Home |
Search |
Today's Posts |
#1
|
|||
|
|||
Is differential phase error important?
If all frequencies are equally delayed, the phase response is a
straight line (on a linear frequency scale). If I plot my delay it is flat down to about 4k, then reduces in a curve. What does it do to the sound? What should a design aim for? Is this why pedal steel guitars swim about? Are they meant to do that? Can someone point me in the direction of enlightenment please. cheers, Ian |
#2
|
|||
|
|||
Hi RATs!
Charting data points is interesting. It allows insight into phenomena. Listening is fun. It allows the experience of music. Trying to correlate charts and music is an interesting challenge. Some people would rather think than listen. Not me Listen and think, that's the ticket! Happy Ears! Al Alan J. Marcy Phoenix, AZ PWC/mystic/Earhead |
#3
|
|||
|
|||
"TubeGarden" wrote
Charting data points is interesting. It allows insight into phenomena. Listening is fun. It allows the experience of music. Trying to correlate charts and music is an interesting challenge. Some people would rather think than listen. Not me Listen and think, that's the ticket! Music is a social enterprise. Fidelity is a moral imperative. Thinking needs talking in order to make sense. Music does its own talking. cheers, Ian |
#4
|
|||
|
|||
Hi Ian . . .
I think that uniform phase response is important . . . but of course people do have different opinions . . . . . . but it's definately a given that the phase response and group delay is extremely important when applying feedback. There are a number of volumes on the subject, both with regard to vacuum tubes and solid-state circuits. In this instance, it is necessary to understand both frequency-dependant and frequency-independant delay to predict and optimize the amplifier's operation when feedback is applied. When feedback is not being used, then the frequency-independant delay is generally regarded as inconsequential . . . what is important is frequency-dependant delay. If I read your post correctly, you have a difference in delay, and consequently phase, of mid-band frequencies and high-band frequencies . . . this would definately be something I would want to correct . . . especially given the high sensitivity of human hearing in the 3-4KC area. Regards, Kirk Patton "Ian Iveson" wrote in message ... If all frequencies are equally delayed, the phase response is a straight line (on a linear frequency scale). If I plot my delay it is flat down to about 4k, then reduces in a curve. What does it do to the sound? What should a design aim for? Is this why pedal steel guitars swim about? Are they meant to do that? Can someone point me in the direction of enlightenment please. cheers, Ian |
#5
|
|||
|
|||
Music is a social enterprise. Fidelity is a moral imperative.
Thinking needs talking in order to make sense. Music does its own talking. cheers, Ian Hi RATs! Sales are a social enterprise. Some bits of music can be bought and sold. Fidelity is just a cheap sticker to put on the bits of music we have for sale. The greater glory of music, which cannot be bought, nor sold, no matter how many lawyers we hire, does not require any silly sticker. People buy music, with and without stickers. Stickers are not the big deal. Music is the big deal. The only people who think the stickers are the important thing are the sticker salesmen, and their feeble customers Happy Ears! Al Alan J. Marcy Phoenix, AZ PWC/mystic/Earhead |
#6
|
|||
|
|||
Ian Iveson wrote: If all frequencies are equally delayed, the phase response is a straight line (on a linear frequency scale). If I plot my delay it is flat down to about 4k, then reduces in a curve. What does it do to the sound? What should a design aim for? Is this why pedal steel guitars swim about? Are they meant to do that? Can someone point me in the direction of enlightenment please. cheers, Ian By differential phase error, do you mean having a different phase angle in the signal applied to the output stage? Say one signal was at +90 degrees, and the other at -90 degrees, then since the two sigs would normally be 180 degrees apart, the signa; applied to the output stage would be 0 degrees, ie a common mode sig is applied, and there is almost zero output. So if the applied signal to the output stage drifts from 180 degrees difference, the signal is attentuated at the output, but the amp works just as hard dissipating heat withing itself, and the imd of any signal passing through is increased. Patrick Turner. |
#7
|
|||
|
|||
"Patrick Turner" wrote
