Post by newey on Apr 26, 2009 8:43:38 GMT -5
In another thread, Ashcatlt summed up some issues concerning phase shift and phase reversal in audio signals. I have reproduced his post here so that further discussion can be had, and so it can be easily located.
Ash's original post:
"Apr 24, 2009, 4:05pm, sumgai wrote:
A variable all-pass filter? Errr, isn't that just a fancy name for an amplifier? ;D
No. I was talking about an all-pass filter with a variable cutoff frequency. Most of these do not offer any actual gain. I guess it might be considered an amplifier in the same sense that a unity gain buffer is called an amplifier…
An all-pass filter passes all frequencies (surprised?) without any gain or attenuation. What it does is to apply a frequency dependent phase shift. That is, different frequencies coming through seem to be delayed by different amounts. This phase shift is not generally a hard stair-step type of cutoff from 0 to 180 degrees, but rather a fairly gradual slope. In fact, for a one or two pole filter, this slope will cover a significant portion of the audible frequency range.
There are (as I’ve mentioned) those who believe that they can hear the effect of a frequency dependent phase shift in a single given signal. An example of this might come from the fact that the tweeters in your home stereo (or studio monitor) speakers don’t inhabit the same physical space as the woofers. Therefore, there’s some amount of delay between the two. I personally think that this effect is beyond the notice of most casual (and even most critical) listeners. However, there is a very well known company (starts with a B, ends with E, has a B in the middle) which seems to have made a large amount of money off a very carefully designed all-pass filter meant to correct for this effect. It is marketed as adding “clarity”, “punch”, and other terms, which are vaguely descriptive of completely subjective phenomena.
There are also folks out there selling all-pass filters with variable cutoff frequency as “phase correction tools”. These are generally used for mixing two (or more) similar copies of a source that happen to be delayed from one another. One example would be if you were to record a bass guitar through both a DI and a mic in front of the amp. Another might be something like a snare drum being recorded through a close mic and a room mic.
The sound waves take time to travel through the air, significantly more time than the electrons take to flow through the wires. Therefore, the DI is going to hit the recorder a cut hair earlier than the signal which comes from the speakers, through the air, to the mic, and then down the wire. In this instance the entire signal is delayed by the same amount at the distant mic. However, when we compare the two signals, we’ll find that the direct signal will have “moved on” further (relative to the whole period) for higher frequencies than for lower at any given delay time. This creates a frequency dependant phase shift. When we try to combine the two signals some frequencies will be more in (or out of) phase than others. Some will interfere constructively (come out louder) and some destructively (come out quieter). This creates what we call a Comb Filter. On some graphs, with appropriate scaling, the peaks and dips in the frequency response look like the teeth of a comb…
Now, there are two “right ways” to fix the issue:
1) Align the waveforms (Nudging). In the old days of tape this was accomplished by applying a short delay to the earlier signal. The delay time was varied while listening to the mix until it was “as less bad” as possible. Nowadays, in DAWland, we can view the two waveforms and slide one or the other until they line up as perfectly as possible. Interestingly, if you delay one signal far enough to where the earlier no longer sounds much like the later at any given instant, phase ceases to be an issue. It sounds like an echo, rather than a filter.
B) The 3:1 Rule. This is better termed the “At Least 3:1 Suggestion”, and is even better named the “At Least 9db Consideration”. Let’s say we’ve got a snare drum and two mics. One of the mics is up close, about 1’ away. The other mic is further - say 3’. Well, the sound pressure level of the snare should be about 9db greater when it reaches the close mic than when it reaches that distant mic. Assuming that the two mics have equal sensitivity and are fed through identical preamps set for the same amount of gain, mixing these signals should not have much noticeable phase interference, despite the delay in the distant mic. There will still be comb filtering, because of the delay, but because the distant mic is so much quieter than the close mic, the dips and peaks will span somewhat less than 1db (the accepted threshold of noticeable volume change).
1 + ( –1) = 0 but
1 + (-0.25) = close enough to 1
There are some problems with both of these methods:
A) Aligning waveforms changes the delay between the two sources
. In the case of a drum kit, the distant mic is likely meant to give a sense of space and ambience, and changing the delay between the direct and reflected versions of the signal can change the sound and feel of the mix. It will tend to make it sound like a smaller room. Of course, if you’re using the distant mic only for ambience, rather than for its contribution to the direct sound, you can often apply the 3:1 Rule.
