-
Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Compensating for mains and sub delay
- Thread starter ErLan
- Start date
ernestcarl
Major Contributor
I believe this is already the very latest version I'm using. However, I have not kept pace with any of the newer changes/feature updates.
As you probably know by now, I'm still quite apprehensive/leery of extremely detailed attempts at phase inversion despite using some such techniques myself.
dasdoing
Major Contributor
try this on your own system with REW beta:
do not correct the speakers combined, but seperatly
put a 5fdw (this seam to work for mine, and the OP, so it might universily work)
invert phase completly (new function)
restrict the window of the created filter to 200ms left and right
export windowed IR
also on the spectogram of the resulting filter there shoud be no sharp shifts
on the spectogram of the convoluted IR there should be no red or orange pointy stuff into the negatives
dasdoing
Major Contributor
I have been experimenting a lot with phase correction, but it is still hard for me to forsee group delay on convulted graphs.
I just tested a bigger window inversion on my system, looked at spectogram and thought "this should work"
it doesn't. pre-delay listening to the IR via phones. I went to online tone generator to identify the frequency and I am pretty sure it is this -5ms in the negatives at 433Hz on excess group delay:
there will always be kind of single frequency long spikes in the negative, but these seam to be artifacts. but tese little triangular negatives in the midrange seam to be always audible
I also tried to identify what predelay I am hearing in OPs right speaker IR, and found it
compared to the left IR which has no pre-delay
Attachments
dasdoing
Major Contributor
two sharp negative spikes in the bad filter also seem to indicate that there will be problems
ernestcarl
Major Contributor
@dasdoing
Hmmmn... noticed that there's quite a bit of negative excess phase and GD in your REW generated corrections as well.
I did try this with the raw un-EQ'd measurements of my nearfield desk's front LCR and single sub (basic MCH only) where the measurements are cleaner. But, eh... this whole business gets complicated pretty quickly the more channels you have to individually correct -- i.e. with bass managed multichannel FIR outs. It's hard to create individual MCH FIR corrections where the summed results are all in harmony with each other.
This may turn to be a bit more complicated/involved than my prior REW+rePhase workflow. But, maybe when I have more time to experiment.
Hmmmn... noticed that there's quite a bit of negative excess phase and GD in your REW generated corrections as well.
I did try this with the raw un-EQ'd measurements of my nearfield desk's front LCR and single sub (basic MCH only) where the measurements are cleaner. But, eh... this whole business gets complicated pretty quickly the more channels you have to individually correct -- i.e. with bass managed multichannel FIR outs. It's hard to create individual MCH FIR corrections where the summed results are all in harmony with each other.
This may turn to be a bit more complicated/involved than my prior REW+rePhase workflow. But, maybe when I have more time to experiment.
dasdoing
Major Contributor
it is very complicated. pretty sure automatic stuff like Dirac doesn't get even near to the limits....aka will only correct up to a point.
you see, one big problem with the GD is that it obviously is noise poluted....plus in the bass region you reading will be influenced by the type of window.
here is another corrected IR I made of OPs left speaker today (this time I also corrected FR). here is excess group delay with the window he chose 500/500
tons of noise.
if we look at the impulse we can make the window much smaller and filter out part of the noise.
here is 30/200
Much cleaner. We still cannot be sure what the noise is.
Another thing that is confusing, at least for me, is the relationship between excess phase and peak energy time.
A perfect impulse will have both of them flat.
However, in-room response can only be one or the other. We can create a perfectly flat peak energy time by completely inverting the phase, but this introduces a lot of pre-delay, which is visible in the excess phase as well.
Also, the graph I posted clearly shows negative excess group delay below approximately 55Hz. However, the phase and peak energy time are flat down to 22Hz. Which reading is correct? It's puzzling.
peak energy time is free of noise, but the peak beeing at zero doesn't garantee there is nothing coming in to soon, isn't it?
In the end, when we apply phase correction to the extreme, we have to verify it by ear since there doesn't seem to be a clear indication. Of course, we can play it safe by applying less correction and keeping everything above zero. But for this, we can use automatic tools like Dirac. Unfortunately, there seems to be a lack of a graph specifically for this purpose. What does the group delay show, after all? It shows a single "line," while we know there are reflections. The wavelet kind of shows reflections, but then again, the reading is very confusing and heavily dependent on the smoothing. And smoothing itself is problematic since it obscures reality, doesn't it?
