MazProjects 3 Way Active (pix and crude measurements included)

[mazurek]

I decided that I've made enough progress to start a speaker thread. I'm nearly done with my two diy four channel amplifiers, and I recently purchased a DCX 2496. I have two sealed 2 cubic foot bass bins with 2x RS265HF each. I have two sealed PE 0.5 cubic feet enclosures with a ScanSpeak 18W4531G and Peerless HDS tweeter. By the way, I love 4 pole speak on connectors, they really clean up my system.

The boxes are not damped yet, but I have some polyfill and acoustic foam for the bass bins and bookshelves. The rationale here is that I think I don't need to resort to fiberglass in a sealed enclosure. The subwoofers are divided internally into one cubic foot sections, I have not installed additional bracing but I think I will be putting in bracing like Zaph did in his RS315 sub build.

My current crossover points are 140 and 1700 Hz LR4, I tried to make it LR4 acoustic with combinations of shelving and butterworths based upon The Edge, Zaph, and MarkK but I need to spend some time measuring. Sub is LT'ed to -3dB at 30Hz, -6dB at 24Hz.

I have not done any far field measurements yet, and have only done limited distortion measurements. One thing I learned is that distortion measurements in box and with the driver free on the carpet can differ a bit.

I will be posting some pictures later today. Hopefully I will get some linear and non-linear distortion measurements in the coming weeks and I will be posting them.

Currently I am considering that the duel ten inch woofers may allow me to get away with a 5.25" mid for either improved polar response or less tweeter loading. I don't know if anyone has any suggestions for a mid, I'd prefer that it is paper and 4 ohm nominal (or high sensitivity 8 ohm). Some replacements I have in mind are Peerless Exclusive, or SS 15M4531K, I really can't afford to spend more than 200 a piece.

Here's a near field graph of the RS265HF.

[dlneubec]

Sounds like a great project!

FWIW, I'm using a single RSS265HF in a 31L TL in the HOSS project. It is crossed at 180hz to a pair of RS150s-4's (LR4 both ways), which are wired in series. They crossover to an RS28AS-4 at about 1580hz, also LR4.

I posted some comments in the HOSS thread about how the combo sounds.

https://www.htguide.com/forum/showthread.php?t=26065

I'm very happy with the RSS265HF and bet it would work up quite a bit higher in an all passive 3way. Here are some nearfield driver and TL terminus measurements I posted of the RSS265HF, which show the same type of extended FR you are getting.
RSS365HF in 31L TL

[Jed]

SS 15M4531K


I wouldn't recommend the above, as I think it is the worse performing Scan Speak 5" driver if considering it's highish THD throughout much of its range compared to the W series, which has a more advanced motor.

[mazurek]

SS 15M4531K

I wouldn't recommend the above, as I think it is the worse performing Scan Speak 5" driver if considering it's highish THD throughout much of its range compared to the W series, which has a more advanced motor.

I didn't know that the W series have a different motor, the only review I saw of the drive was on Troel Gravesen's site, and he came to a similar conclusion that it didn't perform as well as he wanted.

I downloaded ARTA to so some measurements, it is very easy to use.

Attached are the Stepped Sine wave distortion results for my drivers, testing distance was about 2 feet, level was around 90 dB but very poorly controlled (same volume for all drivers of different sensitivities).

[mazurek]

Based on those distortion plots, which roughly coincide with some of my manual TrueRTA harmonic distortion testing, it seems like 160Hz and 1700Hz are good xover points.

[mazurek]

Here's a picture of my new speakers...

[mazurek]

Here's some TrueRTA data with respective distances, slightly off axis (and in the case of the subs, between them)

[mazurek]

I have been tweaking with the DCX and my system is starting to gel.

I have been tossing around some further development ideas. One would be to modify this to linkwitz's version of a Duelund crossover. I realized my favorite 3 way crossover topology is nearly the same as what he describes in the Duelund section of his site. What would the tradeoffs be in terms of group delay, harmonic distortion and polar response?

I've also been thinking of ways I can move this to 2 channels per speaker instead of four, Jed's 2x Auras in series would seem to be part of the equation on that one. I've never done a passive crossover, with my current bookshelf driver selection that would be a really low impedence (which I think raises the THD on my amp), I could try a series crossover or move to the 8531G.

