Envelope & Crest-Factor Detection for Overload and Dynamic Limiting
← Back to:Active Filters & Signal Conditioning
Envelope and crest-factor detection turn “overload” into measurable events (Env/CF/flags) and drive stable limiting (attack/hold/release) that protects headroom without pumping, chatter, or missed bursts. The focus is a production-ready method: define thresholds and time constants from the headroom budget, then verify with burst-realistic tests and data-based pass criteria.
This page explains how to detect overload reliably using envelope and crest factor metrics, and how to convert detection into fast, stable dynamic limiting without corrupting the passband.
Scope & boundary: what this page owns (and what it refuses)
What this page covers (owned content)
- Envelope detection: extraction architectures, bandwidth/latency trade-offs, robustness vs false triggers.
- Crest-factor (CF) estimation: windowing, multi-timescale CF, burst realism, repeatable measurements.
- Overload criteria: event definitions, thresholds + hysteresis + duration rules, severity grading (warn/limit/fault).
- Limiter control laws: attack/release/hold behavior, stability against chatter and pumping, practical pass/fail criteria.
- Verification: stimulus sets, what to log, how to define “recovery time”, false-trigger rate, and audible artifact limits.
What this page refuses (to prevent topic overlap)
- RMS-to-DC internal design: no deep circuit derivation or IC internals (owned by the RMS-to-DC subpage).
- Precision rectifier / peak-hold topology derivations: only interface-level implications (owned by Precision Rectifier / Peak Hold).
- AGC loop theory and comm standards: no system-level AGC stability deep dive (owned by an AGC/RF baseband page if present).
- ADC/DAC sampling theory: no sampling theorem math; only interface-level measurement traps and criteria.
Rule: if a question belongs to a sibling page, only provide a one-line interface takeaway and keep the detailed explanation out of this page.
Deliverables (what a reader should be able to implement)
- A concrete overload event taxonomy (clip / soft compression / slew limiting / recovery-induced overload).
- A metric pipeline (envelope + RMS estimate → CF → decision → limiter action).
- A decision template (thresholds + hysteresis + duration + severity grading).
- A time-constant template (attack / hold / release) with testable criteria and placeholder thresholds.
- A verification recipe with repeatable stimulus sets and “pass/fail” fields suitable for production.
Vertical fields (kept consistent across all sections)
| Field | Typical values | Why it matters |
|---|---|---|
| Input domain | Audio / RF baseband / IF | Sets bandwidth, latency budget, and artifact tolerance. |
| Overload risk type | Clip / soft compression / slew limiting / slow recovery | Determines which metric is sensitive and what action is safe. |
| Decision output | flag / gain reduction / limiter state | Enables production bins and deterministic field behavior. |
| Budgets | latency / false triggers / artifacts | Converts “looks OK” into measurable thresholds and guardbands. |
Why envelope & crest factor: what overload really means in practice
Overload is a family of events (not only clipping)
“Overload” should be defined as observable, testable events that threaten headroom, distortion, or recovery. Different events require different metrics and actions.
| Event type | Observable symptom | Most sensitive metric | Quick check | Safe action |
|---|---|---|---|---|
| Hard clip (rail hit) | Flat tops; strong harmonics; immediate headroom collapse. | Envelope peak threshold | Capture max envelope during a burst; compare to headroom margin. | Fast limiting; optional fault if duration exceeds N ms. |
| Soft clip / compression | Distortion rises before rail; “loud” without obvious clipping. | Envelope + severity grading | Sweep level and observe THD jump near the onset region (no deep theory required). | Soft-knee limiting; avoid aggressive attack that creates clicks. |
| Slew-rate limiting | Slope saturates; HF distortion; may occur below rails. | Envelope + burst sensitivity | Check rise/fall slope at high frequency; compare against expected linear slope. | Limit HF bursts; add hold to prevent chatter at the edge. |
| Recovery-induced overload | Slow release/settling makes the next burst worse; “pumping”. | Envelope sequence + CF windowing | Two-burst test: measure recovery to within 1 dB before the second burst. | Tune release and hold; add state gating to avoid oscillation. |
Engineering rule: Envelope is the fast protection metric; crest factor classifies “burstiness” and improves decision stability through windowing and severity grading.
Why RMS is not enough for overload protection
- Equal RMS does not imply equal peak: headroom is consumed by peaks, not by average energy.
- Waveform structure matters: bursts and multi-tones can create high peaks even when RMS is moderate.
