Erin's Audio Corner gets a Klippel NFS!

[richard12511]

Please, don't go on 'popular opinion' on this one. Science is not a democracy. With normalization polar plots are just 'pretty graphs' - you can't see useful information (resonances, reflections, diffraction/HOMs)

Stereophile's graphs are useless because of the normalization. If you don't want to listen to me, or don't beleive that 'normalization is evil', go by the works and words of Geddes and choose a colour scheme accordingly. Choosing it was pretty involved and went pretty deep on perception. You can search the diyaudio archives on that if you don't beleive me.

Both normalized and non normalized serve a purpose, so both should be included imo.

 

[MZKM]

I prefer normalising to the LW for the same reason Harman calculates the DI based on the LW, using the on-axis can show 'resonances' which simply aren't there under normal circumstances

Those would be diffraction elements (cabinet, waveguide, etc.). A resonance shows up in all (so DI is smooth).

For the bootleg heatmaps I generate, they are normalized to the calculated sensitivity. Sort of a middle ground.

 

[richard12511]

Yet a single, non-normalized polar map can tell a whole story of a loudspeaker. As presented here. I don't get your reasoning.

Yet it omits so much information, that is necessary to evaluate a loudspeaker..

Showing incomplete information doesn't make one 'respected'. That one makes information available, what manufacturer dares not to do, makes them respected. Doesn't change the fact, that normalization is hiding valuable information.

I have shown you how a single graph can tell so much about a loudspeaker, yet you call me 'simply wrong'. Look - I have nothing to do with anything. Presenting data optimally, possibly with a single graph - that is the issue.

Why not just show both, though? The Geddes plot you posted is imo harder to read and compare small differences, and thus it's less useful to me than the beamwidth and polar plots Erin provides.

 

[MZKM]

Why not just show both, though?

Erin has stated he wants to limit the graphs to keep website storage needs down.

 

[TimVG]

Those would be diffraction elements (cabinet, waveguide, etc.). A resonance shows up in all (so DI is smooth).

For the bootleg heatmaps I generate, they are normalized to the calculated sensitivity. Sort of a middle ground.

Well I put it between brackets, but it would be all the same. Diffraction elements etc would be shown as resonances in the off-axis plots. When normalised to the LW however true resonances would still be shown, but other local phenomena wouldn't.

 

[richard12511]

Erin has stated he wants to limit the graphs to keep website storage needs down.

Or we could just donate more .

I personally think both are useful. The non normalized beamwidth is easier for my brain to assess general dispersion width in comparison to other speakers. The normalized beamwidth makes it easier for my brain to see (in a general sense) how closely a speaker's off axis angles track its own on axis(or LW) response. For more precise 1-2dB careful comparisons, I like the off axis SPL graphs.

 

[MZKM]

Or we could just donate more .

But that’ll reduce his budget for buying speakers.

 

[jtwrace]

I have 24TB...maybe I need to host the data.

 

[headshake]

keep website storage needs down.

Offsite storage? Get a google/whatever drive or talk to jtwrace. Dumping on amazon is another option. You only pay for the data transfer and not the file size.

 

[Jdunk54nl]

Offsite storage? Get a google/whatever drive or talk to jtwrace. Dumping on amazon is another option. You only pay for the data transfer and not the file size.

There are issues with quite a few of those. Hosting the files on your own computer, well your IP address changes so every-time it does you would need to update and you'd have to leave the device on and connected to the internet (hope you never lose internet) so $$ there)

Google Drive (or whatever cloud based storage) could work, but then again something could get broken with the links with a simple switch google decides to do (like make photos take up storage space instead of free space). Sure they may give you warning but that means Erin would have to go and change all of that stuff in every review he did that way. Even then, storage is cheaper with these cloud storages, but not free. Once you get beyond like 100gb, it gets expensive.

Easiest by far is host on the same platform (but also most costly to use others storage), that way if any issues, it is usually with the whole platform.

