advertisement


Measuring and interpreting the in-room phase response and time delay of loudspeakers

Tony L said:
I don’t know which would be more use, near or far. I’m just curious as to what degree of phase/time error one can expect from a typical sealed-box mini-monitor. To my mind conceptually such a speaker should be as good as any multi-driver box could be as there is no time domain error from a port and the drivers are as close together as possible other than a coaxial driver such as a Tannoy, LS50 or whatever. I’d love to see a plot from a time-aligned two-way box such as as a Spica TC50.

I remember the late-period Rank owned Leak playing about with time alignment too with stepped baffles. I can’t remember if they were any good, but probably not as Rank era kit was a bit cost-cut as I recall.

PS I have a feeling time-alignment is the reason I tend to prefer being below flat-baffle speakers such as JR149s, effectively I’m tilting them back by moving the seat lower, i.e. from my low listening position they should almost be behaving like a TC50 or SBL.
Click to expand...
I've just checked the measurements and don't think they'll be of much use to you, Tony, as the mic was 1.5cm higher than the axis of the T27 (I had measured my JR149mk2's immediately prior to the JR149s and had set the mic height to the axis of the Scanspeak tweeter and didn't want to move the mic between measurements or else the bass responses would not be comparable. I measured two pairs of mk1 and two pairs of mk2 that day; the phase responses of the two pairs of mk2 are very closely matched, but one pair of mk1 have a lot more 'scribbling' around the crossover frequency than the other pair. Funny thing is the 'scribbled' pair happen to be the ones I prefer the sound of(!) - perhaps I need to re-evaluate...
 
davidsrsb said:
Back to the 66, 25ms at 80Hz is two cycles. Blauert and Laws report on threshold of group delay audibility has been interpreted as putting the critical delay as between 1 and two cycles, so this is too high.
https://trueaudio.com/post_010.htm
Click to expand...

That's an interesting discussion in the reference you give. Only skimmed it thus far. I'll read it more and I may follow up and look at the papers it refers to.

However having done the results for the ESL63 I suspect pretty much all speaker designs will tend to give large dispersions as you go down into the real LF region. And of course, non-DC coupled amps may also do this, and we have no idea at present what studio kit and mics may do! So need to be wary of worrying about only one aspect of a wider problem. How much phase dispersion is a part of the sound of a double bass? :)

I wonder if this has passed un-noticed because it 'hides behind' the 'phase scribbles' seen in standard phase plots and people then don't push past that... I'd certainly tended to assume it was basically down to speaker-mike distance. But there is clearly more to it once we start investigating.
 
@Jim Audiomisc, in REW's sine wave sweep, there is an option to increase the length of the log sweep from the default "256k 5.5 seconds" to either "512k 10.9 seconds" or "1M 21.8 seconds". I'm presuming a longer sweep provides greater resolution but I won't know how this resolution will be apportioned across the frequencies until I run tests at the new settings and export the data to txt.

There's also an option to increase the number of sweeps from 1 to 8, presumably that's simply repetition to allow identification of external noise contamination? I usually just manually repeat each single measurement I take two or three times, and try to wait until there's a lull in external noise from passing traffic etc.
 
ToTo Man said:
@Jim Audiomisc, in REW's sine wave sweep, there is an option to increase the length of the log sweep from the default "256k 5.5 seconds" to either "512k 10.9 seconds" or "1M 21.8 seconds". I'm presuming a longer sweep provides greater resolution but I won't know how this resolution will be apportioned across the frequencies until I run tests at the new settings and export the data to txt.
Click to expand...

The key need is for a *LINEAR sweep* with the resulting (nominal) spot frequencies sufficiently closely resolved. This is likely to be a very slow process and I wouldn't be surprised if they cannae be bothered as few people would use it. The LF results from the log sweep you sent are OK up to around 2kHz. Beyond that they get increasingly 'iffy' (to use a technical term :) ) because they're too far apart and 'averaged' in some way over a spread in frequency that is too large.

Doing a single FFT or adding them over the span tends to blur details unless you're very careful about how things are processed.

I guess REW uses an FFT on the entire sweep. Again, efficient in terms of time, but isn't ideal for this. I used a 'PSD' approach which nails each frequency and its phase quite independently of the other data points.

