Most engineers reach for M/S processing to apply mid/side EQ or tighten low-end width. Few realize they are running a 1931 linear transform that Alan Blumlein filed in a British patent in December of that year. And fewer still understand why that transform provides a mathematical guarantee — not an approximation, a proof — that the mono fold contains exactly what it should, with zero information loss.
The Matrix
The core is a 2x2 linear transform:
M = (L + R) / 2
S = (L - R) / 2Invert it, and you get the decode:
L = M + S
R = M - SVerify with numbers. If L = 0.8 and R = 0.4:
- M = (0.8 + 0.4) / 2 = 0.6
- S = (0.8 - 0.4) / 2 = 0.2
- L_out = 0.6 + 0.2 = 0.8
- R_out = 0.6 - 0.2 = 0.4
Round-trip: perfect reconstruction. The transform is lossless by definition — it is an orthogonal matrix rotation with a scaling factor.
The Gain Trap That Catches Engineers
Some DAW implementations and hardware units skip the /2 normalization in the encoding step:
M = L + R
S = L - RIf you then decode with the un-normalized version as well:
L = M + S = (L + R) + (L - R) = 2L
R = M - S = (L + R) - (L - R) = 2RThat is a 6 dB gain increase on every signal after a round-trip. This is where the "apply -3 dB at encode and -3 dB at decode" advice in older hardware manuals comes from: they were correcting for the un-normalized convention. Modern plug-ins handle this internally, but if you are building an M/S processor in a modular environment or writing DSP code, mixing conventions from different sources will silently add 6 dB everywhere.
The clean approach is to use the normalized form — dividing by 2 in the encode step — so that L = M + S and R = M - S require no further scaling.
Why Mono Compatibility Is Guaranteed
When a mono playback system receives only the M signal:
M = (L + R) / 2That is the perfect mono mix by definition. It contains no S content at all. The S channel holds everything that is different between L and R — stereo width, room tails, panned delays, doubled guitars. When you fold to mono, that content does not cancel and does not leak: the M channel simply never held it.
This is narrower than it might sound. M/S encoding does not protect you from phase-cancellation problems already present in the stereo signal. If you recorded a guitar twice and panned the two takes hard left and right with a slight timing offset, the comb filtering lives inside L and R as a frequency-response problem. M will contain the combed average, and S will contain the combed difference. The transform faithfully represents what is there — it does not clean up bad stereo practice.
What it guarantees is the round-trip. If you start with clean stereo audio, process the M and S channels independently, and decode back to L/R, you get exactly what the processing produced — no artifacts from the conversion itself. That guarantee makes M/S processing a standard tool in mastering precisely because you can target center content and stereo content separately without any interaction from the matrix.
What the Correlation Meter Is Actually Showing
Phase correlation meters display the Pearson correlation coefficient between L and R:
- r = +1: L = R, pure dual-mono, no stereo information
- r = 0: L and R are statistically independent, maximum stereo spread
- r = -1: L = -R, fully inverted dual-mono, the Mid channel carries zero energy
The relationship to M/S is direct. When r = +1, the S channel has zero energy: (L - R)/2 = 0 for identical signals. As r drops toward 0, energy distributes more evenly between M and S. At r = -1, the M channel has zero energy and S holds everything. A steady correlation reading below zero means the mono fold produces silence at those frequencies. You can see this on a meter before you hear it.
The non-obvious implication: a wide stereo image from simple hard panning still reads positive correlation (if the panned content is similar). A narrow image can read negative correlation if there is phase divergence at the center. Correlation measures phase relationship, not spatial width. M/S width controls are adjusting the S-to-M ratio, which changes spatial perception, but the correlation meter responds to the phase math, not your ears.
Mid/Side EQ in Practice
When you cut 200 Hz from the Mid channel in FabFilter Pro-Q or any M/S-capable EQ, you are cutting from the summed center image. That affects kick drum, bass guitar, lead vocal — the elements that mix engineers anchor to center. The Side channel at 200 Hz contains whatever room reflection, reverb, or stereo-panned low content exists at that frequency.
Cutting Side below 150 Hz is one of the most common mastering moves precisely because of what it does not touch: the bass guitar and kick, which live almost entirely in the Mid. You tighten the low-frequency stereo image without touching mono-compatible bass energy. The math explains why: a Side-only cut leaves M intact at that band, and M is what reaches mono listeners.
M/S Compression
A compressor operating on the Mid channel responds to center loudness, not stereo loudness. When the kick drum peaks — center, mono, high energy — a Mid compressor reacts. The Side channel, carrying reverb and width information, is completely unaffected. This is why M/S mastering compression holds the stereo field stable under gain reduction in a way that L/R compression cannot: L/R compression reacts to the loudest channel, which pumps both channels together and moves the stereo image.
The Microphone Technique
Blumlein's 1931 patent covered the microphone application directly. An M/S recording setup uses a cardioid (Mid, pointing at the source) and a figure-8 (Side, oriented exactly 90° to the mid mic, typically placed directly above it to minimize acoustic separation). The figure-8 captures the left side of the room in its positive lobe and the right side in its negative lobe.
Decoding works by copying the Side signal, inverting the polarity of one copy, and panning the original and the inverted copy hard left and right. The output is mathematically equivalent to two cardioids aimed 45° left and 45° right — a virtual coincident X-Y pair. What is on-axis appears center. What is 90° off-axis appears at the stereo edges.
The critical practical advantage: stereo width is adjustable after recording by changing the Side-to-Mid ratio during decode. With an X-Y or ORTF array, the stereo image is fixed at recording time. With M/S, you can narrow the room ambience in post without losing the direct sound, or widen it without adding artificial reverb.
The Transform at the Bottom of Everything
The M/S matrix separates two orthogonal dimensions of a stereo signal: what both channels share versus what makes them different. Process those dimensions independently and decode. Blumlein derived this in 1931 from first principles, motivated by the problem of making cinema sound follow the actor across the screen. The matrix he wrote down is the same one running inside every M/S-capable plug-in shipping today. Understanding it at the level of the equations — not just as "wide" and "narrow" knobs — makes the gain staging, the correlation meter, and the microphone decode follow from the same few lines of arithmetic.