FIR, IIR, And Room Correction

What You Will Be Able To Do

• Choose IIR, FIR, all-pass, delay, or no correction from the measured problem.
• Separate FIR architecture from linear-phase design.
• Separate IIR architecture from minimum-phase classification.
• Interpret minimum phase, excess phase, and excess group delay without one fixed frequency boundary.
• Choose filter length and phase from transition, time-support, latency, and hardware constraints.
• Prepare, run, export, and independently validate automated room correction.

Three Filter Roles

The Rule That Replaces “FIR Is Better”

1Prepare A Known Baseline

1. Verify gain structure, routing, polarity, channel identity, and protective crossovers.
2. Disable undocumented correction, dynamics, loudness, and factory processing where possible.
3. Capture repeatable individual-channel and combined measurements.
4. Save the baseline preset, raw measurements, and listening notes.
5. Define the seat or listening region the correction must serve.

2Measure The Electrical Filter First

Before blaming the cabin, measure the DSP's electrical input-to-output transfer at safe line level. Record filter type, order, coefficients or software settings, sample rate, precision, block order, peak gain, magnitude, phase, group delay, and latency.

Then measure the installed acoustic result. The acoustic response is the electrical filter multiplied by the driver, enclosure, mounting, propagation, and reflection behavior.

3Build The IIR Reference

1. Create a unity bypass path.
2. Add one biquad and verify its electrical magnitude and phase.
3. Increase Q and inspect decay, group delay, and internal level.
4. Compare single- and double-precision modes if the platform provides them.
5. Test coefficient changes, bypass, reset, and preset switching with mute or slew protection.

4Build The FIR Reference

Use a short, known coefficient set first:

y[n] = h[0]x[n] + h[1]x[n-1] + ... + h[N]x[n-N]

1. Confirm coefficient order, format, scaling, and sample rate.
2. Load with outputs muted.
3. Send an impulse and verify the output matches the coefficient set.
4. Measure magnitude, phase, impulse response, group delay, peak gain, and total latency.
5. Compare a symmetric linear-phase and a minimum-phase version of the same magnitude target.

ADAU1701 / Wondom Bench Path

The enclosed Wondom two-input/four-output ADAU1701 is a strong platform for IIR, routing, summing, dynamics, and short resource-verified FIR experiments. It should not be described as a long-FIR room-correction engine without checking the actual compiler, instruction count, data memory, supported block, sample rate, and hardware revision.

Use a more capable convolution engine or processor for long mixed-phase room correction when the ADAU1701 resource report cannot support the required design.

5Separate Filter Length From Resolution

• tap count sets finite time support
• filter order is taps minus one
• transition width depends on design constraints
• DFT bin spacing is not a promise that every feature can be corrected
• linear-phase group delay is order divided by two samples only for the appropriate symmetric design
• block, transport, converter, and other latency must be added separately

6Generate Minimum And Excess Phase

1. Use a stable timing reference and document any bulk-delay removal.
2. Measure beyond the band being evaluated and remain above the noise floor.
3. Record impulse windows and calibration handling.
4. Generate the minimum-phase response from the measured magnitude.
5. Inspect excess phase and excess group delay.
6. Repeat after any window or timing change.

Do not assume everything below 300 Hz is minimum phase. Evaluate the actual frequency regions and positions.

7Classify The Problem Before Inverting

8Define The Correction Contract

Before generating a filter, write down:

• target magnitude and phase
• correction frequency range
• boost and cut limits
• spatial positions and weights
• regularization or inversion limits
• sample rate, tap count, coefficient format, and device resources
• latency budget and mute behavior
• driver, amplifier, converter, and protection limits

9Inspect Time-Domain Energy

Compare linear-phase, minimum-phase, and mixed-phase impulse responses with the same time reference. Record energy before and after the chosen main event, total delay, group delay, and peak gain. Do not call a graph pre-ringing until the aligned main event is defined.

10Run Automated Correction Deliberately

1. Follow the current product and device instructions for position count and geometry.
2. Document software, firmware, license, channel groups, and measurement template.
3. Review every measured channel before filter design.
4. Set a realistic target, correction range, and gain limits.
5. Export to a named filter slot or preset.
6. Retain the raw project and target data.

Do not hard-code one nine-position pattern, one 18-inch radius, one ±12 dB limit, or one branded target curve as a universal rule.

11Independently Verify The Export

1. Start at low level and verify channel identity and polarity.
2. Measure with a tool and workflow independent of the correction prediction.
3. Compare bypass and corrected magnitude, phase, impulse, group delay, and latency.
4. Repeat at the calibration positions and nearby points.
5. Check DSP headroom, analog output, amplifier clipping, distortion, compression, excursion, and thermal behavior.
6. Listen with representative material and preserve a known-good bypass preset.

Troubleshooting Matrix

Continue

• Print the FIR And Room-Correction Workflow
• IIR Filter
• FIR Filter
• Minimum-Phase Response
• Automated Room Correction