Edit* I just looked up godaddy web hosting and it is only $8/month for unlimited storage compared to $6 for the first tier plan. So maybe Erin can look into that. Not sure what he is using for his website. I know wix only has 50gb max plans advertised, and they get expensive.

 

[617]

How does a normalized plot help evaluate a finished product? I'm not clear on the dispute here. The performance represented by a normalized plot cannot be attained except by incredibly involved equalization, which I assume Erin is not going to provide (nor should he.) It's sort of interesting to see how good a speaker could be with equalization but it says nothing about the actual performance.

We look the directivity plots to see if their are large peaks and nulls off axis which affect the reflected sound of a speaker. Normalizing a plot will exaggerate some nonlinearities and obscure others. Again it is a very useful tool for product development but what is the use for product evaluation?

 

[MZKM]

As pointed out, the raw and normalized off-axis plots tell us the same data, heat maps are just to be pretty.

But to point out differences between variants:

Normalized to on-axis: Only shows directivity.

Normalized to listening window: Shows realistic directivity, major resonances, and diffraction in the on-axis.

No normalization: Shows non-linearity, resonances, diffraction, and dispersion.

Referenced to avg(SPL): Shows major non-linearity, major resonances, major diffraction, and (perceived?) dispersion.

______
For the above, dispersion and directivity are different. Dispersion is how wide/narrow the soundstage is, directivity is how the off-axis compares to direct sound. Directivity shows dispersion if using intense EQ.

 

[617]

How does a normalized plot show directivity? It shows hypothetical directivity of a speaker after equalization.

I mean I can see the value of some of the other forms of normalization, such as to the listening window. Raw sonograms have two flaws in my eyes - one, they tend to look weird at low frequencies, and two, they tend to show a lot of warts which shouldn't be too concerning.

To help illustrate your point, can you post the same data with no normalization, listening window normalization, on-axis normalization and SPL normalization?

I think we both agree that the purpose of these plots is not to investigate the acoustic design of the speaker without reference to FR, but to show issues which may indicate poor performance. One of the reasons I favor a strict, no normalization plot is that there are speakers out now, like the Genelec coaxial cardiods, and the D+D units, which have outstanding and really game changing smoothness off axis, and super well controlled low frequency dispersion. The plots should be constructed in such a way to show how amazing these speakers are compared to conventional monopoles/ported monopoles.

 

[Sancus]

I feel like people are arguing about ~the ideal graph~ but that's totally missing the point. I personally use the beamwidth/normalized contour plot to compare a) beam width and b) beam smoothness and to a lesser extent how low the directivity control goes.

The reason I raised the initial concern about the heat map coloring being very different(in terms of how many dB were covered by red) is that many sources with substantial data(Amir, S&R, Neumann/Genelec, Princeton, etc) have a very similar graph, so it makes it really simple to tell at a glance: What does the dispersion look like on this speaker compared to some other speaker that was reviewed?

For a speaker that is actually good(ie free of resonances and reasonably flat/even), that's one of the biggest variables in how it will sound, if not the single biggest, so being able to compare at a glance is handy.

If you want to analyze a speaker in isolation, hunt for resonances or lower-amplitude discontinuities in the frequency response both on and off-axis, Erin includes the colored 2d plots that are ideal for this. So I don't see any reason that ALL of his charts need to be non-normalized, but honestly I don't have a strong opinion about normalization. I just want graphs that are similar to other sources so that it is easy to compare!

I don't care at all how good each individual graph is for analyzing multiple attributes at once, to me that's 100% irrelevant. We have many graphs for a reason.

 

[napilopez]

How does a normalized plot show directivity? It shows hypothetical directivity of a speaker after equalization.

I mean I can see the value of some of the other forms of normalization, such as to the listening window. Raw sonograms have two flaws in my eyes - one, they tend to look weird at low frequencies, and two, they tend to show a lot of warts which shouldn't be too concerning.

To help illustrate your point, can you post the same data with no normalization, listening window normalization, on-axis normalization and SPL normalization?

