Frequency Band Splitters Shootout

[Ostermilk]

Wrong!

What? The simple fact that he stated this led to the discussion which in turn led to the speculations about adobe which in turn led to end- and pointless back and forth. Only thing I did is assert that in the professional audio world a "splitter" is not a crossover and that the use of the term for that confused me as well.

Hey Normen, do you know what? I've just realised from looking at the earlier posts that the thread title was indeed "splitter shootout" rather than "band splitter" which is something I didn't get the significance of when you first mentioned "splitter" as an unqualified term. I too would have been thinking the same as you, i.e a spider or y splitter, had I seen that original title.

I now get myself where the original confusion arose.

I also think for it to be more accurate a title it should read Frequency Band Splitters shootout or perhaps even more accurately given the way it turned out Frequency Crossover Multiplexers shoutdown...

[selig]

OK, first It was not "wrong" to repeat that I was only pointing out that crossover was the conventional term. Those that keep insisting I was annoyed or "insisting" that the conventional term is used are for some reasons confusing what I said - probably my fault at some level.

Again, I was wondering how this new term came into use, and why. Language is indeed flexible, but introducing new terms CAN confuse things, as is evidenced by the original title of this thread. I apologize for derailing this thread, but as more folks joined the discussion it seems I had to keep clarifying my point, thus prolonging the tangent.

Hope things have calmed down now, as this was merely idle speculation and curious banter on my part and certainly not a subject I'm annoyed over!
As for the actual thread, I don't own or use most of the products here or I would be happy to show high res frequency and phase plots. These would be helpful to show the accuracy of the crossovers, and that was probably the original intent of the thread.

[dioxide]

Do you have a different method for doing this or do you use the MClass when you need multiband compression?

[selig]

Do you have a different method for doing this or do you use the MClass when you need multiband compression?

Don't know if you were asking me specifically or not, but I'll put in my 2 cents.

I love the MClass crossover for any multi band processing I'm likely to need in Reason. It's perfectly transparent and has well chosen slopes IMO.

That being said, I don't tend to use as much multi band processing these days as I used to do with it first became popular. I go through cycles, probably like many others, but my overall arc has been to use less and less processing as a general rule. That is to say that while I may still use some multi band processing, I don't "work" it as hard or feel the need to alter the input signal as drastically as in the past with my processing. I HOPE this is because of not only making better choices from the start, but also hopefully from refining my listening skills.

IMO the biggest limitation in Reason as it relates to multi band processors is the inability of controls to be "aware" of other controls. This is true of Combinator devices AND Rack Extensions. For example, it is quite possible to have a lower crossover frequency move ABOVE the higher crossover frequencies. In other multi band devices there is a "bumper" of sorts that keeps the crossover frequencies from getting too close to each other (and keeps them from "crossing over" each other…). Even before they are on top of each other, they are going to interact in often undesirable ways. You may typically want to keep the crossover frequencies roughly an octave away from each other, depending on the slope of the filters used.

This is one of the main reasons I've not bothered to develop any multi band devices myself, as I feel multi band devices are already quite easy to accidentally abuse - why make it any easier to do so?

[dioxide]

IMO the biggest limitation in Reason as it relates to multi band processors is the inability of controls to be "aware" of other controls. This is true of Combinator devices AND Rack Extensions. For example, it is quite possible to have a lower crossover frequency move ABOVE the higher crossover frequencies. In other multi band devices there is a "bumper" of sorts that keeps the crossover frequencies from getting too close to each other (and keeps them from "crossing over" each other…). Even before they are on top of each other, they are going to interact in often undesirable ways. You may typically want to keep the crossover frequencies roughly an octave away from each other, depending on the slope of the filters used.

This is one of the main reasons I've not bothered to develop any multi band devices myself, as I feel multi band devices are already quite easy to accidentally abuse - why make it any easier to do so?

I think it's only a matter of time before we see a splitter that uses the new(er) interactive Displays. I'm unsure if this helps the problem with crossover points being able to overlap though.

[bgold]

Hi, just noticed this thread - neat to see people interesting in testing stuff!

Just in case you're interested, two other tests I spent a lot of time on when developing Splex were:

• (amplitude) complementarity: "split" white noise into lo/hi bands, recombine, and make sure you get white noise back out. I imagine all these devices are designed to do this very well; it's generally the most important thing people want from a crossover.
• phase: sort of like the above, but invert the final combined output and see if it nulls the original signal. If you don't have an inverter you can still get some idea by just mixing the output with the completely unprocessed input. Probably everything except 4Dyne's "perfect" filters and Splex's "BSL" will show interesting stuff going on.

I'm curious now as to how much latency 4Dyne's "perfect" filters introduce if you use them at low freqs (100 Hz or so). I thought about implementing something like that but didn't want to deal with the latency issues.

