JBL HDI-3800 Floorstanding Speaker Review

[sweetchaos]

So the Klippel NFS might not be perfect? Buahahah! *Laughs as he clutches his Umik-1*

I fell off my chair laughing...

 

[pierre]

and-behold, the result shows the same trend. I know I looked at the results before but I can tell you that seeing this data never entered my brain. And I definitely wouldn't have remembered something from nearly a year ago (I can barely remember what

Here is the zipped data for you guys to do what you want to with it:
https://www.dropbox.com/s/fhicwsk4hsfh6ke/JBL HDI-3800.7z?dl=0
(@MZKM @pierre)


And, that's it. I'm exhausted. It's only Wednesday and this has been an exhaustive week between trying to get this review completed, my day job and my family. So, I hope you guys appreciate the effort and get something useful out of it. If you don't like the info I have provided or you want something I didn't do this time, wait a couple days before complaining about it.

Hi Erin,

thanks for the data. It took some times my main computer crashed.

Here is an EQ (I would keep the first 13)

EQ for JBL HDI-3800 computed from ASR data
Preference Score 5.1 with EQ 5.9
Generated from http://github.com/pierreaubert/spinorama/generate_peqs.py v0.6
Dated: 2021-03-26-12:31:29

Preamp: -3.4 dB

Filter  1: ON PK Fc 10321 Hz Gain +3.17 dB Q 5.97
Filter  2: ON PK Fc  2122 Hz Gain -1.27 dB Q 12.00
Filter  3: ON PK Fc   426 Hz Gain +2.11 dB Q 7.85
Filter  4: ON PK Fc   338 Hz Gain +1.99 dB Q 12.00
Filter  5: ON PK Fc  3676 Hz Gain +0.59 dB Q 0.42
Filter  6: ON PK Fc  2249 Hz Gain -0.91 dB Q 7.53
Filter  7: ON PK Fc  1055 Hz Gain +1.03 dB Q 2.03
Filter  8: ON PK Fc  2587 Hz Gain -0.36 dB Q 0.10
Filter  9: ON PK Fc  9421 Hz Gain +0.67 dB Q 12.00
Filter 10: ON PK Fc  4582 Hz Gain +0.53 dB Q 8.56
Filter 11: ON PK Fc   738 Hz Gain +1.34 dB Q 5.51
Filter 12: ON PK Fc   391 Hz Gain -0.54 dB Q 12.00
Filter 13: ON PK Fc  1575 Hz Gain +0.90 dB Q 7.72
Filter 14: ON PK Fc   578 Hz Gain -1.30 dB Q 12.00
Filter 15: ON PK Fc  1414 Hz Gain -0.40 dB Q 12.00
Filter 16: ON PK Fc  1302 Hz Gain +0.81 dB Q 12.00
Filter 17: ON PK Fc   517 Hz Gain -1.17 dB Q 12.00
Filter 18: ON PK Fc   544 Hz Gain +1.42 dB Q 12.00
Filter 19: ON PK Fc   497 Hz Gain +1.21 dB Q 12.00
Filter 20: ON PK Fc   638 Hz Gain +0.39 dB Q 12.00

Results on the spin (corrected up to 16k Hz):

And on the PIR:

 

[Bent Wookiee]

Thanks to Erin for taking the time and effort for putting this together. I was especially interested in this one as a current owner of the Studio 5 series, as I am sort of looking for "that, but more of it" as the next upgrade. Not thrilled with these results but not totally dissuaded either. Was not expecting Revel results, but also wondering what else might do the trick. JTR looks good, but bigger in both size and cost. Also interested in both Arendal and Perlisten but unsure of both the overall quality as well as if they can deliver the clean, dynamic sound I'm looking for. Guess it is better that I am uncertain as otherwise I might do something dangerous to my wallet.

Will be looking forward to the center channel review for sure!

 

[richard12511]

Thanks to Erin for taking the time and effort for putting this together. I was especially interested in this one as a current owner of the Studio 5 series, as I am sort of looking for "that, but more of it" as the next upgrade. Not thrilled with these results but not totally dissuaded either. Was not expecting Revel results, but also wondering what else might do the trick. JTR looks good, but bigger in both size and cost. Also interested in both Arendal and Perlisten but unsure of both the overall quality as well as if they can deliver the clean, dynamic sound I'm looking for. Guess it is better that I am uncertain as otherwise I might do something dangerous to my wallet.