- Overload criteria must be window-aware: CF depends on time window and smoothing; windows should match the protection goal (fast vs slow).
- Production stability requires budgets: false-trigger rate and recovery time must be turned into pass/fail thresholds, not opinions.
Practical split: use envelope peak to guard headroom; use CF to decide whether a peak is expected (bursty content) or suspicious (noise/spike), then grade severity.
What corrupts the passband (and how to control it)
Passband corruption is usually caused by how detection drives gain changes, not by the metric names. Three knobs dominate real-world behavior:
- Detection bandwidth: too wide → noise spikes trigger; too narrow → short bursts are missed.
- Time constants (attack/hold/release): too aggressive → clicks/chatter; too slow → pumping and poor recovery.
- Leakage paths: control feedthrough and bypass coupling can modulate the noise floor even when “not limiting”.
Debug order: pumping → check release/hold → check hysteresis → check detection bandwidth. False triggers → check bandwidth/hysteresis → check spike sources → check threshold drift.
Metric definitions that engineers can implement
Metric data model (signals, intermediate values, outputs)
Definitions should map to implementable variables that remain consistent across detection, decision, and limiter control. The table below lists a minimal set that supports production logging and repeatable verification.
| Name | Meaning | How it is formed | Used for |
|---|---|---|---|
| x(t) / x[n] | Input signal | Measured or sampled waveform | All paths |
| abs[n] | Magnitude proxy | abs[n] = |x[n]| | Envelope path |
| p[n] | Energy proxy | p[n] = x[n]^2 | RMS estimate |
| Env_est | Envelope estimate (linear) | LPF(|x|), sliding peak, or quasi-peak detector | Fast overload detection |
| RMS_est | RMS estimate (linear) | sqrt( average( x^2 ) ) over a defined window/τ | CF calculation, severity grading |
| Peak_est | Peak measure (linear) | Sample peak, quasi-peak, or peak over a window | CF, clip risk |
| Env_dB | Envelope level (dB) | 20·log10(Env_est / Ref) | Thresholding, logging |
| CF_dB | Crest factor (dB) | 20·log10(Peak_est / RMS_est) | Burstiness classification, stability |
| Overload_flag | Decision output | Rule-based state with threshold + hysteresis + duration | Limiter state machine |
| Gain_reduction_dB | Control output | Limiter mapping from Env/CF severity to target gain | Dynamic limiting action |
Envelope and peak variants (implementation-first definitions)
Envelope (Env_est)
- Abs + LPF: Env_est = LPF(|x|). Lowest cost, predictable latency; bandwidth must balance burst capture vs false triggers.
- Sliding peak: Env_est = max(|x|) within T_window. Maximizes burst sensitivity; requires duration/hysteresis to reject glitches.
- Hilbert-based envelope (optional): used only when magnitude tracking must be less dependent on carrier phase; implementation cost and latency rise (no deep math in this page).
Peak variants (Peak_est)
- Sample peak: peak of discrete samples in a window. Fast and simple; may underestimate true peaks in some systems (inter-sample peak risk).
- Quasi-peak: fast charge + slow decay (or hold+dop) behavior. Better for burst capture and stable decisions; less sensitive to single-sample noise spikes when paired with a duration rule.
Interface rule: if Peak_est is derived from sampled data, include a guardband placeholder (e.g., “+X dB”) when setting peak-based overload thresholds.
Crest factor and windowing (make time scale explicit)
Crest factor must be defined together with the measurement window. Without an explicit window/τ, CF values cannot be compared across tests or platforms.
- CF definition: CF_dB = 20·log10(Peak_est / RMS_est).
- Window requirement: Peak_est and RMS_est must share a stated time scale (T_window or equivalent τ).
- Two-timescale practice: use a short window for protection and a longer window for “content classification” and decision stability.
Window ↔ time constants (engineering mapping)
- T_window defines how much history is included in Peak/RMS statistics.
- Attack is the fastest allowed response to a rising envelope (protect headroom).
- Hold prevents chatter at the threshold by enforcing a minimum state duration.
- Release controls recovery speed and pumping risk after overload ends.
Verification rule: any overload criterion must specify both a level and a time requirement (e.g., “Env_dB > threshold for ≥ T_dur”).
Envelope extraction architectures: choose by bandwidth, latency, distortion
Architecture profiles (A–D) with decision fields
A) abs + 1-pole LPF
- Best for: general protection, low cost, predictable behavior.
- Knobs: τ_env / fc_env; optional separate attack vs release shaping.
