oscura 0.7.0__py3-none-any.whl → 0.10.0__py3-none-any.whl

oscura/analyzers/signal_classification.py

@@ -0,0 +1,529 @@
+"""Signal classification and waveform type detection.
+
+This module provides automatic signal classification to detect periodicity,
+waveform shape, and signal characteristics. Classification results are used
+to determine which measurements are applicable to a given signal.
+
+Example:
+    >>> from oscura.analyzers.signal_classification import classify_signal
+    >>> from oscura.core.types import WaveformTrace, TraceMetadata
+    >>> import numpy as np
+    >>> # Create test signal
+    >>> t = np.linspace(0, 0.01, 1000)
+    >>> data = np.sin(2 * np.pi * 1000 * t)
+    >>> meta = TraceMetadata(sample_rate=1e5, units="V")
+    >>> trace = WaveformTrace(data=data, metadata=meta)
+    >>> result = classify_signal(trace)
+    >>> print(f"Waveform: {result['waveform']['waveform_type']}")
+    Waveform: sine
+
+References:
+    IEEE 181-2011: Standard for Transitional Waveform Definitions
+    Signal Processing Fundamentals (autocorrelation methods)
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any
+
+import numpy as np
+
+if TYPE_CHECKING:
+    from numpy.typing import NDArray
+
+    from oscura.core.types import WaveformTrace
+
+
+def detect_periodicity(signal: NDArray[np.floating[Any]], sample_rate: float) -> dict[str, Any]:
+    """Detect periodic vs aperiodic signals using autocorrelation.
+
+    Uses normalized autocorrelation to identify periodic patterns in the signal.
+    A strong peak at non-zero lag indicates periodicity, while absence of peaks
+    suggests aperiodic signals (impulse, DC, random noise).
+
+    Args:
+        signal: Input signal array.
+        sample_rate: Sample rate in Hz.
+
+    Returns:
+        Dictionary containing:
+            - periodicity: "periodic", "aperiodic", or "quasi-periodic"
+            - confidence: Float between 0.0 and 1.0
+            - period: Period in seconds (None if aperiodic)
+
+    Example:
+        >>> # Periodic signal
+        >>> signal = np.sin(2 * np.pi * 1000 * np.linspace(0, 0.01, 1000))
+        >>> result = detect_periodicity(signal, 100000)
+        >>> print(f"Type: {result['periodicity']}, Period: {result['period']}")
+        Type: periodic, Period: 0.001
+
+        >>> # Aperiodic signal (impulse)
+        >>> impulse = np.zeros(1000)
+        >>> impulse[500] = 1.0
+        >>> result = detect_periodicity(impulse, 100000)
+        >>> print(f"Type: {result['periodicity']}")
+        Type: aperiodic
+    """
+    if len(signal) < 16:
+        return {
+            "periodicity": "aperiodic",
+            "confidence": 0.0,
+            "period": None,
+        }
+
+    # Remove DC offset for better autocorrelation
+    signal_centered = signal - np.mean(signal)
+
+    # Check if signal is essentially constant (DC only)
+    if np.std(signal_centered) < 1e-10:
+        return {
+            "periodicity": "aperiodic",
+            "confidence": 1.0,
+            "period": None,
+        }
+
+    # Compute normalized autocorrelation
+    # Use only first half of signal to avoid edge effects
+    max_lag = min(len(signal_centered) // 2, 1000)
+    autocorr = np.correlate(signal_centered, signal_centered, mode="full")
+    autocorr = autocorr[len(autocorr) // 2 :]  # Take only positive lags
+    autocorr = autocorr[:max_lag]
+
+    # Normalize by zero-lag value
+    if autocorr[0] > 1e-10:
+        autocorr = autocorr / autocorr[0]
+    else:
+        return {
+            "periodicity": "aperiodic",
+            "confidence": 1.0,
+            "period": None,
+        }
+
+    # Find peaks in autocorrelation (skip first 2 samples to avoid DC)
+    min_period_samples = 3
+    search_region = autocorr[min_period_samples:]
+
+    if len(search_region) < 3:
+        return {
+            "periodicity": "aperiodic",
+            "confidence": 0.5,
+            "period": None,
+        }
+
+    # Find first significant peak
+    peak_idx = -1
+    peak_value = 0.0
+
+    # Look for local maxima
+    for i in range(1, len(search_region) - 1):
+        if (
+            search_region[i] > search_region[i - 1]
+            and search_region[i] > search_region[i + 1]
+            and search_region[i] > peak_value
+        ):
+            peak_value = search_region[i]
+            peak_idx = i + min_period_samples
+
+    # Classify based on peak strength
+    # Strong peak (>0.5) = periodic
+    # Moderate peak (0.3-0.5) = quasi-periodic
+    # Weak/no peak (<0.3) = aperiodic
+    if peak_value > 0.5:
+        period_samples = peak_idx
+        period_seconds = period_samples / sample_rate
+        return {
+            "periodicity": "periodic",
+            "confidence": float(peak_value),
+            "period": float(period_seconds),
+        }
+    elif peak_value > 0.3:
+        period_samples = peak_idx
+        period_seconds = period_samples / sample_rate
+        return {
+            "periodicity": "quasi-periodic",
+            "confidence": float(peak_value),
+            "period": float(period_seconds),
+        }
+    else:
+        return {
+            "periodicity": "aperiodic",
+            "confidence": float(1.0 - peak_value),
+            "period": None,
+        }
+
+
+def classify_waveform(signal: NDArray[np.floating[Any]], sample_rate: float) -> dict[str, Any]:
+    """Classify waveform shape using spectral and time-domain analysis.
+
+    Analyzes harmonic content, duty cycle, edge sharpness, and DC component
+    to determine the waveform type.
+
+    Args:
+        signal: Input signal array.
+        sample_rate: Sample rate in Hz.
+
+    Returns:
+        Dictionary containing:
+            - waveform_type: One of "sine", "square", "triangle", "sawtooth",
+                           "pwm", "impulse", "dc", "noise", "unknown"
+            - confidence: Float between 0.0 and 1.0
+            - characteristics: Dict with additional features
+
+    Example:
+        >>> # Square wave
+        >>> signal = np.sign(np.sin(2 * np.pi * 1000 * np.linspace(0, 0.01, 1000)))
+        >>> result = classify_waveform(signal, 100000)
+        >>> print(f"Type: {result['waveform_type']}")
+        Type: square
+    """
+    if len(signal) < 16:
+        return {
+            "waveform_type": "unknown",
+            "confidence": 0.0,
+            "characteristics": {},
+        }
+
+    # Remove DC offset
+    signal_centered = signal - np.mean(signal)
+    dc_component = float(np.mean(signal))
+
+    # Check for DC signal (no variation)
+    signal_std = np.std(signal_centered)
+    if signal_std < 1e-10:
+        return {
+            "waveform_type": "dc",
+            "confidence": 1.0,
+            "characteristics": {
+                "dc_level": float(dc_component),
+                "noise_level": 0.0,
+            },
+        }
+
+    # Compute FFT for harmonic analysis
+    n = len(signal_centered)
+    fft_mag = np.abs(np.fft.rfft(signal_centered))
+    fft_freqs = np.fft.rfftfreq(n, 1 / sample_rate)
+
+    # Skip DC component
+    fft_mag = fft_mag[1:]
+    fft_freqs = fft_freqs[1:]
+
+    if len(fft_mag) < 3:
+        return {
+            "waveform_type": "unknown",
+            "confidence": 0.0,
+            "characteristics": {},
+        }
+
+    # Find fundamental frequency (strongest component)
+    fundamental_idx = np.argmax(fft_mag)
+    fundamental_freq = fft_freqs[fundamental_idx]
+    fundamental_mag = fft_mag[fundamental_idx]
+
+    # Check if signal is essentially noise (no dominant frequency)
+    mean_mag = np.mean(fft_mag)
+    if fundamental_mag < 3.0 * mean_mag:
+        return {
+            "waveform_type": "noise",
+            "confidence": 0.8,
+            "characteristics": {
+                "snr_estimate": float(fundamental_mag / mean_mag) if mean_mag > 0 else 0.0,
+            },
+        }
+
+    # Analyze harmonics (multiples of fundamental)
+    harmonics = []
+    for harmonic_num in range(2, 6):  # Check 2nd through 5th harmonic
+        harmonic_freq = fundamental_freq * harmonic_num
+        # Find closest frequency bin
+        freq_idx = np.argmin(np.abs(fft_freqs - harmonic_freq))
+        if freq_idx < len(fft_mag):
+            harmonic_mag = fft_mag[freq_idx]
+            # Normalize by fundamental
+            harmonic_ratio = harmonic_mag / fundamental_mag if fundamental_mag > 0 else 0.0
+            harmonics.append(float(harmonic_ratio))
+
+    # Calculate total harmonic distortion (THD)
+    harmonic_power = sum(h**2 for h in harmonics)
+    thd = np.sqrt(harmonic_power) if harmonics else 0.0
+
+    # Classify based on harmonic content
+    characteristics: dict[str, Any] = {
+        "fundamental_freq": float(fundamental_freq),
+        "thd": float(thd),
+        "harmonics": harmonics,
+    }
+
+    # Early check for impulse: Very brief pulse (check time domain first)
+    # For impulse detection, we need to check if signal is mostly near zero
+    # with only brief excursions. Use absolute signal, not centered.
+    abs_max = np.max(np.abs(signal))
+    # Count samples that are significant (>10% of peak)
