Adding a Mid-derived harmonic signal to the Side path can preserve the mono sum while still shifting stereo energy toward one speaker. The problem is not simply excessive width. It is the mathematical relationship between the new Side material and the Mid signal that created it.
In a conventional Mid/Side decoder, left is rebuilt from Mid plus Side while right is rebuilt from Mid minus Side. That means any new signal inserted into Side appears with opposite polarity across the reconstructed channels.
This is why Panoramic's Mid-derived harmonic path needs more than a saturation stage and a blend control. Creating useful harmonics is only half the job. Their relationship with the source decides whether the stereo image stays balanced.
Stereo balance is different. If T stays correlated and in phase with related content in M, the left reconstruction gains energy while the right loses it. Reverse that relationship and the bias can move toward the other speaker.
That distinction is easy to miss because a mono check can pass. The added Side component disappears when left and right are summed, yet the stereo playback can still lean. Mono compatibility therefore proves that the added difference signal cancels as intended, not that left and right carry equal energy during stereo playback.
A saturation stage makes the situation more interesting because its output is not an unrelated texture. Nonlinear processing generates components whose timing and harmonic structure are tied to the input. If that processed residual comes from Mid and is dropped directly into Side, its connection to Mid can become part of the decoded left-right balance.
This is a different failure from simply boosting an existing Side channel. Existing Side information came from differences already present between left and right. Mid-derived harmonics are newly manufactured stereo information, so their phase relationship has to be managed before they can behave like a stable difference signal.
A 90-degree shift needs a quarter of one cycle. At 100 Hz it lasts 2.5 milliseconds, at 1 kHz 0.25 milliseconds, and at 10 kHz just 25 microseconds. No single delay value can satisfy all three frequencies at once.
That is why broadband quadrature processing uses frequency-dependent phase behavior rather than one ordinary delay line. An ideal Hilbert transform creates a quadrature companion while preserving magnitude. That gives engineers a phase-shifted version without applying a conventional EQ curve.
Real-time implementations usually approximate that behavior over a useful bandwidth. Panoramic documents a two-branch IIR all-pass network with eight cascaded second-order all-pass sections in each branch. All-pass filters leave magnitude nominally unchanged while shaping phase, allowing the branches to approach the required relationship across many frequencies.
The practical target is a broadband quadrature relationship between the relevant signals, not a magic delay time. Near quadrature, correlated energy no longer adds preferentially to one decoded channel in the same way. The result can contribute stereo difference without the obvious left-right tilt produced by an in-phase or polarity-reversed relationship.
That matters when widening material with sharp attacks. A decorrelated harmonic layer may be quieter than the dry Mid signal, but aggressive processing can soften localization or make attacks feel less precise. More decorrelation is therefore not automatically better.
The useful comparison is between image balance, width, and transient integrity at matched loudness. Push the generated Side contribution until the width becomes obvious, then listen for whether centered material appears to lean or whether attacks lose focus. Reducing the blend can be cleaner than demanding increasingly extreme phase manipulation from the decorrelator.
A vectorscope and left-right level meters can expose persistent bias, while a mono fold-down checks a different property. If mono remains stable but the stereo image shifts, the problem is not cancellation in the summed output. It is how the new difference material combines with Mid before that sum ever happens.
In a conventional Mid/Side decoder, left is rebuilt from Mid plus Side while right is rebuilt from Mid minus Side. That means any new signal inserted into Side appears with opposite polarity across the reconstructed channels.
This is why Panoramic's Mid-derived harmonic path needs more than a saturation stage and a blend control. Creating useful harmonics is only half the job. Their relationship with the source decides whether the stereo image stays balanced.
Correlated harmonics can pull the image sideways
Imagine a centered signal that produces a new harmonic component called T. Put T into the Side path, and the decoder produces M plus T on the left and M minus T on the right. The mono sum still becomes 2M because the two T terms cancel.Stereo balance is different. If T stays correlated and in phase with related content in M, the left reconstruction gains energy while the right loses it. Reverse that relationship and the bias can move toward the other speaker.
That distinction is easy to miss because a mono check can pass. The added Side component disappears when left and right are summed, yet the stereo playback can still lean. Mono compatibility therefore proves that the added difference signal cancels as intended, not that left and right carry equal energy during stereo playback.
A saturation stage makes the situation more interesting because its output is not an unrelated texture. Nonlinear processing generates components whose timing and harmonic structure are tied to the input. If that processed residual comes from Mid and is dropped directly into Side, its connection to Mid can become part of the decoded left-right balance.
This is a different failure from simply boosting an existing Side channel. Existing Side information came from differences already present between left and right. Mid-derived harmonics are newly manufactured stereo information, so their phase relationship has to be managed before they can behave like a stable difference signal.
One delay cannot hold a quarter-cycle shift
A fixed delay produces a phase shift that changes with frequency. One millisecond equals 36 degrees at 100 Hz, 360 degrees at 1 kHz, and several complete rotations by 10 kHz. The time offset stays fixed while the phase relationship keeps moving.A 90-degree shift needs a quarter of one cycle. At 100 Hz it lasts 2.5 milliseconds, at 1 kHz 0.25 milliseconds, and at 10 kHz just 25 microseconds. No single delay value can satisfy all three frequencies at once.
That is why broadband quadrature processing uses frequency-dependent phase behavior rather than one ordinary delay line. An ideal Hilbert transform creates a quadrature companion while preserving magnitude. That gives engineers a phase-shifted version without applying a conventional EQ curve.
Real-time implementations usually approximate that behavior over a useful bandwidth. Panoramic documents a two-branch IIR all-pass network with eight cascaded second-order all-pass sections in each branch. All-pass filters leave magnitude nominally unchanged while shaping phase, allowing the branches to approach the required relationship across many frequencies.
The practical target is a broadband quadrature relationship between the relevant signals, not a magic delay time. Near quadrature, correlated energy no longer adds preferentially to one decoded channel in the same way. The result can contribute stereo difference without the obvious left-right tilt produced by an in-phase or polarity-reversed relationship.
Decorrelation has its own audible trade-offs
Phase manipulation is not free just because the frequency-response magnitude stays flat. All-pass networks have frequency-dependent phase and group delay, which means different frequency components can move through time differently. The transient shape can change when that behavior becomes strong enough.That matters when widening material with sharp attacks. A decorrelated harmonic layer may be quieter than the dry Mid signal, but aggressive processing can soften localization or make attacks feel less precise. More decorrelation is therefore not automatically better.
The useful comparison is between image balance, width, and transient integrity at matched loudness. Push the generated Side contribution until the width becomes obvious, then listen for whether centered material appears to lean or whether attacks lose focus. Reducing the blend can be cleaner than demanding increasingly extreme phase manipulation from the decorrelator.
A vectorscope and left-right level meters can expose persistent bias, while a mono fold-down checks a different property. If mono remains stable but the stereo image shifts, the problem is not cancellation in the summed output. It is how the new difference material combines with Mid before that sum ever happens.