I'm super interested in Perlisten.

 

[richard12511]

Well, now that everyone has decided where they stand regarding the LF... is there anything else here that you guys feel is worth talking about?

Really sorry if I came across as being too critical of your review. I bring this up only to hopefully help make the data even better. My hope is that you (or Amir) will be able to find edge case bugs with the NFS that Klippel can then fix and make the machine even better. I don't mean to criticize you, or your process. It's not you that I doubt at all, it's the NFS, and only in the bass frequencies, and only for very large towers. Overall, I think the NFS is the best measurement tool/method there is(even better than an anechoic chamber), but I'm wondering if Klippel still have some improvements to make for these edge cases. Also, when I say I "doubt" the NFS, it's not like I'm 100% convinced it's wrong, but more like I'm not 100% convinced it's right. I believe the NFS 100% for bookshelf speakers, but this review and Amir's Revel F328Be review have me thinking it might be getting the bass extension wrong for those 2 speakers.

Below are a few of the reasons I'm thinking the NFS might have a "bug":

- JBL reports the -6dB point for the 3800 as 37Hz. The NFS reports the -6dB point as 46Hz.
- Revel reports the -6dB point for the F328Be as 26Hz. The NFS reports the -6dB point as 48Hz.
- The NFS reports the -6dB point for the JBL HDI 1600 as 38Hz, 8Hz lower than it reports for the much larger tower speaker in the same series.
- Your ground plane measurements of the Revel F226Be report a -6dB of 43Hz, which is 3Hz lower than Revel spec, and 5Hz lower than what the NFS reports for the F328Be. This is probably the thing that makes me doubt the NFS most. I just can't believe the NFS when it says the F328Be has 5Hz less extension than its little brother. Same deal for the JBLs. I find it hard to believe the NFS when it says the HDI 1600 has better extension than the 3800.

I definitely think the NFS is wrong for the F328Be, but I wouldn't be surprised if it's wrong here a bit too.

Would be interesting to see JBL/Revel respond, though I doubt they will. Either they're lying/wrong, or the NFS is wrong. Both can't be right.

Also would be interesting to hear from Klippel and why they think their results are correct in this case. They confirmed your measurements, but how? Do they have ground plane measurements to compare against for this speaker? Or, are they simply confirming that you used the machine correctly? I'm sure you did use it correctly, but if the machine itself has bugs, then that's out of your control.

Frankly, I'm surprised there isn't more interest. Everyone wants to see floorstanders tested but we are circling the drain on the LF. LOL
But, it seems to be the case that when a concern is raised we as a forum tend to latch on to it and circle the subject and overall interest fizzles. I always hate seeing that "echo chamber" attitude in these threads because I feel good discussion is foregone after people grow tired of going over the same subject and lose interest. Not complaining. Just speaking the truth. And it's no different when it happens in one of my own reviews.

Again, sorry to bring this up. Not at all trying to throw shade at your review. Your reviews are the best there are atm, and you're the only youtube channel for which I have notifications turned on. I only bring this up in hopes of making Klippel aware of potential issues with the NFS that could potentially be improved upon.

 

[richard12511]

OTOH, both Harman and AH data above have their own issues. Both are heavily smoothed (not by choice in the case of AH as the response is gated to the low hundred hertz resolution).

I don’t believe James does vertical response measurements for tower speakers. So the AH data is missing that component. And thus no SPIN data.

So, there’s that.

Indeed, your measurements are much more detailed than the Harman official spin. You don't really see any of the midrange issues that your measurements clearly show. Makes me wonder if they have the more detailed measurement, but smoothed it before releasing it to the public. Or, does their measurement system just not have the resolution that the NFS does.

 

[amirm]

Indeed, your measurements are much more detailed than the Harman official spin. You don't really see any of the midrange issues that your measurements clearly show. Makes me wonder if they have the more detailed measurement, but smoothed it before releasing it to the public. Or, does their measurement system just not have the resolution that the NFS does.

