KEF LS50's converted to active with brickwall linear phase crossovers via rephase and camilladsp!

[Razorhelm]

Thought I would tell you guys about the speaker system I have built over lockdown.
Particularly as I have used camilladsp for the crossovers and eq and I didn’t find much about that online outside a thread on diyaudio.com by the amazing author of the software helping people with it and the github he stores it on.

https://www.diyaudio.com/forums/pc-...iir-fir-engine-crossovers-correction-etc.html

https://github.com/HEnquist/camilladsp
It’s a pair of LS50’s converted so I can bi amp and bypass the crossovers with a couple of 12” subs for the low end

I started with some KEF LS50’s, which I opened up, disconnected the crossovers and individually wired the drivers, I ran the wires out of the port on the back. This was the most nerve racking part as I wondered if I would destroy the best speakers I have ever owned! Thankfully all went well! Added some protective capacitors for the tweeters.

For the low end I used two Kenwood KFC-W112S 12” subwoofer drivers in sealed enclosures.

To provide the 6 channels of amplification required I am using a second hand Harmon/Kardon AVR 255, I was really pleased this worked as it is a great looking AVR!

The camilladsp program runs on a Raspberry Pi 3 B+. It applies linear phase filters based on filters calculated by REW and converted into convolution files in rephase for my room correction and phase correction, the linear phase crossovers are also created in rephase. I cross at 150hz and 2200hz.

Input is via an optical in on a cheap usb sound card, the output is via a Sound Blaster X-Fi Surround 5.1 Pro V3 addressed by camilladsp at the alsa layer in linux so hopefully free of any OS interference!

Here is the Frankenstein's monster itself!

It sounds fantastic! Best system I have ever had by a long way and measures pretty well too!
This is L and R at the listening position (REW with var smoothing), not perfect but I'm pretty pleased! (except that bloody null on the right!)

Curious as to what people think and what tweaks people think I could do to improve?

 

[q3cpma]

Very cool! Do you have distortion (THD and/or IMD) measurements before and after, to see if the active crossover did something here?

 

[abdo123]

Is there something wrong with the LS50 for you to ditch its own crossover? It's a very well performing speaker on its own.

It's not very clear to me what's the point of this project.

 

[q3cpma]

Is there something wrong with the LS50 for you to ditch its own crossover? It's a very well performing speaker on its own.

It's not very clear to me what's the point of this project.

I wouldn't call it "well performing" when seeing this https://www.audiosciencereview.com/forum/index.php?threads/kef-ls50-bookshelf-speaker-review.11144/
Distortion is impressive, though; except that blip at the crossover point without any 2nd or 3rd peak, implying high order harmonics.

 

[dasdoing]

the linear phase crossovers are also created in rephase.

I also use Rephase for linear phase corssovers. what settings do you use? brickwall with limited taps?

 

[andreasmaaan]

Curious as to what people think and what tweaks people think I could do to improve?

Well done Very nice idea crossing the speaker over to a woofer and re-designing the XO.

Just a few thoughts:

Firstly, have you taken quasi-anechoic on- and off-axis measurements? If the only measurements you've used to design the crossover are steady-state measurements taken at the listening position, I would suggest taking quasi-anechoic measurements of the raw driver responses and then basing the crossover and EQ at mid-high frequencies on that. The direct sound is perceptually more important than the reflected sound field at these frequencies.

Secondly, have you experimented with gentler (but still relatively steep) crossover slopes? Admittedly, being a coaxial design, the degree of pre-ringing likely won't be so extreme, particularly between the mid and tweeter, but any kind of brickwall filter always seems unnecessary to me, with possible drawbacks and no clear benefit in comparison to say 48 or 96dB/octave (-ish) filters, for example. (EDIT: there are also various filter types that use a relatively shallow overlap region around the XO frequency and then roll off sharply after that, i.e. seeking to strike a balance between minimising potentially audible pre-ringing and at the same time steeply filtering out-of-band content.)

In particular, there's just no need to brickwall the responses at 150Hz. The wavelength down there is over 2m, while the C2C distance you have there looks to be <30cm. You could use virtually any crossover slope at that frequency/C2C distance and be just about guaranteed not to have to worry about off-axis cancellation. In terms of limiting midwoofer excursion, any slope 24db/octave or greater should be perfectly adequate at that frequency and for a midwoofer that size, IMO.

 

[Razorhelm]

Very cool! Do you have distortion (THD and/or IMD) measurements before and after, to see if the active crossover did something here?

