Understanding Phase plots in REW etc

[almikel]

(edit 21/4/17) - I've made some edits below based on finding some errors - noting each as an edit just makes it messy,

I've left errors due to my lack of understanding - these are addressed by Dave's comments.

==========================================================================

 

I've posted similar information a couple of times recently, but thought it worthy of its own thread.

 

Tools like REW, ARTA, Holm Impulse can show the phase response of speakers, rooms etc.

 

Often phase plots look like this (my tapped horn sub measured indoors at the listening position)

The dotted vertical lines do not represent instantaneous changes in speaker phase - they represent a way to keep the graph on the page known as "wrapped phase".

 

The left axis scale is dB and is irrelevant for this graph - phase is on the right axis in degrees

When the graph hits (in this case) +180 degrees on the right hand scale, the dotted vertical line drops to -180 degrees to keep it on the page, and the curve continues to rise.

The peak around 35-38Hz, and the peak just under 60Hz are  360 degrees apart (ie 38Hz is 180 degrees phase lead, and 60Hz is 540 degrees phase lead) - the graph is "phase wrapped" to keep it on the same page for easier viewing.

This allows more detail to be shown rather than having a phase plot covering over 1000 degrees - to know the actual phase you need to count the vertical lines - each one represents 360 degrees.

The same graph shown "unwrapped" below

same measurement, just unwrapped shows a steadily rising phase from 30Hz - note the scale on the right hand side- 1080 degrees - REW wouldn't let me zoom out further.

 

Research has demonstrated that smoothly changing phase is in-audible - lucky really, if we could hear the phase change above we wouldn't have hi-fi as we know it.

 

However phase changes stop being smooth when we add multi-way speakers, crossovers and rooms

 

Lots has been said about "minimum phase" and I'll attempt to add some value to the discussion on this - but some of my measurements confuse me - so hoping the SNA community can help me also 

 

A dead flat frequency (amplitude) response across the audible range (say 20Hz to 20kHz) is an unachievable target.

Anytime you have a wiggle in the frequency response, you have phase changes.

Look at measurements of speakers - where their response deviates most from flat is where the phase changes most.

 

outdoor close miked measurement of my sub

 

apologies for similar colours - top trace is FR (dark blue) and light blue trace is phase.

Note the phase plot (light blue) is more steep where there are larger  changes in FR.

 

Minimum Phase - what is it?

I still grapple with an easy to understand definition...and always fail 

Consider our stereo systems:

• You can define them in the time domain (we listen to our systems in the time domain) - the normal measurement is an impulse response - this describes the system in a single measurement - any ringing in the time domain will cause wiggles in the frequency domain
• You can define them in the frequency domain based on amplitude and phase - the classic Frequency Response curve (more correctly known as Amplitude Response), only describes part of the system. To properly define a system in the frequency domain requires both an amplitude and a phase response.

 

You can convert between the time domain and frequency domain using a Fourier Transform.

From an impulse response you can calculate both the frequency (amplitude) response and phase response using the Fourier Transform.

To calculate the impulse response you need both the amplitude response and the phase response (still using a Fourier Transform).

 

In a Minimum Phase system (and only in a minimum phase system), you can calculate the Phase Response from the Amplitude Response, and you can calculate the Amplitude Response from the Phase Response.

Remember you need both Amplitude Response and Phase Response to calculate (via the Fourier Transform) the Impulse Response.

So with minimum phase systems:

• if you have either the amplitude response or the phase response you can determine the impulse (time) response

 

With Minimum Phase systems, a peak in the frequency response, can be EQ'd by an opposite "cut" using minimum phase EQ (Parametric EQ - PEQ) and correct both amplitude and phase.

If treating minimum phase effects, dips can also be boosted via PEQ and correct both the amplitude and phase response. 

 

BUT is a room minimum phase?

Unfortunately in normal rooms, regions where the response is "minimum phase" is unpredictable.

Tools such as REW can measure minimum phase.

Read here : https://www.roomeqwizard.com/help/help_en-GB/html/minimumphase.html

 

Excess Phase is a measurement in REW that calculates the difference in phase between the calculated phase of a minimum phase system based on the measured Impulse Response and the actual phase response of the measured system.

Excess Phase shows which areas are "minimum phase" - flat areas indicate "minimum phase".