- Risks: τ too small → noise spikes trigger; τ too large → short bursts missed.
- Test hook: burst response curve + false-trigger rate under noise-only.
B) precision rectifier + RC (analog)
- Best for: low-latency analog protection ahead of sensitive stages.
- Knobs: rectifier bandwidth, RC time constant, loading conditions.
- Risks: conduction artifacts, device capacitance shifting bandwidth, amplitude-dependent behavior.
- Test hook: small-signal linearity + large-signal distortion + temperature repeatability.
C) peak detector + droop control
- Best for: catching fast peaks and burst edges with strong sensitivity.
- Knobs: charge/discharge rates, hold time, droop slope.
- Risks: over-sensitivity to glitches unless paired with duration + hysteresis decision rules.
- Test hook: glitch-injection test + hold/release behavior across burst repetition.
D) digital envelope (DSP) + analog interface
- Best for: multi-mode products needing configurable profiles and field updates.
- Knobs: window length, filter order, pipeline latency.
- Risks: detection can be late; it cannot protect a stage that already clips before digitization.
- Test hook: latency budget verification + “protect object” check (which stage must never overload).
Key trade-offs (make selection deterministic)
- If an analog stage must be protected (before ADC or before a non-linear block), prioritize B or C because the action must happen with minimal latency.
- If false triggers are the top risk (noise/spikes), start with A and pair it with hysteresis + duration rules; widen detection bandwidth only as needed.
- If burst edges dominate overload, use C or A with a shorter τ, then enforce a minimum duration or hold to prevent chatter.
- If product profiles must be field-upgradable, use D, but document the protection target and validate that pre-digitization headroom is not violated.
Stability rule: detection bandwidth and limiter time constants must be tuned together; changing only one side often creates pumping, chatter, or missed bursts.
Typical pitfalls (written as observable failures)
Rectifier conduction artifacts (B)
Failure mode: small signals show an envelope bias or delayed turn-on, causing missed “early warning” triggers. Quick check: sweep input level and compare Env_est vs expected scaling; look for a kink near low amplitudes.
Amplitude-dependent bandwidth (B)
Failure mode: detection bandwidth shifts with amplitude and temperature, moving the effective threshold. Quick check: run the same burst test at two levels and compare trigger latency; large deltas indicate bandwidth drift.
LPF lag misclassifies burst edges (A)
Failure mode: the first burst cycle is underestimated, so overload triggers late. Quick check: two-burst test; measure delay between the burst start and Env_est reaching the threshold.
Peak detector over-reacts to glitches (C)
Failure mode: single-sample spikes or switching noise causes repeated false limiting. Quick check: inject a controlled glitch train and verify the decision includes a minimum duration and hysteresis.
Crest-factor estimation: windowing, robustness, and “burst realism”
Two time scales: fast_CF for protection, slow_CF for stable decisions
Crest factor is only meaningful when its time scale is stated. A practical implementation separates a short-window estimator for rapid burst awareness and a long-window estimator for stable classification and mode control.
fast_CF (short window)
- Role: burst awareness and early warning, aligned with protection time constants.
- Key fields: T_fast, update step, optional hold to ignore glitches.
- Failure to avoid: reacting to single-sample spikes as “real burst.”
slow_CF (long window)
- Role: stable classification (bursty vs steady vs noisy) and mode switching.
- Key fields: T_slow, mode-hold time, percentile/logging strategy.
- Failure to avoid: window mismatch that makes CF incomparable across tests.
Output strategy: Env drives fast actions; CF drives pre-warning and mode switching. This separation reduces chatter and improves burst realism.
RMS estimation for CF: sliding mean-square vs IIR smoothing
Sliding mean-square (windowed)
- Definition: RMS_est = sqrt( average( x² ) over T_window ).
- Strength: explicit window; easy to reproduce in production tests; aligns peak and RMS windows.
- Risk: window too short inflates RMS during bursts (CF appears too low); window too long lags (CF appears too high).
- Use: prefer for fast_CF when thresholds must be repeatable across benches and factories.
IIR smoothing on x² (time-constant based)
- Definition: smooth x² with τ, then RMS_est = sqrt( smoothed(x²) ).
- Strength: smooth behavior; compact implementation; stable mode decisions.
- Risk: mismatch between τ and burst structure creates “tail memory,” biasing slow_CF after bursts end.
- Use: prefer for slow_CF to reduce mode flapping; document τ and update rate.