I have THD, Active is red (with eq and sub), passive is green (just LS50, no eq or sub), one thing to note the actives were about 5db louder and the amp noise is louder in the actives as there is no crossover to attenuate it (I think it would be interesting on amp with adjustable gain to reduce it and see if that noise is reduced). Overall it seems it doesn't improve THD(ignore the 600hz spike, that's a light fitting ringing!) .

How do you measure IMD in speakers? I could measure it in the new system if you like?

Is there something wrong with the LS50 for you to ditch its own crossover? It's a very well performing speaker on its own.

It's not very clear to me what's the point of this project.

Nothing wrong with the LS50 great driver and cabinet! That's why I picked it! I like the flexibility of active, I can experiment with crossovers, linear phase etc I also think separating up the frequencies my amps are producing is beneficial.
I want to build a good system but it's about learning too.

I also use Rephase for linear phase corssovers. what settings do you use? brickwall with limited taps?

Brickwall with 16384 taps, is unlimited taps possible?

Well done Very nice idea crossing the speaker over to a woofer and re-designing the XO.

Just a few thoughts:

Firstly, have you taken quasi-anechoic on- and off-axis measurements? If the only measurements you've used to design the crossover are steady-state measurements taken at the listening position, I would suggest taking quasi-anechoic measurements of the raw driver responses and then basing the crossover and EQ at mid-high frequencies on that. The direct sound is perceptually more important than the reflected sound field at these frequencies.

Secondly, have you experimented with gentler (but still relatively steep) crossover slopes? Admittedly, being a coaxial design, the degree of pre-ringing likely won't be so extreme, particularly between the mid and tweeter, but any kind of brickwall filter always seems unnecessary to me, with possible drawbacks and no clear benefit in comparison to say 48 or 96dB/octave (-ish) filters, for example. (EDIT: there are also various filter types that use a relatively shallow overlap region around the XO frequency and then roll off sharply after that, i.e. seeking to strike a balance between minimising potentially audible pre-ringing and at the same time steeply filtering out-of-band content.)

In particular, there's just no need to brickwall the responses at 150Hz. The wavelength down there is over 2m, while the C2C distance you have there looks to be <30cm. You could use virtually any crossover slope at that frequency/C2C distance and be just about guaranteed not to have to worry about off-axis cancellation. In terms of limiting midwoofer excursion, any slope 24db/octave or greater should be perfectly adequate at that frequency and for a midwoofer that size, IMO.

Yes I have but I actually just used the crossover point KEF give for LS50 drivers, I have heard that the woofer cone can fail catastrophically so I didn't feel comfortable messing with it high SPL in its breakup zone felt like a good way to kill it!, I will have to give that a go in terms of eq'ing, i have only based on listening position so far.

I have experimented with gentler crossovers but honestly I can't hear any difference! I'm not blessed/cursed with golden ears
I used brickwall because that way I don't have to worry much about driver driver cancellation in the frequency overlap, from what your saying having the drivers close enough together prevents this?. I'll take another look at 150hz crossover maybe there is preringing and I just don't know what I'm listening for! Any hints?

 

[Music1969]

I started with some KEF LS50’s, which I opened up, disconnected the crossovers and individually wired the drivers, I ran the wires out of the port on the back. This was the most nerve racking part as I wondered if I would destroy the best speakers I have ever owned! Thankfully all went well! Added some protective capacitors for the tweeters.

Hi do you have photos of the inside ?

And can you share a photo of how the LS50 looks at the back now?

Wires going through the port seems like a good idea - don't need to drill into cabinet anywhere.

Is there easy access to drivers , to install longer wires if needed? Replacing the original wires

 

[dasdoing]

Brickwall with 16384 taps, is unlimited taps possible?

no it is not, because when you move your head you will have pre-delay.
I actualy use 4096, because that is close to the default shape I saw in Acourate. you never had problems with pre-delay with 16384 taps?

 

[Music1969]

Sound Blaster X-Fi Surround 5.1 Pro V3

Hi, how do you get 6 channels from this sound card?

I can see 4 channels for the 2 x RCA jacks and stereo rear 3.5mm jack.

But the C/sub 3.5mm jack is only single channel ?
So 5 channels ?

Or that is stereo also?

 

[Razorhelm]

Hi do you have photos of the inside ?

And can you share a photo of how the LS50 looks at the back now?

Wires going through the port seems like a good idea - don't need to drill into cabinet anywhere.