[see below] No. Sorry, I should be more clear. I'm looking at an ordinary bode plot, and homing in on the phase error. It has a log frequency scale, and shows phase angle on the Y axis. This view is useful for determining phase margin for purpose of stability (see kirk's post, but you know that anyway of course). It may not very useful for wondering how it might sound, or at least how the mono sound sounds...let's leave out stereo effects for the moment. If all frequencies are delayed by the same *time*, then the sound should be the same...the concert just starts a little sooner. The same delay applied to all frequencies would result in a shift in phase angle that would be proportional to frequency. So if you plotted phase angle v frequency on a linear scale, you would get a straight line, sloping down to the right. As an example, if the delay is 1us, then at 1k the phase error would be 360/1000 degrees, and at 2k it would be 720/1000 degrees, which is twice the angle. Now, being a straight line, its differential will be a constant, so if you plot da/df, where a is phase angle and f is frequency, you should get a horizontal line. (it follows that da/df is proportional to delay, which makes sense if you think about it, I hope). Well, if you examine the bode plot of a typical valve amplifier, especially with no global negative feedback, you will see the phase plot shows phase lead at low frequencies, curving down to zero error at around 1k perhaps, and then an increasing delay, sloping down to -180 degrees somewhere beyond the point where gain is zero, hopefully, around 1MHz? If I translate that into delay, with my amps and probably with most, it is constant from around 5k to perhaps 100k. Below 5k it falls to zero and then accumulates an increasing negative value, curving down sharply below 100Hz. Negative delay is lead. Lead, or negative delay, is tricky to get my head round. It has two important features. Firstly, there is the matter of how the sound gets ahead of itself in the first place. In the transient domain you can follow the response to the beginning of a steady, low-frequency tone, and see that the wave marks time for a cycle before coming in ahead of the next. So in the frequency domain it begins with a slightly higher frequency until the shift has taken place. Secondly, you must see all this as relative. But for the first few cycles, the situation is as if the bass had happened just before the mid frequencies, which amounts to the same thing as mid frequency delay. At the moment I can't see why that should make the sound different. Not if the orchestra is playing a sustained chord, anyway. It should only make a difference to the attack of each note, and I dunno under what circumstances that would be audible. But what if the same instrument plays a run of notes spanning several octaves? Or, in the extreme, does what a pedal steel guitar does, and slide whole chords? Does shifting delay applied to a sliding tone sound like movement? What do SS amps do? What do op-amps do? Does all that feedback eliminate the phase error? cheers, Ian |
#8
|
|||
|
|||
"Ian Iveson" ( snip load of confused puppy talk) Does shifting delay applied to a sliding tone sound like movement? What do SS amps do? What do op-amps do? Does all that feedback eliminate the phase error? ** Ian - do try to get your facts straight. The phase shifts you see on a Bode plot are ACCOMPANIED by GROSS response errors !!!! Once NFB is applied the response gets nice and flat and the phase shifts disappear too !!!!! Phase shift of the output signal compared to the input signal in hi-fi power amps is there as a direct result of sub-sonic and super-sonic roll offs in the overall response. The shift follows the order of the filters used - typically they are first order so you get 45 degress shift at the 3 dB down points. If you really want something to fret about - go measure all the phase shifts in a 3 way loudspeaker !!!!! ............ Phil |
#9
|
|||
|
|||
Ian Iveson wrote: "Patrick Turner" wrote [see below] No. Sorry, I should be more clear. I'm looking at an ordinary bode plot, and homing in on the phase error. It has a log frequency scale, and shows phase angle on the Y axis. This view is useful for determining phase margin for purpose of stability (see kirk's post, but you know that anyway of course). It may not very useful for wondering how it might sound, or at least how the mono sound sounds...let's leave out stereo effects for the moment. If all frequencies are delayed by the same *time*, then the sound should be the same...the concert just starts a little sooner. The same delay applied to all frequencies would result in a shift in phase angle that would be proportional to frequency. So if you plotted phase angle v frequency on a linear scale, you would get a straight line, sloping down to the right. As an example, if the delay is 1us, then at 1k the phase error would be 360/1000 degrees, and at 2k it would be 720/1000 degrees, which is twice the angle. Now, being a straight line, its differential will be a constant, so if you plot da/df, where a is phase angle and f is frequency, you should get a horizontal line. (it follows that da/df is proportional to delay, which makes sense if you think about it, I hope). All that makes sense. Well, if you examine the bode plot of a typical valve amplifier, especially with no global negative feedback, you will see the phase plot shows phase lead at low frequencies, curving down to zero error at around 1k perhaps, and then an increasing delay, sloping down to -180 degrees somewhere beyond the point where gain is zero, hopefully, around 1MHz? If I translate that into