I think we both agree that the purpose of these plots is not to investigate the acoustic design of the speaker without reference to FR, but to show issues which may indicate poor performance. One of the reasons I favor a strict, no normalization plot is that there are speakers out now, like the Genelec coaxial cardiods, and the D+D units, which have outstanding and really game changing smoothness off axis, and super well controlled low frequency dispersion. The plots should be constructed in such a way to show how amazing these speakers are compared to conventional monopoles/ported monopoles.

I guess the way we look at polar plots is quite different. For me the only reason to even consider them is for a quick overview of directivity characteristics and nothing else-- SPL plots are just much clearer for precise information about what's happening at every angle. My problem with non-normalized polar plots is that when evaluating speakers with notable peaks and dips in the frequency response, they make it look like there speaker has a directivity problem, when this is often not the case.

Those in favor of non-normalized plots appear to like having the polar plots tell them something about the frequency response of a speaker. But my impression, both intuitively and from reading comments on ASR, is that most people look at polar plots for directivity. I may be wrong, but this is my anecdotal observation.

To use one of my own set of measurements as an example that stood out to me a while back, here's the Sonos Faber Sonetto II:

This isn't exactly the prettiest spin, and even the DI lines suggest mediocre soundstage performance (which I knew wasn't the case subjectively). This is partly a weakness of the spinorama not prioritizing horizontal information in the DI curves. Indeed, if you look at the non-normalized polar plot, at a glance it gives the impression of serious directivity issues (smoothed here to 1/6 octave for clarity)

If you only have the spin and the above non-normalized plot, people will see the dips between 1 and 2k and dismiss the directivity performance immediately.

But the normalized plot shows that horizontal directivity is actually quite well controlled, other than the narrowing about 6kHz(not necessarily a bad thing anyway). In fact, this speaker actually had some of the best waveguide-less directivity I've measured:

Indeed, subjectively the soundstage performance of this speaker was impeccable. It's just the tonality that was off.

The other problem with non-normalized plots is that it is often hard to compare among different sources because their appearance can vary quite dramatically depending on your reference SPL. Because of some rather large FR dips and peakspresent at every angle, the non-normalized plot can also look like this:

Further muddying the picture.

In the absense of other measurents, non-normalized plots obviously tell you more information. But with speakers like these(and we've seen quite a few speakers with bad FR but good directivity!), puzzling directivity properties out of a non-normalized polar plot requires so much scrutiny as to be effectively obscured. But if you already have non-normalized information elsewhere, the normalized plots instead give a clearer and more intuitive look at directivity.

But again, if it were just me I'd stick to SPL plots...

 

[617]

I guess the way we look at polar plots is quite different. For me the only reason to even consider them is for a quick overview of directivity characteristics and nothing else-- SPL plots are just much clearer for precise information about what's happening at every angle. My problem with non-normalized polar plots is that when evaluating speakers with notable peaks and dips in the frequency response, they make it look like there speaker has a directivity problem, when this is often not the case.

Those in favor of non-normalized plots appear to like having the polar plots tell them something about the frequency response of a speaker. But my impression, both intuitively and from reading comments on ASR, is that most people look at polar plots for directivity. I may be wrong, but this is my anecdotal observation.

To use one of my own set of measurements as an example that stood out to me a while back, here's the Sonos Faber Sonetto II:

View attachment 114667

This isn't exactly the prettiest spin, and even the DI lines suggest mediocre soundstage performance (which I knew wasn't the case subjectively). This is partly a weakness of the spinorama not prioritizing horizontal information in the DI curves. Indeed, if you look at the non-normalized polar plot, at a glance it gives the impression of serious directivity issues (smoothed here to 1/6 octave for clarity)

View attachment 114675

If you only have the spin and the above non-normalized plot, people will see the dips between 1 and 2k and dismiss the directivity performance immediately.

But the normalized plot shows that horizontal directivity is actually quite well controlled, other than the narrowing about 6kHz(not necessarily a bad thing anyway). In fact, this speaker actually had some of the best waveguide-less directivity I've measured:

View attachment 114678
Indeed, subjectively the soundstage performance of this speaker was impeccable. It's just the tonality that was off.

