To crossfeed or not to crossfeed? That is the question...

[castleofargh]

my timid attempts at playing with mid/side was really not worth the effort. can't say it's wrong or bad in general, only that it was in my hands ^_^.
I thought about applying an EQ to the middle signal so that I would perceive the singer in front of me, not up or down. and then try different options to make something that would still retain channel separation with the rest and apply some close enough HRTF profile for 30° speakers.
most attempts ended in clear mistakes from me being too ignorant. and getting delays and/or level issues have indeed been some of the issues, that and having the same instrument at 2 places at once in my mind when failing to completely filter it out of the mid channel... all clearly noob mistakes, and I went for most of them.

soon enough I was thinking something well known to people with no willpower and no achievement, "F it, I give up!". and since I've been waiting for the realiser A16(typical lack of brain compensated with money). what I use ATM is to simply apply the HRTF impulses for 30° with a true stereo convolver. it gives marginally better results than Xnor's crossfeed with settings for my big empty head, but it was such a pain just to try a bunch of HRTF impulses one after the other until one felt ok for my subjective image of instrument placements. if I was to start from scratch again, I'd enjoy basic crossfeed while eating pringles instead.

 

[71 dB]

And what about if real signals are more like this:
L(t)=|L(t)|ejwL(t)
R(t)=|R(t)|ejwR(t)
In other words, what about phase?

To be clear I have no knowledge in cross-feed. Maybe by simplification, maybe for other reason you are not taking into account phase.
I am just curious

Real signals aren't like that. You get that if you do Hilbert transformation to make the signal complex. Phase however is of course real. What do you mean taking phase into account?

 

[71 dB]

Ok, understandable, but averaging does not differentiate fixed or dynamic. I'm still trying to figure out what you guys have in mind when you say "dynamic crossfeed" or more specifically what variables you want crossfeed to respond to.

All normal cross-feeders (Linkwitz, Meier etc.) are constant and they do their job. Pinnahertz has build an analog dynamic cross-feeder long ago and it didn't work that well. I have thought about dynamic cross-feed in theoretical level and found it troublesome and iffy. Since constant cross-feeders do their job (remove/reduce spatial distortion), there is no real need for dynamic cross-feeders, is there?

 

[Arpiben]

Real signals aren't like that. You get that if you do Hilbert transformation to make the signal complex. Phase however is of course real. What do you mean taking phase into account?

Well my mistake, I was at first glance thinking you were only dealing with real part of L&R signals and letting aside phase.

 

[Strangelove424]

I have thought about dynamic cross-feed in theoretical level and found it troublesome and iffy. Since constant cross-feeders do their job (remove/reduce spatial distortion), there is no real need for dynamic cross-feeders, is there?

I don't know if there is a real need or not, since the concept of a dynamic crossfeed is purely theoretical at this point, and I'm still trying to get an idea of how it would behave hypothetically. Since the Meier plugin was a recent discovery for me, and the amps upon which they're based relatively new as well, I am open to the idea that crossover technique is something that can still be improved upon greatly.

 

[pinnahertz]

...and so it goes.

 

[pinnahertz]

I don't know if there is a real need or not, since the concept of a dynamic crossfeed is purely theoretical at this point, and I'm still trying to get an idea of how it would behave hypothetically. Since the Meier plugin was a recent discovery for me, and the amps upon which they're based relatively new as well, I am open to the idea that crossover technique is something that can still be improved upon greatly.

Yes, improved upon. The problem we seem to have is that the current common (some say normal) are pretty huge approximations and therefore imperfect. What results is something that covers the range from subjectively objectionable to miraculous. I'd say there's room for improvement in there.

 

[pinnahertz]

The Meier algorithm itself is fixed, but the amount of crossfeed (and delay) is dynamic to the signal. If you wanted the algorithm itself to vary, what variable would it change according to?

Backtracking again... I said it didn't work. I tried varying crossfeed based on program dynamics and separation.

Well, I didn’t push for schematics. You could easily mention what input variable your gain responded to, but it’s ultimately your design, and I’m certainly not trying to force you to reveal anything you don’t want to. As long as you’re not referring to it here as a firm example of dynamic crossover, its success or failure is neither here nor there.

When you ask for details where else would I go? Block diagrams and schemaitics. Why would I publish that if it didn't work?

