REW Measures Changing Phase Slopes for the Same Speaker and Same Position

[doalt]

Hello guys,

In an effort to measure and improve the phase response of my speakers, I have followed several tutorials on https://rephase.org/. But the results all sound weird.

After repeating the processes several times, I noticed the speakers' phase slopes seem to change constantly, even for the same speaker at the same position. I am not sure if this is an issue with the speakers, my system (M1 macbook + headphone jack or USB out to the speakers), or my measurement settings.

Below are some measurements I took from the same mic position with the same settings. The speakers support line-in and USB-in, so I tried both.

- SPL graphs for the same speaker align well, except line-in and USB-in are not volume matched.

- But the phase graphs vary significantly starting from 1.5khz. So it seems I can't depend on them for phase correction.

- And these are the REW settings. I attached the measurements below.

 

[ernestcarl]

Probably just a clock drift issue. Use the same channel as an acoustic timing reference when doing your sweeps. Sometimes, even with that, glitches will still happen… just repeat the measurements.

 

[doalt]

Probably just a clock drift issue. Use the same channel as an acoustic timing reference when doing your sweeps. Sometimes, even with that, glitches will still happen… just repeat the measurements.

Thanks. It does seem like a clock drift.

I used the same channel as reference every time, but still got these results. It's worth noting that I don't have an audio interface, so the USB mic and the speakers may follow different clocks. I will do some more troubleshooting.

 

[ernestcarl]

Thanks. It does seem like a clock drift.

I used the same channel as reference every time, but still got these results. It's worth noting that I don't have an audio interface, so the USB mic and the speakers may follow different clocks. I will do some more troubleshooting.

Manually applying time offsets for each measurement should be tried as well. Overall, the right speaker is more linear and useable enough after vector averaging post individual time shift adj. The left speaker is too janky and a few more repeats will be necessary.

 

[ebslo]

Nothing stands out. I usually do sweeps 6-10dB higher level than yours, so noise could be a factor.

For some perspective, the difference in group delays (same channel and between channels) is < 0.15ms above 1kHz, which corresponds to about 2 inches. Considering you have two ears 6 inches apart, and there are several inches of head position variability, what point is there in correcting differences this small?

 

[sam_adams]

It's worth noting that I don't have an audio interface, so the USB mic and the speakers may follow different clocks.

That is exactly why you are having this issue.

 

[ebslo]

That is exactly why you are having this issue.

Yes, sort of. But REW attempts to use the acoustic timing reference to measure and correct for the clock skew. The correction, in PPM, is in each file the OP posted, and the corrections do not look unreasonable. Of course this relies on acoustic measurement of the timing reference pulses, as well as both oscillators having stable frequency throughout each measurement, so it's definitely prone to error. I usually see better results than these. Consider that everyone who uses a USB mic (ie. the ubiquitous UMIK-1) is taking measurements with the output and input devices on different clocks.

 

[doalt]

Nothing stands out. I usually do sweeps 6-10dB higher level than yours, so noise could be a factor.

For some perspective, the difference in group delays (same channel and between channels) is < 0.15ms above 1kHz, which corresponds to about 2 inches. Considering you have two ears 6 inches apart, and there are several inches of head position variability, what point is there in correcting differences this small?

Thanks. I am still trying to learn what rePhase does and how to use it.

After seeing many threads talk about how important it is to correct phase, it seemed like a good thing to do. But I had no prior understanding about impulse response, group delay, crossover etc so I just followed the tutorials blindly. Judging from the results, I am likely doing more harm than good.

If I want to understand what really happens in the time domain, where do I start?

 

[ernestcarl]

After seeing many threads talk about how important it is to correct phase, it seemed like a good thing to do. But I had no prior understanding about impulse response, group delay, crossover etc so I just followed the tutorials blindly. Judging from the results, I am likely doing more harm than good.

I usually think about this in the context of crossovers and their resulting summed responses at the measurement position.

The most "linear" matching measurements in your set is probably "Line in - L3" and "USB - R1" -- and so if I wanted to improve their overall summed response, I would probably apply a single phase PEQ to the right channel only:

*I entered 45ms in the centering to minimize overall delay if lip sync is a potential issue (depends on the application).

**caveat emptor: this correction only makes sense if speakers were positioned properly (normal listening triangle) in the room already and mic was fixed at the main listening position.

 

[doalt]

The most "linear" matching measurements in your set is probably "Line in - L3" and "USB - R1" -- and so if I wanted to improve their overall summed response, I would probably apply a single phase PEQ to the right channel only:

Thank you for this interesting example. It looks very different from what the tutorials are trying to do, and this is making me rethink what the real goal is.

For reference, I mainly followed these links:

https://rephase.org/tutorials/Bear_REW_rePhase_tuto.pdf

https://rephase.org/tutorials/rePhase-tutorial-fr.pdf

And they have these general guidelines:
- make the phase close to flat at 0.
- avoid phase eq in the bass. Instead, use the filters linearization tab.
- have the same correction on both speakers.

