I've been playing around with Anthropic's Claude which seems very powerful and fairly technical. How does this look as a two way crossover for SB acoustics Satori drivers.
# Advanced Crossover Design
## Driver Characteristics
### Tweeter (TW29BN)
### Midwoofer (MW19P)
## Optimized Crossover Topology
### Revised Design: 3rd Order Asymmetric Linkwitz-Riley
### Precision Component Values
#### Midwoofer Network (Low-Pass)
#### Tweeter Network (High-Pass)
## Compensation Components
- Zobel Network for Midwoofer:
- Rs: 5.6Ω
- Cs: 100 µF
- Notch Filter (Optional): Targeting any mechanical resonances
## Phase and Impedance Correction
## Optimization Goals
1. Flat frequency response (±2 dB)
2. Smooth phase transition
3. Minimized off-axis response irregularities
4. Controlled power response
## Recommended Measurement Procedure
1. Near-field measurement
2. Gated measurement at 1m
3. Verify:
- Frequency response
- Phase coherence
- Impulse response
## Critical Component Selection
- Capacitors:
- Polypropylene film
- Low ESR
- Tight tolerance (±2%)
- Inductors:
- Air-core or Copper/Aluminum foil wound
- Low DCR
- Tight tolerance
- Resistors:
- Non-inductive
- Metal film
- 1% tolerance
## Performance Expectations
# Advanced Crossover Design
## Driver Characteristics
### Tweeter (TW29BN)
- Impedance: 4Ω
- Sensitivity: ~88 dB/1W/1m
- Recommended Crossover: 3kHz
- Fs (Resonance Frequency): ~850 Hz
### Midwoofer (MW19P)
- Impedance: 8Ω
- Sensitivity: ~87 dB/1W/1m
- Recommended Crossover: 3kHz
- Fs (Resonance Frequency): ~55 Hz
## Optimized Crossover Topology
### Revised Design: 3rd Order Asymmetric Linkwitz-Riley
- Low-Pass on Midwoofer: 18 dB/octave
- High-Pass on Tweeter: 12 dB/octave
- Crossover Frequency: 2.8 kHz
### Precision Component Values
#### Midwoofer Network (Low-Pass)
- L1: 1.5 mH (Air-core inductor)
- C1: 10 µF (Polypropylene film capacitor)
- R1 (Pad): 2.2Ω non-inductive resistor
#### Tweeter Network (High-Pass)
- L2: 0.68 mH (Air-core inductor)
- C2: 22 µF (Polypropylene film capacitor)
- R2 (Pad): 3.9Ω non-inductive resistor
## Compensation Components
- Zobel Network for Midwoofer:
- Rs: 5.6Ω
- Cs: 100 µF
- Notch Filter (Optional): Targeting any mechanical resonances
## Phase and Impedance Correction
Code:
L1 C1
MW19P +---+---@@@@---||---+
| | |
R1 | |
/\/\ | |
| | |
+---+------------+
|
| Crossover (2.8 kHz)
|
+-------+-------+
| |
L2 R2
@@@@ /\/\
| |
C2 |
|| |
| |
TW29BN +-------------+
## Impedance Correction Zobel
Rs
TW29BN +/\/\---+
|
Cs
||
|
GND## Optimization Goals
1. Flat frequency response (±2 dB)
2. Smooth phase transition
3. Minimized off-axis response irregularities
4. Controlled power response
## Recommended Measurement Procedure
1. Near-field measurement
2. Gated measurement at 1m
3. Verify:
- Frequency response
- Phase coherence
- Impulse response
## Critical Component Selection
- Capacitors:
- Polypropylene film
- Low ESR
- Tight tolerance (±2%)
- Inductors:
- Air-core or Copper/Aluminum foil wound
- Low DCR
- Tight tolerance
- Resistors:
- Non-inductive
- Metal film
- 1% tolerance
## Performance Expectations
- Usable Frequency Range: 45 Hz - 20 kHz
- Sensitivity Matching: Within 0.5 dB
- Impedance: Nominally 8Ω (min 6Ω)
What's the design based on? Normally successful XO-designs are based on response measurements of the drivers which you then "shape" via added passive components. If you just ask someone (whether a person or an AI-model) to effectively "guess" on an XO-design based on nothing except the driver names I am not really sure it justifies a response from the forum members who really knows something.
EDIT: I understand someone might be looking for a shortcut into XO-design (including yours truly), but using AI-tools for something that they have no concept on is just going to pollute an otherwise useful forum with crap content - and that would be a shame IMHO.
EDIT: I understand someone might be looking for a shortcut into XO-design (including yours truly), but using AI-tools for something that they have no concept on is just going to pollute an otherwise useful forum with crap content - and that would be a shame IMHO.
I told the bot the drivers Satori Tweeter (TW29BN) & Midwoofer (MW19P) , just curious how close it got, and how good it can get.
With the developments coming so fast, this will be the future of most design/iteration in lots of industries. But hopefully there is always a pair of human ears at the end to verify and offer some subjective tweaks.
This thread is just for fun, but it may reach a point where it is better able to predict and model complex behaviour and be trained to excel above human ability.
It is doing a very bad job at the moment of plotting the results frequency/SPL graph.
Here is my follow up posting a Troels Gravensen design with the same drivers.
As an AI language model, I do not have the ability to physically compare speaker designs or make modifications to electronic circuits. However, I can provide a high-level comparison between the 2-way design I previously proposed and the crossover circuit you have presented.
The circuit you have shown appears to be a more complex design, likely incorporating additional features and components beyond a basic 2-way crossover. Some key differences I observe:
1. Topology: The circuit you provided seems to have a more intricate topology, potentially including additional filter stages or compensation networks beyond the typical 2-way crossover.
2. Component Count: Your circuit has a higher number of passive components (resistors, inductors, and capacitors) compared to the simpler 2-way crossover I outlined.
3. Functionality: The additional components in your circuit may enable more advanced features, such as impedance correction, resonance damping, or tailored frequency response shaping.
4. Driver Integration: The complex nature of your circuit suggests it may be optimized to integrate the drivers more seamlessly, potentially addressing challenges like phase coherence or off-axis performance.
