dual band splitter

[JohnRoberts]

John:
Did you actually try this? or is this only theory? Sure you have some phase-shift at the crossover-points and maybe a ripple, depending on the accuracy of the components and who's curve you pick. What are we doing in Equalizer sections is a complex adding and subtracting phase behavior too, we do not bother to make a point of it.

grT

grT

grT

Yes, I designed and sold a L-R crossover commercially back in the early '80s (LOFT).

The L-R alignment makes a damn good (general purpose*) loudspeaker crossover.

For studio effects,,, YMMV.

JR

** Speaker engineers can do better than simple L-R with specific knowledge of box and drivers, and... other stuff.

 

[Balijon]

Hi John,

First of all, I have a deep respect for your knowledge and experience. I really enjoy your work and contribution in this forum. (I like and share your vision on current summing)

On the topic;
I am revering to using electronic crossovers, so not passive loudspeaker crossovers.
Basically the concept is not much different from parallel equalizers as the early Urei's.
This suggested setup compares, for a 3-way electronic crossover, to an parallel equalizer with low-shelving + band-pass + high-shelving, the only difference is the selected slope being higher than 6db/octave.

I had the same doubts at first, if this would work (back in the 80's...).
What I did was test it. I used a 4-way electronic-crossover, connected the outputs into a line-summing-amp and ran a square-wave signal through it.
It came out square-wave again. It is fun to switch-off certain output-bands to see the direct effect on the wave-patern.
Triggered by the effect of switching bands off, I did the same with a 27-band equalizer, this resulted in comparable wave-forms.

Yes, I agree there are phase-effects and level bumps, but no worse than parallel equalizers.
The fun of this solution is that you can process every band individually in compressor/limiter behavior, or as Erik is planning with coloring effects.
i used this mostly on drum-tracks where you like to compress/limit the low-end, low-mid-end and high-end, differently in attack and decay times.
The big advantage is that you sound image shift less if a certain band hits the compressor/limiter.
We also used it for 'band' panning selectively in the stereo image, with the low-band section mono and build a stereo image going up in frequency. (great for bass-gitar too).

I am eager to hear Erik's findings.

grT

 

[JohnRoberts]

I guess a continuous square wave could line up again with the HP output from one cycle added to the LP output from a different cycle, 360' apart.  While this sounds better than the longer persistent tones not matching up, the leading edge of such tone bursts, and transient musical content will not line up for the first cycle period (and last).

Music contains tones but they don't persist for all time. They actually start and stop, so only derived frequency dividing circuits will accurately sum back to the original for the entire waveform.

I agree this may be academic if there is significant processing going on in the middle, but know your paint. I prefer to start with as clean a canvas as possible.

JR

PS: If you don't believe me, perform a null test subtracting your recombined crossover signals from the dry original. If you are correct, they will null to silence. I predict there will be significant deviation from the original, resulting in a measurable null product.

 

[erikb1971]

John, Balijon,

Although I do not completely understand the technical background of your discussion, I enjoy reading it. Things got extremely confusing for me since I initially thought the LR filter was a  sort of derived or subtracted filter since it was in the article. A typical case of reading what you want to read I guess. Just now I see that is was as a comparison to the subtracted filters. Which means at least I do not have to build a LR filter myself, since that is basically the (behringer) crossover. After I finish the effects I want to apply, I will borrow buy or rent one of those crossovers to do some experimenting.
It would be cool though to be able to compare it with a derived filter. So letting go of the LR filter for now. So john, in your opinion, the 4th order subtractive filter network would be the best for splitting the signal?

Cheers

erik

 

[JohnRoberts]

I don't know what you need, (you probably don't know either).

I am surely repeating myself but the derived crossover only has a one pole roll off in one direction, no matter how many poles in the other.

I would experiment first with determining what the two different effects are that you want to apply to the HP and LP. Then determine how severely their inputs need to be separated. For example if HF leaking into the low pass doesn't suck as much as LF leaking into the high pass output you know which direction gets the one pole roll off and which direction gets the steeper (the real multipole filter has the steeper roll off). 

You may end up subtracting a HPF from unity instead of a LPF, depending on which direction needs the steeper attenuation? Since you only get one pole one way, do you really need 4 poles the other way?