I don't view my experiments as a failure, they provided valuable information. As an improvement to fixed crossfeed, no I didn't think it accomplished the goal universally.

But successful implementation aside, on a completely theoretical level, what would a truly dynamic crossfeed function like, and how would it improve crossfeed performance?

I have no idea. My particular goal was to create a processor that would do the crossfeed thing well but provide more consistent results when handling widely varying material. I still think there's a point to doing that, but I also think there's a relatively microscopic market.

Do you design for Meier?

Assuming you meant that as a compliment...thanks, but no.

There are actually a number of broadcast audio processors that perform a sort of crossfeed on a dynamic basis. The goal is more consistent on air sound. I've tried them, and pretty much hate them all. The now even include algorithms for "fixing" mp3 compression artifact! And they work just about as well.

We definitely live in a "just because you can doesn't 'mean you should" world now.

 

[71 dB]

Well my mistake, I was at first glance thinking you were only dealing with real part of L&R signals and letting aside phase.

You can think phase with simple signals such as a sinusoidal, but music is a different story.

If phase difference between L and R increases, (L+R) decreases and (L-R) increases.

 

[71 dB]

I don't know if there is a real need or not, since the concept of a dynamic crossfeed is purely theoretical at this point, and I'm still trying to get an idea of how it would behave hypothetically. Since the Meier plugin was a recent discovery for me, and the amps upon which they're based relatively new as well, I am open to the idea that crossover technique is something that can still be improved upon greatly.

The problem with improving cross-feed is that you jump from simple circuits to DSP processors convoluting HRTF files with the music and what not. That's not something everyone are willing to do. Simple cross-feed removes spatial distortion taking headphone listening to a whole new level.

The correct way to improve things would be to have real binaural recordings for headphones, but we don't have that many, because binaural recordings sound weird on loudspeakers. So, the recordings are mostly for loudspeakers and we cross-feed them for headphones to get rid of spatial distortion.

 

[pinnahertz]

The problem with improving cross-feed is that you jump from simple circuits to DSP processors convoluting HRTF files with the music and what not. That's not something everyone are willing to do. Simple cross-feed removes spatial distortion taking headphone listening to a whole new level.

The correct way to improve things would be to have real binaural recordings for headphones, but we don't have that many, because binaural recordings sound weird on loudspeakers. So, the recordings are mostly for loudspeakers and we cross-feed them for headphones to get rid of spatial distortion.

This... after all we've posted....

Well, your opinion is not shared by everyone. Simple cross-feed does not remove spatial distortion, it may reduce it. The "whole new level" may be higher (better) or lower (worse) than a given recording without it.

Binaural recordings are correct only for the listener matching the specific HRTF that the recording was made with. Other listener results vary quite widely all the way to unacceptable.

We don't "get rid of spatial distortion" with cross-feed because the solution is only a general approximation of the inverse of the problem to begin with, but we may reduce it in some cases.

Every signal DAP has at least a little DSP capability, certainly enough for improved cross-feed, on board. It goes without saying PCs could handle the process. I'm surprised we aren't inundated with advanced, capable, and fully adjustable cross-feed plug-ins. It can't be because of complexity, and the headphone listening market is huge and growing. So why is cross-feed not universal at this late date? Could it be it's not universally desired or accepted? I don't know, but I'm in the "not universally desired" camp.

 

[71 dB]

Wide modification of Linkwitz using fixed -3 dB cross-feed level and additional treble cross-feed (channel difference limiter) at -25 dB works well with varying material. The wider you make the cross feeder, the smaller is the need to vary cross-feed level for different material. For 30° it is about -11 dB … -1 dB. For 60° it is probably -5 dB … -2 dB and for 90° it's just -3 dB.

Wide Linkwitz is Meier-like and actually pans sounds similarly, but it avoids the "aggressive" nature of Meier. The only downside seems to be that the soundstage isn't that deep and forward, but well-recorded music contains enough spatial cues to create a feeling of depth. Those who don't like cross-feed because it reduces the apparent width of the sound may find wide Linkwitz pleasing, because it does hardly anything to the width, just removes spatial distortion.