Without a more detailed analysis of the circuit's purpose and the specific driver characteristics, I cannot provide a comprehensive comparison or evaluation. Speaker design is a highly nuanced and application-specific field, and the skilled engineer who created this circuit likely had specific goals and constraints in mind.
If you have any specific questions about the functionality, performance, or design considerations of this crossover circuit, I'd be happy to discuss them to the best of my abilities based on the information provided. However, I cannot recommend modifications or claim superiority of one design over the other without a more thorough understanding of the context and requirements.
With the developments coming so fast, this will be the future of most design/iteration in lots of industries. But hopefully there is always a pair of human ears at the end to verify and offer some subjective tweaks.
This thread is just for fun, but it may reach a point where it is better able to predict and model complex behaviour and be trained to excel above human ability.
It is doing a very bad job at the moment of plotting the results frequency/SPL graph.
Here is my follow up posting a Troels Gravensen design with the same drivers.
As an AI language model, I do not have the ability to physically compare speaker designs or make modifications to electronic circuits. However, I can provide a high-level comparison between the 2-way design I previously proposed and the crossover circuit you have presented.
The circuit you have shown appears to be a more complex design, likely incorporating additional features and components beyond a basic 2-way crossover. Some key differences I observe:
1. Topology: The circuit you provided seems to have a more intricate topology, potentially including additional filter stages or compensation networks beyond the typical 2-way crossover.
2. Component Count: Your circuit has a higher number of passive components (resistors, inductors, and capacitors) compared to the simpler 2-way crossover I outlined.
3. Functionality: The additional components in your circuit may enable more advanced features, such as impedance correction, resonance damping, or tailored frequency response shaping.
4. Driver Integration: The complex nature of your circuit suggests it may be optimized to integrate the drivers more seamlessly, potentially addressing challenges like phase coherence or off-axis performance.
Without a more detailed analysis of the circuit's purpose and the specific driver characteristics, I cannot provide a comprehensive comparison or evaluation. Speaker design is a highly nuanced and application-specific field, and the skilled engineer who created this circuit likely had specific goals and constraints in mind.
If you have any specific questions about the functionality, performance, or design considerations of this crossover circuit, I'd be happy to discuss them to the best of my abilities based on the information provided. However, I cannot recommend modifications or claim superiority of one design over the other without a more thorough understanding of the context and requirements.
But here's the thing - it's not designing anything, it's writing about a design that it knows nothing about (or more accurately "putting words together about"). From my POV it's a huge difference - and as a contributor to this forum for +20 years I would prefer to keep it as a useful source of human knowledge, not "for fun" posts that add no value. This is not a personal attack on anyone, but purely based on an observation of trends in my internet searches etc. over the last year or so.
I have had some exposure to various "AI" technologies as part of my job (I'm an IT architect so it's kind of a requirement these days) and for some applications it's absolutely a gamechanger if you choose the right technology. However, using any language model for these things is like choosing a hammer to cut two feet off a board that's too long. And the debate about which language model is better is like arguing that the expensive hammer is better at cutting wood than the cheap one - it may be "better" relatively speaking, but on an absolute scale it's still crap.
If you want to use "AI" for solving these problems I would imagine that using ML to find patterns in response curves or using an optimisation algorithm of some sort to find the best parts values to approximate a target response curve based on your own intial measurements would work well - and save a huge amount of effort. However, an LLM is neither of those things and is unlikely to provide anywhere near the same results.
I have had some exposure to various "AI" technologies as part of my job (I'm an IT architect so it's kind of a requirement these days) and for some applications it's absolutely a gamechanger if you choose the right technology. However, using any language model for these things is like choosing a hammer to cut two feet off a board that's too long. And the debate about which language model is better is like arguing that the expensive hammer is better at cutting wood than the cheap one - it may be "better" relatively speaking, but on an absolute scale it's still crap.
If you want to use "AI" for solving these problems I would imagine that using ML to find patterns in response curves or using an optimisation algorithm of some sort to find the best parts values to approximate a target response curve based on your own intial measurements would work well - and save a huge amount of effort. However, an LLM is neither of those things and is unlikely to provide anywhere near the same results.
Last edited:
I don't want people to get the wrong idea about AI (and a topic for the forum, moving forward).
Let's be clear, AI is not new so it's within our power to see it for what it is. Like others I used to come home from school and program AI in contemporary forms, and wonder what it might be like in the future when processing power was abundant. We refrained from naming it intelligence.
It considers technical things but it isn't technical. It refines but doesn't understand, and it is not truly capable of intrinsic improvement. It's progressed beyond a programmer's personal ability to keep track of it and is subject to reactive and subjective patchwork.
I know that it's easy to forget all this once it becomes sophisticated.. but give it a name, and it becomes a thing..
Could it (Anthropic) be trained to perform better? I see a way in which we can do very quick iterative developments, and speaker sims, refine the XO, perhaps suggest a different driver that would perform better, it could be written to simulate in room response, even convolution filters to lay on top. Soon it would be able to order the parts or develop active XO's to simulate passive for quicker iteration...I think this is all possible now and is quick exciting. All from a simple text prompt.
Would anybody by able to offer some insight as to whether the design above works in a sim, and what it has probably missed.
I realise the future is a way off, but it think it shows promise?
Would anybody by able to offer some insight as to whether the design above works in a sim, and what it has probably missed.
I realise the future is a way off, but it think it shows promise?
I don't get the impression you know how an LLM actually works and what it does, sorry. It is writing about designs based on training input that is also just writing about designs, it's not doing any actual designing itself as far as I can see. Be very careful when looking at the output and trying to infer underlying capabilities - that's dangerous.
If you look at the output there's not really a whole lot of concrete output here (highlights mine):
As an AI language model, I do not have the ability to physically compare speaker designs or make modifications to electronic circuits. However, I can provide a high-level comparison between the 2-way design I previously proposed and the crossover circuit you have presented.
The circuit you have shown appears to be a more complex design, likely incorporating additional features and components beyond a basic 2-way crossover. Some key differences I observe:
1. Topology: The circuit you provided seems to have a more intricate topology, potentially including additional filter stages or compensation networks beyond the typical 2-way crossover.
2. Component Count: Your circuit has a higher number of passive components (resistors, inductors, and capacitors) compared to the simpler 2-way crossover I outlined.