Correcting frequency per frequency in Rephase is probably the most accurate way to do it. However, I believe it should be possible with the tools provided by REW.
As I mentioned, I am experimenting. I think it's advisable to work with different frequency-dependent windows over the spectrum. After all, the frequency-dependent window in REW is limited compared to other DRC (Digital Room Correction) software where you have more control over the windowing. We can merge different windows together.
I don't even think this is all worth it. the diferences are subtle. but we are all here for the subtle diferences in this forum lol
ernestcarl
Major Contributor
Several hours later.... I may be getting somewhere with my own bass managed LCR test with this new methodology/phase inversion feature in REW -- that is, IF, I manually correct the center channel and other remaining side and rear surrounds with the supplemental help of rePhase.
I limited the left window to 60ms which is the most FIR delay that I find acceptable when streaming online video content e.g. youtube, Netflix, etc. on a PC
Raw un-EQ'd measurements:
IIR+FIR Bass managed channels including the sub/LFE:
Bass managed channels and LFE are all phase compliant:
Rest of the measurements should be good invidually and in relation to one another:
Doesn't look like there's any significant pre-ringing including negative excess phase and GD.
drive.google.com
I limited the left window to 60ms which is the most FIR delay that I find acceptable when streaming online video content e.g. youtube, Netflix, etc. on a PC
Raw un-EQ'd measurements:
IIR+FIR Bass managed channels including the sub/LFE:
Bass managed channels and LFE are all phase compliant:
Rest of the measurements should be good invidually and in relation to one another:
Doesn't look like there's any significant pre-ringing including negative excess phase and GD.
REW_REPHASE_BM_MCH_LCR_TEST.zip
drive.google.com
I really appreciate that invaluable discussion experiment simulation and observation are going on this thread focusing on time/phase alignment(s) between subwoofer and woofer; I am learning a lot from these.
By the way, however, I assume all of these approaches should be properly reflected in actual air sound at our listening position in our individual acoustic environment/room which would be almost always somewhat echoic (with reflection and reverberation) and asysmmetric in room shape as well as furniture alignments.
I believe, therefore, not only the REW's wavelet delay observation but also wave-shape and 3D-energy-distribution measurement/observation (and tuning) using tone burst signals (8-, 3-, 1-wave) of suitable sound Fq for subwoofer-only/woofer-only/subwoofer+woofer should be performed on recorded air sound at our listening position, just for example I have done and shared here and here on my project thread. Let me share here just two diagrams (among many) from there;
As shown in the above diagram, this kind of wave-shape and 3D-energy distribution measurements/observations would greatly contribute for understanding the transient behavior of SPs around XO Fq, optimization of XO Fq and the slopes, delay (phase continuity) optimization, gain adjustments, and hence giving the better/best transient sound around the XO Fq.
By the way, however, I assume all of these approaches should be properly reflected in actual air sound at our listening position in our individual acoustic environment/room which would be almost always somewhat echoic (with reflection and reverberation) and asysmmetric in room shape as well as furniture alignments.
I believe, therefore, not only the REW's wavelet delay observation but also wave-shape and 3D-energy-distribution measurement/observation (and tuning) using tone burst signals (8-, 3-, 1-wave) of suitable sound Fq for subwoofer-only/woofer-only/subwoofer+woofer should be performed on recorded air sound at our listening position, just for example I have done and shared here and here on my project thread. Let me share here just two diagrams (among many) from there;
As shown in the above diagram, this kind of wave-shape and 3D-energy distribution measurements/observations would greatly contribute for understanding the transient behavior of SPs around XO Fq, optimization of XO Fq and the slopes, delay (phase continuity) optimization, gain adjustments, and hence giving the better/best transient sound around the XO Fq.
ernestcarl
Major Contributor
Right… if you take into account the room acoustics, speaker design, and distance between speaker and main listening position the phase is almost always never going to be zero.