Is there easy access to drivers , to install longer wires if needed? Replacing the original wires

the drivers are easy to solder to once you have opened it, no pic of the back of the driver I'm afraid, the crossovers connect via little "spade clips" (not the correct term im sure!) i disconnected them and just soldered where they were.

 

[Razorhelm]

no it is not, because when you move your head you will have pre-delay.
I actualy use 4096, because that is close to the default shape I saw in Acourate. you never had problems with pre-delay with 16384 taps?

Could you explain this? I'm not familiar with pre delay.

 

[Razorhelm]

Hi, how do you get 6 channels from this sound card?

I can see 4 channels for the 2 x RCA jacks and stereo rear 3.5mm jack.

But the C/sub 3.5mm jack is only single channel ?
So 5 channels ?

Or that is stereo also?



View attachment 109409

The C/Sub is stereo. this was actually a bit of a challenge, the pulse audio layer of ubuntu (my test system before the pi) treated it as mono but when i addressed it at the alsa layer it was stereo full range.

 

[Music1969]

The C/Sub is stereo. this was actually a bit of a challenge, the pulse audio layer of ubuntu (my test system before the pi) treated it as mono but when i addressed it at the alsa layer it was stereo full range.

You've done a really superb job here. Congrats.

It doesn't look like the AVR 255 is rated for 4 ohm drivers but you can play a good volumes for long duration without issues?

The AVR 255 doesn't get hot?

 

[andreasmaaan]

Yes I have but I actually just used the crossover point KEF give for LS50 drivers, I have heard that the woofer cone can fail catastrophically so I didn't feel comfortable messing with it high SPL in its breakup zone felt like a good way to kill it!, I will have to give that a go in terms of eq'ing, i have only based on listening position so far.

The freqeuncies you've chosen seem fine I was talking only about the slopes, not the frequencies.

I have experimented with gentler crossovers but honestly I can't hear any difference! I'm not blessed/cursed with golden ears
I used brickwall because that way I don't have to worry much about driver driver cancellation in the frequency overlap, from what your saying having the drivers close enough together prevents this?

Exactly. Relative to the wavelengths at the crossover frequencies, all drivers are essentially at the same point in space. For this reason, I can see no advantage in using brickwall filters. Even much shallower filters will not result in off-axis cancellation in the first place.

Meanwhile, the reasons not to use brickwall filters include:

1. They will maximise any directivity mismatches between the drivers. For example, if the off-axis radiation of the midwoofer is quite different from that of the tweeter at the crossover point, a brickwall filter will result in the sharpest possible discontinuity in the off-axis response. Shallower slopes, on the other hand, will tend to smooth out any discontinuities.
2. Although the acoustic centres are close enough for pre-ringing to probably not be of much concern, the chances of audible pre-ringing are certainly higher with brickwall filters. The chances are especially high if you haven't determined the acoustic centres of the drivers and adjusted the relative delays so that the outputs from each driver are arriving at the listening position at the same time. So my thinking is : Given there's no benefit in using brickwall filters in the circumstances, why risk it?

Could you explain this? I'm not familiar with pre delay.

This relates directly to my second point. Linear-phase FIR filters result in each driver emitting a rising mini-crescendo of sound prior to the signal. This is called "pre-ringing" or "pre-echo".

The steeper the filter, the longer this pre-ringing will last, i.e. brickwall filters produce the longest possible pre-ringing.

Fortunately, the pre-ringing of two drivers that are crossed over with each other will cancel out when the acoustic centres of each driver are equidistant from the listener, or when the drivers are suitably delayed relative to each other such that their output arrives at the listening position at the same time.

However, if you haven't taken measurements to determine each drivers' acoustic centre and then adjusted the delays accordingly, this will not happen at the listening position. And even if you have adjusted the delays correctly so that the pre-ringing cancels out, such cancellation will become less and less effective as you move off-axis.

For all of the above reasons, I'd recommend one of two things. (1) Simply say: Who cares? What I've done sounds great, I'm sticking with it. Or (2) take quasi-anechoic measurements of the drivers, both on- and off-axis, plug the measurements into loudspeaker design software, and then develop EQ settings plus crossover topologies and slopes that achieve the flattest on-axis response and the smoothest off-axis response, re-design the crossover filters and implement them, then EQ the system in the low and lower-mid frequencies to correct for the effects of the room.

 

[Razorhelm]

You've done a really superb job here. Congrats.

It doesn't look like the AVR 255 is rated for 4 ohm drivers but you can play a good volumes for long duration without issues?

The AVR 255 doesn't get hot?

Yeah there was little information about how it would handle 4ohm drivers so i just took a punt!