2) The 3:1 Rule requires that one of the signals be significantly quieter than the other
. Doesn’t really matter which, either. You can just as happily crank up the miked track and just mix in a touch of the DI. But what if the 50:50 mix is exactly what you’re looking for? Then you’ve got to nudge.
Or, you could buy one of these all-pass filters, put it on one of the tracks, and twist the knob till it sounds “as less bad” as possible. This seems like a kludge to me…or maybe a hack. I know of some audio engineers who are seemingly intelligent and definitely well-paid who use these things all the time, and swear by them.
I’d like to mention that this all-pass filter thing is how your phase shift pedal works. The signal from the guitar is split. One of the copies goes through a series of all-pass filters, shifting various frequencies by different amounts, and is then mixed back with the original, where it interferes and causes some frequencies to be attenuated while others are accentuated. The cutoff frequency of these filters is modulated by an LFO, which causes the sweeping sound we hear. Some choruses use this same technology, but most flangers nowadays are created by a very short, frequency independent delay.
After all this I’d like to mention that there is an instance where I know I’ve seen - and believe I’ve heard – a noticeable effect from frequency dependent phase shift on a single signal.
Some waveforms are asymmetrical. Maybe it swings further in one direction than the other (tubes and output transformers sometimes do this, as does the Harmonic Perculator). Maybe it spends more time on one side of 0 (the duty cycle, or pulse width, often found tweakable on analog-style synths). Or maybe there’s just more high frequency content riding the lower frequencies when they swing one way. One place this (apparently) happens a lot is in male voices like my own.
An all-pass filter applied to an asymmetrical waveform will redistribute the energy of the overall signal to where it becomes much more symmetrical. This is easily visible in DAWland, where I’m happy to reside. It’s also subtly but definitely audible. It can make a voice sound a bit more full, more musical, and somehow bigger than life. What was that about vaguely descriptive of completely subjective phenomena? You hear this type of thing all the time on the radio. In broadcast processing, it’s called a “Phase Rotator”, but it’s really nothing more than the all-pass filter I’ve been talking about. It comes out like a very subtle form of compression, and can open up headroom.
Okay, I’m done for now…"
Ash's original post:
"Apr 24, 2009, 4:05pm, sumgai wrote:
A variable all-pass filter? Errr, isn't that just a fancy name for an amplifier? ;D
No. I was talking about an all-pass filter with a variable cutoff frequency. Most of these do not offer any actual gain. I guess it might be considered an amplifier in the same sense that a unity gain buffer is called an amplifier…
An all-pass filter passes all frequencies (surprised?) without any gain or attenuation. What it does is to apply a frequency dependent phase shift. That is, different frequencies coming through seem to be delayed by different amounts. This phase shift is not generally a hard stair-step type of cutoff from 0 to 180 degrees, but rather a fairly gradual slope. In fact, for a one or two pole filter, this slope will cover a significant portion of the audible frequency range.
There are (as I’ve mentioned) those who believe that they can hear the effect of a frequency dependent phase shift in a single given signal. An example of this might come from the fact that the tweeters in your home stereo (or studio monitor) speakers don’t inhabit the same physical space as the woofers. Therefore, there’s some amount of delay between the two. I personally think that this effect is beyond the notice of most casual (and even most critical) listeners. However, there is a very well known company (starts with a B, ends with E, has a B in the middle) which seems to have made a large amount of money off a very carefully designed all-pass filter meant to correct for this effect. It is marketed as adding “clarity”, “punch”, and other terms, which are vaguely descriptive of completely subjective phenomena.
There are also folks out there selling all-pass filters with variable cutoff frequency as “phase correction tools”. These are generally used for mixing two (or more) similar copies of a source that happen to be delayed from one another. One example would be if you were to record a bass guitar through both a DI and a mic in front of the amp. Another might be something like a snare drum being recorded through a close mic and a room mic.