The more “detailed” and long/extended the correction, we risk making the response far worse the further away we move from our microphone reference point.
One way of avoiding this may be by preemptively limiting the time length of the correction, say, 20 to 40 ms the most — and by allowing a bit of a more natural phase and/or excess roll-off.
Ideally, pre-ringing (perhaps even some other resonance??) between adjacent driver bands should equally cancel each other out. This especially true when we’re trying to achieve “zero phase” or apply time domain FIR correction. When playing looking at burst test tones per driver, we might hear/see some odd artifacts or pre-ringing induced that’s not at all evident when listening with the summed multiway speaker response.
Although, if you look at the sub/LFE only corrected IR I posted, there really is not any significant ringing and so the transient response is already reasonably “tight” by default — at least in this one case.
Last edited:
Some info related to your comments:
1. The frequency where a phase discontinuity occurs is not the same as the GD or PD peak frequency, and subsequent time delays are also quite different and very much frequency dependent:
2. Calibration mics don't come with phase correction files and they are far from being zero phase tools. For at least some accuracy, I use minimum phase version of the calibration file. I haven't checked properly yet but when you compare OP's measurements against Umik-2 mic min phase response below, it's possible that his subs are not even delayed in reality. Hence, I generally find correcting excess phase outside 100Hz-1000Hz a bit meaningless since it's also widely accepted that outside this range phase deviations are not detectable by human.
Last edited:
Your right speaker measurement is distorted and is much louder below 100Hz (and everywhere else to a lesser extent) than the left which is quite odd given both speakers are crossed over with sub at 100Hz. This happens sometimes despite your 1M length measurements. I guess you need to remeasure.
ernestcarl
Major Contributor
Somehow, I doubt that:
Depends on the test/experimental design...
OP's subwoofer frequencies playing a bit late is guaranteed to be audible. One way of testing would be by comparing low frequency transient test tracks (may help if audible bandwidth spans up to the midrange or higher) between closed cans/IEMs and one's own speakers.
I had thought "it's possible" and I'm more convinced now after I checked the data a bit more. Here are some supporting facts:
- The flat delay from 70Hz to above 200Hz in your wavelet graph contradicts with a sub induced delay below 100Hz
- The delay up to 300Hz is even harder to explain
- Right speaker has considerably more GD than the left below 100Hz
- Subwoofer filter delays almost never exceed 5ms
Gradually falling GD graph from around 15ms @ 20Hz to around 2ms @200Hz is not only the norm for an idealized minimum phase speaker but also way below "extrapolated" detectability limits. Longer wavelength bass signals will travel slower than the short wavelength high frequency signals in the air and you cannot change that! There were extensive studies about this by Blauert and Laws, Flannigan and Liski back in the 70s:
Last edited:
I would like to suggest, therefore, the simple and straightforward "time-shift tone burst method" should be also applied...
upon simple PM request to me, I would be happy to share the tone burst signals I prepared for those studies done on my project thread.
ernestcarl
Major Contributor
These in-room measurements aren’t particularly pristine. I would much rather see sweeps of the individual channels. Left and right without sub. And just the sub(s).
I have no energy to argue about what amount and type of delay is audible. Though, I am rather surprised you (of all people) arguing that it’s of no consequence here.
I concentrated on the "detection" of GD because you mentioned it. I cannot comment on whether it is good or bad. Personally, I am not sure. Is it audible? Possibly. Is it audible with around 40dB of typical backgorund noise? Probably, not.
Group delay is a measurement of time distortion in a system. More specifically, group delay is the delay of the envelope of the sinusoid at that frequency and phase delay is the delay of the sine wave itself. If all frequencies are delayed equally, the waveform is reproduced correctly despite the fact that the delay introduced a major phase shift. The requirement is that the system have both flat frequency response and flat time delay. That is, all frequency components must arrive with the correct amplitude and time alignment. Flat phase response is just one of the cases where the time alignment criteria is met.
What's important for me (and I know from your posts also for you) here is to give accurate guidance to people who want to learn. If one tries to completely remove GD at the LP, one will be fighting against the laws of physics and a steady decay GD graph from low to high frequencies is generally a healthy response.
I've never tried tone burst method. I'd like to see them please.
