What this chapter controls

In ultrasonic/acoustic sensing, Rx instability is most often caused by Tx leakage and ring-down that drive the receiver into saturation or non-linear recovery. This chapter locks two design variables: isolation (how much Tx energy is prevented from reaching Rx) and blanking (how long the system waits until metrics become trustworthy).

T/R switching options (selected by voltage, speed, and recovery)

Selection is driven by measurable recovery behavior, not by component name. The decision target is: after Tx stop, Rx must return to linear operation before Listen start.

Blanking window: a measurable gate (not a guessed constant)

• Definition: the period after Tx where Rx sampling and/or feature computation is disabled to avoid corrupted metrics.
• Listen start criterion: begin sampling only when both are true:
  • Linearity: ADC clipping is absent (clip_flag cleared; sat_cnt ≈ 0).
  • Settled baseline: residual ring-down envelope is near noise floor with margin (no structured peaks inside blanking).
• Practical guard: treat listen timing as a bound: listen_start = tx_stop + max(recovery_time, ringdown_to_floor, settling_margin).

A short blanking window increases near-range sensitivity only if it does not reintroduce saturation and false peaks.

Crosstalk paths (and fast isolation checks)

• PCB coupling: Tx edges capacitively inject into high-impedance Rx nodes → early-window spikes.
• Ground return / common impedance: Tx current modulates Rx reference → baseline wobble and phase drift.
• Cable / connector coupling: harness becomes an antenna → inconsistent false peaks by cable routing.
• Enclosure / structure coupling: mechanical vibration couples energy directly into the transducer → “electrical isolation” alone is insufficient.

Fast checks: (1) overlay Rx waveform with Tx gating to confirm time-locked spikes, (2) inspect blanking statistics to ensure no structured peaks exist before listen start.

Evidence chain & acceptance checks

• Rx saturation / recovery time: time until sat_cnt stops increasing post Tx stop.
• Blanking noise floor: distribution of amp_rms inside blanking should resemble noise, not peaks.
• False-peak rate: probability of corr_peak above threshold during blanking should be near zero (or tightly bounded).
• Window integrity: keep logs of blanking_us, listen_us, and window index used for feature extraction.

AFE targets derived from f0 / Q / τ and blanking

Rx AFE design must remain valid under real ring-down and crosstalk. Targets are not “typical” values; they are derived from measured resonance and recovery behavior. The acceptance goal is stable metric distributions (amp_rms, corr_peak, phase_std) while keeping clipping absent in the valid listen window.

• Noise floor: input-referred noise must allow far-range detection above noise statistics.
• Dynamic range: prevent near-range saturation without losing far-range sensitivity.
• Band definition: align filtering to measured f0 and expected drift.
• Observability: expose sat_cnt, clip_flag, gain_trace, and ADC peak statistics for field diagnosis.

LNA: noise + bandwidth (engineerable, testable)

• Noise focus: use input-referred noise as the sizing anchor for far-range capability.
• Bandwidth discipline: limit bandwidth around the target band to avoid integrating environmental noise.
• Recovery readiness: ensure the LNA returns to linear behavior before Listen start (post-blanking).
• Protection impact: clamps and input protection must not introduce long recovery tails that mimic echoes.

BPF: center frequency and bandwidth matched to measured resonance

• Center frequency: set near measured f0 under final mounting (not only bench conditions).
• Bandwidth: wide enough for the chosen excitation, narrow enough to suppress out-of-band noise and interferers.
• Stability: avoid filter behaviors that distort phase metrics or move correlation peak location.

PGA / AGC: avoid near-range saturation without “gain pumping”

• Two-region strategy: use predictable base gain plus controlled gain adjustments to cover range.
• Gating: freeze AGC during Tx and blanking; allow updates only in valid listen windows.
• Telemetry: record gain_trace so threshold shifts and detection drift remain explainable.
• Hard stop: clipping removal has priority; clipped data invalidates phase/correlation reliability.

ADC: sampling, ENOB, and input-range matching

• Sampling rate: support the target band and feature pipeline (envelope, correlation, phase) without aliasing.
• ENOB: size for noise-floor resolution; insufficient ENOB collapses separability of distributions.
• Input range: match PGA output so peaks remain below full-scale in the valid listen window.
• Flags: expose clip_flag and track adc_peak statistics to prove headroom.

