JBL 708i Monitor Review (Passive: Part 1)

[3ll3d00d]

Yes I mentioned that variable Q was used long ago, e.g. https://www.ranecommercial.com/legacy/note101.html (you can see this was originally written in the early 80s). I have never personally encountered such filters in any remotely modern DSP device either. It's not that obvious why storm even mention them, seems basically irrelevant to their point (which is about adjusting high frequency filters that come from other modern dsp implementations).

 

[jdubs]

I understand JBL don't actually come out and state this up front, but 708i is an DSP-equalized and limited, passive-crossover speaker. It has a passive two-way crossover inside that can be used either with a single amplifier plus DSP or two amplifiers plus DSP. It can be bi-amplified, but the only difference is the DSP EQ is now split between LF and HF, the amps are full-range and crossovers are all passive. That's how Amir is able to operate the speaker with a single amplifier, and he mentions the passive crossover in the review. Careful reading of JBLs literature confirms what I am writing to you.

In summary of what the speaker actually is: It's designed to use one or two amplifiers per customer's choice. It has passive crossovers only. It uses DSP for shelving and notch filtering only.

So that you don't misinterpret my words, here is a picture of the DSP filters directly from London Architect EQ files for HF and LF:
View attachment 254307

I think this doesn't account for the input filters or the compressor.

-Jim

 

[pierre]

Interesting pdf, I didn't know about differences between constant-Q and non constant-Q. I use parametric filters a lot for EQ'ing headphones & speakers using REW software to model the effects of the parametric filters on the measured frequency responses of the headphones & speakers, and that uses constant-Q from my understanding of your pdf. In fact, every piece of EQ software I've ever used stipulates Q in this same way - as in the filters are universal & interchangeable (give the same results) between all the different EQ software I've used. I'll list the different EQ software and "systems" I've used: REW / EqualiserAPO (& Peace) / Neutron Player / AutoEQ (github) / miniDSP - it's also the same definition of filters used by everyone in the headphone EQ arena - Oratory1990 / Crinacle / Resolve / Innerfidelity / jaakkopasanen(@AutoEQ). I've never come across any implementation that uses non-constant Q. (I think you made a mistake or typo in your first or second sentence, because you're referring to constant Q in both so there's a logical mismatch happening there regardless.) And I think you must be wrong in saying miniDSP / Roon / Pierre don't use constant Q filters.

EDIT: unless there's a 3rd type of Q that isn't described as constant-Q or non-constant-Q. Either way all the parametric filters I've ever used behave like the following pic in your pdf you linked (note within each graph below the Q value is the same between each of the different graphed lines (within a graph), and the only difference is the amount of Gain, which is where the misunderstanding could be happening with those who have glossed over the pdf):
View attachment 254748
and they've never behaved like the following which is non-constant Q (note that the pdf is saying the Q value of the following two graphed filters are the same, only the Gain is different):
View attachment 254749

EDIT #2: I noticed @pierre put a "Like" on your post, I think he didn't read the pdf in-depth as he's an experienced guy using EQ, so I think he just made an assumption that "constant-Q" meant a Q of 1.41 rather than a person being able to stipulate their Q-value, but that's not what the pdf is referring to in terms of what is meant by "constant-Q".

EDIT#3: There might be a third type called "proportional Q" (after I googled the topic), which might be what most EQ software uses, but that is certainly not the same as the "Figure 2" graph above, and instead looks more like the "Figure 3" graph above (when I model it in REW) which is what Storm Audio is saying is Constant-Q. I'm not sure that Storm Audio have covered off all the different types of Q. @oratory1990 , could you clear this up for us, any misunderstandings that are happening in my post & the one I replied to? (Meanwhile & also @GaryH, I know you like understanding & using EQ, have you come across any of these definitions? I was surprised I didn't know about them.)

I did read the paper. That's confusing.

So we have (at least):
- constant Q like in graphical EQ (with Q varying between equalizers either 1/3 octave or 2/3 octave bandwidth see below)
- constant Q like in Storm paper Q(storm) = Q * K with K = 2pi f_c/fs / sin(2pi fc / fs)
- proportional Q (what most things used nowaday)
- (band)width (and you can compute a corresponding proportional Q = sqrt( 2^W)/(2^W-1) ; that's the width you see in the EQ above from JBL)

there is still a slope parameter that I don't know what to do with. I am also not sure why it is easier to optimise with the storm Q that with a parametric one.

 

[Robbo99999]

I did read the paper. That's confusing.

So we have (at least):
- constant Q like in graphical EQ (with Q varying between equalizers either 1/3 octave or 2/3 octave bandwidth see below)
- constant Q like in Storm paper Q(storm) = Q * K with K = 2pi f_c/fs / sin(2pi fc / fs)
- proportional Q (what most things used nowaday)
- (band)width (and you can compute a corresponding proportional Q = sqrt( 2^W)/(2^W-1) ; that's the width you see in the EQ above from JBL)

there is still a slope parameter that I don't know what to do with. I am also not sure why it is easier to optimise with the storm Q that with a parametric one.

Well, it certainly is confusing, and the terminology doesn't help. The Storm pdf refer to Constant & Non-Constant Q, and that's nothing to do with whether the Q-value has to remain the same or not within your parametric filter (you can still choose any Q value you want for your filter), they're in fact saying the definition of Q is different when referring to Constant vs Non-Constant Q. The main takeaway for me being that Non-Constant Q means that for two filters that have the same Q value then the filter that has a larger Gain (negative or positive) will effect less frequency range either side of the central point, as per the following graph they used:

Now that kind of definition of Q I have never seen before (above), which makes me think their definition of Constant Q is the common version that we see, as per the following graph that shows a number of filters all with the same Q value but different Gain:

And the above graph is the visual behaviour I'm used to seeing when it comes to keeping the Q-value the same and increasing the Gain - so this leads me to believe most EQ implementations are of the Constant Q variety. Note this has nothing to do with the idea that the Q value that you input into the filter has to remain the same for each filter you use (it doesn't have to remain the same), as explained in the bolded part at the beginning of this post.


minor possibly unprobable EDIT: or the Storm pdf is a load of baloney and they don't even understand it and are confusing some simple principles which is leading us down the garden path.

EDIT #2: for what it's worth, I don't think this matters too much, as all the implementations of EQ that I've seen through software programs as well as devices like miniDSP all use the same type of the EQ that is translateable from one to another: ie for example a 30Hz, +6dB, Q1 filter will look the same between all the software & devices I've ever used.

 

[jhaider]

Has anyone converted this data for use with REW?

 

[Robbo99999]

Has anyone converted this data for use with REW?

There's the usual zip file that Amir attaches to all his speaker measurements that contains text files that contain the spinorama data. I think you can just import that into REW, then you'd be able to devise some of your own EQ filters and then use whatever "EQ applying" program or solution you want (EqualiserAPO, parametric EQ in music players, miniDSP, etc).

EDIT: you might need to remove some of the columns from text file. Importing a measurement into REW when I've done it is just two columns, frequency & the SPL(dB). I did this in the past by copying & pasting into a spreadsheet, then deleting the unwanted columns, then copy & pasting the remining two columns into a notepad text file editor (word processing program) and then saving it as a new text file - it now has just the two columns you need to import as a measurement into REW. There might be more time efficient ways of cracking the same egg.