Announcement

Collapse
No announcement yet.

Problems with software summation of responses

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Problems with software summation of responses

    Apologies for this rather dry question. Let me describe a problem I've been having.

    - Woofer measured in arta (gated, measured phase included)
    - Tweeter measured in arta ("") using the same mic position and settings
    - Both drivers measured using same set up.

    When I sum them in software (VituixCAD, but the problem seems to happen in other packages), the simulated summed response doesn't match the measured summed response. Simulation of the effects of passive components seem to be accurate - if I measure with a certain capacitor, and then simulate the same circuit, the two responses match up fine.

    This is driving me crazy, because it means the simulated crossovers I design have pretty big deviations from the measured responses when I go ahead and build them. These deviations are not in the responses of the individual drivers, but in their summation.

    Now, notice that I am measuring from one position, not extracting minimum phase or anything like this. I suspect there is some inaccuracy in the measurements cropping up somewhere, but I can't figure out where.

    Any tips? Should I just go ahead and measure minimum phase and find the offset and everything else?

  • #2
    Two channel setup to get the in flight timing? Sent from my SM-G960U using Tapatalk
    Electronics engineer, woofer enthusiast, and musician.
    Wogg Music

    Comment


    • #3
      Originally posted by wogg View Post
      Two channel setup to get the in flight timing? Sent from my SM-G960U using Tapatalk
      I've found the process of setting up 2 channel in arta to be really intimidating and annoying. I'm wondering if I should switch to another measurement package.

      Comment


      • #4
        I've done some more digging and the null is actually not at the crossover point, but rather the first diffraction dip in the tweeter response. This makes sense, as with a LR4 crossover, a little time misalignment shouldn't cause a huge dip in the crossover region. I remeasured and checked the acoustic offset and it was only 7mm anyway with these drivers.

        This is very frustrating. Why should the diffraction dip not figure into the summation of the drivers? I change the first tweeter cap and it doesn't really do anything to fill in the dip.

        Comment


        • #5
          Well, I surrounded the tweeter with felt insulation and it didn't effect the dip at all. Moving off axis, the dip does go away, like one would expect.

          This still doesn't make sense. The software summed response and the measured summed response are way off. I'm out of ideas on this one.

          Comment


          • #6
            Originally posted by civit View Post
            I've found the process of setting up 2 channel in arta to be really intimidating and annoying.
            Originally posted by civit View Post
            Should I just go ahead and measure minimum phase and find the offset and everything else?
            I find your posts quite confusion, and am having a hard time following what you are doing. Assuming you're using a single channel measurement, then yes of course you should do the minimum phase, determine offset, etc... Your error in the simuilation summation versus reality summation is an error in phase. Where is the start of the FFT window when you measure in single channel mode? Likely determined as the start of the impulse or the peak, which is removing the "time of flight" that determines the phase interaction between drivers. This is the reality of a single channel measurement, there is no reference input so there is no information to tell you how long it took the signal to reach the mic. You need to determine the driver offsets to accurately represent phase.

            "I just use off the shelf textbook filters designed for a resistor of 8 ohms with
            exactly a Fc 3K for both drivers, anybody can do it." -Xmax

            Comment


            • #7
              Well, this diagram shows it as clearly as I can show.
              Click image for larger version

Name:	summation.jpg
Views:	1
Size:	129.5 KB
ID:	1379449 All measurements taken at the same mic position, gated exactly the same, no smoothing. The measured sum - drivers hooked together in parallel - is in red. The calculated sum (woofer + tweeter) is in green. The individual responses are in gray. As can be seen, there is a huge deviation from the calculated sum and the measured sum.

              Relative acoustic offset doesn't play into this, since the small offset between these drivers wouldn't create a deviation this big (I checked that.) The drivers' phase is correct - if I reverse the tweeter phase I get a null at the crossover frequency - this is the best summation.

              I suspect I conked my measurement mic - a UMM-6 (calibration file used on all measurements, naturally.) I ordered a new Umik-1 to see if that works any better.

              What confuses me the most is this - if there was a deviation in the microphone's response, it would apply to both the measurements and the sum of the measurements. The deviation between the software sum and the measured sum remains. This is going to drive me crazy.

              Comment


              • #8
                Originally posted by dcibel View Post


                I find your posts quite confusion, and am having a hard time following what you are doing. Assuming you're using a single channel measurement, then yes of course you should do the minimum phase, determine offset, etc... Your error in the simuilation summation versus reality summation is an error in phase. Where is the start of the FFT window when you measure in single channel mode? Likely determined as the start of the impulse or the peak, which is removing the "time of flight" that determines the phase interaction between drivers. This is the reality of a single channel measurement, there is no reference input so there is no information to tell you how long it took the signal to reach the mic. You need to determine the driver offsets to accurately represent phase.
                Dcibel, I apologize if I am causing confusion; I am confused myself.

                - When I did my initial measurements, I did single channel measurements in ARTA, which centers the impulse at sample #300. I then gated from that sample to my first reflection (6ms or whatever)
                - I used arta to generate minimum phase for these responses.
                - I simulated a crossover with flat response using these measurements (lr4 around 2k)
                - When I built the crossover, the cancellation around 3k was evident upon measurement.
                - I then used winPCD to find the acoustic offset, which turned out to be about 6mm. To do this, I followed the directions calling for two driver responses, minphase, and one sum response, with no phase information
                - Inputting this offset into my crossover had minimal effect, and did not duplicate the deviation seen here.

                HOWEVER. I think you may be right. I just played with delays in my simulation, and delaying the tweeter by 105 microseconds (36mm) resulted in a response very similar to what I am seeing. As a result, I think the lack of timing reference is at the root of my problems.

