Straight Line CBT array

Jump to Latest
G

goldyrathore

Member
Joined 2002
#1
How to implement a straight Line CBT array?

Hi All

I am contemplating a small straight line passive CBT array for personal home stereo use. I ve read the Harman patent explaining the straight line passive CBT at https://docs.google.com/viewer?url=...com/pdfs/dbc4f6e813a18db71b87/EP2247120A2.pdf.

The patent provides a brief explanation of the passive component values as quoted below

"The configuration of the stages in FIG. 2 is recognizable to those of ordinary skill in the art to be a low pass filter. While the topology is the same as a low pass filter, the values of the components are radically different. The component values are mistuned. That is, the component values are sized to create flat group delay with frequency, which is not done with low pass filters. The component values are also sized to create relatively flat attenuation over a broad frequency range. As shown in FIG. 5, the first 4 or 5 transfer functions (from the center out) are flat. The group delay along the ladder is cumulative as is seen in FIG. 4."

However, the below is still not clear
a) group delay
b) L and C values
c) What does it mean to say that the LPF is mistuned

I would appreciate if someone can explain how to implement the CBT.

Thanks,
Goldy
 
Last edited:
#2
While it would work in theory I see mainly one problem:

I predict that this array will have the same problem as the R2R Dacs of the ye olden days. The error % of the inductor / cap are propagated further down the line. If my naive math is correct and you have say 10% cap / inductor then the accumulated error at the 9th section is 236%. The most tolerant inductors I've found with reasonable price are at 3% but that's still 30%, and the caps I've found are usually worse. The group delay depends on the inductors / caps being the right numbers so just this might mess it all up.

I think your best bet is to get say a MiniDSP2x8 or a Hypex DLCP then a lot of cheap amps and run them all active. Say something like a https://www.hypexshop.com/DetailServlet?detailID=4930 for amps. You could possibly do a hybrid though. If you split the line of say 9 into 3 parts of 3 passives, then the accumilated error of the passive XOs shouldn't be as big an issue as with one long.

Or, if you have a single ended array you could just make it curved of course, much cheaper and much more simple.

EDIT: Lets not forget that the actual speaker drivers vary a lot too, unless we are talking about a SEAS array but then we are talking about big money right there =) If you do it curved then the problem isn't significant anymore as you combine lots of drivers into a large section and the individual differences will probably cancel each other out.
 
Last edited:
#3
goldyrathore said:
However, the below is still not clear
a) group delay
b) L and C values
c) What does it mean to say that the LPF is mistuned
I would appreciate if someone can explain how to implement the CBT.
Click to expand...

With all those passive components it might be cheaper to simply make a floor to ceiling (infinite due to reflections) line array. This also has constant beamwidth.

Mistuned means that the inductor and capacitor values don't correspond to the same frequency. In this case, it probably means the capacitors are smaller than normal. In a sense this design is very much like a 2.5 way speaker, although in this case it is more like a 1.5.5.5.5.5.5.5.5.5.5. All of the drivers are used at low frequencies, and progressively fewer are used as the frequency goes up.

In order to design this, you would need to buy or write software for designing arrays. You set directivity by how many of the drivers are operating at the frequency of interest and scale it from there.
 
#4
Ron E said:
With all those passive components it might be cheaper to simply make a floor to ceiling (infinite due to reflections) line array. This also has constant beamwidth.
Click to expand...

And to build upon this: as long as it is shaded the only advantage of a delayed CBT is that it has more wide dispersion above and below the line array length. But if is floor to celing then like Ron says it won't matter. I'd take a look at Don Keeles keynote speech and look at his data, it's an interesting presentation.
 
#5
OllBoll said:
The error % of the inductor / cap are propagated further down the line. If my naive math is correct and you have say 10% cap / inductor then the accumulated error at the 9th section is 236%. The most tolerant inductors I've found with reasonable price are at 3% but that's still 30%, and the caps I've found are usually worse. The group delay depends on the inductors / caps being the right numbers so just this might mess it all up.
Click to expand...

I intend to hand wind the inductors so can achieve good tolerance. Also for each cap, if two are used paralllel then a tight tolerance can be achieved
 
Last edited:
#6
OllBoll said:
Or, if you have a single ended array you could just make it curved of course, much cheaper and much more simple.
Click to expand...

I cant do this since to achieve narrow horizontal directivty, I need to put it in a waveguide. Similar to this image A new center channel for the Octagon. Its really hard to curve a waveguide hence the need for a straight array. Didnt mention this earlier thinking its not relevant.
 
#7
goldyrathore said:
I intend to hand wind the inductors so can achieve good tolerance. Also for each cap, if two are used then a tight tolerance can be achieved
Click to expand...

That would solve that issue.

As long as you are OK with tinkering with values and have a measurement microphone it seems a doable venture. I guess the last problem is that one would need some simulation software for the delay, but at least some program must be able to do that.

I've even thought of some ideas myself of doing a partial array with the middle driver active on one channel and then 3 banks of drivers on another channel. Then I would only need to implement two banks of passive delays instead of the long line as in the paper. If you find a good way to discover the component numbers for a given delay I'd be very interested to hear of it 😀

Though in the first stage I'd probably go active on all banks to discover what delay I want and then try to reproduce it with passive filters.
 
#9
goldyrathore said:
I cant do this since to achieve narrow horizontal directivty, I need to put it in a waveguide. Similar to this image A new center channel for the Octagon. Its really hard to curve a waveguide hence the need for a straight array. Didnt mention this earlier thinking its not relevant.
Click to expand...

In my idea it is similar, I would want a dipole and so it would be akward to make it curved. It could be possible to do it curved and stuff damping material but it would probably perform better if it was flat.
 
