• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Gerzon's Trifield

gnarly

Addicted to Fun and Learning
Joined
Jun 15, 2021
Messages
988
Likes
1,390
Early in the Speaker Testing: Why mono is better thread in the ASR Video Chanel forum, Gerzon's Trifield technique came up, by KSTR and a few others.

I would like to try to implement the technique, and have started this new thread towards that goal. Seems too off-topic to the mono thread, to pursue it there.
@KSTR, i hope you see this and will chime in....and i apologize for quoting you below, jumping threads (and even forums). Not sure what the correct protolcol is here....

Anyway, I have Gerzon's AES paper http://decoy.iki.fi/dsound/ambisoni...for Multispeaker Stereo (TRIFIELD)_Gerzon.pdf

And Meridians paper

So now I'm trying to implement the technique with an open architecture processor, which I believe should be able to handle it pretty easily.

Here's snip of the schematic from the Meridian paper:
Trifield schematic.JPG


My question at this point is how are the MS MATRIX components made, specifically the Summed (M) and the Difference(S).
In Gerson's AES paper, he show's them as:
trifield matrix.JPG


I'm not sure how "matrix form" translates to 'loudpeaker feed signals form'.........

Especially after reading KSTR's quote (from earlier linked page)
"It is actually very simple to implement, You do not need three source channels for that. Just set L and R according to the simple formulas L' = L - R/2 and R' = R - L/2. The center channel is simply the average and can be obtained passively at the power amp input.
Maybe I should open a thread for this...."


Sorry if you did open a thread, and i missed it. Sure will be grateful to you and anyone else who can help get me rolling.
 

dwkdnvr

Senior Member
Joined
Nov 2, 2018
Messages
418
Likes
698
Timely post, as I've been thinking about Trifield recently as well.

Be careful in interpretation - what KSTR describes in your quoted section is one potential 'derived center' approach, but it is not Trifield. The M/S calcs should be straightforward. M = 0.4 (L + R) and S=0.4 (L-R). (Note that in matrix form it ends up not actually being required to compute these depending on the implementation details)

I implemented the Trifield algorithm in Reaper JS a decade or so ago, and found that it worked pretty well. The setup I used it in was specific to my office setup at the time though, and when we moved I never re-created it. I"m now curious about potentially setting it up in the garage/shop and potentially the living room, although placing the C speaker might be a challenge.

I believe this is very close to what I came up with: https://www.diyaudio.com/community/...els-out-of-stereo-signal.193973/#post-2661028

I'm now thinking about trying to replicate this in CamillaDSP. You can use biquads for the various high/low pass filters, but it might be better to use linear phase FIR filters (e.g designed in Rephase) to avoid unpredictable delay. Those filters plus simple mixer config to re-merge the channels is all that should be needed, I think. It lacks the interactive flexibility of the JS implementation, though.
 
OP
G

gnarly

Addicted to Fun and Learning
Joined
Jun 15, 2021
Messages
988
Likes
1,390
Timely post, as I've been thinking about Trifield recently as well.

Aah good. And thanks for the reply.
Be careful in interpretation - what KSTR describes in your quoted section is one potential 'derived center' approach, but it is not Trifield. The M/S calcs should be straightforward. M = 0.4 (L + R) and S=0.4 (L-R). (Note that in matrix form it ends up not actually being required to compute these depending on the implementation details)
That's what I thought after reading more....a 'derived center' approach compared to true Trifield.
Question: where does the 0.4 coefficient come from?
My math skills have long fled....Excel gives 2^(-1/2) (as used in the above snip from Gerzon's AES paper) as 0.707. ????
I implemented the Trifield algorithm in Reaper JS a decade or so ago, and found that it worked pretty well. The setup I used it in was specific to my office setup at the time though, and when we moved I never re-created it. I"m now curious about potentially setting it up in the garage/shop and potentially the living room, although placing the C speaker might be a challenge.

Cool.
I can already see i will have to have keep my center speaker in line with L & R (no equidistant arc) , so it will need a delay and possibly/probably some mild attenuation.

