Updates

analysis/waveform.py

@@ -0,0 +1,401 @@
+"""D-PHY timing extraction and Lane 0 packet decode from scope waveforms.
+
+All voltage thresholds in this module are POST-attenuation values (i.e. what
+the scope sees after the 19.2× probe divider). Don't rescale them back to
+wire voltages — the divider is calibrated and the thresholds were chosen
+to give clean LP/HS state separation at probe output.
+"""
+
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass
+from typing import Optional
+
+import numpy as np
+import pandas as pd
+
+from config import DPHY_SPEC
+
+log = logging.getLogger(__name__)
+
+# Post-attenuation thresholds (volts at scope input, after 19.2× divider).
+LP_HIGH_V = 0.040       # "above" → LP-1   (~770 mV on wire)
+LP_LOW_V = 0.010        # "below" → LP-0 / HS-0  (~190 mV on wire)
+HS_DIFF_V = 0.008       # |CLK_P − CLK_N| above this means HS burst is active
+
+
+@dataclass
+class LaneStateSpan:
+    """A contiguous run of single-ended-detected lane state."""
+    state: str          # "LP-11" | "LP-01" | "LP-10" | "LP-00" | "HS"
+    t_start: float
+    t_end: float
+
+    @property
+    def duration_ns(self) -> float:
+        return (self.t_end - self.t_start) * 1e9
+
+
+# ---------------------------------------------------------------------------
+# Signal reconstruction
+# ---------------------------------------------------------------------------
+
+def differential(lane_p: pd.DataFrame, lane_n: pd.DataFrame) -> pd.Series:
+    return pd.Series(lane_p["voltage_v"].values - lane_n["voltage_v"].values)
+
+
+def common_mode(lane_p: pd.DataFrame, lane_n: pd.DataFrame) -> pd.Series:
+    return pd.Series((lane_p["voltage_v"].values + lane_n["voltage_v"].values) / 2.0)
+
+
+# ---------------------------------------------------------------------------
+# Lane state machine
+# ---------------------------------------------------------------------------
+
+def _classify_sample(vp: float, vn: float, vdiff: float) -> str:
+    """Classify a single (p, n) sample into a D-PHY lane state."""
+    if abs(vdiff) > HS_DIFF_V and vp < LP_HIGH_V and vn < LP_HIGH_V:
+        return "HS"
+    p_high = vp > LP_HIGH_V
+    n_high = vn > LP_HIGH_V
+    p_low = vp < LP_LOW_V
+    n_low = vn < LP_LOW_V
+    if p_high and n_high:
+        return "LP-11"
+    if p_low and n_high:
+        return "LP-01"
+    if p_high and n_low:
+        return "LP-10"
+    if p_low and n_low:
+        return "LP-00"
+    return "TRANS"  # in-between, not yet a settled state
+
+
+def classify_lane(lane_p: pd.DataFrame, lane_n: pd.DataFrame) -> list[LaneStateSpan]:
+    """Walk both single-ended traces and emit consecutive state spans.
+
+    Spans labelled "TRANS" are dropped — they are sub-sample edge transitions,
+    not real D-PHY states. Adjacent same-state spans are merged.
+    """
+    t = lane_p["time_s"].values
+    vp = lane_p["voltage_v"].values
+    vn = lane_n["voltage_v"].values
+    vd = vp - vn
+
+    spans: list[LaneStateSpan] = []
+    cur_state: Optional[str] = None
+    cur_start = t[0]
+
+    for i in range(len(t)):
+        s = _classify_sample(vp[i], vn[i], vd[i])
+        if s == "TRANS":
+            continue
+        if cur_state is None:
+            cur_state = s
+            cur_start = t[i]
+            continue
+        if s != cur_state:
+            spans.append(LaneStateSpan(cur_state, cur_start, t[i]))
+            cur_state = s
+            cur_start = t[i]
+
+    if cur_state is not None:
+        spans.append(LaneStateSpan(cur_state, cur_start, t[-1]))
+
+    return spans
+
+
+def _first_span(spans: list[LaneStateSpan], state: str,
+                start_idx: int = 0) -> Optional[tuple[int, LaneStateSpan]]:
+    for i in range(start_idx, len(spans)):
+        if spans[i].state == state:
+            return i, spans[i]
+    return None
+
+
+# ---------------------------------------------------------------------------
+# Per-parameter measurements
+# ---------------------------------------------------------------------------
+# Each function returns nanoseconds, or NaN if the relevant state span is not
+# present in the capture window.
