oscura 0.0.1__py3-none-any.whl → 0.1.1__py3-none-any.whl

oscura/inference/adaptive_tuning.py

@@ -0,0 +1,453 @@
+"""Adaptive parameter tuning for analysis functions.
+
+Auto-configures analysis parameters based on signal characteristics,
+reducing the need for manual parameter specification.
+
+
+Example:
+    >>> import oscura as osc
+    >>> trace = osc.load('signal.wfm')
+    >>> tuner = osc.AdaptiveParameterTuner(trace.data, trace.metadata.sample_rate)
+    >>> params = tuner.get_spectral_params()
+    >>> print(f"NFFT: {params.get('nfft')}")
+    >>> print(f"Window: {params.get('window')}")
+    >>> print(f"Reasoning: {params.reasoning}")
+
+References:
+    Harris, F. J. (1978): On the use of windows for harmonic analysis with DFT
+    Oppenheim, A. V. & Schafer, R. W. (2010): Discrete-Time Signal Processing
+"""
+
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass, field
+from typing import TYPE_CHECKING, Any
+
+import numpy as np
+
+if TYPE_CHECKING:
+    from numpy.typing import NDArray
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class TunedParameters:
+    """Container for auto-tuned parameters.
+
+    Attributes:
+        parameters: Dictionary of parameter names to values.
+        confidence: Confidence in parameter tuning (0.0-1.0).
+        reasoning: Dictionary mapping parameter names to reasoning strings.
+    """
+
+    parameters: dict[str, Any] = field(default_factory=dict)
+    confidence: float = 0.5
+    reasoning: dict[str, str] = field(default_factory=dict)
+
+    def get(self, key: str, default: Any = None) -> Any:
+        """Get parameter value with default.
+
+        Args:
+            key: Parameter name.
+            default: Default value if parameter not found.
+
+        Returns:
+            Parameter value or default.
+        """
+        return self.parameters.get(key, default)
+
+
+class AdaptiveParameterTuner:
+    """Auto-configure analysis parameters based on signal characteristics.
+
+    This class analyzes signal characteristics and provides intelligent
+    parameter suggestions for various analysis domains (spectral, digital,
+    timing, jitter, pattern recognition).
+
+    Attributes:
+        data: Input signal data array.
+        sample_rate: Sample rate in Hz.
+        signal_type: Optional signal type hint (digital, analog, etc.).
+
+    Example:
+        >>> tuner = AdaptiveParameterTuner(signal_data, sample_rate=1e6)
+        >>> spectral_params = tuner.get_spectral_params()
+        >>> print(spectral_params.parameters)
+        {'nfft': 8192, 'window': 'hann', 'overlap': 0.5}
+    """
+
+    def __init__(
+        self,
+        data: NDArray[np.floating[Any]],
+        sample_rate: float = 1.0,
+        signal_type: str | None = None,
+    ):
+        """Initialize tuner with signal data.
+
+        Args:
+            data: Input signal data.
+            sample_rate: Sample rate in Hz.
+            signal_type: Optional signal type hint (digital, analog, etc.).
+        """
+        self.data = data
+        self.sample_rate = sample_rate
+        self.signal_type = signal_type
+
+        # Pre-compute signal characteristics
+        self._characteristics = self._analyze_signal()
+
+    def _analyze_signal(self) -> dict[str, Any]:
+        """Analyze signal characteristics for parameter tuning.
+
+        Returns:
+            Dictionary of signal characteristics including statistics,
+            noise estimates, frequency content, and signal type indicators.