[JonMarsh]

A properly implemented Duelund can have very low group delay. But depending on the filter coefficient you use, the bandwidth requirement for the mid can be quite high- which is why I'm experimenting with relatively lower coefficients. The example as posted by SL is a Duelund squared, which in principle gives higher ultimate roll of rates. But not significantly so within the -20 to -30 dB range of the drivers, so the standard one seems just fine to me.

The problem I see with the specific example SL shows is the near impossibly wide bandwidth demands on the midrange- it must track on amplitude and phase to about the -18 dB level for it to work correctly, and for SL's posted example, at that filter coefficient that requires 50 Hz to 5 kHz, with a working power bandwidth I'd recommend at -12 dB (for 106 dB peaks, good behavior and low distortion at 96 dB). In the posted example, that's 80 Hz to 3.2 kHz (and keep in mind the CTC requirement at the top end). I don't know of a single driver that does that particular range well, the closest I can think of might be the Accuton C82-T8 (higher distortion than I would prefer for a "high end" solution), or a C173-T6-95, which is a little large physically for the top part of the range.

Reducing the filter coefficient and going for a slightly narrower range on the mid might be something to consider.

[mazurek]

After imagining the driver requirements and digging around Linkwitz's site, I decided that I'm going to see if I can simulate correction for the cascaded second order allpasses that happen for my LR24 3 way using Matlab. This could give me a better idea of whether it is worthwhile to pursue lower group delay, or transient perfect response.

I have an EMU sound card, and was checking to see whether it has any useful allpass delay correction plugins but I don't see any obviously. For all the supposed DSP processing power, the plug ins don't seem very useful for anything.

[SQdude]

Wouldn't it be better to aim for a transient perfect XO and then deal with the group delay via driver offset or time delay circuitry? I'm still learning so please excuse me if I'm totally wrong.

[mazurek]

I only mean to remark that it makes sense to me to explore potential benefits before adopting unfamiliar crossover topologies that present their own design conundrums, and so a DSP experiment may help to understand whether all those resources are necessary.

The problem setup assumes that I am starting with a well designed loudspeaker that is already correctly in phase, and the drivers are suitably delayed with analog all pass. I have no problem getting everything in phase (in an analog active crossover, close to in phase for a DCX because I can't cascade low/high pass filter sections), I want to see if the overall 720 degree phase wrap over time is audible.

[mazurek]

I was forwarded to this program at :

http://www.thuneau.com/

called Arbitrator

It does pretty much what I wanted to test, it requires an ASIO sound card I think. You can compensate the phase delay of standard loudspeakers using your PC, and it works as a pass through allowing you to play standard material. I measured it in ARTA, and it worked, bringing the group delay curve flat. I'm having a hard time hearing the difference between the compensated and uncompensated version though. This might have to wait until I can find an audiophile to test.

[mazurek]

My roommate, who is a normal person that enjoys listening to and playing music, could reliably tell their was a difference for 60Hz saw waves. For the sawtooth wave, I just asked him if their was a difference, and asked him to tell me which was A and B a couple of times. I then played a track he was familiar with, and I asked him to say which sounded better but it was very difficult for him, and toggled between the processed and unprocessed version. Both times he told me the version without DSP sounded better (again he did not know which was which). My short conclusion is that I can probably live with a 3 way with LR24 filters. I will try this test again on my audiophile friend later.

[Jonasz]

Very interesting speakers indeed! :T

How do you like the HDS-tweeter?

[mazurek]

Here's an extended response to your question about the HDS tweeter. I have measurements for it in system crossed at LR4 acoustic. These are with the microphone 2 feet away. I tested at 66, 85, and 95 dB/m.

As far as frequency response goes, I think its alright, the small blip shown in Zaph's and MarkK's response gets sort of washed out after cabinet diffraction and the room have their say.

By the way, I have no idea what is going crazy in the 95 dB/m test at 180 Hz. Both my loudspeakers do it, I'm guessing it the room interacting, or stuff going crazy inside the cabinet.

Overall, I'd say the HDS tweeter is a good match for the drivers I have in terms of distortion limits.