+
+def measure_t_lpx(data_lane_p: pd.DataFrame, data_lane_n: pd.DataFrame) -> float:
+    """Duration of LP-01 (Dp low, Dn high) on data lane — HS Request."""
+    spans = classify_lane(data_lane_p, data_lane_n)
+    hit = _first_span(spans, "LP-01")
+    return hit[1].duration_ns if hit else float("nan")
+
+
+def measure_t_hs_prepare(data_lane_p: pd.DataFrame, data_lane_n: pd.DataFrame) -> float:
+    """Duration of LP-00 on data lane immediately before HS-0 entry."""
+    spans = classify_lane(data_lane_p, data_lane_n)
+    for i in range(len(spans) - 1):
+        if spans[i].state == "LP-00" and spans[i + 1].state == "HS":
+            return spans[i].duration_ns
+    return float("nan")
+
+
+def measure_t_clk_prepare(clk_p: pd.DataFrame, clk_n: pd.DataFrame) -> float:
+    """Duration of LP-00 on clock lane immediately before HS clock starts."""
+    spans = classify_lane(clk_p, clk_n)
+    for i in range(len(spans) - 1):
+        if spans[i].state == "LP-00" and spans[i + 1].state == "HS":
+            return spans[i].duration_ns
+    return float("nan")
+
+
+def measure_t_clk_zero(clk_p: pd.DataFrame, clk_n: pd.DataFrame) -> float:
+    """Duration of HS-0 on clock lane before first clock toggle.
+
+    Implementation: find the LP-00 → HS transition, then walk the differential
+    until the first edge crossing in the opposite polarity (clock toggle).
+    """
+    t = clk_p["time_s"].values
+    vd = clk_p["voltage_v"].values - clk_n["voltage_v"].values
+
+    spans = classify_lane(clk_p, clk_n)
+    hs_start: Optional[float] = None
+    for i in range(len(spans) - 1):
+        if spans[i].state == "LP-00" and spans[i + 1].state == "HS":
+            hs_start = spans[i + 1].t_start
+            break
+    if hs_start is None:
+        return float("nan")
+
+    start_idx = int(np.searchsorted(t, hs_start))
+    initial = vd[start_idx]
+    sign = -1 if initial >= 0 else 1   # look for opposite-polarity crossing
+    for j in range(start_idx + 1, len(vd)):
+        if (sign > 0 and vd[j] > HS_DIFF_V) or (sign < 0 and vd[j] < -HS_DIFF_V):
+            return (t[j] - hs_start) * 1e9
+    return float("nan")
+
+
+def measure_t_clk_prepare_plus_zero(clk_p: pd.DataFrame, clk_n: pd.DataFrame) -> float:
+    a = measure_t_clk_prepare(clk_p, clk_n)
+    b = measure_t_clk_zero(clk_p, clk_n)
+    if np.isnan(a) or np.isnan(b):
+        return float("nan")
+    return a + b
+
+
+def measure_t_hs_zero(data_lane_p: pd.DataFrame, data_lane_n: pd.DataFrame) -> float:
+    """HS-0 preamble on data lane before SoT sync byte (00011101 = 0xB8 LSB-first).
+
+    Approximated as duration from HS entry until first differential transition
+    (i.e. first clock-edge-aligned bit flip).