+    threshold = 0.1 * abs_max if abs_max > 0 else signal_std
+    significant_samples = np.sum(np.abs(signal) > threshold)
+    pulse_width_ratio = significant_samples / len(signal) if len(signal) > 0 else 0
+
+    # True impulse: very few significant samples AND mostly zero baseline
+    signal_range = np.max(signal) - np.min(signal)
+    near_zero_count = np.sum(np.abs(signal - np.min(signal)) < 0.1 * signal_range)
+    near_zero_ratio = near_zero_count / len(signal) if len(signal) > 0 else 0
+
+    if pulse_width_ratio < 0.15 and near_zero_ratio > 0.7:
+        return {
+            "waveform_type": "impulse",
+            "confidence": 0.85,
+            "characteristics": {
+                **characteristics,
+                "pulse_width_ratio": float(pulse_width_ratio),
+            },
+        }
+
+    # Sine wave: Low THD (<0.1), single dominant frequency
+    if thd < 0.1:
+        return {
+            "waveform_type": "sine",
+            "confidence": 0.9,
+            "characteristics": characteristics,
+        }
+
+    # Early check for PWM: estimate duty cycle first
+    duty_cycle = _estimate_duty_cycle(signal)
+
+    # PWM: Extreme duty cycle (far from 50%) with digital-like behavior
+    if duty_cycle < 0.35 or duty_cycle > 0.65:  # Duty cycle far from 50%
+        # Additional check: signal should be mostly at two levels (digital-like)
+        unique_count = len(np.unique(signal))
+        if unique_count <= 10:  # Few discrete levels = digital PWM
+            return {
+                "waveform_type": "pwm",
+                "confidence": 0.8,
+                "characteristics": {**characteristics, "duty_cycle": float(duty_cycle)},
+            }
+
+    # Square wave: Strong odd harmonics (1, 3, 5, 7...)
+    # Theoretical ratios: 1/3, 1/5, 1/7, 1/9
+    if len(harmonics) >= 2:
+        # Check for strong 3rd harmonic and weak 2nd harmonic
+        if harmonics[1] > 0.2 and harmonics[0] < 0.15:  # 3rd strong, 2nd weak
+            # Additional check: duty cycle near 50%
+            if 0.4 < duty_cycle < 0.6:
+                return {
+                    "waveform_type": "square",
+                    "confidence": 0.85,
+                    "characteristics": {**characteristics, "duty_cycle": float(duty_cycle)},
+                }
+            else:
+                # Square wave with duty cycle != 50% = PWM
+                return {
+                    "waveform_type": "pwm",
+                    "confidence": 0.8,
+                    "characteristics": {**characteristics, "duty_cycle": float(duty_cycle)},
+                }
+
+    # Triangle wave: Strong odd harmonics with rapid decay (1/9, 1/25, ...)
+    if len(harmonics) >= 2:
+        # Check for rapidly decaying odd harmonics
+        if (
+            harmonics[1] > 0.05 and harmonics[1] < 0.15  # 3rd harmonic weaker than square
+        ):
+            return {
+                "waveform_type": "triangle",
+                "confidence": 0.75,
+                "characteristics": characteristics,
+            }
+
+    # Sawtooth: Both odd and even harmonics (1/2, 1/3, 1/4, ...)
+    if len(harmonics) >= 2:
+        if harmonics[0] > 0.3 and harmonics[1] > 0.15:  # Both 2nd and 3rd strong
+            return {
+                "waveform_type": "sawtooth",
+                "confidence": 0.75,
+                "characteristics": characteristics,
+            }
+
+    # Unknown waveform type
+    return {
+        "waveform_type": "unknown",
+        "confidence": 0.5,
+        "characteristics": characteristics,
+    }
+
+
+def classify_signal(trace: WaveformTrace) -> dict[str, Any]:
+    """Perform complete signal classification.
+
+    Combines periodicity detection, waveform classification, and signal
+    quality assessment to provide comprehensive signal characterization.
+
+    Args:
+        trace: Input waveform trace.
+
+    Returns:
+        Dictionary containing:
+            - domain: "analog", "digital", or "mixed"
+            - periodicity: Dict from detect_periodicity()
+            - waveform: Dict from classify_waveform()
+            - signal_quality: Dict with SNR, clipping detection, etc.
+
+    Example:
+        >>> from oscura.core.types import WaveformTrace, TraceMetadata
+        >>> import numpy as np
+        >>> # Sine wave example
+        >>> t = np.linspace(0, 0.01, 1000)
+        >>> data = 3.3 * np.sin(2 * np.pi * 1000 * t)
+        >>> meta = TraceMetadata(sample_rate=1e5, units="V")
+        >>> trace = WaveformTrace(data=data, metadata=meta)