3. Functionality: The additional components in your circuit may enable more advanced features, such as impedance correction, resonance damping, or tailored frequency response shaping.
4. Driver Integration: The complex nature of your circuit suggests it may be optimized to integrate the drivers more seamlessly, potentially addressing challenges like phase coherence or off-axis performance.
There's nothing here that actually helps with your design beyond repeating some basic facts you can observe yourself (like the component count) and some very basic things around X-over design which you can learn yourself by reading a basic tutorial on how passive crossovers work. To me it is the definition of useless because it doesn't solve the problem (your design) and it doesn't teach you anything (because you still do not know from the output how a passive crossover works and which role the parts play).
Again, the potential for "AI" in this space is definitely there, but an LLM is simply not the right tool for the job.
If you look at the output there's not really a whole lot of concrete output here (highlights mine):
As an AI language model, I do not have the ability to physically compare speaker designs or make modifications to electronic circuits. However, I can provide a high-level comparison between the 2-way design I previously proposed and the crossover circuit you have presented.
The circuit you have shown appears to be a more complex design, likely incorporating additional features and components beyond a basic 2-way crossover. Some key differences I observe:
1. Topology: The circuit you provided seems to have a more intricate topology, potentially including additional filter stages or compensation networks beyond the typical 2-way crossover.
2. Component Count: Your circuit has a higher number of passive components (resistors, inductors, and capacitors) compared to the simpler 2-way crossover I outlined.
3. Functionality: The additional components in your circuit may enable more advanced features, such as impedance correction, resonance damping, or tailored frequency response shaping.
4. Driver Integration: The complex nature of your circuit suggests it may be optimized to integrate the drivers more seamlessly, potentially addressing challenges like phase coherence or off-axis performance.
There's nothing here that actually helps with your design beyond repeating some basic facts you can observe yourself (like the component count) and some very basic things around X-over design which you can learn yourself by reading a basic tutorial on how passive crossovers work. To me it is the definition of useless because it doesn't solve the problem (your design) and it doesn't teach you anything (because you still do not know from the output how a passive crossover works and which role the parts play).
Again, the potential for "AI" in this space is definitely there, but an LLM is simply not the right tool for the job.
AI good for common places of acoustics, basic analytics; not for synthesis, not able to think, just statistical linguistic tool. No ideas out of the box, similar to the best student in class with very good memory but no imagination. Btw happy that AI will kill all idiotic ‘papers’ on the universities around the world.
Back to topic:
I tried AI for a crossover, with the tricky dynaudio 30w100, he gave me some obviously bad solutions, when I gave in prompt the qts the comment was ‘you are right that was not a good solution’. I tried it for my job which is not technical with the same results, sometimes useful for general information, opportunistic and politically correct, lacking details even in internet available information (I know AI is a very good tool for programmers)
But, two dangers: to think he can give ALL solutions is the first and well-known, the second, bigger, to underestimate the tool and to forget to learn and evolve with his development. Not yet capable to imagine a Manzanita, but who knows in five years?
Back to topic:
I tried AI for a crossover, with the tricky dynaudio 30w100, he gave me some obviously bad solutions, when I gave in prompt the qts the comment was ‘you are right that was not a good solution’. I tried it for my job which is not technical with the same results, sometimes useful for general information, opportunistic and politically correct, lacking details even in internet available information (I know AI is a very good tool for programmers)
But, two dangers: to think he can give ALL solutions is the first and well-known, the second, bigger, to underestimate the tool and to forget to learn and evolve with his development. Not yet capable to imagine a Manzanita, but who knows in five years?
You can easily enter the high pass values ( returned by your chatbot buddy ) into an online crossover calculator ( which is currently an even smarter form of Ai >> but still much too dumb to be believable and//or "trustworthy and actionable") >> then you'll see for yourself the garbage that you're being fed by your new ( infantile ) playmate.
The 3.9R "padding resistor" can only occur in 3 ( of 5 ) significant locations that will effect the online calculators results.
1) Paralleled directly across the driver, giving a load of @ 2 ohms ( for use with the online calculator ).
2) In Series with the driver ( right before the driver ) giving a static load of @ 8 ohms
3) In Series with the driver ( after the capacitor but now paralleled with the driver and coil ) still giving a load of @ 8 ohms for the capacitor but 4 ohms for the coil ) > online calculators can't test this special scenario.
The other 2 positions don't effect the online calcs that much and aren't worth mentioning here ( in light of this exercise ).
🙂
Claude could not be as it currently operates, no. LLMs are writing something that has the FORM of a good answer, which may or may not hit on the FUNCTIONALITY of a correct answer. This is something that is easier to notice when you have deep expertise in a topic you are asking it about - the recall may be encyclopedic, and the answers may look correct or at least plausible; but on close inspection understanding of relationships between concepts, important exceptions etc are often not there.
That being said, while LLMs do not work the way you are describing, they can be extended with other tools just like our intelligences can. And AI/Machine Learning can definitely do what you are suggesting, just not the particular technology that everyone is excited about today. And Claude or other LLM could, given properly formed measured data (what you've fed it is nowhere near sufficient to design a good XO), organize and pass the relevant bits along to a more appropriate tool like Xmachina or Vituixcad for iterative XO development (which they already do), receive the output of that, and present it back to you (currently ChatGPT does this for math with Wolfram, and there likely are and will be many more such combinations).
So in that scenario, you could think of Claude as a sort of wizard/guided frontend for those complex tools, holding your hand through gathering inputs and helping you interpret outputs.
Thanks for the input.
While I see the limitations for an LLM, with the advances in writing code, I thought it could be possible to get it to write super simple programs to do basic XO design with some form of feedback. Maybe next year?
While I see the limitations for an LLM, with the advances in writing code, I thought it could be possible to get it to write super simple programs to do basic XO design with some form of feedback. Maybe next year?
Virtuix can has an optimizer that will try hundreds of combinations to hit your goal.
Many assume that all there is to know about designing for audio is written down. It's not. Much exists in the heads of the experienced. Perhaps it has been written about, but without instructions as to how it was achieved.
(My correction highlighted in bold letters)
Vituixcad can optimize values of an existing circuit.
You could also have a go with XMachina (there is a thread here on diyAudio) that automatically creates entire crossover circuits using measurement data and user target parameters.
