The spatial comb filtering from 3 sources is not as severe as from 2 sources, since it almost never happens all the three signals to have exact amplitude and phase to cancel, whereas with two sources cancellation happens in every wavelength of spatial shift.
I've done comb filtering perception listening tests using mono pink noise (worst possible signal) and 3 speakers are much better than 2.
Here's some of my subjective results I did just recently:
Elias Pekonen Home Page - 2 vs 3 Speaker Stereo listening tests at high frequencies
- Elias
Personally, I am impressed with PLII. The first time I heard it was in Jim Fosgate's laboratory and it was a all analog implementation. It sounded wonderful and had absolutely no tonal coloration at all. The next time I heard it sound as good as the analog version was on a Arcam AVR600 I purchased last year. Once again, no colorations, and the speakers just disappeared, and I was immersed with pure uncolored sound.
These are the only two times I was impressed with PLII
Griesinger has some interesting PPT slides up on his webpage about "Pitch, Timbre, and Source Separation".
Maybe more speakers isn't such a bad idea after all?
I strongly disagree with your statement "center phantom image weak or non existent" for 2-speaker stereo with high pass filtered music.
I do get a stable phantom center from high pass filtered (48dB@1.6kHz) stereo music tracks.
Possible explanations for your findings:
- Low direct/reverberation ratio. Is your room treated in any way? What kind of speaker directivity was used?
- Difference in localization perception. There are reports that some people can't localize phantom images. You've reported similar perceptions in the past. You should have other people listen to your test in your room and report their observations.
Maybe more speakers isn't such a bad idea after all?
Nobody said more speakers are bad (think Ambisonics
) but certainly it is more efficient to use only one stereo speaker It looks like 2 speakers is the worst scenario.
I also don't have problems at 1.6kHz but the problems occur at higher freqs above about 6 kHz.
There is published scientific evidence that humans cannot perceive phantom images at very high freqs (the reference was given in another diyaudio thread). That is the problem of 2 speaker stereo.
3 speaker matrix stereo seems to solve this problem somehow.
- Elias
...in your personal priority list of what sound reproduction needs to be capable of. That list might not be the same for everybody. My priorities are certainly different.
Your page says "center phantom image weak or non existent" for frequency band 1-20kHz. It doesn't say 6-20kHz!? The cutoff in the track I've used is 1.6kHz (not 1kHz as in your example), still I get a perfect phantom center.
You skipped over my questions about room and speaker characteristics in your listening test.
Last edited:
In 2 speaker stereo the phantom image is very weak, for the reason at freqs around 1 kHz there will be a phantom but at higher freqs above 6 kHz the image is split into the speakers. This happens if there is signal energy at both freq ranges.
When there is less signal energy at 1 kHz and more energy above 6kHz there is no phantom imaging and only the two speakers will be perceived.
So, overall the 2 speaker stereo does not work very well in terms of phantom imaging.
That is highly subjective statement
Must be weird kinds of music you are listening to.
I prefer music were - by and large - the harmonics roll off with rising frequency.
Rudolf
Heres my experimental matrix system at proto stage. Three FRS8 at reinforced cardboard dipole baffles Matrixed for coefficient of 0.5. Very old proto speaker has been forced laying on the floor to serve as a dipole bass below about 100Hz.
Sound... amazing
- Elias
Matrixed for coefficient of 0.5. Very old proto speaker has been forced laying on the floor to serve as a dipole bass below about 100Hz.An externally hosted image should be here but it was not working when we last tested it.
Sound... amazing
- Elias
Yes, that's the basic realization that Gerzon plays with. For good reasons he says that gentle blends between the rematrixing bands is needed.
Personally I was thinking about linear phase filters because 1st order analog, while summing linphase, has 90° phase between bands, and acoustic centers of same bands are far apart (speaker-2-speaker distance) so the there could be some off-axis ill effects that might be avoidable with round kneed, gentle slope linphase filters with compact, non-ringing FIR kernels.
I'm FIR based anyway and integrating the Trifield matrixing would not cause any computational resource penalties... still waiting for my new coax drivers (4pcs) but then I'll be eager to try TriField and find out if it significantly betters the simpler Trinaural.
Here's the trinaural matrix
Fullfils matrix equations:
Ls = L - 0.5*R
Rs = R - 0.5*L
Cs = 0.5*L + 0.5*R
- Elias
An externally hosted image should be here but it was not working when we last tested it.
Fullfils matrix equations:
Ls = L - 0.5*R
Rs = R - 0.5*L
Cs = 0.5*L + 0.5*R
- Elias
No question that 2-speaker stereo suffers from severe comb filtering for centered images - specifically, vocalists. I sometimes wonder if that's why some recordings have the vocalist a bit off-center.
The comb filtering noticeably changes the timbral balance of the singer - but then again, just about all recordings have a final mastering on a 2-speaker playback system, and the EQ is intended for 2-speaker playback (with a secondary cross-check for single-driver lo-fi mono playback). Just guessing here, but I'd imagine that vocals sound a bit more direct and "up-front" on a symmetric 3-speaker playback system, due to the absence of comb filtering for the centered singers.
I'm not sure what "direct" and "up front" means, what based on my perceptive observations with 2 speaker stereo the singer is typically projected very close often closer than the loudspeaker distance. It sounds very unrealistic to me. With 3 speaker matrix the situation is different and the singer sounds quite a bit at further distance behind the loudspeakers which sounds much more realistic to me.
I don't know yet is this due to less comb filtering of 3 speaker stereo, or because of increased 'ambiance' due to difference side signal +/- k*(L-R) which is present at both side speakers.
- Elias
As I've said before, doesn't happen for me and the only person I've ever heard reporting such a perception is you.
Sounds like a constructed case. Any specific recording that would contain such signals?
Compared to what? More speakers? Of course more speakers can "work" better.
That is highly subjective statement
Yes, but only in internet forums I have found people that would object. All "real" people I've ever met never had the phantom imaging problems you describe.
I also "suffer" from poor phantom image perception on all the stereo systems I have been listening to (which include some finely calibrated ones): it always appear to me as a diffuse area that gets even more blurry with every tiny movement of the head...
I think it has to do with comb filtering between the two speakers to which I might be "over sensitive" (takes me out of the illusion), so this trinaural approach looks tempting...
Actually THX did a experiment much like Elias did, and they came to the same conclusions. This is why they strongly recommend a center speaker to handle dialog with THX approved systems. Also, if you have ever heard a pre-mixed and pre-mastered recording, vocals do not blend in very well without profound amounts of EQ. Using a dummy head microphone, THX found there was a head related cancellation between 1-4khz that made central images weak and unstable.
I would say unequalized dialog or vocals
A hard speaker position would not have issues with vocal intelligibility like a phantom image would.
This is probably because they have heard the corrected equalized product, and not the pre-EQ'd product.
Elias used unequalized pink noise, so I understand his results. If I am not mistaken(and I could be), Bell Labs came to the same conclusion as Elias has
Looking at some of the articles quoted in the thread, I am not sure I understand: where do 1st order low/high filters come into play? I thought matrixing was done at all frequencies, or is Trifield different ?
Are there actually 3-speakers solutions that do NOT assume that the center speaker is identical to L and R ? The approach sounds promising and I would like to test it, but not if it means another bulky loudspeaker in the middle. How about if the center would only reproduce >100Hz, would the concept still work and would it change the matrixing ?
Interesting thread, btw.
If the conclusion is 3 channels/speakers is better than 2 speaker stereo, then yes. This was 80 years ago
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.