Clipped and Shorn Posted December 5, 2002 Author Share Posted December 5, 2002 Good, I'll get fini to build me some. -c7s Quote Link to comment Share on other sites More sharing options...
Al Klappenberger Posted December 5, 2002 Share Posted December 5, 2002 Hi Guys, I haven't been following this thread but I did see some references to "group delay". I want to point out a trap that MOST people fall into. "Group delay" and "time delay" are two different things. Group delay is a measure of phase linearity ONLY. 20 mSec of "group dealy" can not be thought of as having anything at all to do with 20 mSec of propogation dealy through the air from speaker to you ear. There is NO way to equate one to the other! Group delay is meaningless! It is a measure of the rate of change of phase with a given change of frequency at ONE frequency only. That is the "slope" of phase change, that is, it is the "Derivitave" of phase. It is NOT time delay! AL K. Quote Link to comment Share on other sites More sharing options...
John Warren Posted December 5, 2002 Share Posted December 5, 2002 c Quote Link to comment Share on other sites More sharing options...
Al Klappenberger Posted December 5, 2002 Share Posted December 5, 2002 John, Yes, that's what I am saying! You can't hear group delay! The "two-clicks" thing is TIME DELAY! They are NOT related! Group delay can only be heard through headphones and with very specific test signals. Quite a bit of research has been one on it and group delay is NOT very important when reproducing music! Al K. Quote Link to comment Share on other sites More sharing options...
John Warren Posted December 5, 2002 Share Posted December 5, 2002 c Quote Link to comment Share on other sites More sharing options...
Al Klappenberger Posted December 5, 2002 Share Posted December 5, 2002 John, I too could site a bunch of dueling experts on this subject, but it's too much trouble! I would rather make the point with a simple exercise. If you do not believe what I am saying after this experiment, I will simply go quietly away. This subject is just too controversial to get into a major argument over! Use the following equipment: An oscilloscope having delayed and intensified sweeps, A frequency counter and an audio sine wave signal generator. Set up the oscilloscope to trigger on the input sine wave and set the delayed sweep to intensify a tiny spot after some known time delay that you simply set to that whatever you like. I suggest a delay that spans many cycles of the input waveform. This will represent true "time delay". Adjust the frequency of the input signal so that the intensified spot is at the peak of some displayed cycle and read the frequency from the counter. Move the input signal frequency so that the spot moves to the zero-voltage crossing of the waveform. This will represent a 90 degree change of phase. read the counter frequency again and find the difference. That will be the change in frequency. Using this formula calculate the "group delay": group delay = (degrees change in phase) / (change in frequency * 360) When you do this you will see that the two forms of "delay" are nowhere near equal! They are two entirely different things! There is a long list of "experts" who confuse these two. Group delay should be renamed something like "envelope distortion factor" or something like that. Calling it "delay" is extremely confusing to spite its universal acceptance. As the writer of a network analysis program I experimented with this quite a bit. There is also another form of "delay" to confuse things even more. That's "phase delay". It has a similar shape to group delay when plotted but does not come to the same "peak" at the corner of a filter passband as "group" delay. I have prepared a technical paper on "extreme slope" crossover networks to be published in AudioXpress magazine where I present an oscilloscope photograph of the time delay through the highpass channel of a filter having an over 100 dB / octave slope using a tone burst. The "group delay" was computed and found to be many times larger than the actual "time" delay through the filter. Again -- "Group" delay is NOT "time" delay! To blame "group delay" for hearing two distinct sounds is simply WRONG! Al K. Quote Link to comment Share on other sites More sharing options...
John Warren Posted December 5, 2002 Share Posted December 5, 2002 c Quote Link to comment Share on other sites More sharing options...