Robustness: remove DC/offset and define how CF thresholds are set
DC/offset removal (mandatory for meaningful CF)
- Why: DC and offsets inflate RMS_est, making CF artificially low and hiding burstiness.
- Where: apply on the CF/RMS path (and optionally the peak path if offsets can bias peak detectors).
- How: x_hp = x − LPF(x) or window-mean subtraction, with a documented time scale.
Production check: with “quiet input,” RMS_est should remain below a defined limit and should not drift with slow offsets (limit value is system-specific).
Multi-tone / noise-like content (no probability derivations)
- Do not use a single CF sample: record a short history and extract a robust summary (e.g., a high-percentile or upper envelope).
- Set thresholds by outcomes: define a target false-trigger rate and tune CF_warn to meet it under noise-only and normal content.
- Separate roles: slow_CF gates mode changes; fast_CF assists early warning but should be filtered by duration/hold.
Overload criteria: from simple thresholds to multi-condition decisions
Level 0–1: threshold, then minimum-duration filtering (production minimum)
Level 0 — single threshold
Criterion: Env_dB > Env_th. This is the simplest trigger but is vulnerable to noise spikes and can chatter near the threshold.
Level 1 — add duration
Criterion: Env_dB > Env_th for ≥ T_dur. This is the minimum “production-ready” enhancement to reject glitches and make decisions repeatable.
Verification rule: every overload criterion must specify both a level and a time requirement (level-only rules are not robust).
Level 2: dual thresholds + hysteresis (stop chatter)
- Enter: Env_dB > Env_th_hi for ≥ T_on.
- Exit: Env_dB < Env_th_lo for ≥ T_off.
- Hold: enforce T_hold to prevent immediate toggling after entry.
Production metric: limit the state-toggle count per minute during slow amplitude sweeps and near-threshold burst repetition.
Level 3: combine Env + CF + duration and produce graded outputs
Composite criteria templates
- Fast protect: Env_dB > Env_hi for ≥ T_short → enter Limit.
- Burst pre-warning: Env_dB > Env_warn AND slow_CF > CF_warn for ≥ T_mid → enter Warn.
- Persistent overload: Env_dB > Env_mid for ≥ T_long → escalate to Mute/Fault (system policy).
Severity grading outputs
- Warn: early alert and mode gating (reduce surprise triggers).
- Limit: active gain reduction / limiting action (fast safety).
- Mute: temporary block to avoid audible/visible corruption.
- Fault: latched condition for protection or service policy.
Frequency-aware checks: distinguish rail clip vs slew-rate limiting (methods only)
Overload is not always a rail hit. High-frequency large-swing content may trigger slew-rate limiting first. Use observable checks that guide decision levels without relying on deep sampling theory.
Shape check (time-domain)
- Rail clip: flat-top/flat-bottom plateaus.
- SR limiting: edges collapse to a near-constant slope (triangle-like transitions).
- Soft compression: no plateaus; peaks stop growing proportionally while distortion rises.
Consistency check (same Env, different content)
If overload triggers much earlier with higher-frequency or sharper-edge stimuli at the same Env level, the limiting factor is likely dynamic (slew-related) rather than pure rail headroom. Use this to select grading thresholds and time constants.
Dynamic limiting control laws: attack/release/hold without pumping
Control variable and layering: GR_target → GR_applied → gain element
A limiter becomes stable when “how much to reduce” is separated from “how fast to move.” The gain computer maps detection outputs into a target reduction, and a time-constant shaper turns that target into a smooth, artifact-resistant control signal.
Recommended control variable
- GR_dB: Gain_reduction_dB for logging, thresholds, and production reports.
- g: Linear gain for the gain element interface (VCA/digital multiplier).
- Reference: document the reference point (0 dBFS or full-scale) so GR is comparable across test setups.
Minimal signal set (implementation-ready)
- Inputs: Env_dB, CF_dB, Overload_state (Warn/Limit/Fault).
- Static output: GR_target (knee + ratios + ceiling).
- Dynamic output: GR_applied (attack/hold/release shaped).
- Export: state transitions + GR_applied time trace for production validation.
Gain computer: knee shape and multi-stage action (gentle → hard)
Soft-knee region (mild over)
- Goal: reduce obvious modulation artifacts near threshold.
- Fields: knee_width_dB, ratio_low, ceiling_dB (placeholders set by headroom budget).
- Behavior: gradual GR ramp for small exceedances; avoid “instant clamp” sound/feel.
Hard action region (severe over)
- Goal: guarantee protection under worst-case bursts.