+        """
+        chars: dict[str, Any] = {}
+
+        try:
+            # Basic statistics
+            chars["mean"] = float(np.mean(self.data))
+            chars["std"] = float(np.std(self.data))
+            chars["min"] = float(np.min(self.data))
+            chars["max"] = float(np.max(self.data))
+            chars["range"] = chars["max"] - chars["min"]
+            chars["n_samples"] = len(self.data)
+            chars["duration"] = len(self.data) / self.sample_rate
+
+            # Detect if digital
+            unique_values = len(np.unique(np.round(self.data, decimals=2)))
+            chars["likely_digital"] = unique_values < 10
+
+            # Estimate dominant frequency
+            chars["dominant_freq"] = self._estimate_dominant_frequency()
+
+            # Estimate noise floor
+            median = np.median(self.data)
+            mad = np.median(np.abs(self.data - median)) * 1.4826
+            chars["noise_floor"] = float(mad)
+
+            # SNR estimate
+            signal_power = np.var(self.data)
+            noise_power = mad**2
+            if noise_power > 0:
+                chars["snr_db"] = float(10 * np.log10(signal_power / noise_power))
+            else:
+                chars["snr_db"] = 40.0
+
+        except Exception as e:
+            logger.debug(f"Error analyzing signal: {e}")
+
+        return chars
+
+    def _estimate_dominant_frequency(self) -> float | None:
+        """Estimate dominant frequency using FFT.
+
+        Returns:
+            Dominant frequency in Hz, or None if not detectable.
+        """
+        try:
+            data_ac = self.data - np.mean(self.data)
+            fft_result = np.fft.rfft(data_ac)
+            freqs = np.fft.rfftfreq(len(data_ac), d=1.0 / self.sample_rate)
+            magnitude = np.abs(fft_result[1:])  # Skip DC
+
+            if len(magnitude) > 0:
+                peak_idx = np.argmax(magnitude)
+                return float(freqs[1:][peak_idx])
+        except Exception:
+            pass
+        return None
+
+    def get_spectral_params(self) -> TunedParameters:
+        """Get tuned parameters for spectral analysis.
+
+        Selects FFT size, window function, and overlap based on signal
+        characteristics and quality requirements.
+
+        Returns:
+            TunedParameters with spectral analysis configuration.
+
+        Example:
+            >>> params = tuner.get_spectral_params()
+            >>> print(f"NFFT: {params.get('nfft')}")
+            >>> print(f"Reasoning: {params.reasoning['nfft']}")
+        """
+        params = {}
+        reasoning = {}
+        confidence = 0.8
+
+        n_samples = self._characteristics.get("n_samples", 1000)
+
+        # NFFT - power of 2, balancing resolution and computation
+        ideal_nfft = min(8192, max(256, 2 ** int(np.ceil(np.log2(n_samples / 4)))))
+        params["nfft"] = ideal_nfft
+        reasoning["nfft"] = f"Power of 2 for efficiency, ~{n_samples / ideal_nfft:.0f} averages"
+
+        # Window selection based on signal characteristics
+        snr = self._characteristics.get("snr_db", 20)
+        if snr < 15:
+            params["window"] = "blackman"
+            reasoning["window"] = "Low SNR - using Blackman for better noise rejection"
+        elif snr < 25:
+            params["window"] = "hann"
+            reasoning["window"] = "Moderate SNR - using Hann for balance"
+        else:
+            params["window"] = "hamming"
+            reasoning["window"] = "Good SNR - using Hamming for resolution"
+
+        # Overlap
+        params["overlap"] = 0.5
+        reasoning["overlap"] = "Standard 50% overlap for smooth averaging"
+
+        # Frequency range based on dominant frequency
+        dom_freq = self._characteristics.get("dominant_freq")
+        if dom_freq and dom_freq > 0:
+            params["freq_min"] = max(0, dom_freq / 10)
+            params["freq_max"] = min(self.sample_rate / 2, dom_freq * 5)
+            reasoning["freq_range"] = f"Based on dominant frequency {dom_freq:.1f} Hz"
+
+        return TunedParameters(parameters=params, confidence=confidence, reasoning=reasoning)
+
+    def get_digital_params(self) -> TunedParameters:
+        """Get tuned parameters for digital signal analysis.
+
+        Determines threshold levels, edge detection sensitivity, and
+        baud rate hints based on signal characteristics.
+
+        Returns:
+            TunedParameters with digital analysis configuration.