The sound waves take time to travel through the air, significantly more time than the electrons take to flow through the wires. Therefore, the DI is going to hit the recorder a cut hair earlier than the signal which comes from the speakers, through the air, to the mic, and then down the wire. In this instance the entire signal is delayed by the same amount at the distant mic. However, when we compare the two signals, we’ll find that the direct signal will have “moved on” further (relative to the whole period) for higher frequencies than for lower at any given delay time. This creates a frequency dependant phase shift. When we try to combine the two signals some frequencies will be more in (or out of) phase than others. Some will interfere constructively (come out louder) and some destructively (come out quieter). This creates what we call a Comb Filter. On some graphs, with appropriate scaling, the peaks and dips in the frequency response look like the teeth of a comb…
Now, there are two “right ways” to fix the issue:
1) Align the waveforms (Nudging). In the old days of tape this was accomplished by applying a short delay to the earlier signal. The delay time was varied while listening to the mix until it was “as less bad” as possible. Nowadays, in DAWland, we can view the two waveforms and slide one or the other until they line up as perfectly as possible. Interestingly, if you delay one signal far enough to where the earlier no longer sounds much like the later at any given instant, phase ceases to be an issue. It sounds like an echo, rather than a filter.
B) The 3:1 Rule. This is better termed the “At Least 3:1 Suggestion”, and is even better named the “At Least 9db Consideration”. Let’s say we’ve got a snare drum and two mics. One of the mics is up close, about 1’ away. The other mic is further - say 3’. Well, the sound pressure level of the snare should be about 9db greater when it reaches the close mic than when it reaches that distant mic. Assuming that the two mics have equal sensitivity and are fed through identical preamps set for the same amount of gain, mixing these signals should not have much noticeable phase interference, despite the delay in the distant mic. There will still be comb filtering, because of the delay, but because the distant mic is so much quieter than the close mic, the dips and peaks will span somewhat less than 1db (the accepted threshold of noticeable volume change).
1 + ( –1) = 0 but
1 + (-0.25) = close enough to 1
There are some problems with both of these methods:
A) Aligning waveforms changes the delay between the two sources
. In the case of a drum kit, the distant mic is likely meant to give a sense of space and ambience, and changing the delay between the direct and reflected versions of the signal can change the sound and feel of the mix. It will tend to make it sound like a smaller room. Of course, if you’re using the distant mic only for ambience, rather than for its contribution to the direct sound, you can often apply the 3:1 Rule.2) The 3:1 Rule requires that one of the signals be significantly quieter than the other
. Doesn’t really matter which, either. You can just as happily crank up the miked track and just mix in a touch of the DI. But what if the 50:50 mix is exactly what you’re looking for? Then you’ve got to nudge.Or, you could buy one of these all-pass filters, put it on one of the tracks, and twist the knob till it sounds “as less bad” as possible. This seems like a kludge to me…or maybe a hack. I know of some audio engineers who are seemingly intelligent and definitely well-paid who use these things all the time, and swear by them.
I’d like to mention that this all-pass filter thing is how your phase shift pedal works. The signal from the guitar is split. One of the copies goes through a series of all-pass filters, shifting various frequencies by different amounts, and is then mixed back with the original, where it interferes and causes some frequencies to be attenuated while others are accentuated. The cutoff frequency of these filters is modulated by an LFO, which causes the sweeping sound we hear. Some choruses use this same technology, but most flangers nowadays are created by a very short, frequency independent delay.
After all this I’d like to mention that there is an instance where I know I’ve seen - and believe I’ve heard – a noticeable effect from frequency dependent phase shift on a single signal.
Some waveforms are asymmetrical. Maybe it swings further in one direction than the other (tubes and output transformers sometimes do this, as does the Harmonic Perculator). Maybe it spends more time on one side of 0 (the duty cycle, or pulse width, often found tweakable on analog-style synths). Or maybe there’s just more high frequency content riding the lower frequencies when they swing one way. One place this (apparently) happens a lot is in male voices like my own.
An all-pass filter applied to an asymmetrical waveform will redistribute the energy of the overall signal to where it becomes much more symmetrical. This is easily visible in DAWland, where I’m happy to reside. It’s also subtly but definitely audible. It can make a voice sound a bit more full, more musical, and somehow bigger than life. What was that about vaguely descriptive of completely subjective phenomena? You hear this type of thing all the time on the radio. In broadcast processing, it’s called a “Phase Rotator”, but it’s really nothing more than the all-pass filter I’ve been talking about. It comes out like a very subtle form of compression, and can open up headroom.
Okay, I’m done for now…"