While I'm sorry to not give you a simple direct answer, IMO you need to do a little more homework in the bench to advise how to optimize for direction of real filter and slope.

Sorry to make this more complicated, but good is often a little more complicated. 

JR

 

[erikb1971]

I really appreciate you repeating yourself since Iam too stupid to understand it the first time so stop saying sorry about it.
Iam going to work on the high band effect, and Iam going to start experimenting... I will find you filter specialist when needed.. tnx!

 

[Balijon]

PS: If you don't believe me, perform a null test subtracting your recombined crossover signals from the dry original. If you are correct, they will null to silence. I predict there will be significant deviation from the original, resulting in a measurable null product.
[/quote]

Hello John,

Yes, you are right, they won't null exactly, and you can predict this without testing...

I have seen and read your mythical 'nulling' discussion in many items on the forum. It is a relevant test to some degree if you comparing apples with apples, comparing components in identical circuits.

As soon as you start comparing signal chains with different stages and components, there will always be a difference in delay and phase.
Take for instance a 4 opamp-stage typical state of the art equalizer with dc-decoupling capacitors and dialed in at neutral. Even the best design will not null exactly due to the added delay of the opamps and the phase-shift of the coupling capacitors.
(In fact we used this 'delay effect' and phase-shift-effect to time-align loudspeaker systems like the Urei's and Tannoy's in electronic crossovers in the old days).
So with your suggested nulling test in this specific setup, did you prove that you have a bad circuit if it does not null? No, you proved that adding stages and components introduce time delay and phase-shift, and we know that is a given.... As you have been designing audio consoles, I think you are aware of the effect, choosing your dc-decoupling capacitors right is a tight balance in the low-frequency phase behavior of the design.
I even dare to challenge your nulling method for comparing OpAmp circuits, when the buildup-circuitry of the opamp is not carbon-copy identical (whether it is discrete or not), due to timing-differences and phase-shifts, signals will never null exactly. Again did you prove one or the other is a better design, no you proved they have a different delay and phase behavior..

On the square-wave signal test and transients:
How much more transient behavior can you create compared to a square wave signal?
We have used square-wave testing for decades to tune and compare audio-transformers and compensate ringing and overshoot, just for this reason.
A square-wave signal tells you how a circuit deals with transients where sine-waves do not.

Theo

 

[JohnRoberts]

Active circuitry using negative feedback does not tolerate much delay. The small actual delay in say an opamp, causes the negative feedback to rotate all the way to positive feedback and would cause oscillation if the open loop gain and NF product is not attenuated below unity by that rather high frequency.

So in practice if not oscillating, the delay at audio frequency is nil. That said the compensation (typically a one pole rolloff) network introduces a phase shift (just like a simple integrator) to the open loop transfer function. The actual closed loop response is defined by the NF network and open loop vs. closed loop gain margin. As the open loop gain falls and approaches the closed loop gain (at very high frequency) the resultant phase shift approaches the open loop phase shift.

This is in fact a measurable flaw from running opamps with too much closed loop gain for their ability to deliver open loop gain. Note: one improvement in modern uber opamps is less internal delay and more/higher frequency open loop gain.

The reality is the delay/phase shift from just NF and flat gain is inconsequential, and swamped by HF poles commonly added to the NF networks to reduce out of band noise and/or interference.

So yes even flat circuits will not null perfectly at 1 MHz, but up to 20kHz the null should be dominated by closed loop gain.
------
re: transients, yes the square wave has a fast edge rate, but that isn't my issue. In the context of a L-R crossover with +/-180' of phase shift, the HP and LP outputs that have rotated in phase a full 360' only combine smoothly for a repetitive waveform, because the outputs that line up are actually separated in time by one cycle.  This is useful in practice for loudspeaker crossovers that are juggling multiple constraints (like out of band rejection). Any transient errors are small compared the clean presentation of persistent tones.  Since music starts and stops, this imperfect summation is a known and accepted compromise for professional sound reinforcement which is about delivering SPL not pristine , accurate transient detail.

I submit your good square wave sum, will degrade if presented as short bursts or single pulses.   

JR

PS: Delay is actually much more of an issue in digital consoles and digital processing, While short simple delay is less consequential (IMO) than phase shift unless identical signals interfere, with delay.