When I did this in the past I used a process of step-and-play. i.e. a series of sinewave bursts. Each one quite long. Thus each data point is much better defined. But even then I gave up at about 5kHz.

IIRC each tone burst I used and the following 'gap' to let reverb die was about a second long. Assume 20Hz between bursts and a 5,000 Hz range and you end up needing around 250 seconds. Can't recall the precise values I used off-hand but this is the right ball park. Double it if you want to get to 10kHz!

Can REW do what I did? I suspect no-one expects to need this, so it doesn't.
 
Jim Audiomisc said:
....However having done the results for the ESL63 I suspect pretty much all speaker designs will tend to give large dispersions as you go down into the real LF region. And of course, non-DC coupled amps may also do this, and we have no idea at present what studio kit and mics may do! So need to be wary of worrying about only one aspect of a wider problem. How much phase dispersion is a part of the sound of a double bass? :).....
Click to expand...
There was a comment in the article that the suggested 1 cycle limit of audibility would only apply in otherwise well recorded and amplified cases.

Bass frequency group delay is also a good candidate for why so many phono stage rumble filters are so intrusive, when the music content spectrum is hardly affected
 
davidsrsb said:
There was a comment in the article that the suggested 1 cycle limit of audibility would only apply in otherwise well recorded and amplified cases.

Bass frequency group delay is also a good candidate for why so many phono stage rumble filters are so intrusive, when the music content spectrum is hardly affected
Click to expand...

What does group delay "sound like", in a subjective description?
 
@Jim Audiomisc, if I'm able to locate the farfield measurements I took of my Tannoy Lancaster, Lockwood and Edinburgh loudspeakers, would you mind running your phase stitching and time delay algorithms on these, if it's not too much trouble? I'd be interested to see the difference in low frequency phase shift between a sealed enclosure (Lancaster), aperiodic vented enclosure (Lockwood) and distributed port enclosure (Edinburgh), and I'm sure @Tony L would too. :)
 
I've just discovered that the phase response graph in REW is affected by the amount of smoothing applied to the amplitude frequency response, i.e. less smoothing = even more scribbles!

In my search for the measurements I mentioned in post #75 I have also discovered sine wave sweeps of my Lockwoods before and after applying Dirac DSP. It's very interesting to see the difference (improvement) in impulse response with the DSP engaged. I'll upload graphs of these too.

@Tony L, if it isn't too much of an admin burden, I'm wondering if we should move all posts from #27 onwards to a new thread, perhaps in the DIY section, titled something along the lines of "Measuring and interpreting the in-room phase response and time delay of loudspeakers", so that we can continue to broaden what is turning out to be a very interesting discussion?
 
ToTo Man said:
@Jim Audiomisc, if I'm able to locate the farfield measurements I took of my Tannoy Lancaster, Lockwood and Edinburgh loudspeakers, would you mind running your phase stitching and time delay algorithms on these, if it's not too much trouble? I'd be interested to see the difference in low frequency phase shift between a sealed enclosure (Lancaster), aperiodic vented enclosure (Lockwood) and distributed port enclosure (Edinburgh), and I'm sure @Tony L would too. :)
Click to expand...

Let me have a 'text' or csv file of the data and I can try. As per before, though, the difficulty is that REW tends to muck up the data, so the results may be of limited use because the data points are too far apart, etc.
 
ToTo Man said:
I've just discovered that the phase response graph in REW is affected by the amount of smoothing applied to the amplitude frequency response, i.e. less smoothing = even more scribbles!
Click to expand...

That's the kind of problem I was warning about a while ago. REW is doing some sort of 'averaging' or working with too low a resolution, etc. This tampers with the data and makes it harder (or impossible!) to get reliable time-domain info, particularly at HF.

From what I've seen thus far I guess I can get results for the LF region up to about 2kHz. But beyond that it'll vary from case to case.

The ESL63 results look very good in comparison. But how much that's because the ESL63 is phase linear, and how much its because I started off by getting better measurements, I can't be sure. IIRC Baxendall measured the ESL63 for the book edoted by Borwick. I'll see if that shed any light.
 
