DSP assisted reflex system

The referee named Excursion is the head official. YES.

The most scientific way you can approach this is pay attention to:

1) Fb, which is the highest leverage point of the system. In your system it is around 20-25Hz.

and

2) The lower resonance impedance peak, which is the most vulnerable point of the system. In your models it looks to be around 15Hz. This is where excursion can go wildly out of control.

The trouble is these two points are so close together. This is true for reflex, TL and horns.

The ideal filter from a bandwidth and excursion point of view (not necessarily from a transient response point of view!) is a filter that has a peak at Fb and a dip at the lower resonance. The difference between the peak and dip needs to be 10-15dB.

You can always achieve this with a shelf filter by playing with the Q, frequency and attenuation.

shelf filter Q=5 20Hz -3.5dB.png

This is a high Q of 5 and a gain of only -3.5dB but as you can see, you get +6dB at 23Hz and -10dB at 17Hz at the same time. 16dB spread.

But the filter doesn't apply any more correction than absolutely necessary. (There's hardly any recorded music material below 15Hz, so we don't need any more filtering there. Might be different if this were a PA system.) This keeps group delay from getting excessive.

Have fun dialing this in.
 
That’s exactly right. Nice work. You are picking up 5 or 6 dB and shielding the system from super low frequencies at the same time. When you do that you’ll often be surprised at how loud the system will play and still be clean.

I once built two midbasses for each side of my car, and used a 3.5" Aurasound woofer that had an xmax of something like 5mm

They replaced an 8" woofer that was in a sealed box, and were surprisingly comparable

Obviously processing was one part of the puzzle, but another thing I did, was that I 3D printed it with walls that were about an inch thick, then wraped the enclosure in fiberglass, then slathered the entire thing in liquid nails for subfloors. The idea was to make it rigid and airtight.

It didn't get as LOUD as the 8" sealed box, but it played surprisingly deep and was very "dynamic." This is obviously an "edge case" because it doesn't take a lot of power to produce loud bass in a car, but it was still a bit shocking. Everyone on the car audio forums basically thought it was impossible. But I think that's because most people are accustomed to hearing 3.5" woofers in very flimsy/cheap enclosures with no processing.
 
My Q=2 upward bump (based on this article in AX) assumes that you are using it with a reflex system that has a downward tapering response, and it is already down 6-8dB at Fb.

If you had a conventionally tuned box you would achieve that by lengthening the port or adding weight to the passive radiator and tuning it lower. Which is very interesting and useful. A lot of speakers that only went down to 45 before can now reach 35 or 30.

If you're using a Shelf with a conventionally tuned reflex (butterworth BW3 or similar - maximally flat) then you need to add a -6dB notch filter at Fb with a Q of 1-2 to flatten it out, then you'll still get the sharp attenuation below Fb without the excess group delay.

@mark100 I like your SPL, phase, group delay plots. They really drive home the point of shelf filters having less phase shift and group delay.

I doubt very many people use more than 4th order slopes in home systems. Maybe they do in pro systems. I also know that since the DSPs have 48dB filters, some people go ahead and use them.
This all is much alike the alignment(s) Don Keele designed decades ago. I have used a B6-alignment for more than 20 years, tuned to 25Hz and with analog active crossover with Q= (about) 2. Way back I followed design rules in the Lautsprecher Handbuch, which again were based upon Keele’s articles. If you’re not into DSP, the required analog circuit for this is dead simple.
 
  • Like
Reactions: perrymarshall
The ideal filter from a bandwidth and excursion point of view (not necessarily from a transient response point of view!) is a filter that has a peak at Fb and a dip at the lower resonance. The difference between the peak and dip needs to be 10-15dB.

You can always achieve this with a shelf filter by playing with the Q, frequency and attenuation.

gotcha, makes full sense.

On my existing 190L subs, with Helmholtz freq of 27Hz, I'm going to replace the currently used BW3 hpf, with this shelf.