+    """
+    t = data_lane_p["time_s"].values
+    vd = data_lane_p["voltage_v"].values - data_lane_n["voltage_v"].values
+
+    spans = classify_lane(data_lane_p, data_lane_n)
+    hs_start: Optional[float] = None
+    for i in range(len(spans) - 1):
+        if spans[i].state == "LP-00" and spans[i + 1].state == "HS":
+            hs_start = spans[i + 1].t_start
+            break
+    if hs_start is None:
+        return float("nan")
+
+    start_idx = int(np.searchsorted(t, hs_start))
+    initial = vd[start_idx]
+    sign = -1 if initial >= 0 else 1
+    for j in range(start_idx + 1, len(vd)):
+        if (sign > 0 and vd[j] > HS_DIFF_V) or (sign < 0 and vd[j] < -HS_DIFF_V):
+            return (t[j] - hs_start) * 1e9
+    return float("nan")
+
+
+# ---------------------------------------------------------------------------
+# Aggregate measurement + spec compliance
+# ---------------------------------------------------------------------------
+
+def measure_all(waveforms: dict[str, pd.DataFrame]) -> dict[str, float]:
+    clk_p = waveforms["CLK_P"]
+    clk_n = waveforms["CLK_N"]
+    dat_p = waveforms["DAT0_P"]
+    dat_n = waveforms["DAT0_N"]
+    return {
+        "t_lpx": measure_t_lpx(dat_p, dat_n),
+        "t_hs_prepare": measure_t_hs_prepare(dat_p, dat_n),
+        "t_clk_prepare": measure_t_clk_prepare(clk_p, clk_n),
+        "t_clk_zero": measure_t_clk_zero(clk_p, clk_n),
+        "t_clk_prepare_plus_zero": measure_t_clk_prepare_plus_zero(clk_p, clk_n),
+        "t_hs_zero": measure_t_hs_zero(dat_p, dat_n),
+    }
+
+
+def check_spec_compliance(measurements: dict[str, float],
+                          spec: dict[str, float] = DPHY_SPEC) -> dict:
+    out: dict[str, dict] = {}
+    for name, measured_ns in measurements.items():
+        min_ns = spec.get(name)
+        if min_ns is None:
+            continue
+        if measured_ns is None or np.isnan(measured_ns):
+            out[name] = {
+                "measured_ns": None,
+                "min_ns": min_ns,
+                "pass": False,
+                "margin_ns": None,
+            }
+            continue
+        out[name] = {
+            "measured_ns": float(measured_ns),
+            "min_ns": float(min_ns),
+            "pass": bool(measured_ns >= min_ns),
+            "margin_ns": float(measured_ns - min_ns),
+        }
+    return out
+
+
+# ---------------------------------------------------------------------------
+# Lane 0 DSI packet decode
+# ---------------------------------------------------------------------------
+# Ground-truth fault detector (Falcon prior art, May 2024). The SN65 IRQ
+# register is a hint — packet payload position is the verdict.
+
+DSI_SOT_SYNC = 0xB8     # SoT sync byte after LP-11 → LP-01 → LP-00 → HS-0
+DSI_DT_PIXEL = 0x3E     # Packed Pixel Stream, 24-bit RGB (long packet)
+DSI_DT_HSYNC_START = 0x21
+
+
+@dataclass
+class DSIPacket:
+    burst_idx: int
+    timestamp_s: float
+    data_type: int
+    word_count: int
+    ecc: int
+    payload: bytes
+
+
+def _find_hs_bursts(clk_p: pd.DataFrame, clk_n: pd.DataFrame,
+                    dat_p: pd.DataFrame, dat_n: pd.DataFrame) -> list[tuple[float, float]]:
+    """Return (t_start, t_end) for each HS burst on the data lane."""
+    spans = classify_lane(dat_p, dat_n)
+    return [(s.t_start, s.t_end) for s in spans if s.state == "HS"]
+
+
+def _sample_bits_in_burst(clk_p: pd.DataFrame, clk_n: pd.DataFrame,
+                          dat_p: pd.DataFrame, dat_n: pd.DataFrame,
+                          t_start: float, t_end: float) -> list[int]:
+    """DDR-sample the data lane at every clock edge inside the burst window.
+
+    Returns a list of 0/1 bit values, in clock-edge order.
