SU551-RS28F – a 2-way DEQX system loudspeaker measurement study

[kipman725]

Originally Posted by doug20

Could you comment on the steep linear phase slopes used. Any draw backs? I have wanted the DEQX for 4 years now. I have the DCX, MiniDSP, Hypex DSP solutions and Im just waiting for the best DEQX deal. I can do a 60dB slope with cascading functions on the MiniDSP but I do not think I can do a linear phase Slope

I look forward to your measurements.

as far as I am aware you can use custom biquad functions in minidsp if so you can do a FIR filter. Use matlab according to wikipedia octave should also have the relevent algorithms built in. Parks-McClellan method works quite well for designing FIR filters from my experaince.

[D OB G]

Hi Goran

I only use the Earthworks reference mic.
The baffle step is difficult.
My speakers are dipoles, so it is not an issue.
Floor bounce might be.
It will require different correction for different listening positions (distances), and height of speaker(s).
I think it is best to get at it at source.
I get away with it (I hope!) by using a vertical array of drivers up to 400Hz.
Seems to work (not your 2-way solution though- although it can be if you use mid/tweeter such as the Jordan Jxr6hd or Alpair, or alternatively, maybe, a higher crossover to a waveguide e.g. an econowave).
I'd have to have a bit more of a think about the baffle step issue.
Wayne Parhams's solution would also work, as I believe would Mandrakes's.
I'm not sure that the coincidence issue matters, as when measuring from one position, it still works no matter how far apart the drivers are (they are still after all actually independent measurements). BTW Alan Langford approves.
With baffle diffraction, I don't think DEQX can remedy that.
It has to be corrected at source.
I believe you're right that the speaker MUST be designed so that the off-axis response is consistent (not necessarily identical) with the on axis response i.e. some sort of CD.

David

[boris81]

Originally Posted by kipman725

as far as I am aware you can use custom biquad functions in minidsp if so you can do a FIR filter. Use matlab according to wikipedia octave should also have the relevent algorithms built in. Parks-McClellan method works quite well for designing FIR filters from my experaince.

I'm interested in learning more about this approach. Can you point me to a sample implementation. I'm mostly interested in a flat phase Linkwitz-Riley crossover. What would be the best way to arrive there?

[gornir]

Measurements and off-axis responses:

As I mention earlier in this thread, I made four different (0, 15, 22.5 and 30 degrees) correction filters. Each of them have a ruler flat response when measured at the same angle as the correction filter, but what happens with the frequency response when measured off-axis?

The baffle step and the baffle diffraction have to be taken into account in the design of the loudspeaker.

The baffle step is quite easy to handle in the DEQX. You simply adjust the amount of the baffle step compensation by adjust the lower limit of the correction filter. In this case, 250Hz for a full baffle step compensation and e.g. 500Hz for a half baffle compensation or somewhere in between according to personal taste.

The baffle diffraction is the major issue in this design. The baffle diffraction effect can be reduced by arrange the drivers optimally on the baffle, by enclosure shape and design, but never completely eliminated. Different drivers are also more or less sensitive to diffraction.

The following described measurements make this baffle diffraction and its effects more clear.

0 degrees correction filter:
Picture1: Blue = 0 degrees, Red = 15 degrees, Green = 22.5 degrees
Picture2: Blue = 30 degrees, Red = 45 degrees, Green = 60 degrees

Note: Looks nice in the 0 degrees angle, but the correction filter overcompensates (2500-3500Hz) in the off-axis frequency response.

15 degrees correction filter:
Picture3: Blue = 0 degrees, Red = 15 degrees, Green = 22.5 degrees
Picture4: Blue = 30 degrees, Red = 45 degrees, Green = 60 degrees

Note: Looks a bit better and the correction filter doesn’t overcompensate as much in the off-axis frequency response.

22.5 degrees correction filter:
Picture5: Blue = 0 degrees, Red = 15 degrees, Green = 22.5 degrees
Picture6: Blue = 30 degrees, Red = 45 degrees, Green = 60 degrees

Note: Here the correction filter overcompensates a bit higher up in the frequency response (4000-5000Hz).

30 degrees correction filter:
Picture7: Blue = 0 degrees, Red = 15 degrees, Green = 22.5 degrees
Picture8: Blue = 30 degrees, Red = 45 degrees, Green = 60 degrees

Note: Here the correction filter makes a dip in the frequency response (2500-3500Hz) and with a slight rising response higher up in the frequency.

I listened to each of the four correction filters for a couple of day before I made these measurements to see which correction filter I found sounds the best. In the listening test the speakers had the same setup with no toe-in and a listening angle of approximately 15-22.5 degrees.

Subjectively I found the 15 or the 22.5 degrees correction filter to sound the best and the 0 degrees correction filter to sound the worst. By looking at the off-axis responses I can see why. None of the correction filters did sound bad, but they all had a different sonic character.

The DEQX unit have other features, e.g. a digital EQ that I haven’t used in this test and which can be used to tailor the frequency response and sound further.

Now when I have a deeper knowledge of how these specific driver units in this specific enclosure behave, I can redo the measurement with my recently bought “reference” microphone and tailor the sound with the EQ function and then redo my test measurements to verify the design.

I will be back with a final design, stay tuned….

[jcga]

Originally Posted by kipman725

as far as I am aware you can use custom biquad functions in minidsp if so you can do a FIR filter. Use matlab according to wikipedia octave should also have the relevent algorithms built in. Parks-McClellan method works quite well for designing FIR filters from my experaince.

Well, are you sure ?
custom biquads will only let you design custom IIR filter that's all.
FIR will need a completely different technique

Best from France
Jean-Claude

[kipman725]

Advanced Biquad programming | miniDSP

Pretty sure as a FIR filter is charcterised by polls only a z=0 which is clearly posible from the stated biquad transfer function. If your a* coefficiants are zero you should have an FIR filter.

*for the linear phase characteristic that is desirable the b coefficiants should be symetrical before the filter is decomposed into biquads.

[kipman725]

Originally Posted by boris81

I'm interested in learning more about this approach. Can you point me to a sample implementation. I'm mostly interested in a flat phase Linkwitz-Riley crossover. What would be the best way to arrive there?

A linkwitz-Riley crossover is as far as I'm aware not a linear phase filter therefore there is no advantage apart from stability to using a FIR filter. The Linkwitz-Riley is just a filter that ensures that drivers are in phase at the crossover point and add with 0dB gain.

As for designing FIR filters look at the matlab help for the following commands:
firpm
firpmord

However looking at my lab notes I don't think the mini dsp is powerfull enough to implment good enough FIR filters as I used a 51st order filter to make one pretty poor bandpass filter and the mini dsp only has 68 biquads which would be pretty quickly used up.