Home Theater Forum and Systems banner

1 - 20 of 20 Posts

·
Registered
Joined
·
261 Posts
Discussion Starter · #1 ·
This might be a stupid question but I don't know the answer so I will ask anyway :)

Let's say you have a measurement of a speaker and you want to bring in the right window to remove some reflections from the response. Why can't you just move the left window out by the same amount in order to maintain frequency resolution?
 

·
Premium Member
Joined
·
1,562 Posts
The short answer is that there is normally no data there. It is increasing the total window width to accommodate lower frequencies, but that is not helpful if the LF IR data isn't located there.

I say "normally" because if, for example, a SW distance/delay is set such that its sound starts arriving before the TW then some of the LF data does in fact start before the IR peak. This not normally how the drivers are timed in commercial speakers however. The LF normally starts at about the same time as the HF, but takes more time to complete. The window width for a 10Hz freq takes a 100ms window . A 10kHz freq takes only 0.0001ms window.

In the case that the LF data starts before the IR peak then moving left window out is to the data start is appropriate.

By looking a the step response you can more easily see where the HF arrives Vs the LF.
 

·
Registered
Joined
·
261 Posts
Discussion Starter · #3 ·
Let me give an example of what I am referring to. Here's a close mic measurement of a sub

View attachment R SW.mdat

Tt has an unusual ripple around 100Hz with the default window settings.

uxl_nearfield_fr_default_window.jpg

Here's the IR

uxl_nearfield_ir.jpg

Now lets say I shift the window way over to the left, for example by moving the window ref time. This now gives me this windowed IR

uxl_nearfield_ir_window_shift.jpg

Note the green arrow, I've set the window like this deliberately to reduce that part of the impulse right down.

Here's the corresponding FR

uxl_nearfield_fr_window_shift.jpg

Obviously the level is reduced as the way I set this window reduced the size of the impulse peak however, importantly, the shape is exactly the same and that ~100Hz ripple is gone.

Therefore this suggests to me that that unusual ripple in the response is contributed by that later part of the impulse.

However I know that you can do all kinds of crazy invalid stuff by fiddling with windows so my question here is whether this approach to analysing the IR is valid or not.
 

·
Premium Member
Joined
·
1,562 Posts
Well, that worked out okay, but it is not the way I would recommend the windows to be set.

It helps to turn on the window to get a visual look at the data to be used. Below is the window you applied. We can see there is a lot of zero data included on the left. As a result REW thinks the LF cutoff is 1.65Hz, not a problem in this case since we are not concerned with the LF end of the trace.

We also see that the window is accepting data out 50ms on the right so it is eliminating a good percentage of the reflections as was your intent. It is also trimming off some the peaks of the IR which is reducing the SPL level of the result.

1 Window.PNG

Below for comparison it how I would have set the windows. We can see the reference time was just left at 0ms. The left window was set to just capture the initial rise of the IR. The right window was set to the same 50ms as you used (for consistency to your setting). Also a Blackman-Harris 4 window was used on the right window so that the amplitude of the right side reflections would be minimized in the same general manner that your settings accomplished. Note that the main portion of the IR is not clipped however so the SPL amplitude is not be impacted.

2 ja Window.PNG

Below is the resulting SPL chart. If your trace is shifted up in level about 9dB it follows pretty closely to my trace.

3 SPL.PNG

So I think your settings worked okay to draw the conclusions you made, but the window settings chosen to do it was somewhat questionable.

View attachment ja IR.mdat
 

·
Registered
Joined
·
261 Posts
Discussion Starter · #5 ·
Thanks for looking at this. I did try a variety of settings with a BH4 on the right but it seemed like I was just smoothing the ripple away hence I didn't trust the interpretation. How can you be sure you're not doing that? the sawtooth effect has a resolution of less than 17Hz after all?

though now I look at your settings, it seems more like the reflections are much further out in the +20ms or so area.
 

·
Premium Member
Joined
·
1,562 Posts
I am not sure what the comment/question is, but...

The right window does remove the impact of any reflection that fall outside the right window. That removes ripple.

The windows settings depend on what you are trying to see. I understood that you were just trying to get a sense of the what the SPL response might look like if the impact of some of the later arriving room reflections were reduced. Either your setting or mine accomplished this in this case.

If your question was something else please clarify.

It is always a balance to set the windows to retain the info desired while removing or reducing undesirable effects.

If the right window is too narrow the SPL of the low freq falls off.

Any data that is outside the right window is not used.

In the taper areas the data is reduced accordingly. This allows the LF to still come through, but helps block some the HF ripple of reflections in that time range.

Window settings are up to us and can only do so much. It works well for some questions and doesn't help at all for others. Most questions fall in between.
 

·
Registered
Joined
·
261 Posts
Discussion Starter · #7 ·
My point was that the area of interest was really ~85-120Hz and there was a sawtooth effect maybe 10Hz wide (peak to peak). Your window settings result in a frequency resolution that could elide that detail away and hence it becomes hard to say whether the ripple is not there because of the smoothing or not there because it's simply not in the windowed IR anymore.

It was more an observation albeit somewhat quizzicallystated :) I think what I'm really trying to understand is why my approach of simply rolling the window to the left is questionable? I realise that moving the window like that appears to be a massive hack but still.
 

·
Premium Member
Joined
·
1,562 Posts
My point was that the area of interest was really ~85-120Hz and there was a sawtooth effect maybe 10Hz wide (peak to peak).
Okay I am now on point.

Your window settings result in a frequency resolution that could elide that detail away and hence it becomes hard to say whether the ripple is not there because of the smoothing or not there because it's simply not in the windowed IR anymore.
The settings only eliminate some of the very low freq, i.e. below ~17Hz, and also eliminate late arriving reflections (beyond ~50ms). The windows limit the time window that is analyzed, not the frequency directly. [Although the low freq limit is impacted by the sum of the L/R windows.]

The ripple ~85-120Hz is there as measured. In this case we can also determine that it is due to reflections in the room occurring at ~100ms. To show this just change my right window to Turkey .25 and move it out in steps. The ripples will start to show up around 100ms. That is; the reflected sound in the room arrives back at the mic after ~100 ms and impact the SPL at the mic.

It was more an observation albeit somewhat quizzicallystated :) I think what I'm really trying to understand is why my approach of simply rolling the window to the left is questionable? I realise that moving the window like that appears to be a massive hack but still.
Your right window setting was also 50ms so it also removed that ripple. I set mine to agree with yours so both methods worked about the same in this case.

The only concern I had concerning your settings were:
> Shifting the left window far to the left of the IR data causes REW to plot LF data below the frequency where it becomes meaningless. You weren't looking for information there in this case so it is inconsequential.
> With those setting it appeared to me that the window taper extended all the way down well below the start of the IR (see chart above) so even early arriving data was being somewhat impacted. The SPL result was not very different however except for the 9dB reduction of level. There was only a very small impact on the trace shape.

I was just trying to explain a preferred approach to setting the windows. One where:
> The SPL trace is not showing a trace at lower frequency than there is data for.
> The SPL level is not reduced.
> There is no concern that the heavy taper window shape extending through the initial IR peaks will impact the SPL plot in some negative way.
 

·
Premium Member
Joined
·
1,562 Posts
Well I wasn't quite on point after all as I now see this is a near-field measurement. It is no doubt an attempt at quasi-anechoic SW response. Sorry.

My frame of reference above was thinking of LP measurements for EQ purposes.

It all still applies except that of course I would have said that the ripples do not occur due to the SW anechoic response they are related to room reflections.

I also would have mentioned that it is common to set the right window such to just pass the lowest freq of interest or until the first significant reflection is eliminated; whichever is shorter.

The anechoic response of any properly functioning SW always smooth and defined by the Q of the design and the order of the design. It is not helpful to try to measure it except to confirm that a DIY meets the design intent. Any room will overwhelm the natural SW SPL response in practice.
 

·
Registered
Joined
·
261 Posts
Discussion Starter · #10 ·
The anechoic response of any properly functioning SW always smooth and defined by the Q of the design and the order of the design. It is not helpful to try to measure it except to confirm that a DIY meets the design intent. Any room will overwhelm the natural SW SPL response in practice.
I understand about setting the windows to eliminate 1st reflections and I understand that quasi anechoic in room is impossible however I can't drag my sub outside so this is what I was attempting to establish :) apologies that I was unclear as to what I was doing before. The root of the query is the unusually flat response & the suggestion is that I have actually created a 4th order bandpass design due to the way in which my (sealed) sub is positioned in room.
 

·
Premium Member
Joined
·
1,562 Posts
The root of the query is the unusually flat response...
I would expect a smooth flat anechoic response for any well designed SW in its band-pass so I don't see anything unusual in that regard.

...& the suggestion is that I have actually created a 4th order bandpass design due to the way in which my (sealed) sub is positioned in room.
The LF cutoff is a little more slope than expected, maybe 16-18dB/octave instead of 12dB/octave as expected for second order, but this is a hard measurement to make accurately and it's close enough that it is not unusual for an in-room measurement. The HF roll-off is pretty steep, nearer 24dB/octave or 4th order. This is significantly more than what is expected unless there is some XO or other low-pass filter active during measurement. I doubt the room would do this that cleanly, but that is only a guess. If there is no filter active then even given additional information on the design and test conditions I probably would not be able to identify any other cause.

I expect there was only one SW active during this measurement . If 2 were active that could possibly contribute to this.

I don't really have any good ideas to explain this.
 

