TRIO: active mixer/crossover

[Guest]

Hi gentlemen;
I want to share with you some my design. Friend of mine have Marantz 7200 amp and good full range speakers, but he wants to use that speakers plus a subwoofer. Unfortunately in order to use a subwoofer he has to switch to "small speakers" so only frequencies above 200 Hz go to main stereo channels. If to switch to 'Big speakers" all band goes to main speakers, but no subwoofer signal.

I've decided to build a mixer (left, right, sub) so no matter if "big" of "small" is selected the sub will receive a signal.

Any comments/suggestions?

 

[HerculesVR]

You must use Linkwitz-Riley filter if you wanner high quality corssover .

then you can see AES application which about that.

 

[Guest]

[quote author="HerculesVR"]You must use Linkwitz-Riley filter if you wanner high quality corssover .

then you can see AES application which about that.[/quote]

What do you mean? Something like "One Hand Clapping" from Zen?

 

[lofi]

Grasshopper, relax and breath nature

maybe this would help?

 

[JohnRoberts]

The L-R is a special alignment of crossover filters. Essentially two 2-pole Butterworths stacked in series. Instead of -3 dB at tuning it is now -6dB at the transition point between bandpasses, but since the 4 pole HP and LP sections have rotated through a full 360' of phase shift the two -6 dB signals sum to unity.

This relationship is useful for higher frequency crossover points between adjacent drivers. For a sub crossover there are often other factors. I'm sure the simple filter as drawn will work (150 Hz is a common crossover frequency). It might be useful to remove some of the LF content from small speaker system when sub is in use.

JR

 

[Guest]

[quote author="lofi"]Grasshopper, relax and breath nature

maybe this would help?[/quote]

No, it won't help. In my case it is a LPF only (24 dB/Oct), while Linkwitz-Riley means LPF and HPF pair tuned such a way they give smooth frequency response in cross-over region. the same Sallen-Key filters like in my case may be used, but tuned especially. Theoretically. Practically speakers themselves have dips and peaks higher than 3 dB, also they can't be installed on the same place physically, so anyway phase shifts dependent on listening angle will exist. Though, Linkwitz-Riley is a good approach for passive LC crossovers in speakers.

 

[bcarso]

I think the summer-filter looks fine, and I slightly favor the inverting topologies to minimize common-mode distortion.

With regard to Linkwitz-Riley, besides the 4-pole design shown in the Elliot link, you can also do three and two pole ones. However with the two-pole the filter Qs for the same "perfect" summation need to be 0.5, and one driver needs to be polarity-reversed. Debate rages as to the audibility of one region being so flipped.

And as Wavebourn and Elliott point out, the assumption of time alignment is important for LR to work according to theory, something which will not be likely with a subwoofer.

 

[JohnRoberts]

Yup.. big difference between textbook crossovers and real word.

FWIW the schematic shows a 3 pole filter so in a symmetrical crossover the bandpasses would be 270' apart and not cancel, at least on axis, on paper, etc. :grin:

JR

 

[Guest]

[quote author="JohnRoberts"]
FWIW the schematic shows a 3 pole filter so in a symmetrical crossover the bandpasses would be 270' apart and not cancel, at least on axis, on paper, etc. :grin:
[/quote]

Yes, 3 - pole filter....

...if to remove the 1'st stage :green:

 

[jeth]

Anybody have any links for good reading on ther general subject of crossover networks??...and more specifically tuning crossover networks with narrow bands, something I'm wanting to do but have not found any direct references. I'm aware that there is possible interaction, and comments in an above post just reminded me of the query.

 

[bcarso]

If you raise the Z of the first stage to use 80.3k Rs and a ~6.2n C, assuming equal bias currents out of the inverting inputs the net offset due to bias currents will zero out.

The fact that the second stage has the offset current flowing in a kind of internal loop makes the analysis a bit trickier.

 

[bcarso]

[quote author="jeth"]Anybody have any links for good reading on ther general subject of crossover networks??...and more specifically tuning crossover networks with narrow bands, something I'm wanting to do but have not found any direct references. I'm aware that there is possible interaction, and comments in an above post just reminded me of the query.[/quote]

Do you mean small width passbands, or sharp slopes? And are you talking about line-level processing, or voice-coil-level networks?

 

[Guest]

[quote author="bcarso"]If you raise the Z of the first stage to use 80.3k Rs and a ~6.2n C, assuming equal bias currents out of the inverting inputs the net offset due to bias currents will zero out.

The fact that the second stage has the offset current flowing in a kind of internal loop makes the analysis a bit trickier.[/quote]

Yes, Rs is unpredictable, but anyway offset is compensated more or less (I did not measure, but hear no thump connecting/disconnecting out), plus input currents are very small.

