FUNDAMENT - an open-source, fully-documented subwoofer platform (Build 001: Sealed 18″, BMS 18N862 + FA502)

[fivenp]

Some ideas I've been thinking over/working on that might be relevant:

I personally can't agree more with all of the items you mentioned..... Especially when it comes to cross tool compatibility - we're really missing some very basic standards and everyone seems to cook up "their own stuff". Would really love to see some movement here - and also support you in doing so! (Can we maybe train AI on answering some of those questions better?)

Even just establishing repeatable protocols to validate things (for all of us who don't have a Klippel at home) to be able to still contribute and test

I think the price awareness and alternative suggestion are VERY good points! I'd love to have a list of 5 drivers which all would basically work with some slight EQ twists and potential SPL or distortion limits / alternatives. Would be great for everyone to understand trade-offs / wins vs $ spent.

The question is just – where do you typically start? Do you start with the form factor? With the desired outcome? Do you have one build trying to do everything as good as possible?

My personal feeling is - One project typically starts with a desired outcome and available enclosure / driver size
E.g. in this case – an audiophile 18" for home use which is max 60x60x60cm targeted to music which has REAL grunt to chest slam - especially for more modern music.
This allows us to categorise within those boundaries and propose cheaper drivers, amps, enclosure variations, etc

If you have 300L of room and your target is Cinema LFE – this is not your project (but you could kick one off )
If you want to rather get PA like SPL while staying as audiophile as possible – this might also not 100% be your project

I think the question is – where does someone draw the line for a specific build where do you extend - or where do you branch off?

 

[mbd-00]

Can we maybe train AI on answering some of those questions better? -> To me, this whole 'Can an LLM design a speaker' is really just a tooling, data and context problem, not a fundamental limitation.

I'm going to be rolling out MCP support so we can hook directly into agents and explore how well this works in more detail - but it feels like a more natural starting point to your latter point e.g 'desired outcome and available enclosure / driver size' - define your constraints, and have AI brainstorm them for you, and spend your time on the final 10% of judgement calls, last-mile tweaks - rather than scouring marketplaces, loading in and configuring the cookie-cutter stuff each time. AI will pull from bits to get prices and search by specsheets, then call simulator to setup workspaces and run the simulations to optimise against the user defined constraints, then you can jump in and play around with the results and iterate further.

The question is just – where do you typically start? Do you start with the form factor? With the desired outcome? Do you have one build trying to do everything as good as possible? -> I'd say different for everyone! We could have a higher-order design as a collection of constraints and goals - the local design to achieve those goals is the one which the user can access via their own local constraints (price, availability of drivers etc) - that's just the enclosure, let alone the power/amp requirements, listening room etc.

Headroom is an interesting one too. Different drivers have different 'limits' - e.g xmax is not a standardised definition. PA users might be ok with some linearity loss/violations, and people will be playing back material with varying crest-factors. Power reporting is not too standardised either.

Even just establishing repeatable protocols to validate things -> Yes, and there's a sort of ladder of truth to bring a higher 'resolution' to a result's validity

Simulation:
- T/S based simulation
- 1D TL wave propagation simulation
- FEM/BEM simulation

Driver measurement
- Spec-sheet tracing
- FRD/ZMA measurements (free-air, or in-enclosure)
- Large signal parameters, klippel

Enclosure measurement:
- In-room, free-air etc?

Getting to 'perfect' is probably not the goal, but establishing repositories and standards which can be contributed to over time would be nice!

 

[fivenp]

To me, this whole 'Can an LLM design a speaker' is really just a tooling, data and context problem, not a fundamental limitation.

100% It would be great to capture that intent, context, and all the other things in a centralised place from a user point of view.
Even just being able to diff between "Experienced builder", "Newbie willing to learn", "I can't do anything but screw things in", "I really need something prebuilt" is already a big starting point....

Getting to 'perfect' is probably not the goal, but establishing repositories and standards which can be contributed to over time would be nice!

Can't agree more

 

[Flaesh]

This winter I outlined my thoughts on how to make proper sealed subwoofers here:

Double bass array. Low Q subs. [Quasi]plain wave radiation?

Where is your crossover set to the subs IIRC a little lower 90 Hz. Speakers delayed up to smooth PET. ringing is the same? Not completely. Initial decay improved, while tuning is unperfect, while decay before wasn't very bad. IMO\IME improving initial decay is audible in general case, while...

www.audiosciencereview.com

Some measurements are in the same thread. I'll add more when I have more free time.

 

[Luvchampagne]

Can we maybe train AI on answering some of those questions better? -> To me, this whole 'Can an LLM design a speaker' is really just a tooling, data and context problem, not a fundamental limitation.

