FM radio: https://en.wikipedia.org/wiki/FM_broadcasting

> left and right, but with opposite phase

The classic mid-side mic technique uses two capsules. The first (middle) is a cardioid or omnidirectionsl facing the subject, the latter (side) is a figure-8 turned 90 degrees. The capsules should be as close to each other as possible to avoid low-frequency phase errors. The signals are converted to L+R via a simple Mid/Side network:

L = 0.707 * (M+S), R = 0.707 * (M-S)

which is where "opposite phase" comes into play.

https://www.uaudio.com/blog/mid-side-mic-recording/

Also in FM stereo, if I'm not mistaken. One band contains L+R, the other one, which is not played on a mono radio, L-R.

It's also kinda related to how NTSC works - the color signal is at a frequency and phase such that it falls inbetween consecutive horizontal lines (e.g. pixels), so that when played on a black and white TV that isn't aware of color, the chrominance (as opposed to luminance) isn't visible since it occurs when the beam is advancing.

I’m not sure I agree with that explanation of the color subcarrier.

The color subcarrier is modulated to a higher frequency, but on an old B&W TV, there are no pixels, only lines. The color subcarrier can appear as a pattern superimposed over the picture if it is not filtered out with a low-pass filter—and indeed, B&W TVs made after the advent of color television contain such a filter, which can be removed if you want to use the TV as a higher-resolution monitor for your home computer (ask how I know...)

The two chroma channels are, however, “rotated”, similar to the way mid-side encoding is done. You take a YUV signal, you extract the chrominance UV channels, and then you rotate them to IQ channels. The I and Q channels are then quadrature modulated with different amounts of bandwidth assigned to the I and Q channels. The I channel gets, like, 3x as much bandwidth.

Broadcast NTSC did actually modulate the color signal into a "comb-shaped" frequency spectrum that was overlaid onto the complimentary "comb-shaped" spectrum of the B+W signal, because they didn't want to have to re-allocate higher broadcast channel bandwidths to be able to support color (and therefore have fewer channels available). The comb-shaped nature of the spectrum is an artifact of the discrete lines the picture signal is made up of, and the pause between each line when the electron beam reset itself and prepared to draw the next line. The fact they were able to figure out how to do color in the same bandwidth in a backward-compatible way is actually kind of insane, and I've seen it mentioned somewhere as one of the, if not THE most impressive engineering achievements of the 20th century.

Arguably YUV (luminance plus two chroma dimensions), as used in lots of video codecs.

Yes, that’s a good point. And then in NTSC, the UV is further rotated to produce IQ, which are given different amounts of bandwidth in the color subcarrier.

I thought that was where the article was heading when it mentioned that horizontal was better (and the unstated fact that mono was horizontal), but it still took me a minute to realize that the 45 degree pickups are equivalent to midside. In (digital) mid-side you encode M=L+R and S=L-R. With this, if you breakdown the horizontal and vertical components:

    HL = +L/sqrt(2)    VL = +L/sqrt(2)   L = sqrt(HL^2 + VL^2)
    HR = +R/sqrt(2)    VR = -R/sqrt(2)   R = sqrt(HR^2 + VR^2)
     H = HL + HR        V = VL + VR
       = (L+R)/sqrt(2)  V = (L-R)/sqrt(2)

Note that VR is negative and HR is positive because the sign of the pickup is inverted in the diagrams.