+        >>> result = classify_signal(trace)
+        >>> print(f"Domain: {result['domain']}")
+        >>> print(f"Type: {result['waveform']['waveform_type']}")
+        >>> print(f"Periodic: {result['periodicity']['periodicity']}")
+        Domain: analog
+        Type: sine
+        Periodic: periodic
+    """
+    signal = trace.data
+    sample_rate = trace.metadata.sample_rate
+
+    # Detect domain (analog vs digital vs mixed)
+    domain = _detect_domain(signal)
+
+    # Detect periodicity
+    periodicity = detect_periodicity(signal, sample_rate)
+
+    # Classify waveform shape
+    waveform = classify_waveform(signal, sample_rate)
+
+    # Assess signal quality
+    signal_quality = _assess_signal_quality(signal)
+
+    return {
+        "domain": domain,
+        "periodicity": periodicity,
+        "waveform": waveform,
+        "signal_quality": signal_quality,
+    }
+
+
+# =============================================================================
+# Helper Functions
+# =============================================================================
+
+
+def _detect_domain(signal: NDArray[np.floating[Any]]) -> str:
+    """Detect if signal is analog, digital, or mixed.
+
+    Args:
+        signal: Input signal array.
+
+    Returns:
+        "analog", "digital", or "mixed"
+    """
+    if len(signal) < 3:
+        return "analog"
+
+    # Count unique values
+    unique_values = np.unique(signal)
+
+    # If only 2 unique values, likely digital
+    if len(unique_values) <= 2:
+        return "digital"
+
+    # Check if signal spends most time at discrete levels (digital-like)
+    # Build histogram and check if most samples are at a few discrete levels
+    hist, _ = np.histogram(signal, bins=50)
+    # If >80% of samples are in top 3 bins, likely digital or mixed
+    sorted_bins = np.sort(hist)[::-1]
+    top_3_ratio = np.sum(sorted_bins[:3]) / len(signal) if len(signal) > 0 else 0
+
+    if top_3_ratio > 0.8:
+        # Check if there are transitions (if yes, digital; if no, DC)
+        if len(unique_values) <= 5:
+            return "digital"
+        else:
+            return "mixed"
+
+    return "analog"
+
+
+def _assess_signal_quality(signal: NDArray[np.floating[Any]]) -> dict[str, Any]:
+    """Assess signal quality (SNR, clipping, etc.).
+
+    Args:
+        signal: Input signal array.
+
+    Returns:
+        Dictionary with quality metrics.
+    """
+    if len(signal) < 3:
+        return {
+            "snr": 0.0,
+            "clipping_detected": False,
+            "noise_level": 0.0,
+        }
+
+    # Estimate SNR (simplified)
+    signal_power = float(np.mean(signal**2))
+    noise_estimate = float(np.std(signal) * 0.1)  # Rough estimate
+    snr = 10 * np.log10(signal_power / noise_estimate**2) if noise_estimate > 0 else 100.0
+
+    # Detect clipping (signal at rail)
+    signal_min = float(np.min(signal))
+    signal_max = float(np.max(signal))
+    signal_range = signal_max - signal_min
+
+    # Check if many samples are at min or max
+    threshold = signal_range * 0.01  # Within 1% of rail
+    at_min = np.sum(signal <= (signal_min + threshold))
+    at_max = np.sum(signal >= (signal_max - threshold))
+    clipping_ratio = (at_min + at_max) / len(signal)
+
+    clipping_detected = clipping_ratio > 0.1  # >10% of samples at rail
+
+    return {
+        "snr": float(snr),
+        "clipping_detected": bool(clipping_detected),
+        "noise_level": float(noise_estimate),
+        "dynamic_range": float(signal_range),
+    }
+
+
+def _estimate_duty_cycle(signal: NDArray[np.floating[Any]]) -> float:
+    """Estimate duty cycle from signal.
+
+    Args:
+        signal: Input signal array.
+
+    Returns:
+        Duty cycle as ratio (0.0 to 1.0).
+    """
+    if len(signal) < 3:
+        return 0.5
+
+    # Find signal levels using histogram
+    hist, bin_edges = np.histogram(signal, bins=50)
+    bin_centers = (bin_edges[:-1] + bin_edges[1:]) / 2
+
+    # Find peaks in lower and upper halves
+    mid_idx = len(hist) // 2
+    low_idx = np.argmax(hist[:mid_idx]) if mid_idx > 0 else 0
+    high_idx = mid_idx + np.argmax(hist[mid_idx:])
+
+    low = bin_centers[low_idx]
+    high = bin_centers[high_idx]
+
+    # Calculate threshold at midpoint
+    mid = (low + high) / 2
+
+    # Count samples above threshold
+    above_threshold = np.sum(signal >= mid)
+    duty_cycle = above_threshold / len(signal) if len(signal) > 0 else 0.5
+
+    return float(duty_cycle)
+
+
+__all__ = [
+    "classify_signal",
+    "classify_waveform",
+    "detect_periodicity",
+]

