If I measured my own drive units with REW and then used Xsim to model a crossover, how close would the finished speaker be to the model?
The software is quite accurate. A 'room' where measurements are made might mess up the (end) results though.
Component values can differ (they have a tolerance) too, but one can measure them and compare them to a simulated circuit to check what they need to do.
Impedance traces in measurements can differ too, for instance with ambient temperature or the level with which the measurements were made. Say with a laptop
or an amp as driving signal or with a tester like a DATS.
All in all, it may take some time to gain experience to make it work.
Component values can differ (they have a tolerance) too, but one can measure them and compare them to a simulated circuit to check what they need to do.
Impedance traces in measurements can differ too, for instance with ambient temperature or the level with which the measurements were made. Say with a laptop
or an amp as driving signal or with a tester like a DATS.
All in all, it may take some time to gain experience to make it work.
I expect that both the software programs are quite accurate and extraordinarily powerful tools compared to what loudspeaker designers had only a decade or two ago. But designing a loudspeaekr system to have a flat frequency response at one point in space in an anechoic environment, which is what software like that will allow you to do, is of extremely dubious value. O'Toole and Olive had written volumes on how loudspeakers should behave in the real world to achieve broad acceptance as 'sounding good' and the above is not one of the requirements. Likewise Griesinger has written tomes on perception of audio quality and spaciousness, and again a measured 'perfect score' from a loudspeaker designed for a flat frequency response at an arbitrary distance in an anechoic space doesn't rate as being important of it's own right.
Add Vituixcad to the mix and you can cater to the above knowledge as well. If you take enough measurements under anechoic conditions on and off axis.
If you can't do the off axis measurements, it has tools to simulate/generate the off axis response. Not quite the same but closer than making all decisions
based on on-axis measurements.
If you can't do the off axis measurements, it has tools to simulate/generate the off axis response. Not quite the same but closer than making all decisions
based on on-axis measurements.
I was writing my own Xsim level software closer to 30 years ago, and I wasn't the first.
I meant that in regards to DIY designers of course. Spice was being used to model loudspeakers and crossovers 40 years ago, though by my colleagues not me, I wasn't smart enough.
In 1992 I was using a software called "LMP" by Ralph Gonzalez. It was a DOS-based program that did most of what Xsim does, and it was available to DIY folks. By todays standards, it was comically slow and cumbersome to use. This was pre-Windows gui, 8088 16 bit CPU era. But it worked. I think I paid $50 for a floppy disk and a photo-copied instruction booklet.
It all depends on how accurate and repeatable your measurements are. I use CLIO, LEAP and Bass Box Pro and if you compare the as designed vs the as built measurements they can be spot on within a reasonable tolerance window WRT FR/Phase/Imp and Fb on a reflex box.
Rob 🙂
Yes, good lab practice matters a whole lot (and may be unfamiliar to many users). Do you get the same results in repeated tries or repeats with slight modifications to judge sensitivity of methods?
Let's just deconstruct some of the previous comments about driver modelling software. Room measurement software raises quite a different set of questions.
It is fair to say the apps are spot-on as far as 8th-grade physics and relating measurements to models. Just simple arithmetic. So in that restricted sense, the apps are accurate.
But how close are models to your at-ear experience? Or results with your chosen XO in the middle? Or even to the one-meter out in the park hanging from a crane kind of abstraction they only claim to model?
The T/S computational simplification is now quite old - 40-50 years? And based on models far older. It is a simplified model of a complex mechanical gizmo taken abstractly, not a forever law of physics.*
B.
*Actually, the embarrassing horror is that we are still using Rice-Kellogg drivers that can be modelled however imperfectly by such long-ago theories.
Last edited:
Back in the '90s our section of the AES invited Neville Theile, then in his late '70s, to make a retrospective presentation on loudspeaker alignments using Theile-Small parameters, which went a long way to explaining variable results in the real world. According to Theile some Theile-Small parameters are incredibly sensitive to small errors, but no one seems to do any variance analysis and just assumes the results of their favourite analysis program are valid. If I understood Theile correctly, the way Theil-Small parameters are typically used loudspeaker designers might just as well be playing 'pin the tail on the donkey'.
Ho, ho. Thanks for truly helpful perspective from "the horse's mouth."
But are you alleging members of this forum are donkeys?
B.
But are you alleging members of this forum are donkeys?
B.
Not at all. I think the worst culprits are some of the big HiFi manufacturers who really ought to know better. The take home message is that it is dangerous to assume any process is foolproof. From my engineering background, it is wise to use three completely independent development or verification methods. If all three agree, it's quite likely your result correct. In my experience 9 times out of 10, rather than confirming a simulation or measurement this approach reveals a process mistake or a false assumption.
Imo, both are very accurate. With major caveats.....
Biggest gotcha with modeling software is fully understanding its built in assumptions and limitations. And applying the software correctly.
Biggest gotcha with measuring software, is understand what the measurements mean, what's actually being measured.
....... and what measurements you can act on, along with what should you correct; and conversely what you can't correct, along with what you should not try to correct.
Iow, lot's of learning...that apparently never ends 🙂
Biggest gotcha with modeling software is fully understanding its built in assumptions and limitations. And applying the software correctly.
Biggest gotcha with measuring software, is understand what the measurements mean, what's actually being measured.
....... and what measurements you can act on, along with what should you correct; and conversely what you can't correct, along with what you should not try to correct.
Iow, lot's of learning...that apparently never ends 🙂
You don't input your at ear experience into a simulator. You input your measurement's and driver parameters. Your validation is how close the completed system matches the predicted response. What your are asking is simply not relevant to the accuracy of the simulators.
Rob 🙂
I can't comment on those systems but LEAP / LMS were crazy accurate usually better than .5dB (1/2 dB)!!! And box modeling in LEAP 4.xx was also crazy good, im sure that 5.xx is better, kind of a moot point now but you can still get LEAP running.
G
"Crazy accurate", eh?
gdillon asked a good question - when he tests with REW will it resemble the sim?
If he moves his mic when testing with REW by a few inches, the plot will be a whole lot different than your "crazy accurate" exaggeration.
If predictive sims have a role, it is early in the design exercise.
B.
That's not something to boast about. That's a shortcoming.
In fact, anybody with any sophistication in these matters will recognize that the "ear experience" (which is the criterion for speaker building) isn't even quite the same as an REW plot in-situ.
B.
I have found the results to be very close, if the mic is in the same place of course. But I was also using impedance measurements made by either the WT3 or Woofer Tester II. That's important.