DIY single 18 shelf.JPG




That will give this power response, which is the same as with the BW3.
The shelf above is the only EQ & filter being used. 🙂

DIY single 18 shelf power.JPG



Switching gears.....
What is your ideal filter from a best transient point of view?

The very bottom end is where all my experimentation has been lately. The rest of the spectrum feels pretty easy.
I've tried linear phase high-pass, combinations or IIR and lin-phase, etc. Still undecided on best sound. Seems like outdoors is the only time I can begin to tell any differences.

Gear shift #2...
I see in the Bitches Brew doc that you use Eclipse Audio's software. I started with it back in 2016, as a beta tester for first product rollout. I use the Fir Designer (one-time license version)
Do you use the auto mag, and/or auto phase to target match? Or do you do everything via direct design? Just curious....
 
  • Like
Reactions: perrymarshall
Nice. Excellent implementation of a shelf filter.

The ideal filter for transient response is Bessel, which mathematically is as close as you can get to a time delay that is constant across all frequencies. A sealed box with a Q of 0.577 is an example of a Bessel alignment. Note that you have to achieve a Bessel shape acoustically after everything else, not just use a Bessel filter in your active circuitry.

Just like there is a triad relationship between box size, efficiency and bass extension (Hoffman's Iron Law) there is also a triad relationship between bandwidth, phase response, and harmonic distortion. You can have two, but you pay with the third.

This entire thread and my concept of DSP assisted reflex assumes (without explicitly stating) that we are willing to sacrifice phase response to get low THD and maximum bandwidth. I believe phase delay is the least audible of the three, and since my article examples include 3" full range drivers in a 1 liter box pushed all the way down to 55Hz, that seems like a pretty reasonable compromise. Like you said, the referee named excursion is the head official. Since we are trying to squeeze every last bit of SPL juice out of the orange, transient response goes out the window.

Bass Reflex is inherently problematic because it's a 4th order system and its steep rolloff means it's going to ring. Step response is going to oscillate for a few cycles instead of settling down immediately. (A true Bessel Q = 0.577 has almost no overhang at all, and a Q of 0.50 settles down as fast as possible with zero overhang. It is called "critically damped.")

The 6th order bass reflex you're modeling here is, frankly, going to ring like a bell at around 25-30Hz. That's the price you pay for the out of band protection and the robust bass extension. To be fair, many non-purist listeners are going to like that extra ringing, because it just means more pant leg flapping in the low bass.

The virtue of the Shelf filter is, it has 1/3 to 1/4the the phase shift of an 18 or 24dB high pass filter, so you're getting as little group delay as is physically possible with a minimal-excursion, minimum distortion reflex box. I have an 18" subwoofer that's very similar to the one you just showed me here, with Fb at 30Hz and a high Q shelf filter. It gives fantastic output down to 25Hz and it's super clean.

If you listen to well-recorded bass drums (Circle of Manias and Fear of a Blank Planet by Porcupine Tree are great examples) closely you can hear the "thump" arriving later than the "slap." It's barely noticeable, but once someone points it out, you can hear it.

If you use a 24dB per octave HP filter you can definitely hear it. In fact you can almost time it with a stopwatch! Like I've said before, the thump arrives on a different date, a discernible fraction of a second later.

I don't think you can get around this time delay if you use bass reflex, all you can do is minimize it. (People have been saying that about bass reflex for 50 years, it's nothing new.) Shelf filters are the least compromised way to get the advantages.

It is possible to make a Bessel reflex alignment. I haven't researched it but I'm sure you could find written material on that, see also here. I'll guess it would be tuned to a much lower than typical frequency. The system would be 6-12dB down at Fb with a gentle tapering curve, the opposite of a sharp knee, and to the typical audiophile it would sound bass thin.

However I think the purists would say it gets the timing right.