Al Klappenberger Posted December 5, 2002 Share Posted December 5, 2002 John, All I am saying is that many people confuse the tar out of "time delay" and "group delay". They toss the terms around like they were the same thing, even in technical papers. The real question is the impact of group delay on your hearing. I believe it is very minor in comparison to other factors. The B&L article you mentioned is clearly about "group delay" since they did the tests with "all-pass" filters. These are definitely "group delay" devices. The tests were done by listeners who were actually trained what to listen for. I can site other "experts" who conclude that it does not matter when reproducing music. If you have a tradeoff between group delay error and other factors you should give the group delay lowest billing! I have crossover networks in my Belles that have over 100 dB / octave slopes and have group delay peaks at the crossover frequency that are huge (about 3.6 mSec at 700 Hz). I have carefully listened to square waves, tone bursts and pulses adjusted slowly right across and at the crossover frequency. I could hear no difference in the character of the sound anywhere in the crossover region. I can see a double burst on either side of the crossover frequency using a conventional 12 dB / octave crossover that I built to test the theory. With the extreme slope crossover, there is only one except right at the crossover frequency. The result is a far smoother dispersion with the extreme slope network than with the conventional one. This is to spite the big group delay peak. Two drivers making sound at the same time cause "modeing" in the dispersion as they interact. You can definitely hear that because it causes bad frequency response! It is questionable if a dual burst is audible in itself. I suspect it is not, and that truly is "time delay", not "group delay". Al K. Quote Link to comment Share on other sites More sharing options...
John Warren Posted December 6, 2002 Share Posted December 6, 2002 c Quote Link to comment Share on other sites More sharing options...
Al Klappenberger Posted December 6, 2002 Share Posted December 6, 2002 John, The dealy peak is below the "B&L" limits or I would not have built the network! The fact is that 3.6 mSec IS huge for a crossover network at 700 Hz! Many people would have ruled out such an extreme slope filter just because of the size of that peak compared to the group delay throuout the rest of the passband swearing that the "ringing" would be overwhelming. It isn't! What I am saying is that the ear is nearly deff to "group delay" and phase errors except uner extreme conditions, like through headphones, which is how the writers of the B&L article measured it. At woofer port frequency, I doubt you can hear it at all! Look at his curve and tell me what the group dealy limit would be at 30 Hz! Here's a scan from a letter to the editor that sums up the situation. Phase errors are WAY over rated as a factor when other things are involved. AL K. Quote Link to comment Share on other sites More sharing options...
John Warren Posted December 7, 2002 Share Posted December 7, 2002 c Quote Link to comment Share on other sites More sharing options...
Al Klappenberger Posted December 7, 2002 Share Posted December 7, 2002 John, Nice program you used there. I got it too. It's WinISD.. The simple answer is NO.. Unless you have an accelerometer under you listening chair connected to an electronic differentiator and something to convert 30 Hz up to 1 KHz so you can hear it or some instrument that will operate at 20 to 30 Hz to linsten to it for you, you will NOT her the difference! At 20 to 30 Hz you feal it more than hear it anyhow! You are putting to much faith in published stuff on "group" delay! Cosider this simple point. If there has been extensive research on a subject, then the results must be in question. You do not need to research the obvious. That's why so much research has been done on the importance of phase distortion, group delay and caffeen in you coffee. It MUST be bad, so we are going to prove it, OR ELSE, SOMEHOW! On a second reading of you post.. You ARE still confuisng "time" delay with "group" delay! "fast" and "slow" do not apply here. "Group delay" has NOTHING to do with speed! It has only to do with change is the "slope" of phase! Al K. Quote Link to comment Share on other sites More sharing options...
jnorv Posted December 7, 2002 Share Posted December 7, 2002 I may be missing something here, but isn't that second plot exactly what a base reflex is suposed to do. At 20 Hz, a half cycle is 25 ms. Just what the program predicts. The port is out of phase with the front of the driver. Jim N Quote Link to comment Share on other sites More sharing options...