Evidence chain (minimum signals to log)

• Input-referred noise estimate: noise RMS in a known “no-target” listen window.
• Saturation indicators: sat_cnt, clip_flag gated to listen window only.
• AGC behavior: gain_trace and state transitions (freeze/unfreeze).
• ADC health: peak histogram or max/min per window (adc_peak), overflow markers.
• Metric stability: running stats of corr_peak and phase_std for drift detection.

Engineering boundary (accurate measurement without a “paper”)

Phase/amplitude measurement is only reliable when the receiver is linear and the listen window is clean. This chapter defines a practical measurement pipeline that stays testable: windowed amplitude statistics, I/Q-derived phase stability, and correlation-peak ToF extraction with sidelobe guards.

Amplitude: envelope vs peak vs RMS (choose by failure mode)

Amplitude is not “one number.” Different choices fail differently under residual ring-down, crosstalk spikes, and noise-floor changes.

• Hard gate: any clipping (clip_flag) inside the listen window invalidates amplitude and phase outputs.
• Noise reference: record noise_rms from a known “no-target” region to keep thresholds portable.

Phase: I/Q demod (or correlation-phase) expressed as stability

Phase is most useful as a stability metric, not as a single instantaneous number. The implementation stays practical: demodulate with a known reference, compute phase per sample (or per block), then use phase_std (and optional outlier count) in the listen window.

• I/Q demod path: Mix (NCO/reference) → LPF → I/Q → atan2(Q,I) → phase.
• Correlation-phase option: phase estimate around the correlation peak; only valid if peak quality passes peak_ratio.
• Guardrails: freeze AGC during Tx/blanking; compute phase only when listen gate is enabled and clipping is absent.

Primary evidence: phase_std distribution over temperature points and mounting conditions; phase drift must remain explainable with logged version tags.

ToF / correlation peak: windowing, peak rules, sidelobe handling

• Listen-window anchoring: correlation search starts at Listen start (post blanking), not at Tx stop.
• Peak validity: require both corr_peak above threshold and peak_ratio (main/secondary peak) above a guard limit.
• Search constraints: limit peak search range to physically plausible ToF bins to reduce false positives.
• Sidelobes: coded/chirp excitation may produce sidelobes; the guard is peak_ratio + optional “peak width/shape” check.

Evidence chain & acceptance checks

• Linearity: clip_flag==0 and listen-window sat_cnt not increasing.
• SNR anchor: amp_rms referenced to noise_rms (ratio or dB, either acceptable).
• Phase trust: phase_std remains stable for the same target across temperature points (log temp_point).
• ToF trust: output ToF only when corr_peak and peak_ratio pass; track false_peak_rate.
• Traceability: every measurement record tags algo_ver, threshold_set_id, and window index.

Goal: reproducible decisions under compute and noise constraints

Edge analytics turns measurable features into stable triggers. This chapter defines a repeatable pipeline (features → thresholds → consistency → trigger), with versioned parameters and minimal field logs that can explain every event.

Compute budget & real-time scheduling (keep it bounded)

• Per-window runtime: treat each listen window as a fixed-time budget (feature extraction must complete before next window).
• Tiered workload: run lightweight gates first (energy/RMS), then correlation, then phase stability if needed.
• Deterministic latency: log decision_latency_ms to confirm real-time behavior across firmware builds.
• Gated processing: skip compute when listen_en is false or when clip_flag is set.

Feature vector (small set, high explainability)

Features are grouped by what failure mode they reject. Keeping a compact feature vector improves tuning, logging, and field interpretability.

• Energy / amplitude: amp_rms, amp_ratio (range / surface change).
• Correlation quality: corr_peak, peak_ratio (reject sidelobes and noise peaks).
• Phase stability: phase_std (reject unstable reflections / mechanical noise signatures).
• Band energy: band_energy (reject out-of-band interference).

Decision logic: multi-threshold + hysteresis + time consistency

• Stage 1 (gate): Energy must exceed Th1 to consider a candidate.
• Stage 2 (quality): Correlation must pass Th2 and peak_ratio must exceed a guard limit.
• Stage 3 (stability): Phase stability requires phase_std < Th3 (optional per use-case).
• Hysteresis: separate enter/exit thresholds to prevent chatter; include a hold-off timer for relay outputs.
• Consistency: vote N-of-M windows (or consecutive windows) before asserting TRIGGER.

Every event must reference threshold_set_id and algo_ver so a field trace can be reproduced in the lab.

