I built it myself a long time ago, back when I was still inexperienced.
That's why I asked AI for help this time — to try and improve it properly.
Creating a crossover by ear is doable for a 2 way first order series filter. But for a four way second order filter it's impossible. You could use a set of really good headphones as a reference. You will hear if you need to correct frequency response and phase coherence. But, which LCR combination you need to alter is simply not possible by ear. You need to install room eq wizard and buy a calibrated microphone. You also need to buy something like DATS V3 to measure impedance etc. If you are not willing to do this, then stick to 2-way loudspeakers or kits which includes crossovers.
Given where you seem to be (old crossovers derived solely from basic textbook filter equations?), an intermediate step might be to use one of the tracing tools to gather basic frequency response and impedance data from the driver spec sheets. That data can then be used in a crossover simulator (Xsim is free and pretty easy to use, but there are others with more features like baffle simulation). This will get you part of the way and may let you improve on what you have. But without at least basic measurement capabilities, things are a lot harder.
Measuring quasi-anechoic frequency response accurately enough in a home environment is done routinely, but it's not trivial. There are quite a few details that need to be understood to do it well.
Given that you're starting with a finished speaker, you may want to focus on one aspect at a time. Wherever your biggest complaint is would be a good place to start. Focus on that driver and the one it hands off to, and try to optimize just that pair initially. That would let you get some experience with new tools without undertaking a full 4-way design.
Measuring quasi-anechoic frequency response accurately enough in a home environment is done routinely, but it's not trivial. There are quite a few details that need to be understood to do it well.
Given that you're starting with a finished speaker, you may want to focus on one aspect at a time. Wherever your biggest complaint is would be a good place to start. Focus on that driver and the one it hands off to, and try to optimize just that pair initially. That would let you get some experience with new tools without undertaking a full 4-way design.
I already have the speakers and the crossover network.
I'm simply trying to improve and refine what I already have — not building everything from scratch.
Having your circuits in a simulator is a good start.
At first glance, it looks like you are using default data for impedance and frequency response (flat in both cases?). For improved accuracy, you should at least use data derived from the spec sheets for those two items. That data may be available online from someone that has already done that. With enough prodding, one of your AI friends might be able to generate tabular data as well.
Beyond that, the natural response of the C90-6-079 midrange slopes upward with increasing frequency, and it looks like your crossover does as well. The two of those together may be creating a midrange prominence that's greater than it appears in your plot (~1900 Hz).
At first glance, it looks like you are using default data for impedance and frequency response (flat in both cases?). For improved accuracy, you should at least use data derived from the spec sheets for those two items. That data may be available online from someone that has already done that. With enough prodding, one of your AI friends might be able to generate tabular data as well.
Beyond that, the natural response of the C90-6-079 midrange slopes upward with increasing frequency, and it looks like your crossover does as well. The two of those together may be creating a midrange prominence that's greater than it appears in your plot (~1900 Hz).
Last edited:
The project is simulated with traced curves and all the FRs used are manipulated on a 20cm baffle (guessed from your pictures):
Original filter
The polarity inversion of the mw creates a cancellation at low frequency as there is no hi-pass on the mw. Notice the peak above 100Hz (red curve) caused by the low Fc of the low-pass: the impedance peaks at low frequency need some attenuation).
Filter suggested by ChatGPT
The problems al low frequency still remains but is partly compensated by the 1st order HP (68uF cap). Notice that the multiple zobel networks have little effects as chatgpt is not addressing the problem of the low xover Fc. FR is still bad.
Original filter with non-inverted polarity of the Midbass
The peak above 100Hz is still there and so that of the midrange at 1800Hz. FR is a bit better.
Quick and dirty simulation
Same filter without LCR notches on woofer, midrange and tweeter.
Can a few responses be stacked to show the variance across a range of different listening angles? I'm curious about how consistent the sound will be across the vertical axis. And has Virtuix CAD evolved to make good estimates of the off-axis sound pressure?
Great!
Start with the great job above done by @shadowplay62
Bear in mind his simulation is without the data for "depth" position of acoustic centers of drivers (z axis) - it can be done only with measurements.
Last edited:
Thank you so much, my friend. Thanks to you, I was able to solve it!
Other people kept saying it wasn't possible without showing any real proof, but you showed it clearly with evidence.
I'm really grateful.
Thanks to your help, I think this project will be a big success!
Last edited:
... and they are right!
Although shadowplay62 made а great job, you must be aware of limits of such software simulation with traced curves instead of real-world measured curves of frequency and impedance responses. However, I have no doubt that the crossover designed by shadowplay62 is much, much better then the present one you have (or the one "designed" by AI).