~

It's fitting that you ask about the Bitches Brew dipole. The woofers mounted in the enclosure have an Fs of about 26Hz and a Q of 0.6, which is pretty close to Bessel. Now factor in the dipole loss which is -6dB/octave starting at 45Hz. The DSP corrects that with a 12dB peak filter with a Q of about 1.

When you add all of that together, the phase / transient response is similar to a closed box speaker with a Q of maybe 0.9. Which is to say it's excellent. It sounds very transparent, not thumpy at all, and there is no audible phase delay on transients. The price I pay for that is a very large speaker. Also it doesn't have a sharp cutoff below 20Hz, because the subsonic filtering is modest. I'm mostly relying on the fact that there's very little recorded material that low in the first place, and I have 2 15" high excursion woofers.

I'm optimizing Bitches Brew for transient response. Like Jerry McNutt says, the best way to not sound like a box is to not have a box and that's why they're Open Baffle.

My philosophy of FIR filtering is similarly minimalist. In the Bitches Brew, the 100Hz crossover is passive with series wiring:

1704047040279.png


Notice the curve on the midrange is a 2nd order Shelf filter! It's created by the combination of the series LC circuit and the 26Hz impedance peak from the subwoofer. It has the same gentle phase characteristics as the one you're using for your reflex box.

So the subwoofer + midrange crossover effectively behaves like each has a simple 12dB/octave acoustic slope. That's why I reverse the polarity on the midrange as you can see in the schematic.

That means the FIR filters don't have to do much work, and I mostly use them to correct phase.

The 1100Hz crossover is as close as I can get to 12dB/octave acoustic slopes for both woofer and tweeter. The woofer is attenuated by a shelf filter, not a low pass. The tweeter xover is actually a 6Khz 6db/octave filter that flattens the sloping response of the compression horn, with some extra notch filters to fix a few other problems.

So in total it acts like 12dB per octave acoustic slopes at both Xover frequencies.

The FIR filter adds a 180 degree phase reversal centered at 200Hz and another 180 degree phase reversal at 2.8KHz, rendering the phase response of the system flat from 50Hz to 15KHz.

bitches brew FIR filter.jpeg

I optimized these for cleanest step response, and these were the maxphase crossover points that gave the best measurements.

Then I use the automag and autophase ONLY for the top two octaves, to address breakup modes in the 3" titanium compression driver. Looks like this:
bitches brew FIR for tweeters.png


This is the step response I obtained:
bitches brew step response 27nov2023.png
 
Last edited:
The virtue of the Shelf filter is, it has 1/3 to 1/4the the phase shift of an 18 or 24dB high pass filter, so you're getting as little group delay as is physically possible with a minimal-excursion, minimum distortion reflex box.

If you listen to well-recorded bass drums (Circle of Manias and Fear of a Blank Planet by Porcupine Tree are great examples) closely you can hear the "thump" arriving later than the "slap." It's barely noticeable, but once someone points it out, you can hear it.
"Well-recorded bass drums" often use a condenser mic on the beater head, and a dynamic mic in or near the kick drum port hole.

Screen Shot 2023-12-31 at 1.08.43 PM.png


The dynamic mic lags by 90 degrees from the condenser mic, the LF port output by 180 degrees.
Dynamic mic phase response.jpg

The mics are typically high passed with 2nd (180 degree) 3rd (270 degree) or 4th (360 degree) order filters.

Hearing the "thump" arriving later than the "slap" is quite normal for "well-recorded bass drums".

The low frequency output of a bass-reflex (AKA phase inversion) woofer by definition lags it's upper response by one complete wavelength, phase shifts below the pass band caused by additional filters are not high on my priority list.

Art
 
"Well-recorded bass drums" often use a condenser mic on the beater head, and a dynamic mic in or near the kick drum port hole.

View attachment 1253746

The dynamic mic lags by 90 degrees from the condenser mic, the LF port output by 180 degrees.
View attachment 1253747
The mics are typically high passed with 2nd (180 degree) 3rd (270 degree) or 4th (360 degree) order filters.