It is weird as Harman research clearly stipulates the resolution we are showing and goes as far as saying if it is lower, then it hampers predicting preference. At the same time, the shown graphs are clearly low resolution.

I may have an idea about this. For a while I was working with Harman on writing technical articles based on their measurements. They would always send me these very low resolution graphs as if they used ancient PCs to capture the screen. So it is possible the underlying resolution is higher but the graphs we see are low resolution due to display limitations.

 

[amirm]

Would be interesting to see JBL/Revel respond, though I doubt they will.

At least three times I had communications with Harman to get the identical speaker they measured in anechoic chamber to compare with my results. They even asked me to pay for shipping both ways and I agreed. Still, there was no follow up at the end. I think we could have figured out these issues much earlier if they had cooperated. Maybe there are other reasons I don't understand but the whole thing has been incredibly strange.

 

[Beave]

I find it interesting that the JBL HDI series all seem to have a huge peak right around 20kHz, while the JBL 5xx series all seem to have a huge dip right around the same 20kHz.

Did they put a notch filter in the crossovers of the 5xx series to get rid of a big peak there? And if so, why did they decide to do it for those speakers but not for the HDI series speakers?

 

[Juhazi]

I find it interesting that the JBL HDI series all seem to have a huge peak right around 20kHz, while the JBL 5xx series all seem to have a huge dip right around the same 20kHz.

Did they put a notch filter in the crossovers of the 5xx series to get rid of a big peak there? And if so, why did they decide to do it for those speakers but not for the HDI series speakers?

This kind of sharp peaks and dips above 10kHz are typical for waveguide/horn loaded tweeters - and very much variable regarding fractions of mm difference in postioning of tweeter dome/waveguide edge, mic position etc. Fortunately listening tests can't hear these "issues", wiggles are spatially smoothed and are out of hearing passband for many adults.

 

[thewas]

may have an idea about this. For a while I was working with Harman on writing technical articles based on their measurements. They would always send me these very low resolution graphs as if they used ancient PCs to capture the screen. So it is possible the underlying resolution is higher but the graphs we see are low resolution due to display limitations.

The problem is they just have a 9600 baud acoustic coupler modem for outgoing connections which they all share, so they have to make 320x240 gifs for all images on their 486DX2 before uploading them to their bulletin board system.

At least three times I had communications with Harman to get the identical speaker they measured in anechoic chamber to compare with my results. They even asked me to pay for shipping both ways and I agreed. Still, there was no follow up at the end. I think we could have figured out these issues much earlier if they had cooperated. Maybe there are other reasons I don't understand but the whole thing has been incredibly strange.

Seriously now, it doesn't sound so strange to me any more, Harman has lately often shown that they (ab)use their "science" for their marketing, many good engineering people have left and some of their current generations seem to measure worse than the previous ones (for example Performa Be or 305 MKII) which made me personally from being a Harman fan and having almost a dozens of their products in my home to partially move on, like we say in here there exist also other families with nice daughters.

 

[ctrl]

Well, now that everyone has decided where they stand regarding the LF...

@hardisj, I can well understand that you are pissed. After all the trouble of doing a speaker review, some a-hole comes along and knows better - @amirm can tell you a thing or two about that

But it's really just about checking the NFS measurements for plausibility. Only if errors are detected, they can be corrected.
And the "correct" measured LF frequency response also has an effect on the evaluation/rating of the loudspeaker.

... is there anything else here that you guys feel is worth talking about?

Let's assume that the measurement of the HDI-3800 from Audioholics and the calculated LF frequency response from the near-field measurements are correct - by the way, your impedance measurement also speaks for this, since it shows a tuning frequency of just over 30Hz, so there should still be a relatively high sound pressure level in this frequency range.
Around 30Hz, there is a 9-11dB difference in sound pressure level compared to the NFS measurement.

As said, assuming the other frequency responses with the flat LF curve are correct, then JBL achieves this very flat LF frequency response due to the actually "too low" BR tuning.
For a bookshelf speaker with a small maximum woofer excursion, this would not be a good idea - for several woofers with a large displacement volume together, this is not a problem.