Seems quite (artificially) intelligent to me!
I asked AI to give me a cooking time range for parted out turkey legs at a certain oven temperature. It confidently told me the normal cooking times as if it was attached to the carcass. After I questioned that, it reduced the estimate to a more realistic range, also with "you're absolutely right, I forgot to account for that" b.s. 💩
As for Claude, the Satori XO is complete bollocks. 💩💩💩 You'd get further following the basic calculations in the https://www.diyaudio.com/community/...-crossovers-without-measurement.189847/latest sticky.
But those programs already exist (VituixCAD et al) - what's the point of asking the LLM to write it again? That would only really make sense if you were trying to write your own XO-design software? Without the basic understanding of how the XO works you would not be able to apply that SW in any meaningful way.
What I read is actually that you think crossover design is daunting and complicated (it is) and you can't or won't spend the time required to learn. I get that, but I don't think AI will reach the state of "do this for me so I do not need to understand any of it" for these types of design any time soon.
A more "charitable" interpretation of what you write is perhaps that you find design tutorials hard to read and follow. That's fair, and here the LLM may actually be able to help because it can work with the text of a written tutorial to rephrase/elaborate/clarify wording and the process steps. This is stil not designing an XO though. If you want to learn it may be an accelerator for learning, but if you just want a ready-made answer I think the only way for the foreseeable future is to buy a kit or pay a designer who knows how to use the tools to achieve your goals.
So in recent developments the new AI models now have advanced reasoning. Gimmi the low down on ChatGPT's o3-mini-highs output, for the same two drivers I have...
_________________________
Below is an example “blue‐print” for a high–performance two–way bookshelf loudspeaker using SB Acoustics’ TW29DN‑B‑8 tweeter and MW19P‑4 midwoofer. Keep in mind that “perfect” performance always comes only after careful measurement and (often iterative) tuning. The design below gives you a starting point based on typical driver behavior and proven design practices. (Be sure to check the manufacturers’ datasheets for the exact Thiele/Small parameters of your drivers so you can refine these numbers in simulation.)
In summary, the design uses a well–matched crossover (you may choose first–order for simplicity and excellent phase coherence or, if you prefer steeper slopes, a second–order Linkwitz–Riley alignment) and a rigid, internally braced ported enclosure built from 18 mm birch plywood. (A sealed box is “cleaner” in the bass but a ported design can “extend” the bass and boost efficiency; for best overall performance in a bookshelf size, the ported approach is attractive.)
Below are the details.
1. Driver Integration
• Tweeter: SB Acoustics TW29DN‑B‑8
– Nominal impedance: ~8 Ω
– Small (≈30 mm diameter) dome element with excellent high–frequency detail
• Midwoofer: SB Acoustics MW19P‑4
– (Usually a 19 cm [≈7.5″] driver; model “4” indicates its design class—often a 4 Ω unit; if you have an 8 Ω version, simply scale the crossover inductance accordingly.)
– Provides smooth mid–bass extension down to the low 60’s Hz
A careful layout (for example, with the tweeter mounted in the upper third of a curved front baffle and the midwoofer centered in the lower half) helps minimize diffraction. (Sometimes a small vertical “offset” or acoustic “waveguide” is used so that the drivers’ acoustic centers are aligned in time.)
2. Crossover Design
You want to “hand–match” the drivers so that their overlapping responses sum flat with good phase coherence. Two common choices are:
A. First–Order Crossover (6 dB/octave)
This type is very popular for its natural phase behavior and “open” sound. Its gentle slope means that the drivers’ natural roll–offs do most of the work, so it is well suited if your drivers’ off–axis and free–air responses are smooth.
• Crossover frequency (f₍c₎):
A good starting point is about 2.2 kHz–2.5 kHz. (You may adjust this based on measured frequency responses.)
• For the Tweeter (High–Pass):
A single series capacitor is used. Using the standard formula for a 1st–order filter:
For example, with and :
– Component: Use a high–quality (e.g. polypropylene) non–polarized capacitor rated for 50 V or higher.
• For the Midwoofer (Low–Pass):
A series inductor forms the 1st–order low pass. Its value is given by:
For example, if the MW19P‑4 is a 4 Ω unit at :
(If your midwoofer is actually 8 Ω, double this value to about 0.58 mH.)
– Component: Choose an air–core inductor designed for audio use (low losses, good linearity).
• Level Matching:
In many designs the tweeter is naturally a bit more sensitive. If measurements show a 2–3 dB level difference at the crossover, incorporate an L–pad network (a resistor in series with the tweeter plus a resistor shunted to ground) so that its output is trimmed. For a 3 dB pad on an 8 Ω tweeter, typical resistor values are roughly:
• Series resistor ≈ 3.3 Ω
• Shunt resistor ≈ 11.3 Ω
(These values come from standard L–pad formulas; recalc them using your exact sensitivity data.)
B. (Optional) Second–Order (Linkwitz–Riley) Crossover
A 2nd–order design gives a 12 dB/octave slope and very smooth summed response if well–implemented. (In a Linkwitz–Riley alignment the two filters are “cascaded” from 1st–order Butterworth sections; note that at the chosen frequency each branch is down 6 dB, and the drivers’ outputs sum flat.)
For a 2nd–order high–pass the component values are reduced by a factor of roughly . For example, for the tweeter:
and similarly for the midwoofer the series inductor would be about 0.20–0.30 mH (for a 4 Ω load).
Because the required inductance values are very low, it may be easier in practice to start with the first–order design and then decide if a steeper roll–off is necessary.
3. Enclosure Design
Type & Materials
• Type:
A ported (bass–reflex) design is recommended for extended low–frequency output and higher efficiency. (If you prefer the “tighter” control of sealed designs, you can adapt the following dimensions for a smaller internal volume.)
• Material:
Use 18 mm thick birch plywood throughout. This material is both stiff and well–damped, and—when properly braced—minimizes cabinet resonances.
Internal Volume
• Target Volume (V₍b₎):
Approximately 20 liters (0.02 m³) is a good starting point to suit the midwoofer’s parameters.