Hearing the "thump" arriving later than the "slap" is quite normal for "well-recorded bass drums".

The low frequency output of a bass-reflex (AKA phase inversion) woofer by definition lags it's upper response by one complete wavelength, phase shifts below the pass band caused by additional filters are not high on my priority list.

Art
Art,

Totally fascinating. I had never had knowledge of this recording technique before your post.

The delays imposed by 20 Hz steep filters are much larger than physical delay between one side of a bass drum and the other.

Some people consider this important, others not.

Audiophiles argue and obsess about all sorts of things which are very difficult to measure OR prove audible, as we all know.

But when the thump and the slap arrive on different dates, which are indisputably measurable AND audible; and when the phase delays from the recording AND playback cascade; and when you can fix that with a different filter design, that to me is worth prioritizing.

Most recordings have almost no dynamic range but that's not a reason for designing a speaker with no dynamic range. I don't think "recording engineers usually compromise on X" is a good excuse for compromising on X. Especially if your project is a high end "as few compromises as possible" design.
 
Last edited:
Nice big set of information you shared !

Just like there is a triad relationship between box size, efficiency and bass extension (Hoffman's Iron Law) there is also a triad relationship between bandwidth, phase response, and harmonic distortion. You can have two, but you pay with the third.
Agreed and well said. If fact, a principal design goal for me is to limit the bandwidth span of any particular driver section. (Tis why I'm currently listening to 5-ways.)


This entire thread and my concept of DSP assisted reflex assumes (without explicitly stating) that we are willing to sacrifice phase response to get low THD and maximum bandwidth.

Yep. I get in that spades, that the bottom end acoustic rolloff is going to bring lagging phase, and increased group delay.
And the steeper the rolloff, the bigger the phase phase and group delay issue.
But it takes a steeper rolloff to take drivers lower. Sigh. Back to your triad.

Since acoustic delay is the sum of box/speaker natural rolloff and electrical rolloff, I figure there no getting around the natural rolloff.....unless one's willing to flatten bottom end phase without altering amplitude.
That gets into the whole IIR vs linear phase system high-pass, and whether pre-ring from a linear-phase system high pass is audible.
I still don't have a firm conviction one way or another.

But what I do think is a safe compromise on system high pass, is at least let the electrical high-pass be linear phase.
So In the case of my bass reflex, I have just 4th order natural rolloff, or when running sealed, 2nd order.

I decided to see if I could make a linear-phase shelving filter, the shelf I plan use as a replacement for a BW3 as per #85.
So 19Hz, -16dB, Q=1.18

linear phase diy18 shelf 19Hz.JPGlinear phase diy18 shelf 19Hz GD.JPG

It probably doesn't matter much in my case whether I use this shelf or the BW3, since I use them linear-phase, but I only have 16k taps @48kHz available, and I often wonder how effective they are at the very bottom. So less to linearize the better I guess....in goes the shelf.

I tried to make it with FirD, but FirD's shelving capability is limited. Crosslite+, a program I've had for only a couple of months made the FIR file.
If you haven't encountered it, Crosslite is dual channel measurement, and virtual filter simulation together. And an add on module turns it into a FIR generator too. Crazy capable and feature loaded. (But for anyone interested, not a beginners package imho.)





If you listen to well-recorded bass drums (Circle of Manias and Fear of a Blank Planet by Porcupine Tree are great examples) closely you can hear the "thump" arriving later than the "slap." It's barely noticeable, but once someone points it out, you can hear it.

I listened to the Porcupine Tree tracks. Are there particular versions than standout? I could hear some thump-slap differentiation on 2007 Blank Planet album, but it sounded more like maybe mic timing like Art was talking. Plain dunno
My favorite drum/transient track is probably Joe Walsh's Ordinary Average Guy....maybe because I don't have to wait for the kick drum to hit .
It's one of those tracks you have to have the volume right before you hit play LOl

Oh Hey, and a real dummy question? What's the natural rolloff of an open baffle?
 