Below is an outdoor measurement of my sub - black is "Excess Phase" - not flat is non "minimum phase"

 

My observations:

• I would have expected a flat area of minimum phase between 35-55Hz based on the flattish FR - not the case.
• I did not expect the area between 70 and 80Hz to be flat for minimum phase in the middle of a dip in FR
• the only relatively flat areas of minimum phase are below 25Hz and between 70-80Hz

 

Bringing the sub into the room gives different results

Setup:

• in room measurement at the listening position
• sub + mid bass
• low pass Xover on sub around 50Hz or so 
• no high pass on the 18" mid bass (but loads of linear phase speaker correction to push the 18's lower)
• Dark blue is FR, Light Blue is phase, black is Excess Phase
• Measurement is mic at the listening position, sub and left main 18" driven together (in their range)

 

Observations:

• phase changes most when FR wiggles most (see below 20Hz, 90-100Hz, big dip around 250Hz
• based on Excess Phase, EQ should work well <100Hz, but not above 100Hz

 

All graphs are of measurements taken some time ago - and I did some cherry picking to best describe what I was trying to show.

 

I particularly don't understand how an outdoor measurement of my sub could demonstrate less minimum phase  behaviour than the indoor measurement of the sub integrated with my mid bass speakers - would be great if the SNA community could help me understand why???

 

cheers

Mike

Edited April 20, 2017 by almikel
clarification

[andyr]

Great post, Mike - thanks.

 

Andy

 

[davewantsmoore]

35 minutes ago, almikel said:

Below is an outdoor measurement of my sub (as per above graphs) - black is "Excess Phase" - not flat is non "minimum phase"

 

Not zero is non-minimum-phase.

 

Edited April 15, 2017 by davewantsmoore

[almikel]

11 minutes ago, davewantsmoore said:

 

Not zero is non-minimum-phase.

 

likely correct with measuring a single speaker, but the  the REW page on Minimum Phase quotes "Anywhere the excess group delay plot is flat is a minimum phase region of the response".and using the Excess Phase graph "flat areas" for determining time alignment issues between sub and mains.

https://www.roomeqwizard.com/help/help_en-GB/html/minimumphase.html

 

Is the REW guide wrong?

 

Mike

 

[almikel]

perhaps the "flat but non zero" area of minimum phase in my last REW graph is due to not using the "loopback" time reference? 

 

I've never mucked with this.

 

Mike

[davewantsmoore]

1 hour ago, almikel said:

Is the REW guide wrong?

 

By definition, the "minimum phase", is when there is zero excess phase.

 

Positive, but flat excess phase, is saying that the phase has whatever amount of wiggles it is supposed to have when comparing it to the amplitude response ..... but it is delayed.    This could be telling you EQ is not a good idea   (it's hard to generalise).

 

Moving on to your final chart.    You have altered the phase (independently of the amplitude response) with your filters.   The phase has been corrected flat, even though the amplitude is rising.....  and so becuase of this, the excess phase is not flat.     So your conclusion about EQ above 100Hz becuase of this isn't justified.

 

 

In order to make a "should EQ be ok here?" assessment ....  you should not have any non-minimum phase EQ/filters enabled, because this adjusts the relationship between amplitude and phase, which is what you are trying to see   (to see for example, if something like a reflection is causing the wiggle in your measurement).

[almikel]

cheers Dave,

your insight/knowledge is always an order of magnitude above mine.

 

20 minutes ago, davewantsmoore said:

Positive, but flat excess phase, is saying that the phase has whatever amount of wiggles it is supposed to have when comparing it to the amplitude response ..... but it is delayed.    This could be telling you EQ is not a good idea   (it's hard to generalise).

so potentially useful to show time alignment issues but not much else (and I use different tools to achieve time alignment)

 

25 minutes ago, davewantsmoore said:

Moving on to your final chart.    You have altered the phase (independently of the amplitude response) with your filters.   The phase has been corrected flat, even though the amplitude is rising.....  and so becuase of this, the excess phase is not flat.     So your conclusion about EQ above 100Hz becuase of this isn't justified.

light bulb moment - thankyou - but I'll ask questions after the next quote

 

28 minutes ago, davewantsmoore said:

In order to make a "should EQ be ok here?" assessment ....  you should not have any non-minimum phase EQ/filters enabled, because this adjusts the relationship between amplitude and phase, which is what you are trying to see   (to see for example, if something like a reflection is causing the wiggle in your measurement).

So my linear phase EQ is all about speaker correction which I've always done independently from room correction.

My 18" mid bass have been corrected (and pushed lower) based on "in room" but close miked measurements.