Wide cross-feed is an easy modification to "normal" Linkwitz: You re-calculatale the capacitors in the cross-feed section to lower the cross-feed cut-off frequency to 300-400 Hz (roughly double the capacitance) and add an resistor between left and right channel for the treble cross-feed. If the resistors between output and ground are R then this treble cross feed resistor is about 17*R. Cross-feed level must be tuned to -3 dB of course. Below is the schematic (sorry, quick hand-drawing) of my wide Linkwitz headphone adapter:

R1 = 120 Ω
R2 = 2.2 Ω
R3 = 220 Ω
R4 = 100 Ω
R5 = 70 Ω (82 Ω and 470 Ω in parallel)
R6 = 37 Ω (22 Ω + 15 Ω)
C1 = 1.17 µF (three 390 nF capacitors in parallel)
C2 = 11.5 µF (6.8 µF and 4.7 µF in parallel)
K1 = 2 x ON/ON switch for cross-feed on/off.​

The effective output impedance = R2 = 2.2 Ω which is low enough for practically all headphones in the world (the most demanding cans need 4-5 Ω at most, so this is half of it). One can add another switch in series with R6 to bypass treble cross-feed, but this is the most versatile cross-feeder I know having only on/off switch. This is a type of cross-feeder you can forget existing and just concentrate on enjoying the music because it works with almost any material imo.

 

[71 dB]

This... after all we've posted….

So, should everyone just agree with you 100 % after all we've posted? We have a different approach: You are after perfection or something like that no matter what it takes or costs. I am after the most bang for the buck. Most of us aren't millionaires, so "most bang for the buck" is a rational approach to us.

Well, your opinion is not shared by everyone. Simple cross-feed does not remove spatial distortion, it may reduce it. The "whole new level" may be higher (better) or lower (worse) than a given recording without it.

Increasing cross-level makes the signal more mono-like and mono signal contains zero spatial distortion. Proper cross-feed reduces the channel difference enough to make the sound spatial distortion free. Too little cross-feed only reduces spatial distortion, but that is not proper cross-feed. Some recordings do sound better without cross-feed, but my experience is that that's something like 2 % of recording (of my music collection anyway) and I have the off switch for those cases. So, for the 98 % of my music cross-feed makes things better and in case of those 2 % I bypass cross-feed and my headphone adapter does nothing.

If you have developed better solutions then good for you, but I am very happy with the solutions I have. You haven't presented those better solutions much have you?

Binaural recordings are correct only for the listener matching the specific HRTF that the recording was made with. Other listener results vary quite widely all the way to unacceptable.

That's one problem with them, but in general binaural recordings work quite well for anyone, just not perfectly unless you have the "correct" head shape/size.

We don't "get rid of spatial distortion" with cross-feed because the solution is only a general approximation of the inverse of the problem to begin with, but we may reduce it in some cases.

Why you insist this is beyond me. Proper cross-feed removes the uncomfortable sensation and messiness of stereo image spatial distortion causes, hence I say proper cross-feed removes spatial distortion. Too little cross-feed only reduces spatial distortion (uncomfortable sensations and messiness of stereo image remain, but are weaker), but even that is a big plus. My first cross-feeder was a one-level (-8,7 dB) model and it was too weak for many recordings, but even it did revolutionized my headphone listening. I recognized the need for stronger (and weaker) cross-feed levels and modified it to 3-level before building a completely new 6-level model. My approach is to keep things as simple as possible and only increase complexity if it's something relevant. I believe that is a healthy approach in audio and in life in general.

Every signal DAP has at least a little DSP capability, certainly enough for improved cross-feed, on board. It goes without saying PCs could handle the process. I'm surprised we aren't inundated with advanced, capable, and fully adjustable cross-feed plug-ins. It can't be because of complexity, and the headphone listening market is huge and growing. So why is cross-feed not universal at this late date? Could it be it's not universally desired or accepted? I don't know, but I'm in the "not universally desired" camp.

I'm here to promote all kind of cross-feed. If you have DSP capability then of course it's good to use it to cross-feed. Your posts questioning especially Linkwitz cross-feeder does not have positive effect on people recognizing the benefits of cross-feed. The reality is that most people are completely ignorant about audio. The concept of cross-feed is VERY difficult for people to understand. So, it takes a lot of education.

Nokia Lumia phones had Dolby Phone prosessing, which is effectively cross-feeding, but people wanted iPhones and Android phones so Windows phones lost.

 

[pinnahertz]

So, should everyone just agree with you 100 % after all we've posted?

No, of course not. But you state everything as immutable fact when the entire nature of generalized cross-feed is an approximation, with the results being entirely subjective. Don't expect every to agree with you either!