The other problem with non-normalized plots is that it is often hard to compare among different sources because their appearance can vary quite dramatically depending on your reference SPL. Because of some rather large FR dips and peakspresent at every angle, the non-normalized plot can also look like this:

View attachment 114679

Further muddying the picture.

In the absence of other measurements, non-normalized plots obviously tell you more information. But with speakers like these(and we've seen quite a few speakers with bad FR but good directivity!), puzzling directivity properties out of a non-normalized polar plot requires so much scrutiny as to be effectively obscured. But if you already have non-normalized information elsewhere, the normalized plots instead give a clearer and more intuitive look at directivity.

But again, if it were just me I'd stick to SPL plots...

I see what you're saying. The non normalized plot I agree is not that useful due to the issues with the bass response dropping off.

My contention is that directivity and frequency response are related, and for example if you're looking at what the widest reasonably smooth angle your speaker can work at, a normalized plot will not give you that information.

As can be seen in these examples, the sonus faber appears to have a smooth pattern out to 60 degrees in the normalized plot, but this is not actually the case. In reality the response at +60 degrees is pretty terrible, with the completely predictable narrowing and God knows what else at 1.3khz or so. The best case beamwidth is much narrower than the normalized plot shows, and what it tells you is that to sound best, you'll want to absorb reflections coming from that angle.

The fact that the normalized plot confirmed your impressions of soundstage is not really informative. You didn't hear the speaker in that plot. That speaker does not exist. You heard the speaker with the ****** measurements and it sounded good, which is interesting but not unexpected; Sonus Faber has been making this kind of speaker for a while and knows how to work around the fundamental limitations.

We can agree to disagree on this and I'm not going to have a conniption here but the way I see it, the sonogram is the Truth, the overly informative image of the speakers actual response in space. For a good summary of what matters in terms of audibility, the cea 2034 plots are fine, it has enough information to know if a speaker is good, if it is wide or narrow, and what kind of room it wants. But the sonogram? Either show the warts or leave it out.

Erin is presumably going to put the D+D and who knows what other great, state of the art speakers on the NFS, and these designers deserve to have their products stand out from the crowd. A normalized two way like the Sonus Faber, with a low order crossover and a big tweeter, looks almost as good as the Genelec 8431, apart from maybe some better pattern control down low. However, the reality is that the Genelec approaches perfection for a huge bandwidth, and the Sonus Faber is about as good as a hundred other normal speakers.

My girlfriend is making dinner tonight and there's no Celtics game so I'm prepared to argue about this all evening.

 

[tktran303]

The Sonus Faber example is a great counter-example:

There is clearly a whole lot of things going wrong for over 2 whole octaves- between 600Hz to almost 3Khz. This speaker has a problem on axis.

But the normalised graph is going to appear smooth, because when compared to the on axis- all the off axis are equally and consistently terrible.

 

[617]

I just checked and there is a Celtics game, but even though they beat the Hawks last week, they're on the second game of a back to back and arguing about graphs on ASR might actually be more fun than watching them choke away another 24 point lead and pass to a guy with a quarter billion dollars worth of old knees.

 

[hardisj]

I gotta be honest... this is kind of a hot mess. I do understand everyone's points. This just proves to us that we all have our own preference and sometimes just flat out think another's is dumb.

Though, to be honest, I most agree with @napilopez and feel that the heatmap graphics as typically presented are more just an "eyeball" chart to see how the directivity pattern trends. But, the DI data as well as the off-axis horizontal data also tell us that.

This is also why I came up with the globe polar; because it gives a much more intuitive feel for the radiation pattern of a speaker. You can more easily picture (in your mind) how the speaker's dispersion behaves all the way around and easily spot mismatches in directivity.

 

[MZKM]

How does a normalized plot show directivity? It shows hypothetical directivity of a speaker after equalization.

See what I say towards the bottom for directivity vs dispersion.