- Fields: Env_hi, GR_max, optional mute_threshold (system policy).
- Behavior: fast rise to a bounded GR, with deterministic escalation rules.
Design rule: static mapping defines “how much”; time-constant shaping defines “how it moves.” Mixing these layers is a common cause of pumping and chatter.
Time-constant shaping: fast protection without clicks, stable release without pumping
Attack (protect quickly, avoid abrupt steps)
- Too fast: audible/visible “click” from abrupt gain steps and control feedthrough.
- Too slow: overshoot beyond ceiling during sharp bursts.
- Practical template: two-rate attack (fast for severe over, slower for mild over).
- Fields: T_attack_fast, T_attack_slow, severity split threshold.
Hold (prevent short-cycle toggling)
- Purpose: stop immediate release that creates “gain flutter” between close bursts.
- Template: enforce a minimum hold, then allow release only after stable below-threshold evidence.
- Fields: T_hold_min, hold-extend rule for repeated triggers.
- Production metric: limit state/GR toggles per minute on burst trains.
Release (avoid pumping, avoid long tail)
- Too fast: pumping from gain modulation following the envelope.
- Too slow: tail memory that reduces headroom for the next burst.
- Practical template: piecewise release (slow when GR is large, faster near unity).
- Fields: T_release_hiGR, T_release_loGR, release breakpoint.
Verification set (placeholders): overshoot beyond ceiling, gain-step size, GR ripple rate, recovery time to GR<X dB, and near-threshold toggle count.
Options and pitfalls: look-ahead (optional) and control discretization
Look-ahead (optional)
- When it helps: extremely sharp transients where attack must be effective without abrupt steps.
- Cost: added latency and buffering; tests must record and budget the added delay.
- Scope: treat as a deployment toggle, not a mandatory feature.
Discretization pitfalls (implementation-facing)
- GR step too coarse: audible/visible “staircase” modulation near threshold.
- Update rate too low: slow, chunky control that can create sidebands and instability.
- Minimum fields: update_rate, GR_step_dB, smoothing enable/strength.
Where to place it: analog front-end vs ADC domain vs digital domain
Three placement options and what they can (and cannot) protect
Analog pre-ADC limiting
- Protects: the ADC and analog front-end from saturation.
- Requires: low-distortion variable gain element and careful control-signal isolation.
- Best for: strict “never clip” policies at the ADC input.
At the ADC driver / interface block
- Protects: the immediate drive headroom and common-mode integrity.
- Focus: interface constraints (headroom, distortion, control coupling), not topology derivations.
- Best for: integrated AFEs where the driver is the dominant overload point.
Digital post-ADC limiting
- Protects: downstream digital headroom and output metrics.
- Cannot protect: the ADC itself from clipping/saturation.
- Best for: flexible multi-mode shaping when ADC overload risk is already low.
Interface budget fields: make placement a checkable decision
- Headroom budget: peak reserve required before the first overload point.
- Latency: detect → control delay, especially relevant for burst protection.
- Noise injection: control-signal coupling into the signal path (ground/power/capacitive paths).
- Distortion vs gain: linearity changes across gain states and output swing.
- Control update: step size and update rate that may create modulation artifacts.
Production hooks (mandatory)
- Force overload: deterministic entry into Warn/Limit states.
- Bypass detect: isolate and test gain element behavior without detection logic.
- Log trace: GR_applied + state transitions with timestamps for repeatability checks.
Hybrid approach: analog hard-protect + digital soft-shape
A common production strategy splits responsibilities: analog limiting guarantees the ADC never saturates under worst-case bursts, while digital limiting performs gentle shaping and multi-mode behavior with measurable and updateable parameters.
Analog layer (hard protect)
- Role: prevent ADC clipping deterministically.
- Constraint: set thresholds to preserve a working region for digital shaping.
- Test: force-trigger and validate ceiling + recovery repeatability.
Digital layer (soft shape)
- Role: smooth knee behavior, multi-stage policies, and flexible tuning.
- Constraint: account for detect/control latency and quantization effects.
- Test: measure pumping risk and tail recovery across burst trains.
Quick placement rules (decision-oriented)
- If ADC clipping is unacceptable under any burst → choose Analog pre-ADC or Hybrid.
- If latency must be minimal for burst protection → prefer Analog/Hybrid.
- If multi-mode shaping is required and ADC overload risk is already low → choose Digital post-ADC.
- If control coupling/noise injection is hard to manage in analog → shift shaping burden to Digital, keep analog as hard ceiling only.