delay, with my amps and probably with most, it is constant from around 5k to perhaps 100k. Below 5k it falls to zero and then accumulates an increasing negative value, curving down sharply below 100Hz. Negative delay is lead. Lead, or negative delay, is tricky to get my head round. It has two important features. Firstly, there is the matter of how the sound gets ahead of itself in the first place. Ah, the wonders of sound arriving at the ear, well before it set out on the journey. Everyone can handle phase lag, or sine waves "slipping back" a few degrees, but arriving ahead? Its confusing until you realise there has to be a start time to the stream of sine waves, and the energy is sped up a bit.. In the transient domain you can follow the response to the beginning of a steady, low-frequency tone, and see that the wave marks time for a cycle before coming in ahead of the next. So in the frequency domain it begins with a slightly higher frequency until the shift has taken place. Secondly, you must see all this as relative. But for the first few cycles, the situation is as if the bass had happened just before the mid frequencies, which amounts to the same thing as mid frequency delay. At the moment I can't see why that should make the sound different. Not if the orchestra is playing a sustained chord, anyway. It should only make a difference to the attack of each note, and I dunno under what circumstances that would be audible. The less phase shift, the better. But what if the same instrument plays a run of notes spanning several octaves? Or, in the extreme, does what a pedal steel guitar does, and slide whole chords? Does shifting delay applied to a sliding tone sound like movement? You could play several oscillators all simultaneously, and all producing tones of related harmonics, and providing their F remain related to each other, slowly altering the phases of each, which will have random phase relationships, somewhat different to a string on an instrument, you theoretically shouldn't hear very much. What do SS amps do? What do op-amps do? Does all that feedback eliminate the phase error? Yes, and the phase relationships that exist in the original signal are quite accurately reproduced in a wide BW high NFB amp, with only a tiny error, and the open loop phase errors are corrected just like the other open loop distortions. Patrick Turner. cheers, Ian |
#10
|
|||
|
|||
"Ian Iveson" wrote in message news:HSvAb.957 As an example, if the delay is 1us, then at 1k the phase error would be 360/1000 degrees, and at 2k it would be 720/1000 degrees, which is twice the angle. But this is measuring the frequency-independant phase response of the network, which (without NFB) is inconsequential, unless you care that you are listening to your music 1uS later than it is coming into the amp. Now if there is a 1uS delay at 1KC, and a 2uS delay at 2KC, then you have a frequency-dependant phase error, and this is what we want to avoid. Well, if you examine the bode plot of a typical valve amplifier, especially with no global negative feedback, you will see the phase plot shows phase lead at low frequencies, curving down to zero error at around 1k perhaps, and then an increasing delay, sloping down to -180 degrees somewhere beyond the point where gain is zero, hopefully, around 1MHz? You didn't mention whether you are measuring frequency-dependant or frequency-independant phase error, or the sum of the two. There is a very good and very understandable AES paper on phase measurements with regard to transformers that was done by Deane Jensen, I think it makes these distinctions very clear . . . Jensen Transformers doesn't post it online but their website says that they'll mail you a copy if you ask for it. http://www.jensentransformers.com/apps_wp.html But what if the same instrument plays a run of notes spanning several octaves? Or, in the extreme, does what a pedal steel guitar does, and slide whole chords? Hmmm. This is a tough question . . . trying to describe the sound of a phase shift. One thing to keep in mind is that most all of the pitch fundamentals of instruments are in the midrange and below, it is their overtones that are in the higher frequencies. And even in a sustained note, what we perceive as "timbre" is very closely related to the amplitude and phase of the overtones with relation to the fundamental. Does shifting delay applied to a sliding tone sound like movement? I'm not sure what you mean by "movement". The closest effects device to this phenomenon might be a guitar "flanger". The term "flanging" (not related to, but also not exclusive of, "flogging", esp. among guitarists) originated from the practice of taking two tape recorders that are SMPTE'd together, with the same signal being recorded on both (and maybe being played back simultaneously, I forget) and sticking one's hand on the flange of the supply reel of one machine. This introduces a delay in one signal, and a phase difference that increases with frequency. Whether or not a given phase problem in an amplifier will sound exactly like a flanger, though, would be pure speculation on my part . . . I do think that it would be safe to say that it would have a significant effect on the perception of stereo image, probably room reverb, and in significant doses, on timbre. What do SS amps do? What do op-amps do? Does all that feedback eliminate the phase error? Well, I'm definately not RA!