+
+        Example:
+            >>> params = tuner.get_digital_params()
+            >>> print(f"Threshold: {params.get('threshold')}")
+            >>> print(f"Baud rate hint: {params.get('baud_rate_hint')}")
+        """
+        params = {}
+        reasoning = {}
+        confidence = 0.7
+
+        chars = self._characteristics
+
+        # Threshold based on signal levels
+        if chars.get("likely_digital"):
+            mid = (chars["min"] + chars["max"]) / 2
+            params["threshold"] = mid
+            params["threshold_low"] = chars["min"] + 0.3 * chars["range"]
+            params["threshold_high"] = chars["max"] - 0.3 * chars["range"]
+            reasoning["threshold"] = (
+                f"Midpoint of signal range ({chars['min']:.2f} to {chars['max']:.2f})"
+            )
+            confidence = 0.85
+        else:
+            params["threshold"] = chars.get("mean", 0)
+            reasoning["threshold"] = "Using mean (signal may not be digital)"
+            confidence = 0.5
+
+        # Edge detection sensitivity based on noise
+        noise = chars.get("noise_floor", 0.1)
+        params["min_edge_separation"] = max(2, int(noise * 10))
+        reasoning["min_edge_separation"] = f"Based on noise floor {noise:.3f}"
+
+        # Baud rate hint from dominant frequency
+        dom_freq = chars.get("dominant_freq")
+        if dom_freq and dom_freq > 0:
+            # Common baud rates
+            common_bauds = [300, 1200, 2400, 4800, 9600, 19200, 38400, 57600, 115200]
+            closest_baud = min(common_bauds, key=lambda b: abs(b - dom_freq * 2))
+            params["baud_rate_hint"] = closest_baud
+            reasoning["baud_rate"] = f"Estimated from frequency {dom_freq:.0f} Hz"
+
+        return TunedParameters(parameters=params, confidence=confidence, reasoning=reasoning)
+
+    def get_timing_params(self) -> TunedParameters:
+        """Get tuned parameters for timing analysis.
+
+        Configures time resolution, expected period, and edge timing
+        thresholds based on sample rate and signal characteristics.
+
+        Returns:
+            TunedParameters with timing analysis configuration.
+
+        Example:
+            >>> params = tuner.get_timing_params()
+            >>> print(f"Expected period: {params.get('expected_period')}")
+            >>> print(f"Tolerance: {params.get('period_tolerance')}")
+        """
+        params = {}
+        reasoning = {}
+        confidence = 0.75
+
+        chars = self._characteristics
+
+        # Time resolution based on sample rate
+        params["time_resolution"] = 1.0 / self.sample_rate
+        reasoning["time_resolution"] = f"Based on sample rate {self.sample_rate:.0f} Hz"
+
+        # Expected period from dominant frequency
+        dom_freq = chars.get("dominant_freq")
+        if dom_freq and dom_freq > 0:
+            params["expected_period"] = 1.0 / dom_freq
+            params["period_tolerance"] = 0.2 / dom_freq  # 20% tolerance
+            reasoning["period"] = f"From dominant frequency {dom_freq:.1f} Hz"
+            confidence = 0.85
+
+        # Edge timing thresholds
+        noise = chars.get("noise_floor", 0.1)
+        params["edge_threshold"] = noise * 3  # 3-sigma
+        reasoning["edge_threshold"] = f"3x noise floor ({noise:.3f})"
+
+        return TunedParameters(parameters=params, confidence=confidence, reasoning=reasoning)
+
+    def get_jitter_params(self) -> TunedParameters:
+        """Get tuned parameters for jitter analysis.
+
+        Determines unit interval, histogram binning, and tolerance
+        parameters for jitter measurements.
+
+        Returns:
+            TunedParameters with jitter analysis configuration.
+
+        Example:
+            >>> params = tuner.get_jitter_params()
+            >>> print(f"Unit interval: {params.get('unit_interval')}")
+            >>> print(f"Histogram bins: {params.get('histogram_bins')}")
+        """
+        params = {}
+        reasoning = {}
+        confidence = 0.7
+
+        chars = self._characteristics
+
+        # Unit interval from dominant frequency
+        dom_freq = chars.get("dominant_freq")
+        if dom_freq and dom_freq > 0:
+            ui = 1.0 / dom_freq
+            params["unit_interval"] = ui
+            params["ui_tolerance"] = ui * 0.1
+            reasoning["unit_interval"] = f"From dominant frequency {dom_freq:.1f} Hz"
+            confidence = 0.85
+
+        # Histogram bins based on data range and noise
+        snr = chars.get("snr_db", 20)
+        if snr > 30:
+            params["histogram_bins"] = 256
+        elif snr > 20:
+            params["histogram_bins"] = 128
+        else:
+            params["histogram_bins"] = 64
+        reasoning["histogram_bins"] = f"Based on SNR {snr:.0f} dB"
+
+        return TunedParameters(parameters=params, confidence=confidence, reasoning=reasoning)
+
+    def get_pattern_params(self) -> TunedParameters:
+        """Get tuned parameters for pattern analysis.