For the HDI-3800 with three 8'' woofers, you get a BR speaker that shows a second-order HP filter slope in the low frequencies, which is unusual for BR - a second order Butterworth HP filter @50Hz is shown as the pink target curve:

Thus, the HDI-3800 shows a LF response as usually only sealed loudspeakers do. This has advantages for the placement close to the front wall. Erin also mentioned in his review that placement near the wall is not a problem - the main reason is likely to be the chosen BR tuning.

If the frequency response in the low frequency range corresponds to that of a closed loudspeaker, then the group delay naturally turns out similarly good.
(This is only the simulated GD from the near field measurements)

 

[MrPeabody]

Erin, thanks for another excellent review! I know that it has to be frustrating to put all the time and energy into doing this, and then for so much of the comments to be focused on the uncertainty with respect to the deep bass measurements. I've read through most of the comments, and I thought that comments that @ctrl made were especially productive. This reinforces the suspicion that I've had for a while, that the NFS is not perfect in deep bass, especially when there are multiple sources of deep bass and the speaker is very large. I ask you to not think of this as something that is in any way a fault of yours or of the way you took the measurements. The onus should be on Klippel to figure out exactly what you and Amir and others need to do to get accurate, repeatable results with the system. Your measurements and Amir's measurements should be in full agreement not only with each other, but also with the ground-plane measurements that were taken independently, and also with the sort of synthesized response curve that ctrl produced. Klippel needs to step up and either state in a clear manner the inherent limitations of NFS with respect to deep bass and large speakers, or else publish very precise instructions for what needs to be done when taking these measurements, in order that the measurements will be repeatable by different people in different rooms and in agreement with the other established methods for getting accurate measurements of deep bass response. It's Klippel's problem, not yours.

I don't mean to come across as critical of the NFS. I'm amazed by what it does. Even if it happens that there is a real problem with deep bass measurements with large floor-standing speakers, then it remains true that NFS very greatly improves the accuracy of measurements starting somewhere below 100 Hz. As you pointed out, not even true anechoic measurements taken in anechoic chambers are accurate below a certain frequency. (The reason, I suspect, is that the absorbers are not effective at very low frequency.) NFS does appear to be comparable to, if not better than measurements taken in an anechoic chamber. I particularly like the idea that with splices done with NFS measurements, the splice can done at the middle of the response of one driver, and that there will be broad overlap between the two curves that are being spliced. This seems to me a very good circumstance, a leaps-and-bounds improvement over response curves generated by splicing close-mic measurements taken of multiple drivers independently. If it happens that the best way to obtain a full 20 Hz to 20 kHz response curve using NFS is by splicing the measurement with ground plane measurements, doing the splice at the middle of the woofer range and taking advantage of the wide overlap in the curves, I just do not see this as detracting from the decided benefits of NFS. Rather, I see this as something that is of tremendous value and that is made possible by NFS.

As for the speaker, I think you did a very good job of identifying its strong suit and its weaknesses, and this is exactly what a good test/analysis/review should ultimately do. The major weaknesses that are revealed with any speaker are very often a close consequence of the fundamental design approach. We see this with this speaker just as we have with many other speakers. In the horizontal plane the response is possibly a bit more directional than some people might prefer, but otherwise excellent. Directivity is well managed in the horizontal plane. But ... The 2.5-way approach was chosen in lieu of the 3-way approach that many people believe is inherently superior. The singular advantage of the 2.5-way approach, so far as I've been able to discern, is to permit the erstwhile midrange driver to contribute to the deep bass. But what are the tradeoffs? The upper woofer is large in diameter, such that its directivity is unusually narrow by the time frequency reaches the crossover point. (Less obviously perhaps, are there really any drivers capable of a reasonably flat response over a range spanning six octaves?) To deal with the narrow directivity of the upper woofer near the crossover frequency, the tweeter is placed in a large, deep waveguide. (It looks deep but the pictures might be deceiving.) This of course increases the vertical separation between the upper woofer and the tweeter, and the consequences of this are evident in the polar response in the vertical plane (and in the animated directivity balloons, which I really like by the way), and more importantly, in the observations you made concerning the soundstage when listening to some excellent Joe Walsh. Personally, I'd be happier with just two of those woofers, and for the upper one to be replaced with a midrange with diameter about half as great as the woofer, and for a much smaller waveguide to be used, such that the vertical separation between the tweeter and the driver below it will be less than half what it is with this speaker. For me personally, the design philosophy evident in this speaker just doesn't add up. It seems to me that the 2.5-way approach is potentially advantageous when, and only when, the woofers in the vertical stack are small enough in diameter that there is no reason to use such a large waveguide on the tweeter. And with speakers where the woofers in the vertical stack are not especially large, such that the 2.5-way approach makes better sense, it only makes sense to expect the bass response to be only a slight improvement over a mid-sized bookshelf speaker.