Example External/Effective Internal Dimensions
Accounting for 18 mm walls, you might design an enclosure with approximate external dimensions such as:
• Width: 300 mm
• Height: 250–300 mm
• Depth: 270–300 mm
For example, internal dimensions of about 300 mm (W) × 250 mm (H) × 270 mm (D) give roughly 20–21 liters. (Actual dimensions should be calculated to achieve the precise volume after subtracting wall thickness and any internal bracing.)
Front Baffle & Driver Cutouts
• Baffle Shape:
A gently curved or beveled front baffle minimizes diffraction.
• Driver Placement:
• Midwoofer: Centered in the lower (or central) portion. Its cutout diameter should match the mounting ring (likely around 190 mm).
• Tweeter: Mounted in the upper section (or slightly “in–front” of the woofer) with a small cutout (≈30–40 mm diameter).
A slight vertical offset or use of an acoustic delay (waveguide) may be applied to “time–align” the drivers’ acoustic centers.
Port (Bass–Reflex) Design
• Tuning Frequency:
Aim for a box tuning around 45 Hz to complement the midwoofer’s low–end.
• Port Dimensions:
For a single round port made from a 50 mm diameter tube:
1. Calculate the cross–sectional area (A):
2. Determine Port Length (L):
Using the equation
(with ) and solving for L gives:
Plugging in:
•
•
•
•
You get:
3. End Corrections:
Add roughly 0.85×(port diameter) (≈43 mm) to the physical length to account for port “end effects” (or adjust the tuning with simulation).
Internal Bracing & Damping
• Bracing:
Install internal braces (made from 18 mm ply scraps) to “stiffen” the cabinet and avoid panel resonances.
• Damping:
Lining the enclosure’s walls with acoustic damping material (e.g. poly–fill or foam) will help control standing waves and smooth the bass response.
4. Construction & Final Tuning
• Joinery:
Use high–quality wood glue and screws to ensure all joints are airtight (except for the port).
• Finishing:
Consider a veneer or painted finish on the front, with careful attention to edge treatment.
• Measurements & Adjustments:
After construction, measure the internal volume and, if needed, adjust the port length. Also, bench–test (and if possible, measure) the driver responses with the chosen crossover; slight tweaks (including fine–tuning the L–pad on the tweeter) may be required to obtain a flat summed response and optimal phase coherence.
• Simulation:
Software tools such as WinISD or similar can help simulate both the bass–reflex tuning and the crossover responses—this is highly recommended before finalizing the build.
Summary of Key Numbers (Example Values)
• Crossover Frequency: ~2.2 kHz
• Tweeter High–Pass (First Order): 9 µF (≈8–10 µF polypropylene)
• Midwoofer Low–Pass (First Order): ~0.29 mH (for a 4 Ω driver)
• Tweeter L–pad (if needed for ∼3 dB attenuation):
• Series: ≈3.3 Ω
• Shunt: ≈11.3 Ω
• Enclosure Internal Volume: ~20 liters
• Port: 50 mm diameter; calculated length ≈145 mm (plus ~43 mm end–correction) for tuning near 45 Hz
• Construction: 18 mm birch plywood with internal bracing and damping
By following this plan—and then refining component values and dimensions using measurements and simulation—you should be well on your way to building an optimized two–way loudspeaker that makes the most of the SB Acoustics TW29DN‑B‑8 tweeter and MW19P‑4 midwoofer. Happy building!
CAN YOU MODEL THE RESONSE?
Below is an example of how one might “model” the frequency‐response of the designed two–way system using standard filter equations. Keep in mind that this simulation is based on the idealized electrical (or “netlist”) model of the crossover filters and does not include all the real–world effects such as the drivers’ own acoustical roll–offs, cabinet diffraction, or enclosure resonances. In practice, you’d combine these with measured driver responses (or use simulation software like VituixCAD, XSim, or MATLAB) to get a fully accurate picture. That said, here’s a step–by–step explanation and example calculations for the “first–order” crossover case we described.
1. The Basic Model
For our first–order (6 dB/octave) crossover with a chosen crossover frequency (here, let’s assume 2.2 kHz), the two filter transfer functions (in voltage, ignoring phase shifts from the drivers for the moment) are:
• Midwoofer (Low–Pass):
• Tweeter (High–Pass):
Note: At , each filter’s magnitude is
If the drivers are acoustically “aligned” (that is, their acoustic outputs arrive in phase) and their sensitivities are matched (using an L–pad if needed), the sound–pressure contributions add coherently. In the ideal case the total output pressure is the sum of the two contributions. If the filters’ phases differ by exactly 90° (as they do in a pure first–order filter), then the power sums in quadrature. In our design, however, we aim to adjust the mechanical and electrical delays (or even use a slight acoustic waveguide) so that the acoustic outputs are in phase at the listener’s position. Under that “in–phase” condition, the acoustic pressure response can be approximated as the algebraic sum:
A little algebra shows that when summed in quadrature the net magnitude is nearly constant:
for all frequencies. (This is why a Linkwitz–Riley alignment, which is based on cascaded Butterworth sections, is so popular—it naturally sums to a flat response.)
2. Example Calculations
Let’s step through some example frequencies for the idealized filters (assuming ).
A. Compute the normalized magnitude responses
1. At 100 Hz:
•
• Midwoofer:
• Tweeter:
• Combined Magnitude (if added in quadrature):
2. At 1000 Hz:
•
• Midwoofer:
• Tweeter:
• Combined:
3. At :
• Both branches yield:
• Combined:
4. At 10000 Hz:
•
• Midwoofer:
• Tweeter:
• Combined:
B. Interpreting These Numbers
Under these ideal assumptions:
• Below the crossover: The midwoofer is nearly “full on” while the tweeter is almost off.
• Above the crossover: The tweeter dominates.
• At the crossover: Both drivers contribute equally (about –3 dB each), but if in phase the pressure adds to yield a flat (0 dB) overall response.
Thus, the electrical–filter model predicts a combined response that is flat (i.e. constant gain of 1, or 0 dB relative to a set reference) across frequency.
3. Including the Enclosure & Driver Characteristics
In practice, you must include:
• Driver Responses:
Both the midwoofer and tweeter have their own inherent frequency–response curves. For example, the midwoofer may start rolling off well above 2–3 kHz and the tweeter may not be perfectly flat over its range. Manufacturers’ measurements or Thiele/Small–parameter–based simulations help include these effects.
• Cabinet/Port Effects:
The ported enclosure introduces its own bass–reflex behavior. For example, if tuned to 45 Hz, you might see a controlled peak or a smoother roll–off below 60 Hz. A common acoustic model for a bass–reflex system (for example, a 4th–order Butterworth or Chebyshev response, depending on damping) can be cascaded with the driver’s response.
• Phase Alignment:
Even though the filter network mathematically sums flat in power, proper acoustic alignment (often using physical driver offset or waveguides) is crucial to achieve flat acoustic pressure response at the listening position.
When you combine the crossover filters with the driver and enclosure models, a typical simulation (using, say, MATLAB or dedicated speaker design software) might yield an overall on–axis frequency response similar to the following (numbers are illustrative):
Key points:
• From about 60 Hz to 20 kHz, the combined response is nearly flat (within ±0.5 dB) if the drivers are well–matched and acoustically aligned.
• Below ~60 Hz, the response falls off due to both the driver’s low–frequency limitations and the enclosure’s tuning.
• Near the crossover (around 2.2 kHz), the “slope” of –6 dB/octave in each branch combines to produce an overall flat response when the drivers are in phase.
4. Graphical Representation
If you were to plot the magnitude responses:
• Individual Branches:
• The midwoofer’s low–pass curve would start at 0 dB at low frequencies and roll off gradually past 2.2 kHz.
• The tweeter’s high–pass curve would be near –20 dB at low frequencies, rising at 6 dB/octave until it flattens out above 2.2 kHz.
• Combined Response:
• When the two responses are summed (assuming proper phase alignment), the result is a flat line (0 dB) across most of the passband.
• A small dip or hump might appear in the crossover region if the acoustic centers are mis–aligned or if sensitivity differences aren’t fully corrected.
While I cannot render a live graph here, you can easily plot these equations using a tool like MATLAB. For example, in MATLAB you might use:
This code plots the magnitude (in dB) of each branch as well as the idealized combined response.
5. Final Thoughts
• The “model” above is idealized. In the real world, the drivers’ natural roll–offs, cabinet resonances, diffraction effects from the baffle, and non–idealities in component behavior will modify the response.
• Iterative Measurement & Tuning:
After building the enclosure, many designers measure the in–room frequency response and make small adjustments (such as tweaking the port length or adding acoustic damping) to fine–tune the performance.
• Software Simulation:
For more detailed modeling that includes driver parameters, enclosure Q, and off–axis behavior, consider using specialized speaker–design software.
In summary: The basic filter model predicts that a first–order crossover set at 2.2 kHz will yield two branches whose magnitudes sum to a constant level across the band—resulting in a flat response from roughly 60 Hz to 20 kHz, assuming perfect acoustic alignment and proper sensitivity matching. This is the ideal “model response” of the electrical crossover network. When you combine this with realistic driver and enclosure models, you should expect a very smooth overall response with only minor deviations in the real world.
Happy modeling, and I hope this gives you a solid starting point for both simulation and eventual measurement of your two–way loudspeaker design!
_________________________
Below is an example “blue‐print” for a high–performance two–way bookshelf loudspeaker using SB Acoustics’ TW29DN‑B‑8 tweeter and MW19P‑4 midwoofer. Keep in mind that “perfect” performance always comes only after careful measurement and (often iterative) tuning. The design below gives you a starting point based on typical driver behavior and proven design practices. (Be sure to check the manufacturers’ datasheets for the exact Thiele/Small parameters of your drivers so you can refine these numbers in simulation.)
In summary, the design uses a well–matched crossover (you may choose first–order for simplicity and excellent phase coherence or, if you prefer steeper slopes, a second–order Linkwitz–Riley alignment) and a rigid, internally braced ported enclosure built from 18 mm birch plywood. (A sealed box is “cleaner” in the bass but a ported design can “extend” the bass and boost efficiency; for best overall performance in a bookshelf size, the ported approach is attractive.)
Below are the details.
1. Driver Integration
• Tweeter: SB Acoustics TW29DN‑B‑8
– Nominal impedance: ~8 Ω
– Small (≈30 mm diameter) dome element with excellent high–frequency detail
• Midwoofer: SB Acoustics MW19P‑4
– (Usually a 19 cm [≈7.5″] driver; model “4” indicates its design class—often a 4 Ω unit; if you have an 8 Ω version, simply scale the crossover inductance accordingly.)
– Provides smooth mid–bass extension down to the low 60’s Hz
A careful layout (for example, with the tweeter mounted in the upper third of a curved front baffle and the midwoofer centered in the lower half) helps minimize diffraction. (Sometimes a small vertical “offset” or acoustic “waveguide” is used so that the drivers’ acoustic centers are aligned in time.)
2. Crossover Design
You want to “hand–match” the drivers so that their overlapping responses sum flat with good phase coherence. Two common choices are:
A. First–Order Crossover (6 dB/octave)
This type is very popular for its natural phase behavior and “open” sound. Its gentle slope means that the drivers’ natural roll–offs do most of the work, so it is well suited if your drivers’ off–axis and free–air responses are smooth.
• Crossover frequency (f₍c₎):
A good starting point is about 2.2 kHz–2.5 kHz. (You may adjust this based on measured frequency responses.)
• For the Tweeter (High–Pass):
A single series capacitor is used. Using the standard formula for a 1st–order filter:
For example, with and :
– Component: Use a high–quality (e.g. polypropylene) non–polarized capacitor rated for 50 V or higher.
• For the Midwoofer (Low–Pass):
A series inductor forms the 1st–order low pass. Its value is given by:
For example, if the MW19P‑4 is a 4 Ω unit at :
(If your midwoofer is actually 8 Ω, double this value to about 0.58 mH.)
– Component: Choose an air–core inductor designed for audio use (low losses, good linearity).
• Level Matching:
In many designs the tweeter is naturally a bit more sensitive. If measurements show a 2–3 dB level difference at the crossover, incorporate an L–pad network (a resistor in series with the tweeter plus a resistor shunted to ground) so that its output is trimmed. For a 3 dB pad on an 8 Ω tweeter, typical resistor values are roughly:
• Series resistor ≈ 3.3 Ω
• Shunt resistor ≈ 11.3 Ω
(These values come from standard L–pad formulas; recalc them using your exact sensitivity data.)
