Linkwitz-Riley 24dB/octave Phase Coherent Crossover

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Dear Friends,

I need your opinions about this project :
24dB/octave Phase Coherent Crossover based by Rod Elliot project 09.

This crossover it would be better becouse combine the quality of 24db/octave Linkwitz-Riley filters with best phase-coherency and 24db/octave subsonic and ultrasonic filters.



  • linkwitz-riley  24db octave crossover phase coherent.jpg
    linkwitz-riley 24db octave crossover phase coherent.jpg
    80.7 KB · Views: 2,517
I built the LR Xover of Rod Elliott and it worked fine, however, in the sound department I found it lacking (ie cold though not bright, mostly not very musical or involving), I can't really be more specific than that .
At this time I am ready to concede that it is probably due to the effect of the op amps not to the LR concept itself although I have heard that people temd to like lower order XOVER arrangment better.
Schematics looks very complex , but this is nothing compare with a very good mix console where the sound is routed by 8 - 20 integrated circuits.
In fact , the music (80%) on CD or LP it is not such HI-FI to our expectations , so a 4 or 5 another circuits don't change it so much.

Fake STEREO ( a smart trick using PAN pot.) or,
Fake VOICE tonality maked by EQ, COMPRESOR, etc.

This is what we hear allmost the time when play music.

Anyway , if it is something better ( analog ) , I am very excited to see.

Regards !
I always recommend this crossover Since it comes from Nelson Pass, you can't go wrong with it. It's two way but by combining 2 you can have three way. It's using discreet op amps, and by updating semiconductors you can improve it even more. Actually, the whole idea is not that far from his current design (just few less variations;) ). I've built it and I like it.
I ordered PCB's for Rod Elliot's Project 09 on Friday... Haven't received them yet, but his descriptions say that the same boards may be used for 12db/octave or 24db/octave. I'm hoping that parts substitutions will allow me to avoid the cold sound Grataku reported.

I built a three-way designed by Mr. Linkwitz (as part of the Phoenix project) using OPA2134's and am happy with it. Any comments on the OPA2134? Any suggestions for a better opamp for the next XO? BTW, this XO will be used with rear speakers, not as a replacement for Mr. Linkwitz's boards.
let's not get into the whole argument about how many opamps there are in a mixer, please. That is a real tragedy and we don't need to be reminded of that. I probably own a total of 3 CD that "sound" good the rest of them...never mind you got to listen to something. All my classical music CD are so bad that is almost impossible for me to listen to standing in front of the stereo, I mostly play them as a background. However, when I listen to my "good' CD I need a certain level of performance that quite frankly I wasn't getting from the LR xover setup.
The technology is there to produce good sounding CD the problem is with time and money and good audio engineers.

did you actually build the phoenix? How do you like it? I use some lm6172, I think, There are many other "audio" grade opamps out there. One only needs some money to experiment with various models. There are several lists of recommended OpAmp available on the web and on this site too.
Joined 2001
Paid Member
A LR 4th order filter means that all the drivers are connected in phase, but it is not phase coherent thru the XO point. I have a Marchand which is a state-variable 4th order XO. It definetly adds a veil when in circuit. It uses a lot fewer op-amps than the posted scema. Could you not put a 2nd order filter on each op-amp and get away with half the active parts in the filter stages?

I plan on replacing my Marchand with a home brew subtractive XO (using tubes if i can figure it out). Or with just a 1st order PLLXO (might need more extended woofers thou).

"I have a Marchand which is a state-variable 4th order XO. It definetly adds a veil when in circuit. It uses a lot fewer op-amps than the posted scema. "
Uhh, the high pass output goes through two more op amps in the Marchand design, you just don't want to see them. All the noise, distortion, phase shift, etc of the four op amps in the low pass show up in the high pass when the output of the low pass is summed with the input signal. The Marchand XM16 built as a 24dB LR sounds better than the derived high pass state-variable designs.

"Could you not put a 2nd order filter on each op-amp and get away with half the active parts in the filter stages?"
Uhh, those are all second order filters already.

"My only comment is your use of 1uF and 100nF caps: what kind of tolerence are you aiming for? Sallen & Key filters are rather suseptable to variations in the RC constant caused by component tolerence.
I built something similar over the summer and used 1% polystyrene caps and 1% metal film resistors throughout."
Gee, where can I find some of those 1µF polystyrene caps?

djdan, your crossover is not configured properly.
"Click here
if you really want to see the old version, which is incorrect."
"This version
may not look all that different from the original in Figure 1, but the performance difference is very noticable. "

"For some obscure reason, I've never been able to warm
up to the sound of 24 dB/oct. It can measure great, and
I've tried it many times, but have always gotten better
sonic results with fewer poles. Go figure."
I have found that 99% of commercial units use the derived high pass with the fig1a. midrange output. These can't sound good. The best I have ever heard was a 12dB Bakgaard, followed by the 24dB LR. I have never been able to get a 12 or 18dB Butterworth to sound 'right'.

Even with a textbook perfect fig1b. LR it doesn't always sound 'right'. What the drivers are doing must be accounted for too. Many times I have had to use a 12dB Q=1 lowpass with a 18dB Butterworth high pass to get a 24dB LR transfer function.'4031321'.WKU.&OS=PN/

Bakgaard patent, the JAES article is much more informative.
An active x-over was featured in Electronics World some time ago. It was called PRECISE ACTIVE X-OVER. Perhaps, someone could help with providing the schematic. (I still don't know how to post schematics).

It is not a subtractive filter. It uses a Cauer filter, followed by a second order LR. Together it has a 6th order profile and exhibits phase coherence within the passband. The group delay is minimal. I have taken the High Pass output and again split it to form a three way filter.

The sonics of this filter is of a very high quality. Used in a three way active PA set up the sound is marvellous. I have compared this with commercial offerings which stand no comparison. A 3-way Linkwitz-Riley with All Pass networks for time delay, which shows great measurements and oscilloscope shots pales in performance.

Yes the op-amps used do make a difference to the resolution, just a wee bit. If one can replace each of the opamps with discrete opamps (like the ones described by Nelson Pass) or even with FETs duly taking care of the input buffer stages, in my opinion it would be one of the best x-overs.
Hi sonnya

I assume that these filters might be good and I find the idea of using curent sources as emitter loads quite cool.

But the circuit is definitely not non-feedback !
You have positive feedback from the emitter to the first node of the two cascaded RC-sections (i.e. a classic Sallen-Key topology).
OTOH this is not a bad thing at all because otherwise it would be impossible to achieve higher Q-values than 0.5 when using only resistors and capacitors. If an "active" filter uses capacitors and coils then the higher Q values could be achieved without positive feedback. If someone uses filters with Q values of 0.5 or below he can omit the positive feedback even when using capacitors and resistors only.


Identical cascaded 2nd order sections

Don't know a real lot about filters but I thought it a bit unusual to have identical 2nd order sections cascaded. The unity gain Sallen-Key sections used will each have a -3dB point of say 1kHz, but string two of them together and you get a -6dB point of 1kHz. It eventually rolls off at 24 dB per octave but the initial rollof is very slow. The damping of each section needs to be set differently to get an *overall* damping of 0.707 which is what each individual (Butterworth) section has now.

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.