It is very subjective. You are trying to get enough detail to understand what is going on but not so much that the forrest gets lost in the trees. The 1/6 or 1/12th Octave smoothed phase curves a couple of pages back looked about right to me.
The issue of calculated group delay is a little thornier. This is a differentiation of the phase curve (delay = -dphi/domega) so fine wiggles in the phase curve become huge swings in the group delay curve. Since the fine wiggles look like they are due to a non-anechoic measurement (not inherent in the unit) then the spikes in the group delay curve are meaningless.
Smoothing the phase curve prior to differentiation would give a more meaningful group delay curve.
The spikes in the group delay are not representative of non-minimum phase behavior.
David
Thank you very much Catapult,
bad conditions right now, house is rebuilt ... "wet" room,
lots of glass. But it is the only room quite large enough
to use at the moment.
Nevertheless interesting to see different results from
different software and personal processing preferences.
I guess i still have a more "dry" IR and should ARTA feed
with it to inspect the region >3Khz.
This has gotten real workshop character !
Thanks to everybody for contributing theory and practical
approaches. I could not contribute much, but lots of
information/hints is underway in here !
Kind Regards
HolmImpulse is easy to use (and free) .... since you have a critical view
on measurement software, i think your opinion would be valuable.
Kind Regards
Last edited:
Along similar lines..
I remember Tom Danely stating something like: using TEF because he couldn't truly get accurate phase data from other systems.
If this refers to my plots, they are not group delay, they are excess group delay. Excess group delay is flat (zero slope) in minimum phase regions of the response (more precisely, where the response is minimum phase plus a time delay of the value of the excess group delay in the flat region). In areas where the excess group delay is not flat the response is not minimum phase.
Yes I have a TEF but it does not offer excess time subtraction so the resultant plots have the excess time phase wrapping effect. To bad. The measurements from TEF can also be fooled by incorrect setup and "bad test station" effects so some user intelligence is required.
For an individual driver the IASYS works fine and is used here as a convenience and adequately accurate measuring system with no user intelligence required.
When I want the nitty gritty I use the multi-tone slightly modified sine wave method. This has the HUGE advantage because "excess time" does not enter into the equation measurements. I look for a frequency somewhere around the center of the energy band where phase is zero and reference to that frequency. This gets within just a few degrees of reality as a measurement. Phase, in reality, of that 6.5" inch in the post actually wanders back and forth about +/- 20 degrees from zero over the pass band. This is smaller than the resolution of the IASYS so the IASYS shows no change. This is also much less than all but a few drivers tested and the only midrange with one decade of usable pass band I have found though it really goes haywire up high as can be seen on the post. Same is true for the tweeter though (no haywire) I believe the SEAS one posted may be even better. Most drivers are real &*#% when it comes to phase response. Curved cone woofers are pure sin.
=SUM🙂
Last edited:
Yes, I was refering to your plots.
I've heard of excess phase and group delay but not excess group delay. These are for the planar speaker of post #1090, right? It appears that the curves were taken under non-anechoic conditions. As such the "hash" in the response curves give rise to similar hash on the phase curve. Differentiating the phase curve turns the high slope of the phase curve hash into sporadic spikes. It is a measurement and computational artifact. It is not a characteristic of the driver. (It would totally go away when measured under anechoic conditions.)
Another clue would be the level peaking to 150ms. The curves are flat between the spikes because the autoscaling has been dominated by the extremely high spikes, compressing any true info between the spikes. (At about a ms a foot, where would the sound go for 150ms?)
David S
Well my reality is I can get repeatable measurements using CLIO and more importantly they work just fine when imported into Leap Crossover Shop. So from my own experience using this software there is nothing to validate. I am certainly not going to go chasing after "coherence" as it's defined in the IASYS manual. More an issue with common sense, proper set-up and a consequence of using standard crossover topologies.
Rob🙂
Not to break up the MP/coherence conversation here, but I just wanted to post more measurements of the Little Behringer B2031P and ask an opinion.
Vertical polars from midway between the drivers, 11.25 and 22.5 degrees rotating toward the woofer:
then from our midway point rotating toward the tweeter in 11.25 degree steps out to 45 degrees:
What gather then is that ear height should roughly be tweeter height or a bit higher with this speaker and that it can be mounted on low(ish) stands and possibly under a TV for a center channel. Also it could be mounted upside down and rather high. Does this sound about right to everyone here?
here's the horizontal polars for anyone who missed it:
overlaid impulses:
avg of polars excluding the central axis:
the vertical polars were taken from a different speaker, same model as the horizontal. I didn't keep the horizontal polars from the actual speaker b/c the were too similar to waste the memory. I've measured three of these speakers and they are are very close. The 2 that came as a supposed "matched Pair" are so close to identical it's amazing.
Thanks,
Dan
Vertical polars from midway between the drivers, 11.25 and 22.5 degrees rotating toward the woofer:
then from our midway point rotating toward the tweeter in 11.25 degree steps out to 45 degrees:
What gather then is that ear height should roughly be tweeter height or a bit higher with this speaker and that it can be mounted on low(ish) stands and possibly under a TV for a center channel. Also it could be mounted upside down and rather high. Does this sound about right to everyone here?
here's the horizontal polars for anyone who missed it:
overlaid impulses:
avg of polars excluding the central axis:
the vertical polars were taken from a different speaker, same model as the horizontal. I didn't keep the horizontal polars from the actual speaker b/c the were too similar to waste the memory. I've measured three of these speakers and they are are very close. The 2 that came as a supposed "matched Pair" are so close to identical it's amazing.
Thanks,
Dan
Last edited:
So much verbiage.
DLR - I never said that other software had a "problem" only that the requirement for MP is not "necessary", especially with modern software like HOLM.
Dave S - you are quite correct, as any numerical person will tell you that calculating the derivative from measured data is highly prone to errors. The right way to do it is to fit an analytical function (like a polynomial) to the data and then take the derivative of that function. Point by point differencing is very crude. Using more points helps, but its still an unstable process by nature. Errors get magnified and no measured data is without errors.
Coherence is a real number between 0 and 1 and has no phase. In signal processing the coherence is defined as the magnitude squared of the cross correlation devided by the autocorrelations of the input and output spectrums. Hence it is dimensionless and is the ratio of two powers, the coherent power to the total power. It is 1 if all of the output power is due to the input, and zero if there is no portion of the input power contained in the output power. It cannot be negative nor have a phase.
DLR - I never said that other software had a "problem" only that the requirement for MP is not "necessary", especially with modern software like HOLM.
Dave S - you are quite correct, as any numerical person will tell you that calculating the derivative from measured data is highly prone to errors. The right way to do it is to fit an analytical function (like a polynomial) to the data and then take the derivative of that function. Point by point differencing is very crude. Using more points helps, but its still an unstable process by nature. Errors get magnified and no measured data is without errors.
Coherence is a real number between 0 and 1 and has no phase. In signal processing the coherence is defined as the magnitude squared of the cross correlation devided by the autocorrelations of the input and output spectrums. Hence it is dimensionless and is the ratio of two powers, the coherent power to the total power. It is 1 if all of the output power is due to the input, and zero if there is no portion of the input power contained in the output power. It cannot be negative nor have a phase.
Last edited:
Well Robh that is fine but without first hand validation you really have no idea if the results are accurate or not. The same error can be extremely repeatable! That means the system is "precise" (as defined repeatable) but says nothing about "accuracy." IMHO any test system or modeling software that comes through here gets validated using known circuits, crossovers, or drivers. Even microphones get tested as they are purchased. Same with theories. I don't trust any manufacturer or anyone else to tell me anything other than what the marketing department wants to say. Now I am not at all saying there are not good people and products out there because I have found a few however, there is a whole lot more that belongs in the refuse pile than in use. As for sound reproduction products that goes doubly so. Most of it I find unlistenable with so much distortion and so out of balance that I wonder why the item was even built! Magazines tout "Grade A systems" which have THD at normal listening levels exceeding 10% occasionally. What is that about? Ooooh- a tirade.
Suffice to say- "Assume" *** of U and ME is the golden rule. I test validate everything before attempting to design anything.😱America and the West is not capitalism, it is marketingism. He who markets the most wins and has almost nothing to do with product quality, performance, or functionality.
🙂=SUM
Dan I think you have it. A near field monitor really wants the main lobe of radiation to not point down because it bounces of the console and really messes the sound. This means the main lobe needs to be a little "higher" as in line more with the tweeter than the midrange and maybe even point up slightly. This is not perfect but placement of near field monitor is usually right over the meter bridge or on a table or something less that ideal. Your results cause me to wonder if it was designed sitting on a table or over a meter bridge? Interesting thought. Doubt we will ever know. Mine are...
🙂🙂=SUM
🙂🙂=SUM
I see the benefit of fitting an analytical function, but lacking that would not some minor smoothing be an improvement?
Dave
It gets validated when it gets Calibrated. Because it is calibrated it is accurate. I have also checked it against other measurement software.
Rob🙂
If by "smoothing" you mean averaging, that's not a real good method, but better than nothing. But if by smoothing you use some form of "spline" fit, then thats much better and pretty much the same as what I said.
Take for example a noisy sine wave. No amount of averaging is going to make the error go to zero, but a high order fit to this data CAN theoretically yield zero error. This is what high-end measurement systems like Sound Check actually do. In a slightly different form this is what I do in my Polar Mapping software. In very sophisticated Physics experiments data is never "averaged" its always fit to a model. This technique is far far better than simple averaging could ever achieve. Its also far more complex.
I think most of the group delay stuff that was posted is pretty noisy - based on what happens when its smoothed - so I would give it a lot of credance.
Last edited:
I have explained that already:
Take for example SoundEasy as you have mentioned. You have to gate out the nearby reflections to obtain the level of minimum phase data usable for system design work driver model data. This is known to most people designing speakers. Is this not true? To take the issue further, most people will use a short mic such as an ECM2000 and clamp it to a stand like using a normal mic. However, the clamp itself will generate some reflection/diffraction, do you gate it out or not? In most cases, the data is sufficient for work even if you don't gate it out, and you can still "consider" the data as "minimum phase" data. I find that whenever there are things that you do not seem to comprehend, you just put the blame on someone else. It would be a more mature approach to just drive the question deeper.
George
You say the same thing about me when I disagree with you. At some point a look in the mirror might be beneficial.
Oh, and for the record, a SISO measurement CAN have reflections and diffraction - or do I just "not comprehend"?
It CAN have reflection and diffraction, just to what extend these reflections and diffraction will polute the data enought that you cannot get data usable for your purpose. Is this not correct? If not, please explain why.
BTW, I only respond like that when people start pointing fingers instead of keeping the discussion subject related. If you can point to the specific post which this is not the case, I will openly appologize.
Last edited:
- Status
- Not open for further replies.