Al Klappenberger Posted December 7, 2002 Share Posted December 7, 2002 John, "Group delay" is about the most over-rate bunch of "Bull____" to come along since God knows what. Look at the attached graphic. It is a composite of the computed amplitude response and "group" delay of the high frequency channel of the woofer / squawker crossover I am using in my modified Belles. Look at the amplitude response. It goes down like the side of a barn and the group delay peak is almost exactly at the crossover of 700 Hz. Now look at the actual response take from an oscilloscope screen of a single cycle tone burst at 700 Hz of that network. The oscilloscope is triggered from the input burst making the two upper traces locked in time with the input burst. All channels are "chopped". The group delay is about 2 mSec. If you can find ANYTHING on the middle trace (the high frequency channel output) that corresponds to 2 mSec in "time", I will go away! Impulse response of any waveform, and therefor it's response in "time", is computed by a method called "Fourier analysis". The math to do it involves the summation of the instantaneous AMPLITUDE and ABSOLUTE phase of each and every frequency element in the spectrum. There is NO information required about the "CHANGE" in the phase of each harmonic element, only the instantaneous number of degrees. "Group" delay relates only to the rate of "change" in phase of a single frequency, not to it's delay in time! Al K. Quote Link to comment Share on other sites More sharing options...
Al Klappenberger Posted December 7, 2002 Share Posted December 7, 2002 Jim, Yes.. Who cares about what comes out of a bass reflex speaker below the port frequency anyhow? The woofer is unloading from the box and the sound is shutting down anyhow! The discussion about "group" delay down there is a moot point! AL K. Quote Link to comment Share on other sites More sharing options...
Al Klappenberger Posted December 7, 2002 Share Posted December 7, 2002 Oops - double post. I must have hit the button twice! AL K. Quote Link to comment Share on other sites More sharing options...
Clipped and Shorn Posted December 7, 2002 Author Share Posted December 7, 2002 "Yes.. Who cares about what comes out of a bass reflex speaker below the port frequency anyhow? The woofer is unloading from the box and the sound is shutting down anyhow! The discussion about "group" delay down there is a moot point! AL K." Close enough for jazz, I say, and certainly even closer for latin jazz and Stravinsky. {What about Jesse Crawford's Organ Favorites you say....well I don't know about that, how many of those organ notes go below 24 hz?}. OK, I have armed guards posted, I am cautiously moving the plywood out of van and into the shop. The wind is my sails. Might even buy a new saw-blade for the table saw. "Serenity now." -c7s Quote Link to comment Share on other sites More sharing options...
Al Klappenberger Posted December 7, 2002 Share Posted December 7, 2002 Clipped, Good deal, but watch that group delay. Move the sheets of wood one at a time. If you move them as a "group", they may get there before you move them. Group delay can go negative you know! (Look at the curve of dealy I posted. It goes negative through the "notch" frequency od the filter). Weird stuff! Al K. Quote Link to comment Share on other sites More sharing options...
John Warren Posted December 7, 2002 Share Posted December 7, 2002 c Quote Link to comment Share on other sites More sharing options...
Al Klappenberger Posted December 7, 2002 Share Posted December 7, 2002 John, The derivitave of phase versus frequency plot is "group delay". That is the definition of it. What it tells you is how equally the phase is changing with frequency over a specific frequency range. If a network has flat (all equal) group delay over a particular frequency range, and if all the individual elements of the "group" of elements (sidebands) that make up a particular complex waveform are passing through that network, the elements will then reassemble to make the waveform in an undistorted manor. It will usually not tell you how "fast" it will get through that network. The only thing I know of that has "group delay" equal to "time delay" is a length of transmission line. That is a length of RG-59U coaxial cable at radio frequency for example. I have never seen a direct relationship between "group" delay and "time" delay. In the filter design world, it seems that the wider the bandwidth over which group delay is flat, the closer the two will be to each other. I don't know why or even if it is universally true. To talk about the group delay at a single frequency is totally useless. There is no "group" involved! "Envelope distortion" is often described and specified in two ways. Either flatness of group delay over a given frequency range or in terms of +-X degrees error from a best fit straight line between two frequencies. The only time group delay is consider at one frequency is if you are "phase matching" two separate filters. Even then the matching is usually over a range of frequencies. If you like, I will post sample plots of a linear phase bandpass filter and a Butterworth or Chebyshev filter showing both group delay flatness and phase error from a best-fit straight line on each. I can also use Fourier analysis to show what happens with each in the time domain. It's a pain in the butt to do all the plots and upload them, but I'll do it if you are really interested. Al K. Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.