I listen from about 2m away at about 80-85db at the listening position for hours, it gets warm but not crazy hot, it has a fan in it, so worried about when that fails! I have listened as high as 91db but I'll only do that for one song trying to cut down on that bad habit!

 

[Razorhelm]

The freqeuncies you've chosen seem fine I was talking only about the slopes, not the frequencies.


Exactly. Relative to the wavelengths at the crossover frequencies, all drivers are essentially at the same point in space. For this reason, I can see no advantage in using brickwall filters. Even much shallower filters will not result in off-axis cancellation in the first place.

Meanwhile, the reasons not to use brickwall filters include:

1. They will maximise any directivity mismatches between the drivers. For example, if the off-axis radiation of the midwoofer is quite different from that of the tweeter at the crossover point, a brickwall filter will result in the sharpest possible discontinuity in the off-axis response. Shallower slopes, on the other hand, will tend to smooth out any discontinuities.
2. Although the acoustic centres are close enough for pre-ringing to probably not be of much concern, the chances of audible pre-ringing are certainly higher with brickwall filters. The chances are especially high if you haven't determined the acoustic centres of the drivers and adjusted the relative delays so that the outputs from each driver are arriving at the listening position at the same time. So my thinking is : Given there's no benefit in using brickwall filters in the circumstances, why risk it?

This relates directly to my second point. Linear-phase FIR filters result in each driver emitting a rising mini-crescendo of sound prior to the signal. This is called "pre-ringing" or "pre-echo".

The steeper the filter, the longer this pre-ringing will last, i.e. brickwall filters produce the longest possible pre-ringing.

Fortunately, the pre-ringing of two drivers that are crossed over with each other will cancel out when the acoustic centres of each driver are equidistant from the listener, or when the drivers are suitably delayed relative to each other such that their output arrives at the listening position at the same time.

However, if you haven't taken measurements to determine each drivers' acoustic centre and then adjusted the delays accordingly, this will not happen at the listening position. And even if you have adjusted the delays correctly so that the pre-ringing cancels out, such cancellation will become less and less effective as you move off-axis.

For all of the above reasons, I'd recommend one of two things. (1) Simply say: Who cares? What I've done sounds great, I'm sticking with it. Or (2) take quasi-anechoic measurements of the drivers, both on- and off-axis, plug the measurements into loudspeaker design software, and then develop EQ settings plus crossover topologies and slopes that achieve the flattest on-axis response and the smoothest off-axis response, re-design the crossover filters and implement them, then EQ the system in the low and lower-mid frequencies to correct for the effects of the room.

Cool, thanks I'll have to have play around with all this, really curious if i can hear pre ringing when I have something to compare to, might try doing this on one side but not the other!

 

[voodooless]

This relates directly to my second point. Linear-phase FIR filters result in each driver emitting a rising mini-crescendo of sound prior to the signal. This is called "pre-ringing" or "pre-echo".

The steeper the filter, the longer this pre-ringing will last, i.e. brickwall filters produce the longest possible pre-ringing.

That is a vast oversimplification. You can make a fir filter without any pre-ringing. It’s all about the specific filter design. You can also make a very short filter with massively audible pre-ringing. Also this pre-ringing is only prevalent in very specific cases (if designed well).

longer filters might have more pre-ringing, but usually that is not the point of a longer filter. Also that does not make it more audible. The amplitude is so small that it usually is drowned in noise anyway. The point is the low frequency extension that you get, so you can better EQ down low.

I still would not use brickwall filters for X-overs mainly because of directivity mismatch.

Your listening to brickwall filters every day BTW. Just about every DAC contains them.

 

[andreasmaaan]

That is a vast oversimplification. You can make a fir filter without any pre-ringing. It’s all about the specific filter design.

Yes, but you can't make a linear-phase FIR filter without pre-ringing.

longer filters might have more pre-ringing, but usually that is not the point of a longer filter.

Also agree. But I said nothing about filter length. All I talked about was filter steepness.

Your listening to brickwall filters every day BTW. Just about every DAC contains them.

But (hopefully) not within the audio band.

 

[dasdoing]

3 things I use to identify pre-rining:

1) excess group delay meassurement - everything in the negative above 200Hz-ish is audible (ignoring little constant overpasses in the 0,5ms range when using REW auto delay estimation)
2) a pulse will make any midrange pre-delay audible
3) Michael Jackson - Billie Jean > the bass drum in the beginning will for some reason make any 3 figures-ish Hz pre-delay audible