Sum a stereo mix to mono and listen. If the kick stays solid but the wide synth pad disappears, you have just witnessed the M-S matrix doing exactly what it was designed to do. Most engineers treat this as a test to pass. The ones who understand the underlying math can use M-S processing deliberately, not just defensively.
The idea is almost a century old. On December 14, 1931, British engineer Alan Blumlein filed British Patent 394,325. Among the techniques it described: a microphone recording method using a cardioid capsule aimed at the source and a figure-8 capsule aimed sideways. He called the outputs Mid and Side. When decoded back to left and right, you get stereo. When the Side channel is muted, you get mono with no information loss. This was not a coincidence. It was a geometric property of the encoding.
The Rotation
L/R stereo can be thought of as a 2D coordinate system. Left is one axis, Right is the other. Every stereo signal is a point in that space.
Mid-Side encoding rotates that coordinate system by exactly 45 degrees:
M = (L + R) * (1/√2)
S = (L - R) * (1/√2)The inverse rotation recovers L and R:
L = (M + S) * (1/√2)
R = (M - S) * (1/√2)The factor 1/√2 (approximately 0.7071) is not arbitrary. A pure rotation matrix must preserve the length of every vector in the space, which means it preserves power. Without the 1/√2 normalization, encoding a full-scale mono signal (L = R = 1.0) would produce M = 2.0, which is 6 dB over full scale. With 1/√2, M = 1.0. Power is conserved, clipping is avoided, and the round-trip encode/decode is lossless at unity gain.
Written as a matrix multiplication:
[M] [ 1/√2 1/√2 ] [L]
[S] = [ 1/√2 -1/√2 ] [R]This is the normalized 2x2 Hadamard matrix, the simplest orthonormal transform. Every MPEG-1 Layer III (MP3) stereo joint-coding block uses this same 2x2 rotation to decorrelate channels before psychoacoustic compression. When you stream music at 128 kbps, this matrix runs on every joint-coded frame.
Why Mono Compatibility Falls Out for Free
The Mid channel is (L + R) / √2. Summing your stereo mix to mono divides both channels by 2, giving (L + R) / 2. That is the Mid channel scaled by 1/√2. The geometry guarantees it: when you mute S and listen to M alone, you hear exactly what a mono listener hears (within a constant gain factor).
The Side channel is (L - R) / √2. When a listener with a mono speaker sums L + R, the Side contribution becomes (L - R) + (R - L) = 0. Side energy cancels completely in mono. Every element in your mix that lives predominantly in the Side channel is inaudible on a mono speaker. This is not a flaw in mono playback. It is a direct consequence of where those sounds are in the L-R coordinate system.
A signal that appears only in the left channel sits at coordinates (1, 0). Its Mid value is 0.707, its Side value is 0.707. Mono listeners hear it at 0.707 amplitude, not zero. A signal that appears equally in both channels but with opposite polarity (fully out of phase) sits at (1, -1). Its Mid value is 0, its Side value is 1.414. Mono listeners hear nothing. This is the case that fails the mono test: signals with high phase offset between channels are pure Side energy and disappear.
The Microphone Technique
Blumlein's original application was microphone placement. The M-S technique positions two capsules at the same point in space (or as close as physically possible), stacked vertically:
- The Mid mic faces the source directly. Any pattern works, but cardioid is common because it rejects room reflections from behind.
- The Side mic is a figure-8 (bidirectional), oriented 90 degrees off-axis. Its positive lobe faces left, its null faces forward, its negative lobe faces right.
After recording, the decode matrix produces L = M + S and R = M - S (with √2 scaling). Because both capsules are at the same physical point, the stereo image has no comb filtering from arrival-time differences between the pair. Width is entirely phase coherent.
The commercial advantage over XY or ORTF is that stereo width becomes a post-recording variable. The gain of the Side channel relative to Mid sets the width. Attenuate Side at mixdown and the image narrows. This is adjustable without re-recording. For broadcast applications where mono transmitters were common, M-S was the professional standard precisely because mono compatibility was built in.
M-S in the Mastering Chain
The mastering use case is different: the recording already exists as L/R. You encode it to M/S, process Mid and Side independently, then decode back to L/R.
Two operations that are difficult in L/R become straightforward in M/S:
Side high-pass filter. Bass and sub frequencies below roughly 120 Hz should be in mono on most music. Low frequencies in the Side channel cause phase cancellation on speaker systems that share a subwoofer, and club PA systems routinely sum below 150 Hz. A high-pass filter on Side at 100-120 Hz forces sub energy into the Mid channel. The audible result is a tighter, more focused low end.
Differential EQ. The mid vocals and kick sit in M. Room reverb tails, string pads, and overhead cymbals sit predominantly in S. A high-shelf boost on Side at 6 kHz adds perceived air and width without touching the center image. A dip on Side in the 300-600 Hz range removes boxiness from room tone without dulling the snare. In FabFilter Pro-Q 3, individual bands can be assigned to Mid-only, Side-only, Left-only, Right-only, or stereo, so a single instance handles both regular EQ and M-S EQ simultaneously.
Implementation in C++
The encode and decode are a four-multiply, two-add operation per sample:
constexpr float kInvSqrt2 = 0.7071067812f;
void encodeMidSide(float left, float right,
float& mid, float& side)
{
mid = (left + right) * kInvSqrt2;
side = (left - right) * kInvSqrt2;
}
void decodeMidSide(float mid, float side,
float& left, float& right)
{
left = (mid + side) * kInvSqrt2;
right = (mid - side) * kInvSqrt2;
}SIMD implementations process four or eight stereo pairs per instruction. The SSE2 _mm_addsub_ps instruction computes alternating adds and subtracts in a single cycle, making it a natural fit for vectorized M-S decode. At 44.1 kHz stereo, encode plus decode costs roughly 8 multiply-add operations per sample pair. Even on a 2 MHz Motorola 68000, this runs in real time. The compute cost of M-S is negligible compared to any downstream plugin.
The Phase Correlation Meter
A phase correlation meter shows where a stereo signal sits relative to the M-S axes. The readout ranges from +1 (signal is pure Mid, fully mono-compatible) to -1 (signal is pure out-of-phase, all Side). A reading near zero means Mid and Side energy are roughly equal.
Streaming delivery targets for broadcast and music typically require a phase correlation above 0.5 to ensure adequate mono compatibility. A stereo master that reads consistently between 0.3 and 0.7 across the track represents healthy stereo width without over-correlated dullness or cancellation-prone phase spread.
What Blumlein Actually Gave Us
The 1931 patent solved a cinema problem: sound that followed actors across the screen. The M-S matrix was the mechanism. Ninety-five years later, the same 45-degree rotation appears in MP3 codecs, mastering chains, microphone arrays, and streaming normalization pipelines.
When a mix collapses in mono, the rotation makes the problem visible. The energy is in the Side quadrant of L-R space. Move it toward Mid and the problem disappears. That is not a frequency problem, a level problem, or a dynamics problem. It is a geometry problem. Once you see the coordinate system, the fix is obvious.