Introduction
FIR and IIR filters both appear stable in textbook theory.
In embedded DSP systems, their real-world behavior can be dramatically different.
Engineers frequently discover that designs which simulate perfectly become unstable, noisy, or fragile once deployed.
This article explains the practical stability differences between FIR and IIR filters under real numerical constraints.
Theoretical Stability vs Practical Stability
Mathematically:
- FIR filters are always stable
- IIR filters are stable if poles remain inside the unit circle
Numerically:
- FIR filters accumulate rounding error slowly
- IIR filters amplify coefficient and state errors
The difference matters in finite-precision hardware.
Why FIR Filters Are Naturally Robust
FIR structures:
- contain no feedback
- do not recycle numerical errors
- have bounded impulse responses
This makes them extremely tolerant to:
- quantization
- coefficient rounding
- arithmetic noise
Their downside is computational cost.
Why IIR Filters Are Numerically Sensitive
IIR filters rely on feedback:
[ y[n] = \sum b_k x[n-k] - \sum a_k y[n-k] ]
Any error is re-injected continuously.
High-Q or high-order IIR designs:
- magnify rounding noise
- drift under coefficient error
- approach instability boundaries
This causes ringing, oscillation, or blow-up.
The Embedded Reality
Embedded platforms impose:
- 32-bit float or fixed point
- limited headroom
- saturation effects
IIR sensitivity increases dramatically under these conditions.
FIR remains predictable.
Engineering Tradeoff Summary
| Aspect | FIR | IIR |
|---|---|---|
| Stability | Extremely robust | Sensitive |
| CPU cost | High | Low |
| Precision tolerance | Excellent | Limited |
| Drift tolerance | Excellent | Moderate |
| Implementation risk | Low | Higher |
When IIR Is Still the Right Choice
IIR is appropriate when:
- computational budget is tight
- bandwidth is narrow
- coefficients are carefully constrained
- drift is measured and controlled
Engineering discipline is required.
Engineering Takeaway
FIR offers brute-force robustness.
IIR offers efficiency at the cost of numerical fragility.
Embedded DSP systems favor stability margins over theoretical elegance.
Fixed-point behavior is detailed in: Fixed-Point DSP Filter Stability
Limit cycle effects are discussed in: Limit Cycles in IIR Filters Under Fixed-Point Arithmetic
Back to Stability Pillar: Embedded DSP Filter Stability
Conclusion
Most real-world DSP failures blamed on “filter design” are actually numerical stability problems.
Understanding FIR vs IIR behavior under finite precision prevents fragile deployments.
Robust systems are built with margin — not just math.