Precision Spectral Analysis & Automated Filter Synthesis
SignalForge is a deterministic coprocessor for DSP teams. It turns raw time-domain samples into auditable evidence + normalized IIR/FIR coefficients you can ship to firmware— without a black box.
License: Perpetual per major version. Includes 1-year maintenance.
Example Case (Real Outputs): 1 kHz Tone + Spurs
Hard Evidence: IIR Stability (Pole Radius < 1.0)
For biquad SOS with a0 = 1, a quick (conservative) engineering sanity check is r ≈ sqrt(a2).
Below is from the exported SOS coefficients in filter_coeffs.h (case-matched).
| Notch Center (Hz) | Q | a2 | r ≈ sqrt(a2) |
|---|---|---|---|
| 1212.8906 | 30.0 | 0.99472172 | 0.9974 |
Note: This is a conservative “sanity check” (fast audit). Full response plots and design details are in the PDF.
filter_coeffs.h) — SOS + FIR taps.Open: filter_coeffs.h · report.pdf
Technical Capabilities
Built for recurring engineering pain: “What is actually tonal?”, “Will it drift?”, “What coefficients do I ship?”, and “Can I defend this design?”
Statistical Characterization
Cross-check PSD with STFT to separate stationary tones, drifting tones, and transient events. Designed to reduce false positives and “phantom notch” designs.
Drift Tracking
Uses time-frequency evidence to estimate drift bandwidth and choose practical Q-factors—so the notch keeps working after temperature/time variation.
Direct Code Gen
Exports normalized SOS and FIR taps into a C header. Avoids manual transcription and coefficient normalization mistakes.
Hard Evidence (Not Marketing Claims)
The example case ships with real artifacts you can audit: plots, JSON, a PDF report, and a firmware-ready header.
- Before/After PSD: measurable changes where it matters, without hiding the main tone.
- Tonal evidence panel: STFT + time-coverage / presence visualization for detected tonal candidates.
- Design outputs: explicit parameters (notch center/Q, FIR cutoff/taps) in exported artifacts.
- PDF Report — audit trail & plots.
- filter_coeffs.h — firmware deliverable.
- Before/After PSD (PNG) —
before_after_psd_1k.png. - Tonal Evidence Panel (PNG) —
tonal_debug_1k.png.
Note: artifacts shown are generated outputs (not mockups).
Engineering Limitations
SignalForge is designed to assist engineering decisions — not replace them. The following boundaries are explicit by design.
SignalForge does not infer or optimize for system-specific constraints such as fixed-point word length, saturation strategy, CPU budget, real-time scheduling, or memory layout.
These remain explicit engineering decisions and are intentionally left under user control.
All detected tones, filter parameters, and coefficients are fully exposed in JSON and C header outputs.
Engineers are free to review, modify, or reject any design decision. SignalForge does not enforce acceptance of generated results.
User-Governed Design Parameters
SignalForge operates under explicit user-defined specifications rather than implicit assumptions.
- Design scope: allowed filter types, notch count limits.
- Detection thresholds: prominence and time-coverage criteria.
- Design preferences: conservative vs. sensitive analysis modes.
- Override policy: engineers may accept, modify, or discard results.
All specifications and outputs are fully auditable via exported JSON artifacts and the C header.
Core Specifications
A rigorous pipeline designed for verification-first workflows.
- Data Import: WAV, CSV (time-series).
- Detection Logic: prominence + presence ratio (time coverage), cross-validated by STFT.
- Leakage Handling: window-aware analysis to reduce spectral artifacts.
- Health Checks: clipping, DC offset, sanity checks for signal quality.
- Filter Types: IIR Notch (SOS), FIR Low-pass.
- Deliverables:
design_result.json,noise_analysis.json,filter_coeffs.h,report.pdf. - Normalization: exported SOS uses
a0=1layout for embedded DSP loops. - Designed for: firmware handoff, reviews, QA audits, regression baselines.
The "Audit-Ready" PDF Report
The report is meant to answer “why this notch?”, “what evidence?”, “what parameters?”, and “what exactly do we ship?”—in one artifact.
Evidence Chain
Plots and tracking evidence that justify detection decisions (not just a filter response screenshot).
Parameter Traceability
Shows concrete parameters and the case context so results can be reproduced and reviewed.
Download Sample
Use it directly in technical reviews. It is a real report generated from a real input case.
Production Outputs
Zero-friction handoff to firmware engineering: no re-typing, no hidden normalization steps.
C Header Export
SOS layout: [b0, b1, b2, a0, a1, a2], with a0 = 1. FIR taps included.
JSON Artifacts
Design and analysis JSON can be version-controlled for regression testing and review traceability.
Report Artifact
One PDF you can archive, email, and attach to issue trackers or compliance reviews.
Licensing
Simple, engineer-friendly terms. No hidden subscriptions.
- 14-Day Full Feature Trial
- Offline Execution (Air-gap friendly)
- Node-locked Activation
- 1st Year Maintenance Included
- Optional Maintenance (Year 2+): $45/year
The trial is fully functional for 14 days so you can validate on a real project end-to-end.
Post-Trial State: Analysis remains viewable. Code export and PDF generation require activation.
- Windows 10/11 (x64)
- Import: WAV (PCM), CSV
Technical FAQ
Clarifications for engineering teams.
Is internet access required?
Only for initial license activation. SignalForge runs offline (air-gap friendly) for secure lab environments.
Is there any data telemetry?
No. SignalForge analyzes local files only. No signal data is uploaded.
How does SignalForge avoid “phantom notches”?
Notch candidates are evaluated using both PSD and STFT evidence (time-coverage / presence behavior). This helps distinguish a true stationary/drifting tone from short transients or leakage artifacts.
Can I import the coefficients into ARM CMSIS-DSP?
Yes. The export includes SOS layout compatible with common biquad cascade implementations (including CMSIS-style layouts).
What are the analysis modes (Standard / High Precision / High Sensitivity)?
Standard: balanced default for most signals.
High Precision: tighter analysis settings to reduce false positives (recommended for conservative analysis).
High Sensitivity: detects weaker tones but may require stricter prominence/presence thresholds to avoid over-notching.
What is the maintenance policy?
The license is perpetual for the major version purchased. You get critical fixes and minor updates for 1 year. Extended maintenance is optional at $45/year.
Policies & Legal
Support
Technical Inquiries: [email protected]