B. (Optional) Second–Order (Linkwitz–Riley) Crossover
A 2nd–order design gives a 12 dB/octave slope and very smooth summed response if well–implemented. (In a Linkwitz–Riley alignment the two filters are “cascaded” from 1st–order Butterworth sections; note that at the chosen frequency each branch is down 6 dB, and the drivers’ outputs sum flat.)
For a 2nd–order high–pass the component values are reduced by a factor of roughly . For example, for the tweeter:
and similarly for the midwoofer the series inductor would be about 0.20–0.30 mH (for a 4 Ω load).
Because the required inductance values are very low, it may be easier in practice to start with the first–order design and then decide if a steeper roll–off is necessary.
3. Enclosure Design
Type & Materials
• Type:
A ported (bass–reflex) design is recommended for extended low–frequency output and higher efficiency. (If you prefer the “tighter” control of sealed designs, you can adapt the following dimensions for a smaller internal volume.)
• Material:
Use 18 mm thick birch plywood throughout. This material is both stiff and well–damped, and—when properly braced—minimizes cabinet resonances.
Internal Volume
• Target Volume (V₍b₎):
Approximately 20 liters (0.02 m³) is a good starting point to suit the midwoofer’s parameters.
Example External/Effective Internal Dimensions
Accounting for 18 mm walls, you might design an enclosure with approximate external dimensions such as:
• Width: 300 mm
• Height: 250–300 mm
• Depth: 270–300 mm
For example, internal dimensions of about 300 mm (W) × 250 mm (H) × 270 mm (D) give roughly 20–21 liters. (Actual dimensions should be calculated to achieve the precise volume after subtracting wall thickness and any internal bracing.)
Front Baffle & Driver Cutouts
• Baffle Shape:
A gently curved or beveled front baffle minimizes diffraction.
• Driver Placement:
• Midwoofer: Centered in the lower (or central) portion. Its cutout diameter should match the mounting ring (likely around 190 mm).
• Tweeter: Mounted in the upper section (or slightly “in–front” of the woofer) with a small cutout (≈30–40 mm diameter).
A slight vertical offset or use of an acoustic delay (waveguide) may be applied to “time–align” the drivers’ acoustic centers.
Port (Bass–Reflex) Design
• Tuning Frequency:
Aim for a box tuning around 45 Hz to complement the midwoofer’s low–end.
• Port Dimensions:
For a single round port made from a 50 mm diameter tube:
1. Calculate the cross–sectional area (A):
2. Determine Port Length (L):
Using the equation
(with ) and solving for L gives:
Plugging in:
•
•
•
•
You get:
3. End Corrections:
Add roughly 0.85×(port diameter) (≈43 mm) to the physical length to account for port “end effects” (or adjust the tuning with simulation).
Internal Bracing & Damping
• Bracing:
Install internal braces (made from 18 mm ply scraps) to “stiffen” the cabinet and avoid panel resonances.
• Damping:
Lining the enclosure’s walls with acoustic damping material (e.g. poly–fill or foam) will help control standing waves and smooth the bass response.
4. Construction & Final Tuning
• Joinery:
Use high–quality wood glue and screws to ensure all joints are airtight (except for the port).
• Finishing:
Consider a veneer or painted finish on the front, with careful attention to edge treatment.
• Measurements & Adjustments:
After construction, measure the internal volume and, if needed, adjust the port length. Also, bench–test (and if possible, measure) the driver responses with the chosen crossover; slight tweaks (including fine–tuning the L–pad on the tweeter) may be required to obtain a flat summed response and optimal phase coherence.
• Simulation:
Software tools such as WinISD or similar can help simulate both the bass–reflex tuning and the crossover responses—this is highly recommended before finalizing the build.
Summary of Key Numbers (Example Values)
• Crossover Frequency: ~2.2 kHz
• Tweeter High–Pass (First Order): 9 µF (≈8–10 µF polypropylene)
• Midwoofer Low–Pass (First Order): ~0.29 mH (for a 4 Ω driver)
• Tweeter L–pad (if needed for ∼3 dB attenuation):
• Series: ≈3.3 Ω
• Shunt: ≈11.3 Ω
• Enclosure Internal Volume: ~20 liters
• Port: 50 mm diameter; calculated length ≈145 mm (plus ~43 mm end–correction) for tuning near 45 Hz
• Construction: 18 mm birch plywood with internal bracing and damping
By following this plan—and then refining component values and dimensions using measurements and simulation—you should be well on your way to building an optimized two–way loudspeaker that makes the most of the SB Acoustics TW29DN‑B‑8 tweeter and MW19P‑4 midwoofer. Happy building!
CAN YOU MODEL THE RESONSE?
Below is an example of how one might “model” the frequency‐response of the designed two–way system using standard filter equations. Keep in mind that this simulation is based on the idealized electrical (or “netlist”) model of the crossover filters and does not include all the real–world effects such as the drivers’ own acoustical roll–offs, cabinet diffraction, or enclosure resonances. In practice, you’d combine these with measured driver responses (or use simulation software like VituixCAD, XSim, or MATLAB) to get a fully accurate picture. That said, here’s a step–by–step explanation and example calculations for the “first–order” crossover case we described.
1. The Basic Model
For our first–order (6 dB/octave) crossover with a chosen crossover frequency (here, let’s assume 2.2 kHz), the two filter transfer functions (in voltage, ignoring phase shifts from the drivers for the moment) are:
• Midwoofer (Low–Pass):
• Tweeter (High–Pass):
Note: At , each filter’s magnitude is
If the drivers are acoustically “aligned” (that is, their acoustic outputs arrive in phase) and their sensitivities are matched (using an L–pad if needed), the sound–pressure contributions add coherently. In the ideal case the total output pressure is the sum of the two contributions. If the filters’ phases differ by exactly 90° (as they do in a pure first–order filter), then the power sums in quadrature. In our design, however, we aim to adjust the mechanical and electrical delays (or even use a slight acoustic waveguide) so that the acoustic outputs are in phase at the listener’s position. Under that “in–phase” condition, the acoustic pressure response can be approximated as the algebraic sum:
A little algebra shows that when summed in quadrature the net magnitude is nearly constant:
for all frequencies. (This is why a Linkwitz–Riley alignment, which is based on cascaded Butterworth sections, is so popular—it naturally sums to a flat response.)