I'm going to be rolling out MCP support so we can hook directly into agents and explore how well this works in more detail - but it feels like a more natural starting point to your latter point e.g 'desired outcome and available enclosure / driver size' - define your constraints, and have AI brainstorm them for you, and spend your time on the final 10% of judgement calls, last-mile tweaks - rather than scouring marketplaces, loading in and configuring the cookie-cutter stuff each time. AI will pull from bits to get prices and search by specsheets, then call simulator to setup workspaces and run the simulations to optimise against the user defined constraints, then you can jump in and play around with the results and iterate further.

The question is just – where do you typically start? Do you start with the form factor? With the desired outcome? Do you have one build trying to do everything as good as possible? -> I'd say different for everyone! We could have a higher-order design as a collection of constraints and goals - the local design to achieve those goals is the one which the user can access via their own local constraints (price, availability of drivers etc) - that's just the enclosure, let alone the power/amp requirements, listening room etc.

Headroom is an interesting one too. Different drivers have different 'limits' - e.g xmax is not a standardised definition. PA users might be ok with some linearity loss/violations, and people will be playing back material with varying crest-factors. Power reporting is not too standardised either.

Even just establishing repeatable protocols to validate things -> Yes, and there's a sort of ladder of truth to bring a higher 'resolution' to a result's validity

Simulation:
- T/S based simulation
- 1D TL wave propagation simulation
- FEM/BEM simulation

Driver measurement
- Spec-sheet tracing
- FRD/ZMA measurements (free-air, or in-enclosure)
- Large signal parameters, klippel

Enclosure measurement:
- In-room, free-air etc?

Getting to 'perfect' is probably not the goal, but establishing repositories and standards which can be contributed to over time would be nice!

Can we maybe train AI on answering some of those questions better? -> To me, this whole 'Can an LLM design a speaker' is really just a tooling, data and context problem, not a fundamental limitation.

Can't disagree with you more. There are specific posts related to posting AI garbage. As a new user you may want to read up on it. Personally I just instantly IGNORE any user that does it. Reading their posts almost universally are a waste of time. And in the event there might be a post or 2 that are relevant, well I'm fine with missing those.

 

[Blankenstejner]

Amazing to hear. I think the 10L should not really matter here

That's very interesting. I had them in a similar chain (Dirac, SHD, etc) in an OB at some point playing up to ~150Hz and while I felt they were a little weird at the beginning – I've never heard better upper bass after....

Would be curious to know what boost you had on the FA502?

I remember Troels is using the BMS up to 200hz in his legendary 4-way - maybe there's something to learn from there

The-Loudspeaker

Forgot to check this thread

I believe it's a shelv lowpass, though I can't remember the exact frequency and boost. I based it on the vituixCad simulation and with roomgain they extended down to about 10hz, without about 25-30hz @ -3dB. I haven't touched their filter setup since I set it up a few years ago.

In terms of the upperbass of the bms, I think my experience might partially be from bad dirac live measurement, but I remember that I still missed some punch with dirac off. My new speakers are diy with dual celestion FTR15-3070C in a horizontal orientation. The first temporary implementation are with only the celestion woofer box and a pair of dali kupid as standins for the future opus 12 horns, with a crossover of around 400hz. The dynamics of the FTR15-3070C are fantastic, but when I tried using a quick dirac setup on them the dynamics were severely muted. With a manual approximate eq and no dirac the dynamics were back.

 

[Blankenstejner]

With dirac and a fairly simple boost in the FA502 I had them playing down to 10Hz "flat" at my listening location (4.4mx4.4m room).

Curious at what spl level you reached 10hz flat and at what distortion levels?

I never measured that much detail, mostly because I live in an appartment complex

 

[Flaesh]

FTR15-3070C in a horizontal orientation

That seems not bad!)) What's above and (if subs used) below?

ps: I reread post. Which Opus 12 exactly and with what driver?

 

[Blankenstejner]

That seems not bad!)) What's above and (if subs used) below?

ps: I reread post. Which Opus 12 exactly and with what driver?

I will be using my BMS 18n862 subs again when I get my hands on either a minidsp tide16 or flex 8.

The compression driver I picked on impulse is the Oberton ND100 1.4, for use with the curved front Opus 12 1.4 horn: https://at-horns.eu/opus12.html.

Btw in order to keep this thread on topic I will not write more about my speakers here. A dm is always welcome.

 

[mbd-00]

Hey @fivenp - inspired by this thread to re-vamp the share side of the simulator. You can now edit your share design directly to link to discussion threads, github repos etc.

 

[fivenp]

Hey @fivenp - inspired by this thread to re-vamp the share side of the simulator. You can now edit your share design directly to link to discussion threads, github repos etc.

View attachment 541312

Love it! Thank you

Thinking a little bit ahead - I guess it would be really crazy to be able to collaborate on a design / simulation in a centralized place