We have a different approach: You are after perfection or something like that no matter what it takes or costs. I am after the most bang for the buck. Most of us aren't millionaires, so "most bang for the buck" is a rational approach to us.

You've misunderstood my "approach" entirely. Not surprising.

If you have developed better solutions then good for you, but I am very happy with the solutions I have. You haven't presented those better solutions much have you?

A careful reading of my posts would reveal that I've state clearly that I have not developed a better solution. Given that, I have nothing specific to present. But since I have worked on the problem, I recognize the failings of the approach you advocate, having tried that myself.

That's one problem with them, but in general binaural recordings work quite well for anyone, just not perfectly unless you have the "correct" head shape/size.

And in my opinion they universally present a novel but incomplete perspective unless made with the individual HRTF. I've actually made binaural recordings both with an artificial head and pinna, and with my own head and pinna. The difference is not small, and forms the based of my conjecture re the failings of recordings made with a generalized average HRTF.

Why you insist this is beyond me. Proper cross-feed removes the uncomfortable sensation and messiness of stereo image spatial distortion causes, hence I say proper cross-feed removes spatial distortion. Too little cross-feed only reduces spatial distortion (uncomfortable sensations and messiness of stereo image remain, but are weaker), but even that is a big plus. My first cross-feeder was a one-level (-8,7 dB) model and it was too weak for many recordings, but Your posts questioning especially Linkwitz cross-feeder does not have positive effect on people recognizing the benefits of cross-feed.

I object to subjective opinion being stated as scientific fact, and frankly the arrogance of insisting absolutes when clearly you're working with a very significant approximation with results that could only be subjective, doesn't help either.

 

[71 dB]

No, of course not. But you state everything as immutable fact when the entire nature of generalized cross-feed is an approximation, with the results being entirely subjective. Don't expect every to agree with you either!

I try to base my opinions on science which is more or less objective. I don't want to promote BS. Cross-feed is an approximation and I never said it isn't, but that doesn't mean it doesn't remove spatial distortion because it does. It is science based on human spatial hearing. Cross-feed doesn't give exact stereo image along to lines of instruments being in exactly correct places and angles etc. It puts instruments in approximately correct places instead of those instruments spreading all over the place due to spatial distortion. That is what is relevant and that's what we can relatively easily fix. You can keep whining, but I am happy with it.

I don't expect narrow-minded "purists" agree with me. I don't expect besser-wissers agree with me. I expect open-minded people who are interested in improving their headphone listening experience concentrating on relevant "bang for the buck" things being open to what I say.

You've misunderstood my "approach" entirely. Not surprising.

Sorry. I am new to this forum and I am only learning to know you based on what you say. So far your approach has been rather "harassing." and other members here seem much nicer and more relaxed. 80 % of my energy and time here goes defending myself against your attacks (including this post!).

A careful reading of my posts would reveal that I've state clearly that I have not developed a better solution. Given that, I have nothing specific to present. But since I have worked on the problem, I recognize the failings of the approach you advocate, having tried that myself.

I did mention earlier that I have studied the problem of dynamic cross-feed too. I have also said I think dynamic cross-feed is "going too far" and it's best to settle with "normal" cross-feed methods as those when done properly do fix the relevant problem, spatial distortion and messiness or sound image. In audio you can't have everything 100 % so it's best to know how to compromise in an optimal way. Understanding helps in it and I have used the last 5-6 years with studying cross-feeding so I'd say I know something about the issue. What I don't know is something I want to learn.

And in my opinion they universally present a novel but incomplete perspective unless made with the individual HRTF. I've actually made binaural recordings both with an artificial head and pinna, and with my own head and pinna. The difference is not small, and forms the based of my conjecture re the failings of recordings made with a generalized average HRTF.

Tell me something I don't know. Good luck finding your favorite music recorded using your own head. I know that's impossible so I acquiesce in cross-feeding recordings and having if not 100 % authentic sound immersion something that is enjoyable and free of problems generated by excessive stereo separation.

I object to subjective opinion being stated as scientific fact, and frankly the arrogance of insisting absolutes when clearly you're working with a very significant approximation with results that could only be subjective, doesn't help either.

My subjective opinions on this issue are based on scientific facts/understanding and subjective experiences so you can call them semi objective if you want. Sometimes approximation is all it takes to solve a problem.