DSP as the Last Resort

"As long as we are concerned with the realistic reproduction of sound, the original sound must stand as the criterion by which the reproduction is judged!"

The Appeal and the Problem

Digital signal processing has made equalization more accessible, more precise, and more capable than anything the analogue era produced. A modern convolution engine can apply correction filters with a resolution that would have required a roomful of analogue hardware a generation ago. Room correction platforms such as Dirac Live, Audyssey, and ARC offer automated measurement cycles followed by correction curves that, on a frequency-response plot, look genuinely impressive. The corrected line is flat. The room, apparently, has been tamed.

It has not. What the plot shows is a frequency-domain snapshot at a single measurement point. What the plot cannot show is what was sacrificed to achieve it: phase coherence, transient accuracy, headroom, and the temporal integrity of the signal as it actually arrives at the listener's ears. The correction is real. So are the costs.

This piece argues a specific position, one grounded in physics rather than preference: equalization, whether analogue or digital, is a tool of last resort. It is appropriate when every upstream remedy has been exhausted and a measurable, audible problem remains. It is not a substitute for loudspeaker placement, room treatment, equipment matching, or source quality. Applied prematurely, DSP correction does not solve problems. It rearranges them, often in ways that are harder to identify and harder to undo.

The Acoustical Basics guide on this site addresses the room as the primary acoustic variable and establishes why physical remediation must precede any electronic intervention. This piece goes further into the domain of equalization itself, examining what happens inside the signal when correction filters are applied, and why the frequency domain tells only part of the story.

What Equalization Actually Does

An equalizer, at its most fundamental level, is a filter. It alters the amplitude of a defined frequency range relative to surrounding content. What this description obscures is the nature of that alteration. A filter does not operate on a single frequency in isolation. It operates on a band of frequencies, and the transition regions, the slopes on either side of the target frequency, inevitably affect content the designer did not intend to change.

This is not a limitation of implementation. It is a consequence of the mathematics that govern all linear time-invariant systems. The Fourier transform, which underpins all frequency-domain analysis, describes signals in terms of sine waves that extend infinitely in both time directions. A filter that modifies amplitude at a given frequency modifies the phase relationships of neighboring frequencies as a direct consequence. The two cannot be separated. They are different expressions of the same physical operation.

What is often overlooked is that amplitude response is only the magnitude component of a complex transfer function. The phase response is the complementary component, and the two are linked through the Hilbert transform. Any attempt to alter one necessarily alters the other unless non-causal processing is introduced, which brings its own consequences.

Consider a parametric boost applied at 80 Hz to compensate for a perceived bass deficiency. The filter does not raise 80 Hz alone. It raises a band centered on 80 Hz, with a slope that affects content on either side. The width of that band, described by the Q factor, determines how rapidly the influence falls away. A narrow Q of 4 or higher confines the effect relatively tightly but creates steep slopes that impose significant phase rotation at adjacent frequencies. A wide Q of 0.5 spreads the influence across several octaves with gentler phase consequences but alters far more of the spectrum than intended.

Neither choice is without cost. The engineer selects not between good and bad but between different distributions of the same unavoidable compromise.

The same logic applies in subtraction. A cut at a resonant frequency reduces the amplitude of the problem but affects the phase relationships of surrounding content in the process. The resonance in the acoustic domain does not disappear. The recording contains frequencies that interact with the room and produce the resonance acoustically. The filter changes what the loudspeaker outputs. The room still does what it does. The two processes do not cancel cleanly.

The Phase Consequence

Phase is not an abstract concept. It is a direct description of the timing relationship between different frequency components within a waveform. When a complex sound arrives at the microphone, its constituent frequencies are in a specific temporal relationship with one another. That relationship encodes the identity of the sound.

Phase determines when energy arrives. More precisely, it defines the relative timing of spectral components that together form a waveform. A musical signal is a time-coherent structure in which harmonic components align to produce recognizable events such as attack, texture, and spatial localization.

When phase relationships are preserved, transients retain their original geometry. The leading edge of a percussive event remains sharp. Harmonic structures build naturally. Spatial cues remain intact.

When phase is altered in a frequency-dependent way, these relationships shift. The effect is not a simple delay. It is a redistribution of energy over time. This is best understood through group delay. If different frequency components of a transient are delayed by different amounts, the transient stretches. Its peak amplitude may remain similar, but its slope is reduced. This is perceived as a loss of speed, impact, and articulation.

At low frequencies, the ear is less sensitive to steady-state phase shifts but remains sensitive to timing errors in transients. At mid and high frequencies, phase errors directly affect localization and image stability.

An equalizer distorts these relationships. Every analogue filter introduces phase rotation as a direct consequence of amplitude modification. In a system with multiple bands active simultaneously, the cumulative phase response can deviate substantially from flat across the audible band, even when amplitude appears corrected. The audible consequence is a softening of transient definition, reduced spatial precision, and a subtle loss of immediacy. These effects accumulate. This corresponds to measurable changes in group delay. A flat group delay preserves timing. A varying group delay smears it. What was coherent at the microphone arrives less coherent at the listening position.

Minimum-Phase Versus Linear-Phase DSP

Digital equalization offers a capability that analogue cannot: the choice between minimum-phase and linear-phase filters.

A minimum-phase filter behaves like analogue. Amplitude and phase are coupled. It introduces no pre-ringing and remains causal.

A linear-phase filter applies identical delay to all frequencies. Phase is preserved. Amplitude is corrected independently.

The cost is pre-ringing. Because the filter must be non-causal, energy appears before the transient that caused it. In some material this is inaudible. In high-resolution recordings it can reduce perceived precision.

Linear-phase filters do not eliminate time-domain distortion. They redistribute it symmetrically around the transient. Minimum-phase concentrates smearing after the event. Linear-phase spreads it before and after. The ear is generally more tolerant of post-event smearing than pre-event artifacts.

Hybrid approaches manage this trade-off intelligently. They do not eliminate it.

Why Subtracting Is Safer Than Adding

Cutting is preferable to boosting.

A boost increases energy demand. A 6 dB boost requires four times the power. This reduces headroom and increases the risk of clipping.

A cut reduces energy. It introduces no additional demand and cannot cause clipping.

In rooms, deficiencies are often nulls caused by cancellations. Boosting does not fix cancellations. It increases energy that continues to cancel. The problem remains.

Peaks, however, are real excess energy. Cutting them reduces the severity without introducing new problems.

Psychoacoustically, cuts are perceived as removal of coloration. Boosts are perceived as additions.

The practical conclusion is straightforward. If equalization is required, remove excess before adding anything.

The Time-Domain: What EQ Cannot Touch

Equalization addresses amplitude. It does not resolve time-domain behavior.

Room problems are not static. They are dynamic, spatial, and level-dependent. A resonance stores energy and releases it over time. Reducing its amplitude does not shorten its decay.

A 45 Hz resonance continues to ring after the transient. EQ makes it quieter. It does not make it faster.

This limitation is structural. Equalization assumes steady-state behavior. Room acoustics are not steady-state phenomena. The mechanisms that produce them, set out in Acoustical Basics, must be addressed in the room itself.

Time-domain DSP can address this partially, but it requires precision, stability, and constant recalibration. Physical treatment reduces the problem at its source and remains valid over time.

Dynamic Behavior and Headroom

Boost-based correction reduces available headroom.

When filters are applied before the DAC, boosted content pushes the signal closer to full scale. Without proper gain management, clipping or inter-sample overload can occur.

Even when internal DSP operates in floating point, output stages remain bounded. Inter-sample peaks can exceed reconstruction limits.

The result is subtle distortion that accumulates over time and is perceived as hardness or compression.

Cut-based correction does not introduce this risk.

The Room Is the Problem, Not the Loudspeaker

Room correction systems measure a combined response. They do not isolate the loudspeaker.

The correction applied is therefore correcting the room's influence, not the loudspeaker itself.

This matters because the room response varies with position. A correction valid at one point is invalid elsewhere.

Applying correction treats a spatial and temporal problem as if it were a fixed property of the loudspeaker. It is not.

This aligns with the engineering philosophy that runs through signal integrity and our engineering standard: the signal should remain intact. Corrections indicate upstream problems.

Treatment changes the cause. Correction manages the symptom.

The Correct Order of Remediation

1. Loudspeaker placement
2. Acoustic treatment
3. Equipment matching
4. Equalization

Each step reduces the degrees of freedom of the problem. Equalization applied earlier operates on an underdetermined system and produces locally optimized but globally inconsistent results. Coupling and decoupling decisions, often overlooked, sit alongside placement at step one.

When DSP Is Justified

DSP is justified when physical solutions are not feasible, in integrated active systems, or in very large rooms.

Even then, it must be applied conservatively:

• Prefer cuts
• Use minimal correction
• Manage headroom
• Choose filter types carefully
• Recalibrate regularly

A correction profile is a snapshot. Any change in system conditions alters its validity.

In Summary

Equalization is not an enhancement. It is a compromise.

Every filter introduces trade-offs: phase rotation, pre-ringing, or reduced headroom.

The correct question is whether the benefit exceeds these costs.

Too often, DSP is applied early because the measurement improves. The measurement is incomplete. Phase, time behavior, and spatial validity remain unaddressed.

The governing principle is simple: no intervention without a demonstrable problem. No aggressive solution where a minimal one suffices.

Applied correctly, this leads to a clear hierarchy: placement, treatment, matching, and only then correction.

A well-resolved system needs little correction. A poorly resolved one cannot be fixed by it.

Physics does not permit the shortcut.