Trigger / relay outputs (controlled, debounced, safe)

• Output forms: GPIO trigger pulse, latched output, or relay drive (implementation-specific; the policy remains the same).
• Debounce / minimum pulse: enforce minimum assert width; reject re-trigger within hold-off window.
• Anti-chatter: combine hysteresis + vote + hold-off to prevent rapid cycling in borderline conditions.

Logging for replay (minimal “event packet” for support)

• Ring buffer: keep short pre/post-event windows (raw decimated samples or feature snapshots).
• Feature snapshot: store the feature vector at trigger time: energy, corr quality, phase stability.
• Counters: event_cnt, false_cnt (or “rejected candidate” count) for on-device health tracking.
• Tags: log_schema_ver, threshold_set_id, algo_ver, plus temp_point and supply tag if available.

Engineering pitfalls (reliable action without false triggering)

This chapter addresses the common engineering pitfalls related to output reliability: relay surge, output jitter, cable ESD, and interlocking logic.

Output types: open-drain/relay/isolated DO (select by load)

• Open-drain: for triggering or low-power short-line applications (requires external pull-up).
• Relay: for driving larger loads with isolation, but consider coil surge and mechanical bounce.
• Isolated DO: for long-line, high-interference environments; ensures separation between the load and control electronics.

Each output type has its failure mode: open-drain is vulnerable to voltage instability; relay suffers from mechanical bounce; isolated DO must handle external surge without causing reverse backflow.

Anti-chatter policy: minimum hold time, release condition, bounce handling

• Minimum hold time: ensure output remains stable for a defined period to avoid chatter (typically 20-100ms).
• Release condition: ensure a stronger threshold for output deactivation to prevent premature switching.
• Re-arm delay: prevent multiple triggers by introducing a hold-off period after a release.

The goal is to ensure stable output actions and prevent rapid switching that can lead to mechanical or electronic failure.

Driver & load stress: surge, back emf, and current waveform analysis

• Relay coils: Ensure proper flyback protection and current limiting to prevent damage to the driver or MCU.
• Inductive loads: Protect the driver from inductive kickback and ensure that the output is designed to handle such stresses.
• Bounce handling: Minimize the risk of contact bounce by using debouncing algorithms or physical mechanisms.

Monitoring current waveform during load switching can help identify potential issues such as excessive surge or back emf that could damage the output circuitry.

I/O protection: TVS, current-limiting, and surge suppression (module-level focus)

• TVS: Use TVS diodes near the output to clamp any voltage spikes and protect the device.
• Current-limiting: Ensure that series resistors or current-limiting components are in place to protect the relay and drivers.
• Surge suppression: Use common-mode chokes or other components to protect against high-energy transients from long cables or inductive loads.

Protecting the output from both transient and continuous electrical surges is critical for ensuring long-term reliability and preventing failure.

Evidence chain & acceptance checks: timestamp, jitter count, current waveform, fault code

• Timestamp: Record precise timestamps for output actions, including activation, deactivation, and fault events.
• Jitter count: Count the number of times the output jittered or failed to meet the required time threshold.
• Current waveform: Record the load current waveform during output switching to verify that it matches the expected behavior.
• Fault code: Implement fault codes for various output failure modes such as overcurrent, relay sticking, or ESD reset events.

These logs ensure that all output actions are traceable and that any failures can be debugged and corrected quickly.

Environmental consistency (how to maintain performance across environments)

This chapter focuses on the necessary calibration items for maintaining consistency across temperature, humidity, and aging. It provides the engineering strategies and verification methods for calibration without discussing storage systems or NVM.

Temperature drift: frequency, gain, sound speed

• Frequency drift: change in resonance due to temperature variations.
• Gain drift: changes in system gain as a function of temperature.
• Sound speed variation: distance mapping changes for ultrasonic distance sensing due to temperature fluctuations.

Humidity, wind, rain: effects on the echo

• Echo attenuation: reduced echo return due to environmental conditions.
• Phase instability: increased jitter or inconsistency due to environmental impact.
• Engineering countermeasures: applying filtering or threshold compensation strategies to mitigate environmental influences.

Field self-test: transmission, reception, blockage detection

• Transmission self-test: periodic transmission checks to ensure consistent output under different environments.
• Reception self-test: validate received signal strength and quality.
• Blockage detection: detect obstructions that could affect the sensing accuracy.

Evidence chain & verification: calibration before and after error comparison

• Verification: ensure that calibration before and after application produces consistent, measurable improvements.
• Drift curve: verify and log drift curves for all key metrics (gain, frequency, sound speed).
• Test plan: conduct environmental testing with multiple conditions to assess overall system accuracy and consistency.