diff --git a/rfg_adc_plotter/processing/normalization.py b/rfg_adc_plotter/processing/normalization.py
index 688fee2..6066541 100644
--- a/rfg_adc_plotter/processing/normalization.py
+++ b/rfg_adc_plotter/processing/normalization.py
@@ -20,7 +20,7 @@ def normalize_sweep_simple(raw: np.ndarray, calib: np.ndarray) -> np.ndarray:
 
 
 def build_calib_envelopes(calib: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
-    """Estimate lower and upper envelopes of a calibration curve."""
+    """Estimate smooth lower/upper envelopes from local extrema."""
     n = int(calib.size)
     if n <= 0:
         empty = np.zeros((0,), dtype=np.float32)
@@ -40,32 +40,48 @@ def build_calib_envelopes(calib: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
     if n < 3:
         return values.copy(), values.copy()
 
-    dy = np.diff(values)
-    signs = np.sign(dy).astype(np.int8, copy=False)
-
-    if np.any(signs == 0):
-        for i in range(1, signs.size):
-            if signs[i] == 0:
-                signs[i] = signs[i - 1]
-        for i in range(signs.size - 2, -1, -1):
-            if signs[i] == 0:
-                signs[i] = signs[i + 1]
-        signs[signs == 0] = 1
-
-    max_idx = np.where((signs[:-1] > 0) & (signs[1:] < 0))[0] + 1
-    min_idx = np.where((signs[:-1] < 0) & (signs[1:] > 0))[0] + 1
-
     x = np.arange(n, dtype=np.float32)
 
-    def _interp_nodes(nodes: np.ndarray) -> np.ndarray:
-        if nodes.size == 0:
-            idx = np.array([0, n - 1], dtype=np.int64)
-        else:
-            idx = np.unique(np.concatenate(([0], nodes, [n - 1]))).astype(np.int64)
-        return np.interp(x, idx.astype(np.float32), values[idx]).astype(np.float32)
+    def _moving_average(series: np.ndarray, window: int) -> np.ndarray:
+        width = max(1, int(window))
+        if width <= 1 or series.size <= 2:
+            return np.asarray(series, dtype=np.float32).copy()
+        if width % 2 == 0:
+            width += 1
+        pad = width // 2
+        padded = np.pad(np.asarray(series, dtype=np.float32), (pad, pad), mode="edge")
+        kernel = np.full((width,), 1.0 / float(width), dtype=np.float32)
+        return np.convolve(padded, kernel, mode="valid").astype(np.float32)
 
-    upper = _interp_nodes(max_idx)
-    lower = _interp_nodes(min_idx)
+    def _smooth_extrema_envelope(use_max: bool) -> np.ndarray:
+        step = max(3, n // 32)
+        node_idx_list = []
+        for start in range(0, n, step):
+            stop = min(n, start + step)
+            segment = values[start:stop]
+            idx_rel = int(np.argmax(segment) if use_max else np.argmin(segment))
+            node_idx_list.append(start + idx_rel)
+
+        extrema_idx = np.unique(np.asarray(node_idx_list, dtype=np.int64))
+        if extrema_idx.size == 0:
+            extrema_idx = np.asarray([int(np.argmax(values) if use_max else np.argmin(values))], dtype=np.int64)
+
+        node_idx = np.unique(np.concatenate(([0], extrema_idx, [n - 1]))).astype(np.int64)
+        node_vals = values[node_idx].astype(np.float32, copy=True)
+        node_vals[0] = float(values[extrema_idx[0]])
+        node_vals[-1] = float(values[extrema_idx[-1]])
+        node_vals = _moving_average(node_vals, 3)
+        node_vals[0] = float(values[extrema_idx[0]])
+        node_vals[-1] = float(values[extrema_idx[-1]])
+
+        envelope = np.interp(x, node_idx.astype(np.float32), node_vals).astype(np.float32)
+        smooth_window = max(1, n // 64)
+        if smooth_window > 1:
+            envelope = _moving_average(envelope, smooth_window)
+        return envelope
+
+    upper = _smooth_extrema_envelope(use_max=True)
+    lower = _smooth_extrema_envelope(use_max=False)
 
     swap = lower > upper
     if np.any(swap):
diff --git a/tests/test_processing.py b/tests/test_processing.py
index d45f4f8..dab5d61 100644
--- a/tests/test_processing.py
+++ b/tests/test_processing.py
@@ -44,6 +44,11 @@ class ProcessingTests(unittest.TestCase):
         self.assertEqual(lower.shape, calib.shape)
         self.assertEqual(upper.shape, calib.shape)
         self.assertTrue(np.all(lower <= upper))
+        self.assertTrue(np.all(np.isfinite(upper)))
+        self.assertLess(
+            float(np.mean(np.abs(np.diff(upper, n=2)))),
+            float(np.mean(np.abs(np.diff(calib, n=2)))),
+        )
 
         simple = normalize_by_calib(raw, calib + 10.0, norm_type="simple")
         projector = normalize_by_calib(raw, calib, norm_type="projector")