Please keep it up and please do not become discouraged by things that aren't really even under your control.

 

[Beave]

This kind of sharp peaks and dips above 10kHz are typical for waveguide/horn loaded tweeters - and very much variable regarding fractions of mm difference in postioning of tweeter dome/waveguide edge, mic position etc. Fortunately listening tests can't hear these "issues", wiggles are spatially smoothed and are out of hearing passband for many adults.

Interesting. I had assumed it was a driver characteristic, not a result of the horn loading. I'm far from an expert on horns/waveguides, however.

I agree that it's probably inaudible for most people. I know it would be for me, as I can't hear about 15kHz any more.

 

[rynberg]

Erin, your criticisms match the ones I formed auditioning the HDI-1600 in my home last month. It is funny, one of my first review notes was "resonance in Suzanne Vega's voice?"

Compared directly to a Revel M106, I couldn't get over how small the sound was and also the treble above 5kHz was 2-3 dB lower at the listening position. It was a more forgiving speaker but had less detail than the Revel for sure.

 

[Dave Zan]

Erin, this measurement indicates to me that the full response of the bass is not captured. Whenever you see that inversion in the sub-bass output, it indicates a problem. Did you measure this speaker vertically and if so, what was the mic distance?

I sympathise that Erin is tired of comments on the bass measurement issue but I think I understand this problem, maybe we can put it to bed once and for all.
The Near Field Scanner can remove the effect of room reflections because the sound field is scanned at two different radii and it is possible to separate waves outward from the speaker with reflections from the walls that travel inwards.
Unfortunately there is a complication with reflected waves that then reflect off the speaker itself.
Those speaker reflections mean that we essentially have 2 known measurements but 3 unknowns so we can't solve the equations.
Usually the speaker reflections are small and can be overlooked, the software assumes they are trivial.
When the speaker size is substantial compared to the mic distance this is not true and the calculations are inaccurate.
It's actually a bit more complicated than that, the waves don't really "reflect" off the speaker, it's more that the speaker disturbs the pressure field.
Or you could describe them as evanescent waves that decay exponentially, so extra mic distance has quite a effect.
It's not directly related to the maximum dimension of the speaker but rather how much of the mic's "view" is blocked by speaker panels.
So it should help to place a tall, slim speaker so the scan can be a cylinder around the main axis.
For the Klippel that would seem to mean vertically.

Just to emphasise that this is only what I think is the problem but I did work on this (painfully!) in the DIY Klippel Scanner thread in DIYaudio.
Plus it's consistent with @amirm's experience that increased mic distance is the solution.
Also with @ctrl 's calculations that look correct to me, and the ground plane measurements.
And finally, the shape of the sound-power curve in the first post just looks physically unlikely.
Note that the Klippel Fit Error does not detect this problem, it's a different issue, as Amir commented.
Perhaps this explains the Klippel feedback, they may have looked only at the Fit Error and not for other problems.
It a classic issue in a test review, not exactly laziness but subconsciously you just want it to be fine.

So, sorry if this is a bit of a downer to Erin but it's only to help improve the measurements.
Which is, after all, why we are on this website.
I did appreciate the review of the speaker, thanks for that.
And thanks for the chance to really think about how the maths plays out in practice*.

Best wishes
David

*Like in the credit card advert, that part is priceless.