By the way, no final filter schematic yet, I've achieved something that is reasonably pleasant to listen to, but its been mainly through an iterative process of playing with the DCX, and taking more gated measurements. I'll do a more rigorous version using averaged measurements over a -20 to 20 angle and Matlab.

[mazurek]

I think I finally have my preferred filter down. I don't think I can afford the most functional version of lspCAD, so I programmed tools in Matlab. At least this was a very good refresher on my transfer function math. I can document the process or release some of these tools if there is interest.

It was a real pain correctly programming the function into the DCX, I finally figured out how all the parameters were specified (f1 for shelving lowpass, f2 for shelving highpass, etc). If there is any substantial interest I would post the filter settings to create this project using a DCX.

I'm on to the next stage of making an analog circuit board. I was wondering if I should include a bass equalizer loop, so the <140Hz frequencies could be equalized using an external device. Other options I am thinking about are one adjustable notch filter to target one main room mode. How do other people with full range box speakers deal with low frequency equalization, any devices with enough fidelity to pass a full range signal through? I know the answer people will give me is dipole, but I don't want to make any tradeoffs in low frequency output or increased baffle size.

[chasw98]

It was a real pain correctly programming the function into the DCX, I finally figured out how all the parameters were specified (f1 for shelving lowpass, f2 for shelving highpass, etc). If there is any substantial interest I would post the filter settings to create this project using a DCX.

I would be interested in seeing how you did some programming in your 2496. I have been 'fiddling with mine' to do more than just straight crossover points or shelving filters. I remember seeing a description of how someone did a cauer filter in the 2496.

Chuck

[mazurek]

Here's some brief tips on how I decomposed things. I won't be able to go into full detail until Sunday probably.

As always, things start with Linkwitz's site:

http://www.linkwitzlab.com/filters.htm

http://www.linkwitzlab.com/thor-eq.htm

The most important thing with using the DCX to simulate an analog transfer function is to record the output and make sure the transfer function matches your intended transfer function.

Here are a couple of rules I learned from messing around with the thing, and checking what I got relative to my ideal Matlab filters.

For 6dB shelving lowpass, it is specified the same way roughly as how Linkwitz shows on his site. The lower frequency f1 is specified in the DCX, and Gain=f2/f1. If a negative gain is used, the lower frequency still specifies the location, and the gain still behaves as expected = -(high freq/low freq).

For 6dB shelving highpass, the higher frequency f2 is specified in the DCX, and Gain=f2/f1. Again, this is located in the same place whether a positive or negative gain is used.

For 12dB shelving filters, they are located in a similar way to the 6dB filters of half the magnitude. I forget what I found the Q1 and Q2 to be, I think it is 0.7. I don't use 12dB shelving filters, I think it is easier to get what I want by compounding 6dB shelving filters.

I found it to be a pain to work with their fixed Q highpass and lowpass filters.
To change a 2nd order highpass Butterworth (Q=0.7) to a 2nd order Q=1, add a parametric eq at the crossover frequency of magnitude 3dB and Q=0.707.

Also, I found that on occasion I would want to adjust the Q of a box rolloff, you can adjust the Q of a box rolloff, where initial Q=0.7 to Q=0.5 by putting a parametric eq of gain-3dB Q=0.7.

Sometimes I would want to move the poles of a driver. For example, I wanted a 140Hz crossover when my midwoofer had a 70Hz rolloff. The easiest way to move the poles is if your initial F0,Q0 and target F1,Q1 have Q <=0.5. Then you can decompose the transfer function into two shelving 6dB filters. I suppose you could also work something if your Q0,Q1=0.7 easily, but I liked compounding 6dB filters better.

On one occasion I wanted a first order lowpass at one frequency and a second order lowpass at a different frequency in order to combine inductive rolloff, and driver peaking to make an acoustic 4th order. I hate that you can only specify one frequency for the crossover. In this case I moved the pole of the lowpass moving a shelving filter.

I can get into more of the math later if you want. I used Matlab to solve for and graph transfer functions. The DCX is much more powerful if you can simulate your actual transfer functions. Pretty much the only thing that will be different in my analog crossover filter transfer function is that I will cascade sections (for example I can't include the natural highpass of the subwoofer in the midwoofer filter on the DCX, but it clearly has phase effects).