2. Example Calculations
Let’s step through some example frequencies for the idealized filters (assuming ).
A. Compute the normalized magnitude responses
1. At 100 Hz:
•
• Midwoofer:
• Tweeter:
• Combined Magnitude (if added in quadrature):
2. At 1000 Hz:
•
• Midwoofer:
• Tweeter:
• Combined:
3. At :
• Both branches yield:
• Combined:
4. At 10000 Hz:
•
• Midwoofer:
• Tweeter:
• Combined:
B. Interpreting These Numbers
Under these ideal assumptions:
• Below the crossover: The midwoofer is nearly “full on” while the tweeter is almost off.
• Above the crossover: The tweeter dominates.
• At the crossover: Both drivers contribute equally (about –3 dB each), but if in phase the pressure adds to yield a flat (0 dB) overall response.
Thus, the electrical–filter model predicts a combined response that is flat (i.e. constant gain of 1, or 0 dB relative to a set reference) across frequency.
3. Including the Enclosure & Driver Characteristics
In practice, you must include:
• Driver Responses:
Both the midwoofer and tweeter have their own inherent frequency–response curves. For example, the midwoofer may start rolling off well above 2–3 kHz and the tweeter may not be perfectly flat over its range. Manufacturers’ measurements or Thiele/Small–parameter–based simulations help include these effects.
• Cabinet/Port Effects:
The ported enclosure introduces its own bass–reflex behavior. For example, if tuned to 45 Hz, you might see a controlled peak or a smoother roll–off below 60 Hz. A common acoustic model for a bass–reflex system (for example, a 4th–order Butterworth or Chebyshev response, depending on damping) can be cascaded with the driver’s response.
• Phase Alignment:
Even though the filter network mathematically sums flat in power, proper acoustic alignment (often using physical driver offset or waveguides) is crucial to achieve flat acoustic pressure response at the listening position.
When you combine the crossover filters with the driver and enclosure models, a typical simulation (using, say, MATLAB or dedicated speaker design software) might yield an overall on–axis frequency response similar to the following (numbers are illustrative):
| Frequency (Hz) | Simulated Response (dB, On–Axis) |
|---|---|
| 40 | –3 to –2 (roll–off begins) |
| 60 | 0 to +0.5 |
| 100 | +0.2 |
| 200 | 0 |
| 500 | 0 |
| 1000 | 0 |
| 2200 | 0 |
| 5000 | 0 |
| 10000 | 0 |
| 20000 | –0.5 to 0 |
• From about 60 Hz to 20 kHz, the combined response is nearly flat (within ±0.5 dB) if the drivers are well–matched and acoustically aligned.
• Below ~60 Hz, the response falls off due to both the driver’s low–frequency limitations and the enclosure’s tuning.
• Near the crossover (around 2.2 kHz), the “slope” of –6 dB/octave in each branch combines to produce an overall flat response when the drivers are in phase.
4. Graphical Representation
If you were to plot the magnitude responses:
• Individual Branches:
• The midwoofer’s low–pass curve would start at 0 dB at low frequencies and roll off gradually past 2.2 kHz.
• The tweeter’s high–pass curve would be near –20 dB at low frequencies, rising at 6 dB/octave until it flattens out above 2.2 kHz.
• Combined Response:
• When the two responses are summed (assuming proper phase alignment), the result is a flat line (0 dB) across most of the passband.
• A small dip or hump might appear in the crossover region if the acoustic centers are mis–aligned or if sensitivity differences aren’t fully corrected.
While I cannot render a live graph here, you can easily plot these equations using a tool like MATLAB. For example, in MATLAB you might use:
Code:
f = logspace(1, 5, 1000); % frequencies from 10 Hz to 100 kHz
fc = 2200;
H_LP = 1./sqrt(1+(f/fc).^2);
H_HP = (f/fc)./sqrt(1+(f/fc).^2);
H_total = sqrt(H_LP.^2 + H_HP.^2); % assuming quadrature addition
semilogx(f, 20*log10(H_LP), 'b', f, 20*log10(H_HP), 'r', f, 20*log10(H_total), 'k', 'LineWidth',2)
xlabel('Frequency (Hz)')
ylabel('Magnitude (dB)')
legend('Midwoofer LP','Tweeter HP','Combined')
grid on5. Final Thoughts
• The “model” above is idealized. In the real world, the drivers’ natural roll–offs, cabinet resonances, diffraction effects from the baffle, and non–idealities in component behavior will modify the response.
• Iterative Measurement & Tuning:
After building the enclosure, many designers measure the in–room frequency response and make small adjustments (such as tweaking the port length or adding acoustic damping) to fine–tune the performance.
• Software Simulation:
For more detailed modeling that includes driver parameters, enclosure Q, and off–axis behavior, consider using specialized speaker–design software.
In summary: The basic filter model predicts that a first–order crossover set at 2.2 kHz will yield two branches whose magnitudes sum to a constant level across the band—resulting in a flat response from roughly 60 Hz to 20 kHz, assuming perfect acoustic alignment and proper sensitivity matching. This is the ideal “model response” of the electrical crossover network. When you combine this with realistic driver and enclosure models, you should expect a very smooth overall response with only minor deviations in the real world.
Happy modeling, and I hope this gives you a solid starting point for both simulation and eventual measurement of your two–way loudspeaker design!
Garbage in, Garbage out.
You need to input the impedance (nominal impedance is not ver useful) and you also need to add in the raw frequency response. Baffle shape …more.
dave
You need to input the impedance (nominal impedance is not ver useful) and you also need to add in the raw frequency response. Baffle shape …more.
dave
Last edited:
@Vermouth you seem pretty convinced that "AI". can do this for you despite what everyone tells you, so why don't you just try it and put your own money and time into it?
From my POV: I see more details being added to the replies, but I still don't see any evidence of ability to "design"anything. As I pointed out above, you might save some minutes not having to do you own google searches for formulas etc. but you will not save the hours needed to turn it into a usable design.
From my POV: I see more details being added to the replies, but I still don't see any evidence of ability to "design"anything. As I pointed out above, you might save some minutes not having to do you own google searches for formulas etc. but you will not save the hours needed to turn it into a usable design.