example of its gainful use exists. I therefore recommend its removal. Ohgddfp (talk) 23:13, 5 June 2022 (UTC)
 * Is the concept useful for theoretical analysis? Not everything has to be practically applicable "in real life" to be worthwhile.--Srleffler (talk) 03:20, 6 June 2022 (UTC)
 * If the concept is useful for theoretical analysis, let's use it to theoretically analyze something that will support "Group delay in audio", where using another simpler concept, already covered in this wiki article, is not as enlightening. (Also, the analysis should include a plausable hypothetical example.) Ohgddfp (talk) 03:59, 6 June 2022 (UTC)

First line is wrong
First line claims phase delay and group delay (metrics of phase distortion) describes how frequencies are delayed in time for LTI systems, like microphones and loudspeakers, etc. These (passive) devices contain no means to delay anything in time and the whole statement is plain wrong. In reality, for such passive analog devices, these metrics describe phase distortion, not time delays. Further treatment with carrier and wave packets relates to telegraphy, which is (AFAIK) where phase delay and group delay were first treated in relation to electrical signals (by Henry Nyquist). Later, this article mentions negative group delay and that this does not violate causality, which is a big red flag clearly indicating that positive and/or negative group delay (in these passive analog systems, like microphones and loudspeakers) simply cannot be related to time delays (since negative group delay would then imply time advancement). Cfuttrup (talk) 17:35, 27 April 2024 (UTC)


 * Signals actually may be delayed in time when passing through LTI systems. Passive devices are able to delay signals. A simple example is a transmission line, which sometimes is even used as a delay line (for signals within the transmission line's bandwidth). The equations governing capacitances and inductances (which make up passive devices) are time-dependent (the interval of integration in an inductor's or capacitor's current-voltage relationship integral form includes all time up to the present). So they can be imagined to act as remembering past voltages and currents. They don't just respond to the present voltage and current, but also incorporate past voltages and currents.
 * Negative group delay is a more difficult topic. But since I don't think you get regular delay I don't think you will get negative group delay. Em3rgent0rdr (talk) 18:52, 27 April 2024 (UTC)
 * You mention that you think I don't get (=understand) regular delay.
 * The word regular means your comment is ambiguous because are we talking about time, group, or phase delay? All of them are regular.
 * Positive group delay and negative group delay are not two different issues and can (should) be explained in the one and same way.
 * Let me reiterate that my initial comment only points to passive devices like loudspeakers and microphones, and the false claim that frequencies are DELAYED in time.
 * Time delay of the signal in the real physical sense is not happening in a passive circuit, as you mention capacitors and inductors, or (passive) loudspeakers, but as you also (correctly) mention in your response, what can be observed is the response time = integration time.
 * What we see is a response time where energy is built up. The energy buildup happens at a slower pace than the signal itself and is often shown as the envelope of the signal.
 * It is therefore well understood that the group delay relates to the envelope of the signal, not the signal itself, and therefore - please try to reread the very first sentence of this Wikipedia article again - it is simply wrong that group delay describes the time delay of the signal, or for that matter frequency components of the signal.
 * In my first comment, I already foresaw the issue with telegraphy and the invention of delay lines (by W. E. Thomson). These delay lines take advantage of the integration time and, for example, ensure that it is the same time for a chosen frequency range (the bandwidth of the delay line), but it is not for all frequencies from DC to light, so it is not a true time delay in the physical sense.
 * There are no delay lines built into passive loudspeakers or microphones as a whole (i.e., in the audible range from 20 Hz to 20 kHz), so the whole argument that it COULD BE is moot.
 * You mentioned, and I quote, "So it can be imagined to act as remembering..." and the word imagined is sometimes called apparent, i.e., there is an apparent time delay.
 * I recognize this from other discussions, but as you probably know from discussions about negative group delay, imagining a time advancement is in fact fooling yourself (and others).
 * The apparent (or imagined) time delay is easily proven wrong, simply by cutting off the input signal and observing the response of the circuit (or device). The response of the system to the cut-off of the input signal is immediate and without delay and accumulates as the response (integration) continues.
 * For loudspeakers, group delay is a metric for smearing in time of the input signal, as dictated by the integration time, or rather, the smearing is observed when group delay isn't constant and can be expressed, for example, by the first-order derivative of group delay (named dispersion).
 * Group delay can be different at different frequencies (as by band-limited passive loudspeakers or microphones), or it can be (roughly) constant for certain circuits, e.g., delay lines, and you cannot observe time-smearing of frequencies within the limited frequency band where group delay is constant (i.e., phase response is linear).
 * Therefore, for loudspeakers and microphones, group delay is a metric of phase distortion, not time delays.
 * Please bear in mind that my input is based on my knowledge about loudspeakers (and other band-pass circuits). I am not considering digital systems, for which signal processing tricks exist where phase can be linearized (in exchange for a true time delay), for example, using FIR filters.
 * I find it difficult to formulate one sentence that universally covers everything, unless one reverts to the equation(s) because these define what group delay and phase delay are, and the equations also define what they aren't. For example, there isn't any time delay anywhere in these equations, only metrics for the rate of phase changes.
 * I honestly think we to some extent agree, because in your reply above, you steer in the direction of an imagined time delay, which implies it is not a real one.
 * Can the first sentence be rescued by adding the word imagined? Maybe.
 * Associating group and phase delay with time delay is, in my opinion, dangerous and misleading, and it is the root cause of lots of confusion, as can be readily observed all over the internet. Please avoid facilitating this confusion on Wikipedia. Cfuttrup (talk) 18:48, 4 June 2024 (UTC)