Screenshot from rePhase-tutorial-fr.pdf:

On the other hand, your correction seems to:
- make the phase change more smooth.
- apply phase eq right in the bass.
- affect only R speaker, matching L and R channels.

I have yet made sense of the discrepancies.

 

[ernestcarl]

Thank you for this interesting example. It looks very different from what the tutorials are trying to do, and this is making me rethink what the real goal is.

For reference, I mainly followed these links:

https://rephase.org/tutorials/Bear_REW_rePhase_tuto.pdf

https://rephase.org/tutorials/rePhase-tutorial-fr.pdf

And they have these general guidelines:
- make the phase close to flat at 0.
- avoid phase eq in the bass. Instead, use the filters linearization tab.
- have the same correction on both speakers.

Screenshot from rePhase-tutorial-fr.pdf:
View attachment 267028

On the other hand, your correction seems to:
- make the phase change more smooth.
- apply phase eq right in the bass.
- affect only R speaker, matching L and R channels.

I have yet made sense of the discrepancies.

If one is going to use FIR EQ, why not linearize the remaining excess phase as well? I mean, you could certainly do that… and I do use linearizing filters too — up to a point — mainly at my active monitor’s mid-HF crossovers. But in the bass, yeah, you need to be a bit more careful. There can be weird side-effects to the sound that’s noticeable when playing transient clips/tracks.

When I do this, it’s within the context of using different sets of speakers having different magnitude and phase profiles and me primarily wanting them to sum more effectively.

A nifty trick/result is at the MLP/central mic position where every speaker is time and phase aligned, you can play a sweep to all speakers and not get any sort of deep cancellations. I’ll post an example…

 

[ernestcarl]

Below is a relatively mild FIR magnitude and phase EQ example which I have applied to my small listening/movie couch room's 5.1 MCH setup. I believe it accomplishes the highlighted text in the PDF guide tutorial you linked to.


*In this case, I wanted to find out if I could fit the filters in some (theoretically) "handicapped" loudspeaker management hardware -- something that could only handle very limited amount of taps. Luckily, my speakers didn't need much work to begin with anyway... the front coaxial mains being actively FIR corrected already, and the single full-range driver center channel has a mostly flat phase response out of the box.

The bulk of the EQ is applied separately from the FIR stage and is all minimum phase e.g. high pass and peak IIR filters.

Time alignment and levels adjustment between channels is sort of half set/done in the above graphs, but phase and magnitude EQ have yet to be applied.

After EQ:

Things are looking a bit more "aligned" now across the board... and resulting summed responses follow as well:

-------

Now, I'm going to perform a sweep to all channels simultaneously by copying/re-routing the left channel to the rest of the bass managed channels in JRiver (loudspeaker DSP software I use):

Here's the prior mentioned "nifty" log swept sine measurement result with all that EQ applied when measured at our precisely centered microphone position:

Move or change positions along the couch/room and you will of course get comb filtering and see some amount of energy loss from the changing path length difference between speakers and boundary effects etc.


*BTW, the effect of that FIR phase gain/boost adjustment around +100 Hz is a smoother transition between sub and mains which also reduced one particular GD peak:

In this particular room setup, I have also done other xo alignments without any additional FIR EQ and it has always resulted in less optimal xo summation with a bit more non-minimum phase energy loss.

Overall, we don't see a lot of loss (green trace) after applying considerable frequency dependent windowing -- wherein the direct sound is well-preserved or mostly comes out intact -- which is good. Ideally, there would be no visible interruption or magnitude loss to be seen anywhere. But, one has to deal with real, small listening room acoustics. In this scenario, the issues are minor to start with as I've already worked out the best positioning in the room.

 

[ernestcarl]

I wanted to find out if I could fit the filters in some (theoretically) "handicapped" loudspeaker management hardware -- something that could only handle very limited amount of taps.

Here's just a quick simulation as to what happens when I remove the low tap count handicap/restriction:

FDW 6 cycles applied in the SPL & Phase graphs


Pre-echo or pre-ringing is visible in the wavelet spectrogram at 1/2 resolution. Also, notice I didn't even bother to correct that group delay peak around ~126 Hz or so. That's because my previous heavy-handed attempts in the past to do this resulted in audible artifacts when listening carefully to bass transient clips/test tracks.

 

[ernestcarl]

*Hmmmn... I thought that it could probably be much better so I made several iterative adjustments in the paragraphic phase EQ section until I got the step response just right where I wanted it to be.

*Off-axis (from where I focused the FIR phase correction) at the left and right corner seating areas of the couch as well as desk seat in front of the room, linearity appears very well maintained... but, there is definitely slightly bit more pronounced pre-echo in the measurements -- I kind of already expected this -- though not nearly as bad I thought. Audibility will have to be tested.