 

[richard12511]

I sympathise that Erin is tired of comments on the bass measurement issue but I think I understand this problem, maybe we can put it to bed once and for all.
The Near Field Scanner can remove the effect of room reflections because the sound field is scanned at two different radii and it is possible to separate waves outward from the speaker with reflections from the walls that travel inwards.
Unfortunately there is a complication with reflected waves that then reflect off the speaker itself.
Those speaker reflections mean that we essentially have 2 known measurements but 3 unknowns so we can't solve the equations.
Usually the speaker reflections are small and can be overlooked, the software assumes they are trivial.
When the speaker size is substantial compared to the mic distance this is not true and the calculations are inaccurate.
It's actually a bit more complicated than that, the waves don't really "reflect" off the speaker, it's more that the speaker disturbs the pressure field.
Or you could describe them as evanescent waves that decay exponentially, so extra mic distance has quite a effect.
It's not directly related to the maximum dimension of the speaker but rather how much of the mic's "view" is blocked by speaker panels.
So it should help to place a tall, slim speaker so the scan can be a cylinder around the main axis.
For the Klippel that would seem to mean vertically.

Just to emphasise that this is only what I think is the problem but I did work on this (painfully!) in the DIY Klippel Scanner thread in DIYaudio.
Plus it's consistent with @amirm's experience that increased mic distance is the solution.
Also with @ctrl 's calculations that look correct to me, and the ground plane measurements.
And finally, the shape of the sound-power curve in the first post just looks physically unlikely.
Note that the Klippel Fit Error does not detect this problem, it's a different issue, as Amir commented.
Perhaps this explains the Klippel feedback, they may have looked only at the Fit Error and not for other problems.
It a classic issue in a test review, not exactly laziness but subconsciously you just want it to be fine.

So, sorry if this is a bit of a downer to Erin but it's only to help improve the measurements.
Which is, after all, why we are on this website.
I did appreciate the review of the speaker, thanks for that.
And thanks for the chance to really think about how the maths plays out in practice*.

Best wishes
David

*Like in the credit card advert, that part is priceless.

Interesting. So the fix is measuring a little bit further away? How much further? Also, does measuring far away introduce new issues?

@amirm, did you ever remeasure the F328Be with the mic further away? and if so, what was the -6dB point?

 

[Dave Zan]

...a little bit further away? How much further?... introduce new issues?...

Excellent question! It was the cause of some of my brain strain in the DIY Klippel thread.
It seems it should be possible to scan at three radii and solve the problem completely but I can't see how to do it and presumably neither have the Klippel team.
You can read their patent https://patents.google.com/patent/US20140198921 to see what they have been able to work out and that they are pretty smart.
So it's not an easy problem, @NTK has done nice numerical exploration of the scanner maths, maybe he has some ideas.

Nor could I work out a mathematical expression for the error introduced, so I can't say "keep it this far away to keep the error below this limit"
We want to have the speaker block only a small proportion of the mic's "view", exactly what "small proportion" is comes down to practical experience.
I planned to successively increase the distance until the results stabilised, I am very keen to learn from @amirm's experience what distance he found satisfactory.

One issue is that as we scan further away then the S/N decreases.
At what point this becomes a problem depends ultimately on the mic and electronics noise but probably, in practice, on how noisy the measurement environment is.
That obviously depends on the individual set up so it's hard to provide a recommendation except that quieter is better of course.
In any case the noise can be dealt with statistically, with more sweeps at the expense of more time.
So the main issue may be simply the physical constraints, the reach of the scanner and the size of the measurement space so you don't hit the floor or walls.
The Klippel hardware looks like it would be best to vertically position a tall speaker like the JBL.
My DIY proposal was for spherical scan hardware and part of the reason I paused work was uncertainty about the size.
This new data is a bit of a kick to move it forward.

Best wishes
David

 

[amirm]

@amirm, did you ever remeasure the F328Be with the mic further away?

I did and that was in the review. Prior to that I was measuring at closer distance and that produced quite wrong results. I have since squeezed a bit more distance using my rig so I can maybe improve a hair on what I post. Problem is the effort of lifting this massive speaker and spending time re-measuring it.

 

[MZKM]

Problem is the effort of lifting this massive speaker

Bribe your son