Occurs to me that maybe I should do a better version of the programs I wrote to carry out this as a speaker measurement. The snag for others is that it would be in 'C' and built for Linux. Alternatively, I could generate a stepped test waveform people could play out and capture. But then they'd need to be able to measure the relative *phase* of some sections as well as amplitudes. Does something like Audacity do that? I can't recall seeing it. if not, again, a program to do it would be needed.
 
Jim Audiomisc said:
That's the kind of problem I was warning about a while ago. REW is doing some sort of 'averaging' or working with too low a resolution, etc. This tampers with the data and makes it harder (or impossible!) to get reliable time-domain info, particularly at HF.

From what I've seen thus far I guess I can get results for the LF region up to about 2kHz. But beyond that it'll vary from case to case.

The ESL63 results look very good in comparison. But how much that's because the ESL63 is phase linear, and how much its because I started off by getting better measurements, I can't be sure. IIRC Baxendall measured the ESL63 for the book edoted by Borwick. I'll see if that shed any light.
Click to expand...
I think I have measurements of my ESL 63 too, perhaps that would be good for comparison?

With regards to the three Tannoy models I mentioned, it's mainly the low frequencies we're interested in comparing to see if it reveals much about the different cabinet designs.

My list of measurements to send to you appears to be growing with each post(!). I don't wish to exploit your generosity to process all this data so perhaps I should list what I have and allow everyone to vote on what would be most interesting to analyse?:

- Farfield measurements of Quad ESL 63 (left & right individually from listening seat).
- Farfield measurements of Tannoy Lancaster MG15, Tannoy Lockwood MG15 and Tannoy Edinburgh MG12 (left & right individually from listening seat). EDIT - I do not appear to have farfield sine wave sweeps of my Edinburgh MG12s, only Pink Noise RTA, so sadly no farfield phase information.
- Nearfield measurements of Tannoy Lancaster MG15, Tannoy Lockwood MG15 and Tannoy Edinburgh MG12 (left & right individually from 1 metre).
- Farfield measurements of Tannoy Lockwood MG15 before/after Dirac frequency & time correcting DSP (left & right individually from listening seat).
- Nearfield measurements of two pairs of JR149 and two pairs of JR149 mk2 (left & right individually from 1 metre).

Is it reasonable to assume that is is unnecessary to analyse the left and right channels from each speaker pair? If so then I could just send you measurements for one channel, which would effectively halve your workload.

PS - I've tried exporting the data to .txt with various levels of smoothing applied but the number of data points remains unchanged, so it seems the smoothing only the visual presentation of the data and the number of points in the .txt file is likely to be the highest resolution original captured.
 
Not followed this in detail but a couple of observations:

- if you divide Jim's plots by 2*pi they would look closer to what might be expected. Perhaps worth a check.

- The phase can be calculated from the magnitude for systems that are equalisable ("minimum phase") like loudspeakers. I believe REW has a button to do this so it might be worth pressing (but I am not familiar with the software). If it doesn't then you can calculate it with a python script (or equivalent) and a Hilbert function.
 
h.g. said:
Not followed this in detail but a couple of observations:

- if you divide Jim's plots by 2*pi they would look closer to what might be expected. Perhaps worth a check.

- The phase can be calculated from the magnitude for systems that are equalisable ("minimum phase") like loudspeakers. I believe REW has a button to do this so it might be worth pressing (but I am not familiar with the software). If it doesn't then you can calculate it with a python script (or equivalent) and a Hilbert function.
Click to expand...
Is this the 'Minimum Phase' button? (See post #7)
 
Briefly glancing at the phase graphs in REW measurements I've made in recent years, the measurements with the LEAST number of phase wraps ('scribbles') are those from Tannoy drive units that were measured enclosure-less, i.e. the drivers simply laying on the floor at a 45-degree-ish upward angle towards the mic. This is the opposite of what I expected, - I'd have thought the output from the rear of the cone would cause havoc with the phase?!
 
ToTo Man said:
Is this the 'Minimum Phase' button? (See post #7)
Click to expand...
Probably but to be comfortable with what is actually being done it would first need checking with signals where the answer is known. My point (which I think is confirmed by your plot) is that REW can almost certainly produce sensible phase plots if you master the controls. However, like Jim, if I had your data to analyse I would write it out and analyse it myself because I don't know REW and would probably want to do other things with it that REW doesn't have buttons for.
 
You must log in or register to reply here.


advertisement


Back
Top