-RA! for solid state in general, but most SS stuff doesn't really have exactly these same problems in the audio band. The phase response of most SS amps and opamps are defined by a single-pole Miller capacitor across the voltage amp stage for high frequencies, and at low frequencies by the capacitors that couple in the signal or ground the feedback loop. Just as in a tube amp, "improvements" in measured closed-loop phase response via application of feedback doesn't really fix the problem, it simply trades one thing for another . . . i.e., closed loop stability for bandwidth, etc. And this can be a complex problem with SS topologies because bandwidth above unity gain is usually very high, and open-loop gain is also very high but varies significantly with frequency. The complicating factor in tube amps with phase response is, er, the output transformer. While these things IMHO have many benefits that outweigh their shortcommings, more complex time-domain behaviour is simply part of the package. It's just a question of how much and where (frequency-wise). Anyway, an interesting and thought-provoking subject. Best regards, Kirk Patton |
#11
|
|||
|
|||
Hi RATs!
Interesting! Important? That's up to you Happy Ears! Al Alan J. Marcy Phoenix, AZ PWC/mystic/Earhead |
#12
|
|||
|
|||
When an amp has different phase shift for different freqs it can definately
effect the sound. Some may view this effect as a positive thing others not. I guarantee some strange phase shifts in a guitar amp but whether or not this is causing some specific problem I have no idea. "Ian Iveson" wrote in message ... If all frequencies are equally delayed, the phase response is a straight line (on a linear frequency scale). If I plot my delay it is flat down to about 4k, then reduces in a curve. What does it do to the sound? What should a design aim for? Is this why pedal steel guitars swim about? Are they meant to do that? Can someone point me in the direction of enlightenment please. cheers, Ian |
#13
|
|||
|
|||
"Jimmy" When an amp has different phase shift for different freqs it can definately effect the sound. ** Seeing as phase shifts are produced by response changes that is true - but one hears the response changes. I guarantee some strange phase shifts in a guitar amp but whether or not this is causing some specific problem I have no idea. ** Those last four words are the key here. ........ Phil |
#14
|
|||
|
|||
Thanks for your contributions, Kirk. I have been thinking, slowly.
Trying to cover points you made in both posts not otherwise addressed. Yes I lumped all the phase response together...as I said originally, I was looking at the bode plot of an amplifier. My thought has been about where frequency-independent phase error (hardly response, since it is independent...) fits in. Originally I said that constant delay would appear as a straight line, sloping down to the right, and that any deviation from straightness would be frequency-dependent delay. Since delay is proportional to distance from source, I suggested "movement" because, if the delay varies with frequency, then the source should sound like it is moving forward or backward: into or out of the speakers. For this to be true relies on the assumption that, although we cannot sense distance directly, we can sense *change* in distance. Further, if you were listening to a group of sources playing in time (to themselves) from a distance, perhaps frequency-dependent delay places the notes, rather than the instruments, at different distances. Combined with the frequency-independent delay, the soundstage would be dynamically warped. Perhaps a touch of this movement allows us to locate the centre of motion of each instrument, and so enhances the sense of space. Too much feedback would kill the space, too little would jumble it. I blame all this on Jim. Right, erm, now to the bit I missed out. For a linear phase response to signify *only* constant delay, it must pass through the origin. Hence delay at DC would be indeterminate (0/0) rather than infinite. So a straight line not passing through the origin must have a component of delay which is in linear proportion to frequency. Trouble is, if you have a curved line, the difference between the two forms of non-linear delay is difficult to guess. Let's see...if I extent the straight section of my bode plot (on a linear frequency axis, remember) leftwards, should the extension pass through the origin if there is none of this second form of non-linear delay present? Without reading heaps of stuff, it looks like my phase response could be summarised by: theta = a/f + b +cf So if you take the differential as indicative of delay, you eliminate "b", highlight the effect of "a" by squaring its "f", and show that the "c" is insignificant because constant delay