·
Registered
Joined
·
261 Posts
Discussion Starter · #12 ·
The LF cutoff is a little more slope than expected, maybe 16-18dB/octave instead of 12dB/octave as expected for second order, but this is a hard measurement to make accurately and it's close enough that it is not unusual for an in-room measurement. The HF roll-off is pretty steep, nearer 24dB/octave or 4th order. This is significantly more than what is expected unless there is some XO or other low-pass filter active during measurement. I doubt the room would do this that cleanly, but that is only a guess. If there is no filter active then even given additional information on the design and test conditions I probably would not be able to identify any other cause.
the suggestion is that the way my subwoofer is positioned in room (flush fitted into an alcove with the drivers firing into the wall, 3" gap between driver and wall) has turned it into a 4th order bandpass enclosure with an extremely leaky (low Qa) vented section of the enclosure.
 

·
Registered
Joined
·
261 Posts
Discussion Starter · #13 ·
I wonder if I can trouble you for a view on the way I've applied the windows in this measurement

View attachment mp150_C_1.2m_0H.mdat

this is an in room measurement of the speaker with the mic and speaker positioned so that I had ~1.3m in all directions from both mic and speaker before hitting any surface. The green arrow points at the first serious reflection which suggests I get a bit more time before a strong reflection hits.

speaker_ir.jpg

Should I move the L window further in so it is up against the very start of the impulse?
Should I use BH4 on the R window to taper the response further thus smoothing the HF FR further?
 

·
Premium Member
Joined
·
1,562 Posts
This looks good to me. Any minor fine tuning adjustments from here make no practical difference.

Just comments FYI:
> RE left window location: It's fine as is. The 1ms of zero data before the IR start is small enough to not mislead us too much as to the limit of the LF accuracy. We could set it to the start at -.3ms as you asked, but the impact is probably negligible for your needs.
> RE BH4 on right: It's fine as is. Using BH4 will only smooth the HF a little more as you noted. The SPL and phase response is pretty clear either way.

If you were hoping to see lower freq then using BH4 with a wider right window would allow that without getting too much HF ripple. That is, it helps reduce the impact of the late arriving HF that causes the ripple. If your initial settings show the range you are interested in then all is good as is.

Just for your info I am attaching the adjustments I made:
> A slight manual shift to the location of the IR to align the initial peak at 0ms. This shows the phase in a more conventional way. That is, with the TW falling to near 0°. It's just a habit as I do a lot of phase work and want charts to be comparable for measurement overlays. This not needed for other charts.
> I changed to BH4 and a 10ms right window. A little additional HF ripple crept in as a result, but the LF is a little deeper and a little more accurate. At least I think it is.

This is just a setting I may choose if I were to trying to get as much info as possible on the low end and not disrupt the high end too much. Normally we can get what we need using near field measurements without fussing around this much. These setting become more tricky to establish if we are measuring at the LP and trying to see the phase.

[For some reason my .mdat attachment is being identified as .php in this post and will not open so I will try post it again later.]
 

·
Registered
Joined
·
261 Posts
Discussion Starter · #16 ·
thanks for the extra insight, v useful as always.

as a sidebar, do you think this resolution of data is good enough for XO design? I'm going to be building some speakers soon and the XO is likely be in the region of 1.5kHz, just curious as to whether I need to move outside or not.
 

·
Premium Member
Joined
·
1,562 Posts
No need to move outside for XO related measurements in my opinion. Your measurement above is very good throughout that XO region.

Your measurement above shows the XO in the speaker used there is at about 1.4kHz and it is reasonably well timed for handoff between the drivers. A rough estimate suggest the there may be about a 45° offset of the phase between the 2 drivers at 1.4kHz. The TW should be a little more delayed to be ideal. It is not clear if the entire SPL dip at that frequency is XO timing or not, but a portion of it probably is.

In modeling for a new speaker XO setup just measure the 2 drivers separately as mounted in the box using loopback timing and enter the results into a PCD or other modeling program. For MiniDSP or other DSP XO you can just find one of the "ideal" XO settings experimentally; no need for modeling.
 

·
Registered
Joined
·
261 Posts
Discussion Starter · #18 ·
OK that's good news, thanks for that :)

how did you come up with the estimate of a 45degree offset?
 

·
Premium Member
Joined
·
1,562 Posts
It is just a hint from the phase chart. Without making measurements of the individual drivers it is not possible to know with any real certainty.

3 Hints were:
> Step response suggests the there is a slight time gap in the handoff from the TW to the MR.
> The SPL response has a dip at the XO point.
> The phase chart suggests an extension of the general trend line of the MW below the XO and TW above the XO to be slightly offset from one another.

I just mentally extended those 2 trend lines and used REW to measure the apparent offset - just a fun guess, maybe ill advised.

I could easily be wrong on this, and would only be a little surprised if it proves to significantly different from this. it appears to be as good as many/most passive XOs, it just doesn't appear to be ideal. I guess I would be a little surprised if the offset wasn't someplace between ideal and 90°. This estimate is also based on an assumption that this is a small box speaker with a passive XO.

The only way for me to really know is with some proper measurements.
 
1 - 20 of 20 Posts
Top