+
+        Configures minimum pattern length and maximum distance for
+        fuzzy matching based on signal characteristics.
+
+        Returns:
+            TunedParameters with pattern analysis configuration.
+
+        Example:
+            >>> params = tuner.get_pattern_params()
+            >>> print(f"Min pattern length: {params.get('min_length')}")
+            >>> print(f"Max fuzzy distance: {params.get('max_distance')}")
+        """
+        params = {}
+        reasoning = {}
+        confidence = 0.7
+
+        chars = self._characteristics
+        n_samples = chars.get("n_samples", 1000)
+
+        # Min pattern length based on signal characteristics
+        dom_freq = chars.get("dominant_freq")
+        if dom_freq and dom_freq > 0:
+            samples_per_period = self.sample_rate / dom_freq
+            params["min_length"] = max(3, int(samples_per_period / 4))
+            reasoning["min_length"] = (
+                f"Quarter of estimated period ({samples_per_period:.0f} samples)"
+            )
+        else:
+            params["min_length"] = max(3, n_samples // 100)
+            reasoning["min_length"] = "1% of signal length"
+
+        # Max distance for fuzzy matching based on noise
+        noise_ratio = chars.get("noise_floor", 0.1) / max(chars.get("range", 1), 0.001)
+        params["max_distance"] = max(1, int(noise_ratio * 10))
+        reasoning["max_distance"] = f"Based on noise ratio {noise_ratio:.2%}"
+
+        return TunedParameters(parameters=params, confidence=confidence, reasoning=reasoning)
+
+    def get_params_for_domain(self, domain: str) -> TunedParameters:
+        """Get tuned parameters for a specific analysis domain.
+
+        Args:
+            domain: Analysis domain name (spectral, digital, timing, jitter, pattern).
+
+        Returns:
+            TunedParameters for the specified domain.
+
+        Example:
+            >>> params = tuner.get_params_for_domain("spectral")
+            >>> print(params.parameters)
+            {'nfft': 8192, 'window': 'hann', 'overlap': 0.5}
+        """
+        domain_lower = domain.lower()
+
+        if "spectral" in domain_lower or "fft" in domain_lower:
+            return self.get_spectral_params()
+        elif "digital" in domain_lower:
+            return self.get_digital_params()
+        elif "timing" in domain_lower:
+            return self.get_timing_params()
+        elif "jitter" in domain_lower:
+            return self.get_jitter_params()
+        elif "pattern" in domain_lower:
+            return self.get_pattern_params()
+        else:
+            # Return basic params for unknown domains
+            return TunedParameters(
+                parameters={},
+                confidence=0.5,
+                reasoning={"note": "No domain-specific tuning available"},
+            )
+
+
+def get_adaptive_parameters(
+    data: NDArray[np.floating[Any]],
+    sample_rate: float,
+    domain: str,
+    signal_type: str | None = None,
+) -> TunedParameters:
+    """Convenience function to get adaptive parameters.
+
+    This is a shortcut for creating an AdaptiveParameterTuner and
+    getting parameters for a specific domain.
+
+    Args:
+        data: Input signal data.
+        sample_rate: Sample rate in Hz.
+        domain: Analysis domain (spectral, digital, timing, jitter, pattern).
+        signal_type: Optional signal type hint.
+
+    Returns:
+        TunedParameters for the specified domain.
+
+    Example:
+        >>> params = get_adaptive_parameters(signal, 1e6, "spectral")
+        >>> print(f"Window: {params.get('window')}")
+        >>> print(f"Confidence: {params.confidence}")
+    """
+    tuner = AdaptiveParameterTuner(data, sample_rate, signal_type)
+    return tuner.get_params_for_domain(domain)
+
+
+__all__ = [
+    "AdaptiveParameterTuner",
+    "TunedParameters",
+    "get_adaptive_parameters",
+]