[selig]

Amplitude and phase are the same thing with filters/EQs. If the amplitude is flat after recombining, it means the phase is also "flat" (0° phase shift).
I rely on FuzzMeasure to do testing like this as it's far more accurate showing both amplitude and phase responses. The inversion test will only prove one signal is not like the other, but won't reveal much more that will help you solve any issues.

As a reference, the humble Stereo Imager has a perfect reconstruction response, and has absolutely no latency to boot. Complimentary filters are easy to construct using the low pass as the basis to create the high pass. This will ensure the filters truly are complementary. There may be other ways to ensure this that I'm not aware of, but I'm no expert in this field (but learning quickly!).

[bgold]

I may have not been detailed enough in describing my second test, but I don't entirely agree with you on the phase issue. Analog filters, and IIR digital filters that emulate them, unavoidably introduce some group delay, generally when the cutoff kicks in. It's true that when you sum back both outputs of a crossover, you generally get back to the original phase. But the individual outputs can and often do have phase delays, and this can be an issue if you do time-sensitive effect processing (like compression) on the individual bands before recombining.

I'm honestly not sure what the Stereo Imager does internally. The complement of a low-pass filter is not automatically a nicely behaved high-pass filter. Conversely, summing a low-pass and a high-pass of just any random filter type doesn't necessarily give you a flat response. That's why crossovers aren't trivial.

[selig]

What then is the difference between "phase" and "group delay"? Maybe I've misunderstood what you're trying to say here… apologies if so.

Since the compliment of a low pass filter is a high pass filter, and since it's derived directly from the low pass filter, there is one and only one way for this to work. So I would have to say that whatever the result, it IS automatically the same thing for every example! Make sense?

If you look at the two filters separately, they won't look "right" or "perfect", but they will always re-combine the same exact way since they are derived from the same processes.

Just to be clear, the process I'm referring to is to take the low pass filter and combine it with the inverted "dry" signal, thus giving you a high pass response. The result is always the same, so yes it WILL automatically generate a nicely well behaved high-pass filter.

I would 100% agree that you can' just take any LP and HP filter and build a crossover and expect it to re-combine in anything resembling the original response.

Also, I would not call compression a "Time sensitive" processing, but it IS a level sensitive process. Compression should not take any "time" to accomplish unless you're talking about adding look ahead. Or have I totally misunderstood you here?

[bgold]

What then is the difference between "phase" and "group delay"? Maybe I've misunderstood what you're trying to say here… apologies if so.

I'm a bit sloppy and use both terms, but group delay is the derivative of phase with frequency - sometimes it's a nicer thing to talk about since it's more directly related to how much transients get "smeared" out.

Since the compliment of a low pass filter is a high pass filter, and since it's derived directly from the low pass filter, there is one and only one way for this to work. So I would have to say that whatever the result, it IS automatically the same thing for every example! Make sense? Just to be clear, the process I'm referring to is to take the low pass filter and combine it with the inverted "dry" signal, thus giving you a high pass response. The result is always the same, so yes it WILL automatically generate a nicely well behaved high-pass filter.

Yes and no. The complement of a low pass filter will be something that, roughly speaking, cuts low frequencies and passes high frequencies, so yes, it could be called a high pass filter. But it will not be the same type of filter as the low-pass. For example the complement of a low-pass Chebyshev filter is not a Chebyshev high pass filter (of any order). In fact, to express it in terms of other filters you'll generally need a complex mix of high-pass and band-pass. For certain simple filters things actually work out very nicely, but it's not at all a general thing.

Also, I would not call compression a "Time sensitive" processing, but it IS a level sensitive process. Compression should not take any "time" to accomplish unless you're talking about adding look ahead. Or have I totally misunderstood you here?

Compressors typically have an attack and release, and thus an internal state and "memory" of what the input was doing at earlier times. That's what I mean by "time-sensitive": what it does to an input sample depends on more than just the sample value - it depends on a bunch of previous samples too. It means that even if you use a crossover and use identical compressors on each band before recombining, the result can sound a bit different than if you'd just used an identical compressor on the un-split signal. It's most noticable with sharp transients right at the cutoff frequency.

[Ostermilk]

Splex I've found is a very practical and useful device.

More and more I find I'm having to deal with compound stems, drum loops that have key elements in different frequency bands so actually I find I'm using multi-band devices more, especially for creative tasks along with more traditional uses like dynamics, rather than less often.

What I like about Splex is it's unfussy approacch, the compressor sections work lovely and the choice of slopes for the crossovers has made it a goto tool for me.

It's succinct, flexible and effective and gets regular use in my rack these days. It's got that workmanlike quality, similar to how a well thought out console style eq does, in that it gets you where you want to go really quickly.

Recommended.

[AJTIV]

I have the SPLEX and frankly I think it sounds great. Each different setting can offer some kind of subtle character or can be completely transparent. Thats the only multi band splitter I have needed, aside form the 4Dyne for mastering, which is excellent.