Obviously amplitude and phase have been mucked with independently in this process - the target being a flat frequency response for the speaker (let's say I pushed the 18's down to 40 Hz using linear phase FIRs - close miked).

My sub has no FIRs applied and runs between 20 - 50hz (I think a 48dB LR 50Hz low pass)

 

Now I put the mike back in the listening position to do room measurement and correction using PEQ.

I run a sweep from the listening position using REW to look at excess phase to assist me in determining where EQ may be applied

- are the listening position measurements of excess phase once the TD18s take over invalid (say 60 - 250Hz) to assist in determining where PEQ can be applied?

- if using linear phase FIRs for Xover and speaker correction for all speakers (incl sub), is measuring excess phase useful at all in determining if PEQ can be applied?

 

Modern DSP is heading down a linear phase filter path.

Excess Phase has been a useful tool in knowing when to apply PEQ and when not to.

If linear FIRs muck all that up, what is left to determine how to apply sensible EQ - minimum phase or linear phase?

 

As per other recent threads - well applied EQ can do great good, but poorly applied even greater evil.

 

cheers

Mike

 

[davewantsmoore]

11 hours ago, almikel said:

- are the listening position measurements of excess phase once the TD18s take over invalid (say 60 - 250Hz) to assist in determining where PEQ can be applied?

 

Yes.      However "determining where EQ can be applied" .... is very 'binary'.    It isn't as cut and dry as that.     Positive excess phase doesn't say you can't (or can) use EQ.... it is just an indication that something might be in your measurement that makes it more complex than it might seem.       It usually IS more complex than it seems.... which doesn't mean don't use EQ.  Just means take care.

 

 

 

12 hours ago, almikel said:

If linear FIRs muck all that up, what is left to determine how to apply sensible EQ - minimum phase or linear phase?

 

Turn off all your EQ.    Now your source is (supposed to be) minimum phase.      Measure....   and see where it's not.

 

Where you have excess phase, this can imply that some type of reflection is in the measurement, and maybe complicating the amplitude response .... and hence a peak or dip, is not a "simple" peak or dip that can be knocked up and down with EQ .... at least not without careful consideration (eg. measurements from lots of locations, and time windows).

[davewantsmoore]

PS.    If you are not using a timing reference... then in REW don't forget to click the "estimate IR delay button" to remove delay.

[Keith_W]

This is a beautiful thread, with a beautifully explained post on phase. More people should read it ... so ... bump. 

[Nigel]

Thank Yous chasp,

it will take many reads and my own measurements to begin to understand what you might be talking about. Concepts on concepts....

This thread may become a reference.

[almikel]

On 4/18/2017 at 7:34 PM, Keith_W said:

This is a beautiful thread, with a beautifully explained post on phase. More people should read it ... so ... bump. 

 

7 hours ago, Nigel said:

Thank Yous chasp,

it will take many reads and my own measurements to begin to understand what you might be talking about. Concepts on concepts....

This thread may become a reference.

When I joined SNA way back when, I didn't understand phase much at all.

A lot of what I've learned has been from SNA directly or inspired by SNA to make me look further.

On 4/15/2017 at 5:52 PM, almikel said:

Anytime you have a wiggle in the frequency response, you have phase changes.

To coin a new SNA term - this is a daveism (@davewantsmoore) - reading similar in a dave post a long time ago had a profound impact on my understanding of the relationship between FR and phase - if my recollection is correct Dave even used the technical term "wiggle".

 

I greatly enjoy my membership in the SNA community - adding something is what SNA is all about.

 

cheers

Mike

 

 

 

 

 

 

[Primare Knob]

I thought this would be good place to ask for help in understanding what is going on at 70Hz.

 

Placing the mic 10cm closer to the back wall, I am creating a weird resonant peak at 70Hz which starts to intensify massively after 80ms reading the graph.

Looking into this and if I could EQ the problem I started reading up minimum phase, and the explanation in REW. After that, I stumbled upon this thread and things started to become confusing again in that, that the excess phase needs to be zero to be considered minimum phase, which it never seem to be.

 

If placing the mic closer to the back wall is creating this phase response , does that mean it has to do with a reflection?

[almikel]

11 hours ago, Primare Knob said:

 

 

If placing the mic closer to the back wall is creating this phase response , does that mean it has to do with a reflection?

yes I would expect so - SBIR as opposed to room resonance.

and SBIR is never minimum phase so don't use EQ to try and fix it.

I'd add that any time you have sharp peaks or dips in the FR response it isn't minimum phase - ie above 70Hz in your graphs.