                Do you have any suggested solutions for this? As I see it, I have two options:

                - bite the bullet and figure out how to wire up arta for 2 channel measurements
                - use the acoustic timing reference in REW





                Comment


                • #9
                  Originally posted by civit View Post
                  All measurements taken at the same mic position, gated exactly the same, no smoothing. The measured sum - drivers hooked together in parallel - is in red. The calculated sum (woofer + tweeter) is in green. The individual responses are in gray. As can be seen, there is a huge deviation from the calculated sum and the measured sum.
                  The hint to me would the change in response at 1kHz. The claim is that the drivers are hooked in parallel in the red line, however you can see this measurement is about 0.5dB higher at 1kHz than the measurement of the woofer alone, even though the tweeter output is a good 22dB down from the woofer. This is a red flag that something is wrong with the summed measurement to me. You'd need the tweeter output to be closer to 65-70dB to increase the output by 0.5db here.

                  I would simply retake the measurements. Once you have a measured some that makes some sense compared to each individual measurement, do the min phase extraction of each, determine offsets and enjoy ;) From your post above, it sounds like you haven't determined acoustic offset when designing your crossover, this critical step will be the reason that the simulation of the crossover doesn't match reality.

                  Just FYI, if ARTA is placing the start of the FFT window at the start of the impulse, the meaurements are nearly minimum phase to begin with, but you should still complete the min phase steps (HBT tails then extract) to get as accurate a result as possible.
                  "I just use off the shelf textbook filters designed for a resistor of 8 ohms with
                  exactly a Fc 3K for both drivers, anybody can do it." -Xmax

                  Comment


                  • #10
                    Here are the measurements with no crossover. There is still a slight deviation in the sums:




                    These measurements were done in arta, using the loopback 2 channel method. I checked, and this placed the impulse at different points depending on where I put the mic. I made two measurements 17" apart, and the impulse peaks were offset by around 1.25ms, which checks out.

                    I used the same gate and did not extract minimum phase, so these files should have the same relative timing. Still see a deviation, although it appears to be in agreement around 2k, which is a big improvement.

                    Comment


                    • #11
                      Originally posted by civit View Post
                      - When I did my initial measurements, I did single channel measurements in ARTA, which centers the impulse at sample #300. I then gated from that sample to my first reflection (6ms or whatever)
                      Unless the system is mounted high off of the floor, first reflection is usually 3-4 msec.
                      I've never used arta, so I can't comment on it.

                      - I then used winPCD to find the acoustic offset, which turned out to be about 6mm. To do this, I followed the directions calling for two driver responses, minphase, and one sum response, with no phase information
                      WinPCD does require phase, but for relative offset determination and use, that phase must be minimum-phase. Keep in mind that what you determine for relative offset is highly dependent on how you extend measurements prior to calculating minimum-phase. The slopes at both ends have an impact on the offset value, lowpass generally the most important, so any subsequent measurements must have the same slopes applied if minimum-phase is again generated from a subsequent measurement, such as verifying what you did previously, otherwise you must re-determine offset for the new set of minimum-phase measurements.

                      HOWEVER. I think you may be right. I just played with delays in my simulation, and delaying the tweeter by 105 microseconds (36mm) resulted in a response very similar to what I am seeing. As a result, I think the lack of timing reference is at the root of my problems.
                      If you are using minimum-phase data, the timing reference is immaterial as long as you eliminate any reflections. This goes back to my comment above about 3-4ms for a typical floor (usually) or ceiling reflection.

                      To reiterate a common point I've made over the years, measured phase is not a requirement if you intend to use minimum-phase data. Using the 3-measurement scheme obviates the need for measured phase.

                      dlr
                      WinPCD - Windows .NET Passive Crossover Designer

                      Dave's Speaker Pages

                      Comment


                      • #12
                        One reason that I've previously run into these "prediction vs measured" problems is due to using a trace-made ZMA file.

                        Tracing programs typically don't create a phase trace for export ( just look at the phase column when the file is opened within a text editor > one will see a column of zeros ).

                        A work around is to import the newly created/traced ZMA file into Blender ( renamed as an .frd file ) then extract minimum phase and resave the file and change the file type back to a .zma. A proper phase trace will get created by Blender.

                        Last edited by EarlK; 06-19-2018, 09:17 AM.

                        Comment


                        • #13
                          Originally posted by EarlK View Post
                          One reason that I've previously run into these "prediction vs measured" problems is due to using a trace-made ZMA file.

                          Tracing programs typically don't create a phase trace for export ( just look at the phase column when the file is opened within a text editor > one will see a column of zeros ).

                          A work around is to import the newly created/traced ZMA file into Blender ( renamed as an .frd file ) then extract minimum phase and resave the file and change the file type back to a .zma. A proper phase trace will get created by Blender.

                          To your point - I use my beloved DATS for my impedance sweeps, and the SPL tracing program I use, included in VituixCAD, does generate phase data to my knowledge.

                          Comment


                          • #14
                            Originally posted by EarlK View Post
                            One reason that I've previously run into these "prediction vs measured" problems is due to using a trace-made ZMA file.

                            Tracing programs typically don't create a phase trace for export ( just look at the phase column when the file is opened within a text editor > one will see a column of zeros ).

                            A work around is to import the newly created/traced ZMA file into Blender ( renamed as an .frd file ) then extract minimum phase and resave the file and change the file type back to a .zma. A proper phase trace will get created by Blender.

                            Not so much a "work around" as it is just the way you have to do it. Same applies to tracing frequency response, and many measurements as well. When measuring, it is important to understand how your measurement works and what the values mean to be confident that you have accurate data.
                            "I just use off the shelf textbook filters designed for a resistor of 8 ohms with
                            exactly a Fc 3K for both drivers, anybody can do it." -Xmax

                            Comment

                            Working...
                            X