oscura/analyzers/signal_integrity/sparams.py

@@ -24,6 +24,9 @@ from typing import TYPE_CHECKING
 
 import numpy as np
 
+# Note: load_touchstone removed from re-export to break cyclic import
+# Import directly from: from oscura.loaders import load_touchstone
+
 if TYPE_CHECKING:
     from numpy.typing import NDArray
 
@@ -277,9 +280,6 @@ def _abcd_to_s_single(abcd: NDArray[np.complex128], z0: float) -> NDArray[np.com
     return np.array([[S11, S12], [S21, S22]], dtype=np.complex128)
 
 
-# load_touchstone has been moved to oscura.loaders module
-
-
 __all__ = [
     "SParameterData",
     "abcd_to_s",

oscura/analyzers/statistics/__init__.py

@@ -20,11 +20,13 @@ from oscura.analyzers.statistics.advanced import (
 )
 from oscura.analyzers.statistics.basic import (
     basic_stats,
+    measure,
     percentiles,
     quartiles,
     running_stats,
     summary_stats,
     weighted_mean,
+    weighted_std,
 )
 from oscura.analyzers.statistics.correlation import (
     CrossCorrelationResult,
@@ -99,6 +101,7 @@ __all__ = [
     "kernel_density",
     # Advanced (STAT-012)
     "local_outlier_factor",
+    "measure",
     "modified_zscore_outliers",
     "moment",
     "moving_average",
@@ -114,6 +117,7 @@ __all__ = [
     "seasonal_decompose",
     "summary_stats",
     "weighted_mean",
+    "weighted_std",
     # Outlier detection
     "zscore_outliers",
 ]