The broad hump around 30Hz has a reasonable chance of being minimum phase and may respond well to some EQ cut (unless you like the bass boost in that range, in which case leave it  ).

 

 

 

Mike

 

Edited January 26, 2018 by almikel

[almikel]

16 minutes ago, almikel said:

yes I would expect so - SBIR as opposed to room resonance.

 

Mike

 

Although the frequency doesn't move??? which would suggest modal not SBIR???

[Primare Knob]

Although the frequency doesn't move??? which would suggest modal not SBIR???

What about the different decay impulse over time between 110cm and 120cm. Does that make sense?
At 110cm the 70Hz decay is linear over time, but at 80ms it has a huge rise, which peaks above everything before 80ms.

[Primare Knob]

I wonder if it could be that the back wall or door is resonating. The back wall construction is 70mm studs, 16mm plywood and 13mm plasterboard on both sides with insulation. The large part of that wall is part of a small toilet area.

[almikel]

12 minutes ago, Primare Knob said:

What about the different decay impulse over time between 110cm and 120cm. Does that make sense?
At 110cm the 70Hz decay is linear over time, but at 80ms it has a huge rise, which peaks above everything before 80ms.

 

8 minutes ago, Primare Knob said:

I wonder if it could be that the back wall or door is resonating. The back wall construction is 70mm studs, 16mm plywood and 13mm plasterboard on both sides with insulation. The large part of that wall is part of a small toilet area.

there's really too much going on to determine that easily

It would be interesting (but not necessarily useful and "could" lead to more confusion) to put the mic behind the wall (ie in the rooms behind) - needs a long Mic lead.

 

This would give you some idea of the acoustic transparency of that wall. Given it's light weight, but with insulation, I would expect some "soaking up" of bass, but still reasonably high transmission.

 

33 minutes ago, Primare Knob said:

What about the different decay impulse over time between 110cm and 120cm. Does that make sense?
At 110cm the 70Hz decay is linear over time, but at 80ms it has a huge rise, which peaks above everything before 80ms.

could be at 110cm from the lightweight wall you're getting closer to the modal peak of the 2nd room.

Keep moving the mic closer and see what happens...

 

It's a bit hit and miss, trial and error, in a very serial process guided by non-experts (ie me) - you could chase your tail forever trying to determine what's going on.

 

If you engaged a professional, they would take measurements, and based on what they see, take more measurements, and much more quickly hone in on your problem areas, suggest treatment, measure again etc.

The expense upfront will compress the timeframe and reduce the risk of spending money on poorly conceived treatment.

 

Mike

 

[christosd]

I was once freaked out by a similar response inside the boiler of a power station... The same but different. In my case it was definitely modal, but as my sweep ascended, there was some very non linear responses from the boiler tubes in that the modes were excited by tones at harmonics of higher order than the fundamental. Usually only happens when elements in the system have non linear stiffness.... Such as your door or membrane stiffness vs bending stiffness in a wall panel. I dropped my excitation by 10-20 dB and I changed from a sweep to a carefully stepped sine and only measured once things settled the response changed. In my case I was just looking for acoustic modes rather than re-radiated noise from the tubes so I reverted to simply looking for sudden phase changes vs frequency that matched expected damped response peaks. 

 

Just read mike's post... Sorry for my blah, but I suggest when looking for modes in a room, then keep away from the boundaries, probably for reasons other than I have mentioned :). 

[Primare Knob]

I'll try to move as much discussion not related to phase regarding my room towards my own thread to keep this as clean and to the point as possible.

Am I correct in thinking that, as room nodes are minimum phase, room nodes would not show up in the Group Delay graph?

I have created a single graph from a collection of Excess Group Delay graphs in Photoshop.

 

Red: Subs quarter way the width of the room at front wall, Mic position 98cm from floor, 120cm from back wall

Blue: Subs half way the length of the room, Mic position 98cm from floor, 120cm from back wall

Orange: Subs half way the length of the room, Mic position 110cm from floor, 120cm from back wall

Green: Subs half way the length of the room, Mic position 110cm from floor, 110cm from back wall

Purple: Subs half way the length of the room, Mic position 110cm from floor, 97cm from back wall

 

But peaks that don't move in frequency suggest that they are room nodes and not SBIR.

There is always something happening at 70Hz, which could be a first height (246cm) node, or second length (487cm) node (2x 35Hz)

There is another recurring peak at 105Hz (70 + 35)

Placing the subs at the quarter way the width of the room at front wall (Red) hardly produces any big phase change with the mic placed at the quarter length in the back of the room, regarding the room length node (35, 70, 105, etc) but the SPL graph is showing a different story.