**Decreasing taps further down to 16,384 in order to "economize" is causing tiny bit of phase and magnitude slippage below 5Hz, but it's not enough to make one concerned. Lastly, 171 ms is rather too high of an additional DSP latency for streaming AV content from the web e.g. Netflix, Disney+ and the like.

Spoiler: Quick listening test results

Neumann’s kickdrum train test track was used.

In the couch area the convolved test track sounded louder and way more forward in its attack — I kind of like this assault to my senses, but wished it was just a tad softer. Also, in front of the room at my desk area it sounded off.

The original track itself with the 5ms FIR filter sounds slightly distant or delayed — and kind of softer. Consequently, the repeating bass attack feels more comfortable to listen to for longer periods.

My guess is I probably would much prefer the shorter FIR correction when listening casually in the background; but, then again, I don’t feel that it’s lacking enough to really want anything more.

Because the longer FIR filter makes listening to bass transients more intense — almost annoyingly — I should probably modify the filter to delay things or increase GD a little bit more. Eh, maybe later…

 

[doalt]

wow... I feel like my son looking at his new books. There are so many pictures and words, but I don't understand any of them.

Sorry I I can't provide any meaningful feedback, just... wow.

 

[ernestcarl]

wow... I feel like my son looking at his new books. There are so many pictures and words, but I don't understand any of them.

Sorry I I can't provide any meaningful feedback, just... wow.

What you do with software tools like REW and rePhase is totally up to you. You can even just use the normal FR magnitude PEQ section and forget about fooling around with phase EQ. I think most people are going to be more than satisfied with that kind of use alone. Once you start playing with phase, eh, let’s just say it’s very easy to make things worse. Keeping the tap amount restrictively low also means you are forced not apply as much (phase) EQ in the bass region — consider it as a sort of a self-limiting handicap to decrease the likelihood of messing things up.

Having independent controls for the phase and magnitude response is a useful tool when one is setting up a multichannel system where one is combining speakers not necessarily designed to be paired with each other (my own use case). It’s also very useful for the DIY’er who wants to create his own active, DSP’d multiway speakers from scratch. Not sure if you fall in either category, but if not, then you might simply want an additional layer of time-domain correction (reducing GD perhaps and optimizing speakers to match better in phase) on top of basic FR magnitude adjustments. I think most people simply don’t need this extra layer of complexity in their lives… rather prioritize optimizing room positioning and acoustics more so first. Maybe do all this other stuff as an experiment of sorts for fun only when you have lots of extra time and are bored to death.

 

[doalt]

Not sure if you fall in either category, but if not, then you might simply want an additional layer of time-domain correction (reducing GD perhaps and optimizing speakers to match better in phase) on top of basic FR magnitude adjustments. I think most people simply don’t need this extra layer of complexity in their lives… rather prioritize optimizing room positioning and acoustics more so first.

This is exactly my situation. I didn't know how much improvement it would make, so I was just giving it a try. Thank you for the suggestion.

If you have a few minutes, could your help take a look at my current FR and suggest where I should focus on next? I used REW EQ to knock down a few peaks, but I still see some nasty dips, especially the 80Hz one on R speaker. I don't have a lot of room to move them around because it's a small bedroom.

 

[ernestcarl]

This is exactly my situation. I didn't know how much improvement it would make, so I was just giving it a try. Thank you for the suggestion.

If you have a few minutes, could your help take a look at my current FR and suggest where I should focus on next? I used REW EQ to knock down a few peaks, but I still see some nasty dips, especially the 80Hz one on R speaker. I don't have a lot of room to move them around because it's a small bedroom.

View attachment 268987

I took a quick look, but... I don't think it's a good idea to fill-in such a null. Tried manually filling it in with about 6dB headroom loss, however, the combined response is barely, marginally better -- even in combination with the mild phase PEQ adjustment previously shown here since the null is just too deep.

I'm surprised how your L+R measurement looks better than what one gets from a vector average of the individual left and right measurements. And so doesn't really look like much more needs to be "corrected" after your own EQ. Well, we could reduce the GD a little tiny bit, but IMO, it's going to be too small of a difference to be all that worth it. For certain albums or tracks where you feel something could be improved, you could always adjust the bass or treble with general boost/cut shelving filters to taste. *Oh, and additional acoustic treatment primarily focusing in reducing the bass decay below 200Hz (so thicker panels) is something you should also try to explore.

 

[doalt]

I took a quick look, but... I don't think it's a good idea to fill-in such a null.

Understood. Thank you.

I'm surprised how your L+R measurement looks better than what one gets from a vector average of the individual left and right measurements.

Sorry about the confusion. I just upgraded from some random speakers to Adam T5Vs, based on ASR reviews. The latest measurements are from the new speakers, not the ones at the beginning of the post.