Algorithm Engineer (DSP / Estimation)

About the Role

We’re building a co-timing engine that aligns multiple measurement streams onto a common clock, cancels what is truly shared, and exposes what remains as a measurable signal tied to geometry and medium effects.

This role is for an algorithm engineer who can turn solid DSP/estimation ideas into reliable, testable code— especially around delay estimation, coherence diagnostics, and drift tracking.

What You’ll Do

• Phase-slope delay estimation: implement and harden delay estimators based on cross-spectrum phase (unwrap, robust slope fits, outlier handling, confidence measures).
• Coherence + “window validity” checks: compute magnitude-squared coherence and related metrics; detect when assumptions break (low coherence, nonstationarity, multi-path, band-limited ambiguity).
• Drift tracking over time: build delay-over-time models (piecewise linear, Kalman-style trackers, smoothing) and decide when to re-estimate vs propagate.
• Calibration pipelines: design routines for gain/phase offsets, channel delays, sample-rate mismatch, and sensor idiosyncrasies; produce calibration artifacts that are inspectable and reproducible.
• QA datasets + evaluation harness: create synthetic and bench-derived datasets with ground truth (known delays/drifts), adversarial cases, and regression tests; maintain metrics dashboards (error distributions, failure rates, confidence calibration).
• Algorithm-to-product integration: package estimators as clean modules with stable I/O contracts, clear failure signals, and runtime diagnostics suitable for production.

Concrete Deliverables

• A phase-slope delay module with: phase unwrapping strategy, robust regression, confidence scoring, and clear “refuse to estimate” criteria.
• A coherence-driven gate that labels windows as usable / borderline / invalid and explains why.
• A drift tracker that outputs delay(t) with uncertainty and handles dropouts gracefully.
• A QA suite: curated test corpus + synthetic generator + regression harness that prevents “it worked yesterday” regressions.

Required Qualifications

• Strong foundation in DSP / statistical estimation: FFT/STFT/Welch, cross-spectrum, coherence, windowing/leakage, noise sensitivity, bias/variance.
• Experience implementing estimators in Python (NumPy/SciPy) and/or C++/Rust with attention to numerical stability and testability.
• Comfort with messy real-world signals: clipping, dropouts, nonstationarity, interference, clock mismatch.
• Ability to write tight, measurable specs: what success means, when to abstain, and how to know it’s broken.

Preferred Qualifications

• Prior work on time synchronization, sample-rate offset estimation, PLL concepts, sensor fusion, RF/audio timing, or inertial systems.
• Familiarity with robust statistics (Huber/RANSAC/quantile methods), filtering/smoothing (Kalman, RTS), or uncertainty quantification.
• Experience building data-centric QA pipelines: dataset versioning, metric tracking, automated regression triage.

How You’ll Be Measured (First 60–90 Days)

• Delay estimation accuracy improves on benchmark datasets (lower MAE, fewer catastrophic outliers).
• Coherence/validity gating reduces false “good” windows and provides actionable failure labels.
• Drift tracking remains stable across long runs and degraded conditions (dropouts, SNR swings).
• QA dataset + harness becomes a dependable regression shield for the team.

Working Style

• You build estimators that come with a confidence score and a failure mode, not just a number.
• You like instrumented algorithms: plots, logs, and metrics that explain behavior under stress.
• You’re collaborative—able to translate between math, code, and product needs.

Title & Level

Algorithm Engineer (DSP / Estimation) (mid-to-senior, depending on experience), reporting into the signal processing / applied math lead.