doesn't matter. That is what Menno van der Veen implies in his evaluation of Plitron transformers. BUT it is a mistake to eliminate "b"...constant phase error is a form of frequency-dependent delay too. Wonder why he did that. Perhaps I should ask him. I should also find out the speed of sound and work out what kind of order of magnitude these distances would be. If they are millimetres then I have been wasting my time. Lexicon have made a fortune from sound processing equipment, because someone there knows this stuff. Much more is known than is in the public domain, unfortunately. cheers, Ian "Kirk Patton" wrote in message m... "Ian Iveson" wrote in message news:HSvAb.957 As an example, if the delay is 1us, then at 1k the phase error would be 360/1000 degrees, and at 2k it would be 720/1000 degrees, which is twice the angle. But this is measuring the frequency-independant phase response of the network, which (without NFB) is inconsequential, unless you care that you are listening to your music 1uS later than it is coming into the amp. Now if there is a 1uS delay at 1KC, and a 2uS delay at 2KC, then you have a frequency-dependant phase error, and this is what we want to avoid. Well, if you examine the bode plot of a typical valve amplifier, especially with no global negative feedback, you will see the phase plot shows phase lead at low frequencies, curving down to zero error at around 1k perhaps, and then an increasing delay, sloping down to -180 degrees somewhere beyond the point where gain is zero, hopefully, around 1MHz? You didn't mention whether you are measuring frequency-dependant or frequency-independant phase error, or the sum of the two. There is a very good and very understandable AES paper on phase measurements with regard to transformers that was done by Deane Jensen, I think it makes these distinctions very clear . . . Jensen Transformers doesn't post it online but their website says that they'll mail you a copy if you ask for it. http://www.jensentransformers.com/apps_wp.html But what if the same instrument plays a run of notes spanning several octaves? Or, in the extreme, does what a pedal steel guitar does, and slide whole chords? Hmmm. This is a tough question . . . trying to describe the sound of a phase shift. One thing to keep in mind is that most all of the pitch fundamentals of instruments are in the midrange and below, it is their overtones that are in the higher frequencies. And even in a sustained note, what we perceive as "timbre" is very closely related to the amplitude and phase of the overtones with relation to the fundamental. Does shifting delay applied to a sliding tone sound like movement? I'm not sure what you mean by "movement". The closest effects device to this phenomenon might be a guitar "flanger". The term "flanging" (not related to, but also not exclusive of, "flogging", esp. among guitarists) originated from the practice of taking two tape recorders that are SMPTE'd together, with the same signal being recorded on both (and maybe being played back simultaneously, I forget) and sticking one's hand on the flange of the supply reel of one machine. This introduces a delay in one signal, and a phase difference that increases with frequency. Whether or not a given phase problem in an amplifier will sound exactly like a flanger, though, would be pure speculation on my part . . . I do think that it would be safe to say that it would have a significant effect on the perception of stereo image, probably room reverb, and in significant doses, on timbre. What do SS amps do? What do op-amps do? Does all that feedback eliminate the phase error? Well, I'm definately not RA!-RA! for solid state in general, but most SS stuff doesn't really have exactly these same problems in the audio band. The phase response of most SS amps and opamps are defined by a single-pole Miller capacitor across the voltage amp stage for high frequencies, and at low frequencies by the capacitors that couple in the signal or ground the feedback loop. Just as in a tube amp, "improvements" in measured closed-loop phase response via application of feedback doesn't really fix the problem, it simply trades one thing for another . . . i.e., closed loop stability for bandwidth, etc. And this can be a complex problem with SS topologies because bandwidth above unity gain is usually very high, and open-loop gain is also very high but varies significantly with frequency. The complicating factor in tube amps with phase response is, er, the output transformer. While these things IMHO have many benefits that outweigh their shortcommings, more complex time-domain behaviour is simply part of the package. It's just a question of how much and where (frequency-wise). Anyway, an interesting and thought-provoking subject. Best regards, Kirk Patton |
Reply |
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Forum | |||
Equalizers | Audio Opinions | |||
Doppler Distortion - Fact or Fiction | Pro Audio | |||
Phase Scope - How to read? | Pro Audio | |||
Transient response of actively filtered speakers | Tech | |||
Blindtest question | High End Audio |