Error sources & false triggers: make it production-stable
False triggers: common sources and fast isolation steps
Production stability starts with repeatable fault isolation. False triggers must be classified by observables (timing correlation, duration, recovery tail) rather than by assumptions.
Noise spikes
- Typical symptom: single-sample or very short bursts cause Env/flag jumps.
- Quick check: compare trigger timestamps to the raw x(t) peak duration; count sub-ms events.
- Mitigation fields: duration_min, hysteresis (Env_th_hi/lo), hold.
Switching / digital coupling
- Typical symptom: triggers recur at fixed phase relative to clocks or switching events.
- Quick check: correlate trigger timestamp with known switching markers (log + overlay).
- Mitigation fields: blanking_ms after known events, state-based re-arm rules.
Clamp / protection recovery tail (interface-facing)
- Typical symptom: after a disturbance, Env/RMS_est decays slowly and causes repeated triggers.
- Quick check: detect long recovery tails and repeated state entries without new large input peaks.
- Mitigation fields: Recover state, extended hold, optional estimator freeze/reset policy.
Probe / fixture injection
- Typical symptom: false-trigger rate changes when probes/grounding change.
- Quick check: repeat the same stimulus with a locked fixture config; compare counters.
- Mitigation fields: measurement configuration lock + logged setup metadata.
Missed detections: bandwidth, windowing, and estimator lag
Detection bandwidth too low
- Observable: burst edges are smoothed; Env peak is underestimated.
- Field knobs: T_attack / envelope LPF cutoff / peak-hold enable policy.
- Production test: burst edge sweep (rise time vs trigger latency/overshoot).
Window too long
- Observable: short bursts are averaged away; CF under-reacts to real peaks.
- Field knobs: fast_CF + slow_CF (two time scales) with gated combination logic.
- Production test: burst duration sweep (T_on) and gap sweep (T_off).
RMS estimator lag
- Observable: CF becomes inconsistent at burst entry/exit due to delayed RMS_est.
- Field knobs: RMS IIR constant, state-conditioned freeze/reset behavior.
- Production test: step-in amplitude test and burst-train consistency check.
Drift and consistency: keep thresholds stable across temperature and lots
- Threshold drift: Env_th and CF_th move with temperature, supply, and gain error.
- Offset / DC component: RMS_est inflates, CF is biased low without DC removal.
- Required fields: programmable thresholds + parameter versioning (EEPROM/flash).
- Production alignment: inject known burst patterns to align trigger points and recovery.
Coupling between measurement and control
Limiting changes RMS/CF, which can re-trigger or prematurely release. A state machine with hold and re-arm evidence prevents self-induced chatter and improves repeatability.
Calibration hooks: programmable thresholds and deterministic production injection
Parameter set (minimum)
- Env_th_hi / Env_th_lo, CF_th, duration_min
- T_attack / T_hold / T_release, blanking_ms
- GR_max, knee_width_dB, update_rate
- Parameter version + calibration version (logged)
Production injection (repeatable)
- Burst pattern: A, T_on, T_off, repetition count, and frequency set.
- Targets: trigger point alignment, recovery time stability, false-trigger rate under noise-only.
- Force modes: deterministic entry to Warn/Limit for validation and binning.
Production rule: a stable limiter is defined by repeatable triggers and repeatable recovery under a locked stimulus and a locked measurement window.
Verification & measurement traps: how to test envelope/CF/limiter correctly
Test philosophy: lock stimulus, lock window, lock trigger
- Stimulus repeatability: burst patterns and levels must be deterministic.
- Window alignment: capture windows must match the algorithm windows used for CF/RMS.
- Trigger definition: latency and recovery must use a single, fixed definition.
- Config lock: sampling rate, bandwidth, and computation settings must be logged and fixed for production.
Stimulus library: minimum set that covers real failure modes
Baseline sine
- Purpose: threshold alignment and static knee sanity check.
- Fields: A, f_set, record length.
Burst sine (on/off)
- Purpose: attack latency, overshoot, hold behavior, recovery time.
- Fields: A, f_set, T_on, T_off, repeat count.
Multi-tone / crest-factor stress
- Purpose: CF realism, pumping risk, and state stability near threshold.
- Fields: tones (f_i), per-tone level, duration.
Noise + burst overlay
- Purpose: false-trigger rate under noise-only and robustness under realistic backgrounds.
- Fields: noise level, burst A, burst schedule.
Pulsed envelope
- Purpose: duration gating, blanking, and state re-arm logic validation.