 

[Balijon]

phase shift = frequency depended delay compared to the original source signal

Active crossovers for Urei's and Tannoy's were used this way to compensate the timing difference caused by the physical driver distance offset (between LF speaker and horn driver). The passive crossovers attempted to do the same, only less accurate.
Both factors we used to create time-allignment, delay from the circuits and phase-time-delay.
Early analogue circuit based echo and reverb systems (like good old Dynacord's) were based on this principle, analogue video delay / sync systems as well.

A capacitor does not only have an attenuation effect on the frequency that passes through it, it also changes the phase which is a timing difference between the original source for that frequency.

>>PS: Delay is actually much more of an issue in digital consoles and digital processing,<<
Absolutely agree! Try nulling sources there.... I think you just confirmed the relevance of your suggested nulling testing here.

Theo

 

[JohnRoberts]

While a complete discussion of loudspeaker crossovers will confuse this even further, yes indeed.. All pass filters have been used for decades in analog crossovers to introduce effective in band time delay to partially compensate for drivers effective distance relative to the baffle board. BUT, sound fields are three dimensional so this correction only works on axis and there will still be lobing and other errors off axis. Note: DSP based crossovers, can easily generate real delay, while all pass filters are still used sometimes for driver specific tweaks. 

I understand how capacitors work, and unit delay is the operative actor inside digital filters.

The phase shift in an opamp's open loop transfer function is dominated by the compensation cap forming an integrator.. this is so the actual propagation delay in the circuitry can not cause mischief.

I will stop repeating myself soon.

JR

 

[Balijon]

Could we agree that, if there are time delay and/or phase effects in a circuit, it will not null to silence when subtracting original source signal?

Theo

 

[JohnRoberts]

Could we agree that, if there are time delay and/or phase effects in a circuit, it will not null to silence when subtracting original source signal?

Theo

Of course... That's why we always need to review the null product and not read too much into errors associated with HP and LP skirts. The errors and failure to null that I predict are not at frequency extremes but in band and specifically at the crossover transition frequency.

JR

 

[Balijon]

Could we agree that, if there are time delay and/or phase effects in a circuit, it will not null to silence when subtracting original source signal?

Theo

Of course... That's why we always need to review the null product and not read too much into errors associated with HP and LP skirts. The errors and failure to null that I predict are not at frequency extremes but in band and specifically at the crossover transition frequency.

JR

I have attached an article on active crossover designs and their crossover behavior in summing and phase / time delay behavior.

The most clean solution is a Linkwitz-Riley crossover design:
>>
1. In-phase outputs (0° between outputs) at all frequencies (not just at the crossover frequency as popularly believed by some).
2. Constant voltage (the outputs sum to unity at all frequencies).
<<

however:
>>
A Linkwitz-Riley crossover alignment is not linear phase: meaning that the amount of phase shift is a function of frequency. Or, put into time domain terms, the amount of time delay through the filter is not constant for all frequencies, which means that some frequencies are delayed more than others. (In technical terms, the network has a frequency-dependent group delay, but with a gradually changing characteristic.)
Is this a problem? Specifically, is this an audible “problem?” In a word, no.
Much research has been done on this question with approximately the same conclusions: given a slowly changing non-linear phase system, the audible results are so minimal as to be nonexistent.
<<

Transient response:
>>
We also refer to the step response as the transient response of the circuit. The transient response of the summed out- puts is perfect since their sum is perfectly equal to one.
<<

Where the Linkwitz-Riley result in power-drop of -3db at the crossover frequency in a loudspeaker system, they sum on a line-level perfectly as they are constant voltage.

Due to the non-constant input / output phase behavior, summing with subtracting the original source signal will never null completely to silence. (also relevant and more dominant for Butterworth based filters).

Theo

 

[JohnRoberts]

Could we agree that, if there are time delay and/or phase effects in a circuit, it will not null to silence when subtracting original source signal?

Theo

Of course... That's why we always need to review the null product and not read too much into errors associated with HP and LP skirts. The errors and failure to null that I predict are not at frequency extremes but in band and specifically at the crossover transition frequency.

JR

I have attached an article on active crossover designs and their crossover behavior in summing and phase / time delay behavior.

