IIR Filters vs FIR Filters
Both filter types can implement octave bands, weighting curves, or noise shaping. The right choice depends on phase, CPU, and stability requirements.
Info topics
Core differences
| Aspect | IIR | FIR |
|---|---|---|
| Structure | Recursive, feedback terms | Non-recursive, feedforward taps |
| Phase | Non-linear phase | Can be linear phase |
| Latency | Low | Higher (kernel length) |
| CPU and memory | Low for steep filters | Higher for narrow bands |
| Stability | Depends on pole locations | Always stable |
Why IIR is common for octave bands
Octave and third-octave filters need steep skirts to meet standards. IIR designs achieve this with fewer coefficients, making them efficient for real-time analysis and portable devices. Many published standards are defined from analog prototypes that naturally map to IIR filters.
When FIR is the better choice
- When you need linear phase to preserve waveform shape or transient timing.
- When you already process in blocks (FFT-based) and can reuse convolution infrastructure.
- When exact magnitude control at band edges is more important than CPU usage.
Large FIR kernels can be expensive, but multi-rate processing or FFT convolution can reduce the cost.
Practical design notes
- Use SOS cascades for IIR filters and verify pole radii are less than 1.0 in the z-plane.
- For FIR, choose a window or Parks-McClellan design, then validate stop-band attenuation.
- Always test on real data: band levels, time constants, and aggregation can alter the subjective output.