- Fields: A, duty, repetition rate, burst edges.
Must-capture outputs: record the right signals with aligned windows
- Core metrics: Env_dB, RMS_est, CF_dB
- Control: Overload_flag / State, GR_applied (or g)
- Timing: trigger latency, recovery time (single definition)
- Counters: false trigger rate under noise-only, state toggle count under burst train
- Config log: capture Fs/BW + metric window parameters + parameter version
Measurement traps: avoid incorrect peaks, incorrect windows, and incorrect timing
Peak miss (capture BW / sample rate)
- Symptom: measured peak is low while distortion/overload is visible.
- Fix: increase capture Fs/BW; keep peak definition consistent across tools.
Window mismatch (CF/RMS inconsistency)
- Symptom: CF varies across scripts/tools for the same capture.
- Fix: lock T_window, alignment rules, and estimator warm-up behavior.
Recovery time ambiguity
- Symptom: recovery appears to change test-to-test with the same stimulus.
- Fix: define recovery as either “GR_applied > -1 dB” or “State=Idle,” then never mix.
Setup-induced coupling
- Symptom: pumping sidebands change with probe/ground changes.
- Fix: fixture lock + logged configuration; validate with a known reference capture.
Pass criteria templates (placeholders driven by system budget)
- Trigger latency: overload trigger delay < X ms (burst sine / pulsed envelope).
- Release: time to within -1 dB < Y ms (burst train).
- Pumping: gain-modulation sidebands < Z dBc (multi-tone).
- False triggers: < N per hour under noise-only (locked config).
Report each criterion with: stimulus ID, capture configuration, window settings, parameter version, and a single recovery definition to guarantee reproducibility.
Engineering checklist & parameter table (design hooks you can reuse)
This section turns the envelope/crest-factor/limiter discussion into a reusable, production-friendly template: a minimal parameter set, a non-negotiable tuning order, implementation checks, production data fields, and concrete material numbers to speed up bring-up and vendor lookup.
A) Minimal parameter set (copy/paste table)
Use these fields as the “minimum viable” configuration. Keep them versioned (param_version) and locked for production bins.
| Parameter | What it controls | Starting rule (engineering) | If set wrong | Minimum verification hook |
|---|---|---|---|---|
| Env_th_hi | Upper envelope threshold for entering Warn/Limit | Set from headroom budget: leave guardband to avoid rail/SR limit under worst-case load | Too low → false triggers; too high → real overload slips through | Burst test: trigger delay & miss rate (H2-10 template) |
| Env_th_lo | Lower threshold for de-assert / recovery (hysteresis) | Choose hysteresis wide enough to cover noise + control ripple | Too narrow → chatter; too wide → slow exit / unnecessary attenuation | Noise-only soak: state toggles per minute |
| Warn_th / Limit_th | Graded response levels (telemetry + action) | Warn for early margin loss; Limit for protection. Keep mapping stable across firmware | Wrong grading → noisy UX, unstable system policy, hard-to-debug bins | Bin report: event counts per level |
| T_attack | How fast gain reduction ramps in | Fast enough to protect, but avoid audible clicks / control-edge feedthrough | Too fast → clicks; too slow → overload passes | Overload trigger delay & residual clip metric |
| T_hold | Keeps GR steady before release (prevents chatter) | Cover burst spacing / control quantization; hold until evidence of recovery is stable | Too short → pumping/chatter; too long → sluggish dynamics | State toggle count with periodic bursts |
| T_release | How fast GR returns to 0 dB | Slow enough to avoid pumping sidebands, fast enough to not mask next burst | Too fast → pumping; too slow → tail / “stuck attenuation” | Recovery time to within −1 dB |
| CF_window_fast / slow | Two-time-scale crest-factor estimate | Fast path for protection realism; slow path for statistics / mode gating | Wrong windows → CF “fake high/low”, unstable policy decisions | Burst set: CF repeatability across capture windows |
| duration_min / blanking_ms | Rejects short spikes; avoids re-trigger during switching | Set to ignore EMI spikes but still catch real overload events | Too large → misses; too small → noise-triggered events | Noise-only false trigger rate & burst miss rate |
| GR_max / knee_width_dB | Maximum limiting action + soft/hard knee behavior | Limit max GR to protect without “mute-like” behavior; widen knee to reduce distortion | Too much GR → audible artifacts; too little → insufficient protection | Distortion rise & pumping sidebands (FFT) |
Tip: keep parameter names identical across analog/digital implementations so test scripts and binning stay stable.