 

So, which information do I use where? As it looks, phase changes aren't always creating problems in the SPL graph, but can create problems in the decay time.

 

Edited January 27, 2018 by Primare Knob

[almikel]

On 1/27/2018 at 12:58 PM, Primare Knob said:

Am I correct in thinking that, as room nodes are minimum phase, room nodes would not show up in the Group Delay graph?

have another read of the REW article on minimum phase

https://www.roomeqwizard.com/help/help_en-GB/html/minimumphase.html

I've read it about 100 times over the last few years and think I've just about understood it

The two parts particularly relevant are:

"A common cause of non-minimum phase behaviour in rooms", and "Axial modes in a rectangular room"

 

When modes add the behaviour is no longer minimum phase

 

On 1/27/2018 at 12:58 PM, Primare Knob said:

 

 

the label says Group Delay - did you mean to display Excess Group Delay?

Regardless, the areas of wild swings in Group Delay won't be minimum phase, and you should avoid EQ in these areas.

 

cheers

Mike

[Hipper]

Thanks for almikel for starting this thread and all the contributors. It's the reason I came to this site (found it by Google). Having read it all, the links, and REW pages, I don't pretend to be too much nearer understanding it all but I've moved a bit further forward.

 

I have a well treated room and use REW to measure, to help with positioning and to apply some EQ using a Behringer DEQ2496. Recently I redid my room and took the opportunity to measure things as I put back the room treatment. My speakers are three way full range and I don't use subs.

 

The red graph is an SPL & Phase measurement before any treatment (and no EQ) for the left speaker.

And green is also the left speaker after all treatment (still no EQ) has been added (twenty-one Soffit Bass Traps and four absorbant panels on each of the back and front walls).

Whilst things are clearly not perfect you can see how the room treatment has not only made the frequency response much smoother but it has done the same for phase.

 

The decay times have also been considerably reduced, from typically 300ms to 120ms.

 

As you might imagine this makes things sound much better, especially bass and percussion type sounds.

 

Incidentally, the big dip between 45-70Hz persisted whatever I did with speaker movement or anything else. In the end I thought I'd try just increasing that area using the Behringer (breaking all the rules) and it worked! I don't know why.

 

Anyway, this thread is about phase so I thought I'd offer this info..

 

 

 

 

[davewantsmoore]

1 hour ago, Hipper said:

Whilst things are clearly not perfect you can see how the room treatment has not only made the frequency response much smoother but it has done the same for phase.

This is to be expected.   Fixing frequency, also fixed phase

 

This is true where the response is "minimum phase".   Speakers are minimum phase .... but not all room response effects are.

[davewantsmoore]

1 hour ago, Hipper said:

before

You had a really angry room.   

Nice work.

1 hour ago, Hipper said:

Incidentally, the big dip between 45-70Hz persisted whatever I did with speaker movement or anything else. In the end I thought I'd try just increasing that area using the Behringer (breaking all the rules) and it worked! I don't know why.

Often "filling the hole" won't work... but not always (complicated).

 

What you will find though is that ringing (look at waterfall or decay)  at thee frequencies has probably increased dramatically in that range too (although with all your absorption, perhaps it won't be too bad).

[almikel]

Hi Michael,

welcome to SNA!

what did we do before Google?

 

There's many smart people on SNA I've learnt a lot from.

16 hours ago, Hipper said:

I don't pretend to be too much nearer understanding it all but I've moved a bit further forward.

I feel the same - but it's very gratifying when I get those "light bulb" moments!

 

You've made a big difference to your room - well done! I bet the listening difference before/after is chalk and cheese. 

As Dave says, it's an angry room - are all the boundaries rigid (block/brick/concrete)?

 

14 hours ago, davewantsmoore said:

This is to be expected.   Fixing frequency, also fixed phase

 

This is true where the response is "minimum phase".

Hi Dave,

I think you meant to say using EQ to clean up the frequency response only cleans up the phase response where the response is minimum phase?

Using treatment (not EQ) will help both frequency and phase response regardless of whether the region is minimum phase or not.

Eg using absorption to damp SBIR (which is never minimum phase) will reduce swings in frequency response and phase response.

 

@Hipper - I'd be trying some specialist treatment for that big dip at 50Hz - Helmholtz, or limp mass etc.

 

cheers

Mike