#10
Ron E said:
Mistuned means that the inductor and capacitor values don't correspond to the same frequency. In this case, it probably means the capacitors are smaller than normal. In a sense this design is very much like a 2.5 way speaker, although in this case it is more like a 1.5.5.5.5.5.5.5.5.5.5. All of the drivers are used at low frequencies, and progressively fewer are used as the frequency goes up.

In order to design this, you would need to buy or write software for designing arrays. You set directivity by how many of the drivers are operating at the frequency of interest and scale it from there.
Click to expand...

A 2.5 way does not achieve a constant delay as needed for a CBT. The .5 is only configured with an intended cutoff and slope in mind.

As per Fig 4 and Fig 5 in the patent the LC ladder network introduces a fixed attenuation across the entire band and fixed attenuation just like a normal CBT would. Whats not mentioned is: how to calculate the values?
 
Last edited:
#12
The patent is pretty explicit and gives enough details to copy it, including schematic values.

Mistuned means that you aren't achieving classic corner shapes (i.e. Butterworth). You can see that each section uses a different inductor value but the same capacitor. Q will drop as you work your way across the line from center to end.

I don't see how compounding errors would apply here. Most errors will be random in direction. Also inductor values double every 2 steps from left to right, so each new inductor will tend to swamp the previous ones. If the errors aren't random then you would just be scaling the progressive effect one way or the other.

Group delay refers to the phase shift that accumulates from the center elements to the outer. It is the electrical counterpart to the set back of the usual curved CBT. There is a progressive response roll off tied to the delay line as well. The line gets shorter at high frequencies.

Note that his schematics show an 8 ohm resistor for every tap. The fine print shows that this is R Load: "Actual Load", meaning the resistors should not be used but the drivers should approximate them as 8 ohm loads (or it can be scaled if needed).

Magnitude scaling (alone) can be useful and will both smooth the vertical performance and increase vertical coverage. See my McIntosh paper for details of that.

David S
 
#13
speaker dave said:
The patent is pretty explicit and gives enough details to copy it, including schematic values.

Mistuned means that you aren't achieving classic corner shapes (i.e. Butterworth). You can see that each section uses a different inductor value but the same capacitor. Q will drop as you work your way across the line from center to end.

I don't see how compounding errors would apply here. Most errors will be random in direction. Also inductor values double every 2 steps from left to right, so each new inductor will tend to swamp the previous ones. If the errors aren't random then you would just be scaling the progressive effect one way or the other.

Group delay refers to the phase shift that accumulates from the center elements to the outer. It is the electrical counterpart to the set back of the usual curved CBT. There is a progressive response roll off tied to the delay line as well. The line gets shorter at high frequencies.

Note that his schematics show an 8 ohm resistor for every tap. The fine print shows that this is R Load: "Actual Load", meaning the resistors should not be used but the drivers should approximate them as 8 ohm loads (or it can be scaled if needed).

Magnitude scaling (alone) can be useful and will both smooth the vertical performance and increase vertical coverage. See my McIntosh paper for details of that.

David S
Click to expand...

Hmm, that makes sense that accumulating errors isn't an issue with increasing inductor values.

I just thought of a way to calculate the values to get the wanted delay:

The delay is based on electrical slopes and not values,
  • First we assume that we used impedance correction networks so our drivers are close enough to resistive load.
  • Secondly we use some spreadsheet to figure out what acoustical slopes give us the delay we want.
  • We hook up some simulation software and then reproduce those slopes with inductors and caps.

I guess one difference in this CBT is that the delay network is not true all passes but lowpasses of high frequency which makes it all more easy.
 
#14
I could not find the mapping of the arc delays to the caps and inductors, however I did find the values to an existing 20" tall 40 degree beamwidth array. Its the JBL-50LA CBT array.
The values from the service manual are attached.
 

Attachments

  • CBT-40.jpg
    CBT-40.jpg
    12.8 KB · Views: 510
#15
In order to use the same values of components, I can make mine too of the same height and beamwidth.
However I have a doubt, the JBL uses larger size drivers and hence only 8 of them to fit in the 20" height. So it uses 4 sets with 2 drivers in each set. I intend to use smaller drivers and hence need to use 24 drivers to fit the height which makes 6 drivers in the same 4 sets as the JBL. So 6 drivers with the same delay and attenuation will only loosely emulate the CBT arc. The delays are too coarse. For CBT-50 it probably does not matter much since its a pro product where listening distance would be more. I intend to use mine for home listening at 6' to 12'. The coarsely stepped delays would cause some issues.
On a second thought the no of components are very less and may be not hard to implement.

Any thoughts?
 
Last edited:
#16
Are tapped inductors fine for this application?

I am soon winding the inductors. Is it ok to have a single inductor (1.4mH) with taps for the needed 0.1mH, 0.3mH and 1.0mH just because its easier to do it? Normally the inductors are physically oriented to avoid mutual coupling which is not possible the way I intend to make. Does it matter in this application?

Thanks,
Goldy
 
#19
goldyrathore said:
I am soon winding the inductors. Is it ok to have a single inductor (1.4mH) with taps for the needed 0.1mH, 0.3mH and 1.0mH just because its easier to do it? Normally the inductors are physically oriented to avoid mutual coupling which is not possible the way I intend to make. Does it matter in this application?

Thanks,
Goldy
Click to expand...

Hard to predict what the coupling between inductors will be. The best approach is the X, Y, and Z orientations in a repeating pattern down your board. Otherwise just have adequate spacing.

You can wind a 1.4mH and tap it and then you will have the inductance from end to tap, but I don't see how in the world that is easier? You are going to have to calculate or measure where to put the tap (how many turns from start). Once you find that point then STOP WINDING.

Regards,
David
 
You must log in or register to reply here.
Top Bottom