The most clean solution is a Linkwitz-Riley crossover design:
>>
1. In-phase outputs (0° between outputs) at all frequencies (not just at the crossover frequency as popularly believed by some).
2. Constant voltage (the outputs sum to unity at all frequencies).
<<

however:
>>
A Linkwitz-Riley crossover alignment is not linear phase: meaning that the amount of phase shift is a function of frequency. Or, put into time domain terms, the amount of time delay through the filter is not constant for all frequencies, which means that some frequencies are delayed more than others. (In technical terms, the network has a frequency-dependent group delay, but with a gradually changing characteristic.)
Is this a problem? Specifically, is this an audible “problem?” In a word, no.
Much research has been done on this question with approximately the same conclusions: given a slowly changing non-linear phase system, the audible results are so minimal as to be nonexistent.
<<

Transient response:
>>
We also refer to the step response as the transient response of the circuit. The transient response of the summed out- puts is perfect since their sum is perfectly equal to one.
<<

Where the Linkwitz-Riley result in power-drop of -3db at the crossover frequency in a loudspeaker system, they sum on a line-level perfectly as they are constant voltage.

Due to the non-constant input / output phase behavior, summing with subtracting the original source signal will never null completely to silence. (also relevant and more dominant for Butterworth based filters).

Theo

Thanks for the link, I will look at this more closely later, but it looks like they are saying the LR-2, a 2 pole Linkwitz-Riley variant, is transient correct. I'm not familiar with what a 2 pole L-R is but I'll take their word that it sums. Dennis Bohm is a solid citizen. IIRC he used to write app notes for Nat Semi back in the old days, and he was cooperative when i tried to get AES to define Q (still trying).

I would bless a 2 pole L-R, whatever that is, for this application. Back in the late '80s I messed around with variant SVF topologies where you could in principle sum to unity, but for more than a couple poles it was difficult to make it stable, at least I gave up on trying. 

JR



.

 

[erikb1971]

I just purchased a second hand behringer crossover http://www.behringer.com/EN/Products/CX3400.aspx for 40 euro's. It should arrive somewhere next week, so In a week I hope to be able to provide some test clips...

Cheers

Erik

 

[erikb1971]

;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D ;D

I have no other way to describe the fun I just had the past hour! What happened? I connected my Ipad to my little soundcraft mixing console. With a modified insert cable, I managed to get a copy of that sound in my behringer crossover. The low output from the x-over went through my Rocktron Austin Gold distortion and then into the console. The mid/high went through my Ibanez tube screamer and into the console. On my Ipad I started the DM1 drum-computer app. And then: FUN! and shitloads of it! I had all these knobs to play with : the xover frequencies, the distortion and tone controls of the stompboxes, the levels of the mixing console. I was The prodigy and The chemical brothers in one!  For a moment I was even the one The Prodigy and The Chemical Brothers ask for advice about distorting drums! Give me this set up and I could do main stage on Glastonbury! With just one 4 bar drum loop For three days!

So was it any good you might ask?

You might not believe this, but I have a slight tendency to overreact when Iam enthusiastic about something. I will have to see how useful this will be on real drums mixed in with real music when more subtle effects are needed.  Or on single drum hits. Or on guitar or vocals.  so I am far from sure that this will be the go to unit for colouring drums. But as an inspirational tweeking tool, or a live tool for DJ's or electronic drums based bands.... so far, this ROCKS!

 

[Balijon]

Great to hear Erik!
Have fun!
Maybe you can upload some samples?
On real drums it is big fun to play with compressors/limiters with different settings for different bands.

Theo

 

[erikb1971]

Yeah I should upload some samples.. but first I have get everything set up a bit better.... with actual working insert cables and an output for the third band.. but so far.. very promising!

 

[Tim Ragnur]

That's awesome!
I wonder how the Behringer is built, maybe you can borrow/get inspired by the design?

 

[erikb1971]

I will open the behringer today and make some pics... Thursday I think I will make a soundclip and video, I will have a third stompbox by then so the high freq should be a bit better taken care of...

I wonder if John still reads this thread, maybe he has some specific tests I could do to check his initial "problems" for lack of better word with using the x over filter for this kind of use!