+    """
+    t_clk = clk_p["time_s"].values
+    vd_clk = clk_p["voltage_v"].values - clk_n["voltage_v"].values
+    t_dat = dat_p["time_s"].values
+    vd_dat = dat_p["voltage_v"].values - dat_n["voltage_v"].values
+
+    i0 = int(np.searchsorted(t_clk, t_start))
+    i1 = int(np.searchsorted(t_clk, t_end))
+    if i1 - i0 < 2:
+        return []
+
+    edges: list[float] = []
+    prev_sign = 1 if vd_clk[i0] >= 0 else -1
+    for k in range(i0 + 1, i1):
+        cur_sign = 1 if vd_clk[k] >= 0 else -1
+        if cur_sign != prev_sign:
+            edges.append(t_clk[k])
+            prev_sign = cur_sign
+
+    bits: list[int] = []
+    for et in edges:
+        idx = int(np.searchsorted(t_dat, et))
+        if 0 <= idx < len(vd_dat):
+            bits.append(1 if vd_dat[idx] > 0 else 0)
+    return bits
+
+
+def _bits_to_bytes_msb_first(bits: list[int]) -> bytes:
+    out = bytearray()
+    for i in range(0, len(bits) - 7, 8):
+        b = 0
+        for k in range(8):
+            b = (b << 1) | (bits[i + k] & 1)
+        out.append(b)
+    return bytes(out)
+
+
+def decode_lane0_packets(waveforms: dict[str, pd.DataFrame],
+                         max_payload_bytes: int = 16) -> list[DSIPacket]:
+    """Best-effort DSI Lane 0 packet decode.
+
+    Scope window at 5 ns/div × 500 kpts is ~2.5 µs — enough for SoT + header
+    + first ~200 bytes of payload. We only need the first few payload bytes
+    to classify Fault A (all-zero payload start).
+    """
+    clk_p = waveforms["CLK_P"]
+    clk_n = waveforms["CLK_N"]
+    dat_p = waveforms["DAT0_P"]
+    dat_n = waveforms["DAT0_N"]
+
+    bursts = _find_hs_bursts(clk_p, clk_n, dat_p, dat_n)
+    packets: list[DSIPacket] = []
+
+    for idx, (t0, t1) in enumerate(bursts):
+        bits = _sample_bits_in_burst(clk_p, clk_n, dat_p, dat_n, t0, t1)
+        bs = _bits_to_bytes_msb_first(bits)
+
+        sot_pos = bs.find(bytes([DSI_SOT_SYNC]))
+        if sot_pos < 0 or len(bs) < sot_pos + 5:
+            continue
+
+        header = bs[sot_pos + 1 : sot_pos + 5]
+        data_type = header[0]
+        word_count = header[1] | (header[2] << 8)
+        ecc = header[3]
+
+        payload_start = sot_pos + 5
+        payload_end = min(payload_start + max_payload_bytes, len(bs))
+        payload = bs[payload_start:payload_end]
+
+        packets.append(DSIPacket(
+            burst_idx=idx,
+            timestamp_s=t0,
+            data_type=data_type,
+            word_count=word_count,
+            ecc=ecc,
+            payload=payload,
+        ))
+
+    return packets
+
+
+def classify_packet_fault(packets: list[DSIPacket]) -> dict:
+    """Classify Fault A (zero-payload pixel packet) from decoded packets."""
+    pixel_packets = [p for p in packets if p.data_type == DSI_DT_PIXEL]
+    if not pixel_packets:
+        return {"fault_a_detected": False, "reason": "no pixel packets decoded"}
+
+    first = pixel_packets[0]
+    head = first.payload[:8] if first.payload else b""
+    fault_a = len(head) >= 4 and all(b == 0x00 for b in head[:4])
+
+    return {
+        "fault_a_detected": bool(fault_a),
+        "first_pixel_payload_hex": head.hex(),
+        "n_pixel_packets": len(pixel_packets),
+        "n_total_packets": len(packets),
+    }
+
+
+def detect_lane_stall(data_lane_p: pd.DataFrame, data_lane_n: pd.DataFrame,
+                      stall_threshold_ms: float = 10.0) -> dict:
+    """Fault B: continuous LP-11 longer than threshold during what should be active video."""
+    spans = classify_lane(data_lane_p, data_lane_n)
+    longest_lp11_ms = 0.0
+    for s in spans:
+        if s.state == "LP-11":
+            ms = s.duration_ns / 1e6
+            if ms > longest_lp11_ms:
+                longest_lp11_ms = ms
+    return {
+        "fault_b_detected": bool(longest_lp11_ms > stall_threshold_ms),
+        "longest_lp11_ms": float(longest_lp11_ms),
+        "threshold_ms": float(stall_threshold_ms),
+    }