Thx for that link...will study, but I have zero experience with that software.
I'm now thinking about trying to replicate this in CamillaDSP. You can use biquads for the various high/low pass filters, but it might be better to use linear phase FIR filters (e.g designed in Rephase) to avoid unpredictable delay. Those filters plus simple mixer config to re-merge the channels is all that should be needed, I think. It lacks the interactive flexibility of the JS implementation, though.

I'm using open architecture q-sys which has drag and drop filters.... in any flavor, included various order FIR high and low pass.
If I use FIR for the hpf and lpf, I'm thinking i will need to delay the MATRIX MS "S" output by the amount of FIR delay. Correct?

Seems like the "S" line needs either an all-pass, as talked about in paper, or delay to compensate for use of FIR. ????
 

oivavoi

Major Contributor
Forum Donor
Joined
Jan 12, 2017
Messages
1,721
Likes
1,934
Location
Oslo, Norway
Bumping this thread. Any luck with going the trifield route, @gnarly or @dwkdnvr ? I have also been pondering going trifield for my desktop setup. For the moment the only way of doing so seems to be to get hold of an old meridian processor. I would ideally have liked to do it directly from my computer, surely that shouldn't be impossible? But I'm a complete noob when it comes to doing things myself in DSP software...
 

Tim Link

Addicted to Fun and Learning
Forum Donor
Joined
Apr 10, 2020
Messages
722
Likes
638
Location
Eugene, OR
I'll be watching this thread. I too am curious about the potential of upmixing 2 channel recordings to 3 channel. I tried a simple matrix recently Left = L-R, Center = L+R, Right = R-L. I had 3 matching speakers and set them up without trying too hard and had a listen. I was amazed at how good this sounded everywhere in the room despite being such a crude method. So then I set them up with all 3 speakers right next to each other, figuring it should create an Polk SDA type effect and still produce a wide sound field. This worked as predicted and sounded quite good when sitting directly ahead of the center speaker. Not so good off to the sides. Next I moved the 3 speakers to another room and set them further apart to be wider than my 75" TV. This didn't sound so good, with straight stereo sounding much better. The 3 channel mix sounded muddy and weird with the speakers that far apart. That was a big surprise and disappointment. My simple up-mix hadn't definitively beaten straight 2 channel play back after all.

What did work with the big 75" TV was Dolby Digital content mixed down to 3 channel. That sounded great!

I've also tried Dolby Pro Logic up-mixing from 2 to 3 channel. That does an amazing job of somehow steering sound cleanly into the 3 discreet channels without them seeming to bleed into each other at all, but the process seems to degrade the sound quality in some way, enough that I can't quite be happy with it.
 

Tim Link

Addicted to Fun and Learning
Forum Donor
Joined
Apr 10, 2020
Messages
722
Likes
638
Location
Eugene, OR
This weekend I worked on implementing my simple center channel derivation matrix and realized it can be done with just a two channel amplifier and a stereo signal. The standard L/R stereo signal has to be converted to a L+R / L-R stereo signal. I got the best sounding results doing the summing/difference digitally using Audio Hijack on my Mac Mini. The L+R signal is sent to the center channel. The side channels get the L-R signal and are wired together in parallel, with the right side speaker wired with reverse polarity to create the R-L signal.

This works really well, and produces an amazingly wide sound stage on some material, while retaining perfectly unprocessed purity of center panned vocals and instruments. Instead of the usual stereo phantom center image, you end up with phantom side images and a solid center image. The distance between the speaker drivers is critical and should be adjusted to match the listener's head.

After some further thought it occurred to me that recursive crosstalk elimination could also be implemented mechanically by adding more left and right side channels, all spaced ear distance apart. This array still only requires a two channel amp and a stereo signal, with the only processing being signal summing and difference, and maybe some EQ and attenuation to optimize the channels further outboard for HRTF differences as the sound is hitting the head at increasingly bigger angle differences. I've only tried it so far with one speaker in each side channel and it already works very well but I think I might have to make or buy a bunch of small speakers to see if the recursion enhances the effect. I definitely will be using at least the three speaker array as my desktop setup at work.
trifield crosstalk elimination with recursion.jpg
 
Last edited:
Top Bottom