Last edited:
  • Like
Reactions: perrymarshall
At risk of belaboring the point, I have never heard a real live bass drum where the thump arrived later than the slap.
That's a funny thing about hearing, we all seem to perceive things a bit differently.
When you are on the beater side of the kick drum, the "crack" or "slap" is more obviously leading the resonant "thump" tone.

This waveform (from DETECTING AND DESCRIBING PERCUSSIVE EVENTS IN POLYPHONIC MUSIC by Martin Haro Berois) shows the order of what occurs in a kick (bass) drum waveform, an initial high amplitude, high frequency short duration transient, followed by a semi-stochastic period as the stretched head drops in frequency, then the resonant tail.

Bass drum.png

It all happens "in the blink of an eye" 😉

Art
 
  • Like
Reactions: perrymarshall
Since acoustic delay is the sum of box/speaker natural rolloff and electrical rolloff, I figure there no getting around the natural rolloff.....unless one's willing to flatten bottom end phase without altering amplitude.

ALL speakers are high pass filters, so there's no escaping it.

That gets into the whole IIR vs linear phase system high-pass, and whether pre-ring from a linear-phase system high pass is audible.
I still don't have a firm conviction one way or another.

If you have tons of FIR taps you can try to shift the lowest frequencies forward in time. I've never had a DSP with enough taps to attempt this. Definitely worth trying. More realistic as Moore's law makes hardware cheaper and more accessible. Also, since most recordings have high pass phase shift, you could overcorrect and see if that improves things even more.

The audio magazines use a lot of words like "rhythm" and "pace" and "drive" and we all have at vague sense of what the reviewers are trying to convey. These experiments might make these terms a little more meaningful.

But what I do think is a safe compromise on system high pass, is at least let the electrical high-pass be linear phase.
So In the case of my bass reflex, I have just 4th order natural rolloff, or when running sealed, 2nd order.

I decided to see if I could make a linear-phase shelving filter, the shelf I plan use as a replacement for a BW3 as per #85.
So 19Hz, -16dB, Q=1.18

View attachment 1254129View attachment 1254130

It probably doesn't matter much in my case whether I use this shelf or the BW3, since I use them linear-phase, but I only have 16k taps @48kHz available, and I often wonder how effective they are at the very bottom. So less to linearize the better I guess....in goes the shelf.

FIR DSP may have its own peculiar combination of filter shapes (there are lots of ways to make a shelf filter; you can use multiple notch filters instead for example) and windowing profiles (Hamming, Hanning, Envelope etc) that give you the most acceptable result from the taps you've got.
I listened to the Porcupine Tree tracks. Are there particular versions than standout? I could hear some thump-slap differentiation on 2007 Blank Planet album, but it sounded more like maybe mic timing like Art was talking. Plain dunno

Title track of "Fear of a Blank Planet" is the best, it's got lots of slap AND the thump is very low, between 20-30Hz so if a system has group delay you can hear it easily. Easy experiment: Use your IIR (not FIR) DSP to add a 48dB/octave 20Hz high pass to your system, then listen to this track. Switch the 48dB High Pass in and out and see how much you can hear the group delay.

My favorite drum/transient track is probably Joe Walsh's Ordinary Average Guy....maybe because I don't have to wait for the kick drum to hit .
It's one of those tracks you have to have the volume right before you hit play LOl

Oh Hey, and a real dummy question? What's the natural rolloff of an open baffle?
If you wrap a tape measure around the speaker and measure the distance "D" from the voice coil on the back side of the woofer to the dust cap on the front side of the woofer, in inches,

then the -3dB point of an Open Baffle in Hz will be 2000/D. It will roll off 6-9 dB per octave below that point.