B) Non-negotiable tuning order (avoid “good on bench, bad in system”)
- Fix headroom budget → output: guardband target (rail/SR/distortion limits) and graded response levels.
- Set Env thresholds (Env_th_hi/lo, Warn/Limit mapping) → output: stable state boundaries before time constants exist.
- Tune attack (T_attack) → output: meets trigger-delay target without clicks/control feedthrough.
- Tune hold (T_hold, duration_min, blanking_ms) → output: chatter/pumping suppressed under burst patterns.
- Tune release (T_release) → output: recovery time meets spec while pumping sidebands stay below target.
- Enable crest-factor logic last (CF_window_fast/slow, CF_th) → output: CF used for “mode gating / pre-warning”, while Env remains the fast protection driver.
C) PCB / implementation checklist (stability + repeatability)
Detection path quality
- □ Detection bandwidth covers intended burst edges (avoid systematic miss).
- □ Noise peak and switching spikes remain below Env_th_hi for the chosen duration_min.
- □ DC removal / offset handling is defined (otherwise CF can drift low and hide overload).
- □ Capture/export uses consistent windowing (CF repeatability depends on it).
Control isolation & return paths
- □ Control lines (gain/VCA/DVGA programming) are routed away from sensitive inputs.
- □ Control update rate is bounded; blanking is applied around switching to suppress self-trigger.
- □ State transitions are logged (timestamp, state, GR_dB) to correlate field failures.
Bypass / injection for production
- □ Force bypass exists (diagnose “detector vs limiter” quickly).
- □ Known burst pattern injection exists (repeatable trigger and recovery bins).
- □ Force Warn/Limit exists (verify state machine and output flags without analog stimulus complexity).
D) Production data fields (traceable bins)
Minimum schema (store per unit / per temperature point):
- param_version, cal_version, temp_point
- trigger_count_warn, trigger_count_limit, false_trigger_count
- max_GR_applied_dB, avg_recovery_ms, state_toggle_count
- fault_code / exception_code (fixed enum)
Lock stimulus ID + config lock during production testing to make bins comparable across lots.
E) Reference material numbers (starter parts; verify against targets)
These part numbers are provided to speed up datasheet lookup and prototype validation. Selection must be driven by the parameter table above (worst-case conditions + guardbands).
- ADL5511 (Analog Devices) — envelope + RMS detector for wideband RF; useful when burst/envelope bandwidth matters.
- ADL5513 (Analog Devices) — demodulating log amp for RF burst detection / level monitoring (log output simplifies thresholding).
- AD8317 (Analog Devices) — demodulating log amplifier for RF power measurement/control (dB-scaled output for graded thresholds).
- THAT2180A/B/C (THAT Corporation) — high-performance VCAs for audio dynamics/limiting prototypes.
- THAT4301 (THAT Corporation) — integrated “dynamics processor” (VCA + RMS sensor + op-amps) for fast compressor/limiter bring-up.
- SSM2166 (Analog Devices) — microphone conditioner with limiting/compression blocks (voice-band focused; good reference for integrated dynamics behavior).
- LMH6401 (Texas Instruments) — digitally-controlled variable-gain amplifier (DVGA) for IF/RF/high-speed time-domain; useful for programmable limiter action in wideband chains.
- ADL5330 (Analog Devices) — RF VGA/attenuator (up to GHz-class use cases) for analog-domain gain control experiments.
- AD8338 (Analog Devices) — variable gain amplifier for LF–tens of MHz; useful when differential chain + defined gain law is needed.
- TLV3501 (Texas Instruments) — high-speed comparator for threshold/window flagging in fast overload paths.
- LTC6752 (Analog Devices) — very high-speed comparator family for fast edge/threshold decisions when needed.
- ADG1411 (Analog Devices) — quad SPST analog switch for bypass routing and stimulus injection points.
- MCP4131 (Microchip) — SPI digital potentiometer for trimming thresholds, hysteresis, or gain-law coefficients in prototypes.
- PGA281 (Texas Instruments) — programmable gain amplifier (precision chain) when a controllable front-end gain step is part of the limiting strategy.
Some legacy VCAs (e.g., SSM2164) exist as historical references, but availability/status must be verified before new designs.
Request a Quote
FAQs: Envelope / Crest-Factor Detection & Dynamic Limiting
Short, actionable troubleshooting only. Each answer uses the same 4-line format with measurable pass criteria to keep production decisions consistent.
