Adaptive Filtering

Adaptive Filtering

Implementation guideline — Use DSP System Toolbox System objects to implement adaptive filters. Do not implement manual weight-update loops.

When to Use

• System identification — Model unknown FIR or IIR systems online
• Noise or interference cancellation — Recover signals from noise-corrupted measurement
• Echo cancellation — Suppress acoustic or line echo
• Active noise control — Feedforward ANC with secondary path
• Inverse system identification — Equalization and deconvolution
• Adaptive prediction — Linear prediction and speech coding
• Algorithm evaluation — Compare adaptive filter algorithm performance
• Migrating from deprecated adaptfilt.* objects — Replaced by dsp.*Filter System objects (removed in R2020a)
• Any task involving dsp.LMSFilter, dsp.RLSFilter, dsp.FilteredXLMSFilter, dsp.FrequencyDomainAdaptiveFilter, dsp.AffineProjectionFilter, or maxstep()

When NOT to Use

• Static (non-adaptive) FIR/IIR filter design — Use matlab-design-digital-filter
• Kalman filtering or state estimation — Use Control System Toolbox
• Deep learning-based denoising — Use Deep Learning Toolbox
• Simulink adaptive filter blocks — Use when working in Simulink (different modeling workflow)

Workflow

Every adaptive filtering task follows this five-step workflow:

1. Analyze the Problem

Before writing code, determine:

• Topology — System identification, inverse system identification, noise cancellation, ANC, or prediction?
• Signal characteristics — White or colored input? Stationary or time-varying?
• Constraints — Filter length, latency budget, computational cost, real-time?
• Filter length — Match or slightly exceed the unknown system order

2. Select Object and Method

Use the routing table to pick the right System object:

| Scenario | Object | Method/Config |