 

[jeth]

[quote author="bcarso"]

Do you mean small width passbands, or sharp slopes? And are you talking about line-level processing, or voice-coil-level networks?
[/quote]

Sorry if I didn't make that clear... yes, small passband width and interaction between the slopes is what I meant. I had built a 4 way circuit with very narrow passbands on the two low frequency channels, roughly an octave each. Having used a spreadsheet program to calc. values and plot response it seemed the answer to the anomolies caused by interaction between the slopes of the closely spaced filters should be solved by changing gain on the low mid band. In practise this didn't seem to work so well.

So, any information on how to get better results with such narrow bands would be useful. Steeper slopes would be an obvious improvement but I'd like to stick to 24dB LR if possible.

 

[bcarso]

[quote author="jeth"]Sorry if I didn't make that clear... yes, small passband width and interaction between the slopes is what I meant. I had built a 4 way circuit with very narrow passbands on the two low frequency channels, roughly an octave each. Having used a spreadsheet program to calc. values and plot response it seemed the answer to the anomolies caused by interaction between the slopes of the closely spaced filters should be solved by changing gain on the low mid band. In practise this didn't seem to work so well.

So, any information on how to get better results with such narrow bands would be useful. Steeper slopes would be an obvious improvement but I'd like to stick to 24dB LR if possible.[/quote]

This is a pretty decent source: http://www.linkwitzlab.com/filters.htm

 

[bcarso]

[quote author="Wavebourn"][quote author="bcarso"]If you raise the Z of the first stage to use 80.3k Rs and a ~6.2n C, assuming equal bias currents out of the inverting inputs the net offset due to bias currents will zero out.

The fact that the second stage has the offset current flowing in a kind of internal loop makes the analysis a bit trickier.[/quote]

Yes, Rs is unpredictable, but anyway offset is compensated more or less (I did not measure, but hear no thump connecting/disconnecting out), plus input currents are very small.[/quote]

What I meant: Make R4 80.3k, (80.6k nearest standard value) and C1 6.2n, and with equal bias currents out of the inverting inputs of A1a and A1b, the overall circuit output voltage will be zero at d.c., assuming zero opamp input offset voltages.

That is, my "Rs" meant resistors R1, R2, R3, R4, not "Rsource". I think it was NewYorkDave who urged doing without apostrophes where they are not needed to indicate plurals of capital letters. But since rendering subscripts is a little awkward, there can be ambiguity.

Of course in most systems a little d.c. offset is going to get highpassed out down the line anyway, so this is mostly academic. But you asked at one point (and then deleted I guess) why you didn't have the traditional bias current compensating resistors in the noninverting inputs, which prompted the analysis.

 

[Guest]

[quote author="bcarso"][quote author="Wavebourn"][quote author="bcarso"]If you raise the Z of the first stage to use 80.3k Rs and a ~6.2n C, assuming equal bias currents out of the inverting inputs the net offset due to bias currents will zero out.

The fact that the second stage has the offset current flowing in a kind of internal loop makes the analysis a bit trickier.[/quote]

Yes, Rs is unpredictable, but anyway offset is compensated more or less (I did not measure, but hear no thump connecting/disconnecting out), plus input currents are very small.[/quote]

What I meant: Make R4 80.3k, (80.6k nearest standard value) and C1 6.2n, and with equal bias currents out of the inverting inputs of A1a and A1b, the overall circuit output voltage will be zero at d.c., assuming zero opamp input offset voltages.

That is, my "Rs" meant resistors R1, R2, R3, R4, not "Rsource". I think it was NewYorkDave who urged doing without apostrophes where they are not needed to indicate plurals of capital letters. But since rendering subscripts is a little awkward, there can be ambiguity.

Of course in most systems a little d.c. offset is going to get highpassed out down the line anyway, so this is mostly academic. But you asked at one point (and then deleted I guess) why you didn't have the traditional bias current compensating resistors in the noninverting inputs, which prompted the analysis.[/quote]

Ah yes, what you mean is better, however. I did not bother to calculate compensation of shifts with DC feedback in mind in the 2'nd stage so made an error though it is not significant. Thanks a lot, if I decide to build one more copy I'll take your advise.

 

[Guest]

[quote author="jeth"][quote author="bcarso"]

Do you mean small width passbands, or sharp slopes? And are you talking about line-level processing, or voice-coil-level networks?
[/quote]

Sorry if I didn't make that clear... yes, small passband width and interaction between the slopes is what I meant. I had built a 4 way circuit with very narrow passbands on the two low frequency channels, roughly an octave each. Having used a spreadsheet program to calc. values and plot response it seemed the answer to the anomolies caused by interaction between the slopes of the closely spaced filters should be solved by changing gain on the low mid band. In practise this didn't seem to work so well.

So, any information on how to get better results with such narrow bands would be useful. Steeper slopes would be an obvious improvement but I'd like to stick to 24dB LR if possible.[/quote]

I would go this way: