FIR vs IIR Oversampling: Choosing Between OFF, 2×, 4× and 8×

An oversampling menu contains two different decisions. The factor determines how much the internal sample rate increases. The filter mode determines how the rate conversion is implemented. Neither decision can be reduced to a universal “best sound” setting.

1994 Channel Drive offers FIR and IIR modes, each with OFF, 2×, 4× and 8×. That gives you a useful range for different sessions and sources. To use it well, separate the factor, the filtering trade-offs and the sound you are trying to improve.

What the factor changes

Oversampling surrounds the relevant processing with sample-rate conversion: increase the rate, perform the processing, filter and return to the session rate. Upsampling needs interpolation filtering; downsampling needs filtering to control content that would otherwise alias at the lower rate. Merely inserting additional sample positions is not a complete oversampler. Steven W. Smith, Compression and Expansion, Multirate Methods.

At a 48 kHz session rate, the nominal rate relationships are:

These values are arithmetic examples, not a claim that every stage of 1994 runs at the displayed internal rate. The output remains at the DAW's session rate. OFF disables the oversampling selected here; it should not be interpreted as a promise of zero plugin latency or disabled ADAA.

A higher internal Nyquist gives generated harmonics more room before foldback. It does not give an arbitrarily strong nonlinearity unlimited bandwidth, and it does not repair aliasing already present in an input recording.

FIR: finite impulse response, with design choices

FIR means finite impulse response. A suitably symmetric FIR design can have linear phase, so its phase response represents a constant delay. FIR by itself does not guarantee linear phase: the coefficients still determine the response. Steven W. Smith, Phase Response.

In common oversampling implementations, linear-phase FIR filters trade latency and computation for that phase behaviour. A sharp transition may also produce ringing on either side of the delayed main impulse. Whether any pre-ringing is perceptually relevant depends on the design, frequency region and source—not just the label FIR.

Do not assume that a linear-phase resampling filter turns an entire analogue-style channel strip into a linear-phase processor. The channel's own EQ and nonlinear stages remain part of the path.

IIR: feedback with a different phase-and-latency balance

IIR means infinite impulse response. These filters commonly use feedback, allowing a response to continue beyond a finite set of input contributions. Efficient, steep responses are possible with relatively compact structures. Steven W. Smith, The Recursive Method.

For a concrete library example, JUCE's oversampling class offers linear-phase FIR equiripple filtering and polyphase-allpass IIR filtering. Its documentation describes a latency-versus-phase trade-off, with the IIR option reducing latency and changing phase around Nyquist. This illustrates a common design choice; it does not independently establish 1994's exact coefficients or measured latency. JUCE Oversampling documentation.

The practical distinction is therefore not “FIR is professional, IIR is cheap.” Both can be appropriate engineering choices. The useful question is how the available modes behave at the chosen factor in your session.

Compare one variable at a time

Start with a short passage that exposes the processing: sustained high notes, bright percussion or a driven vocal with clear consonants. Use a realistic Drive amount and keep the EQ and Emphasis unchanged.

First, select one filter family and compare OFF, 2×, 4× and 8×. Keep the output level consistent. Ask whether the higher setting produces a worthwhile improvement in the actual mix, not simply a different sound in solo.

Then compare FIR and IIR at the factor you are considering. Allow the DAW and plugin to settle after a setting change; restarting playback can help make the comparison repeatable. Check session responsiveness and any reported latency as well as the sound.

Finally, revisit the setting once the source is playing with the rest of the arrangement. A difference that matters on an exposed synth lead may be irrelevant on a tucked-away percussion layer. A setting that works on one instance may become inconvenient across a large session.

A sensible way to choose in 1994 Channel Drive

These are suggested tests, not factory defaults or measured performance recommendations. There is no requirement to use 8× everywhere, and no reason to reject it where it provides a useful result.

Keep phase and latency separate

The manual describes latency-compensated bypass and warns that the analogue-style EQ introduces frequency-dependent phase shift. Those statements concern different effects. Host delay compensation can align processing delay; it does not automatically reverse a filter's phase curve or nonlinear changes.

If you blend a driven return with the dry signal, judge the blend. A wet sound that works by itself may combine differently once both paths play together. Changing an oversampling mode can be a reason to recheck the result, particularly when the arrangement depends on close timing relationships.

Choose the mode and factor that help the mix, then keep that decision consistent through the final check. The value of the menu is control over a trade-off—not a number to maximise automatically. Explore 1994 Channel Drive.

Related reading

• Virtual Analog Saturation Without the Hype: From Tanh to Musical Drive
• Sallen–Key Filters: Why Position Matters as Much as Cutoff
• ADAA Explained: A Better Way to Evaluate a Saturation Curve