diff --git a/bellows/ash.py b/bellows/ash.py
index 3c2a4b4f..66c0e516 100644
--- a/bellows/ash.py
+++ b/bellows/ash.py
@@ -3,11 +3,13 @@
 import abc
 import asyncio
 import binascii
+from collections.abc import Coroutine
 import dataclasses
 import enum
 import logging
 import sys
 import time
+import typing
 
 if sys.version_info[:2] < (3, 11):
     from async_timeout import timeout as asyncio_timeout  # pragma: no cover
@@ -55,7 +57,7 @@ class Reserved(enum.IntEnum):
 
 # Maximum number of DATA frames the NCP can transmit without having received
 # acknowledgements
-TX_K = 1
+TX_K = 1  # TODO: investigate why this cannot be raised without causing a firmware crash
 
 # Maximum number of consecutive timeouts allowed while waiting to receive an ACK before
 # going to the FAILED state. The value 0 prevents the NCP from entering the error state
@@ -81,6 +83,23 @@ def generate_random_sequence(length: int) -> bytes:
 # Since the sequence is static for every frame, we only need to generate it once
 PSEUDO_RANDOM_DATA_SEQUENCE = generate_random_sequence(256)
 
+if sys.version_info[:2] < (3, 12):
+    create_eager_task = asyncio.create_task
+else:
+    _T = typing.TypeVar("T")
+
+    def create_eager_task(
+        coro: Coroutine[typing.Any, typing.Any, _T],
+        *,
+        name: str | None = None,
+        loop: asyncio.AbstractEventLoop | None = None,
+    ) -> asyncio.Task[_T]:
+        """Create a task from a coroutine and schedule it to run immediately."""
+        if loop is None:
+            loop = asyncio.get_running_loop()
+
+        return asyncio.Task(coro, loop=loop, name=name, eager_start=True)
+
 
 class NcpState(enum.Enum):
     CONNECTED = "connected"
@@ -463,15 +482,14 @@ def data_received(self, data: bytes) -> None:
     def _handle_ack(self, frame: DataFrame | AckFrame) -> None:
         # Note that ackNum is the number of the next frame the receiver expects and it
         # is one greater than the last frame received.
-        ack_num = (frame.ack_num - 1) % 8
+        for ack_num_offset in range(-TX_K, 0):
+            ack_num = (frame.ack_num + ack_num_offset) % 8
+            fut = self._pending_data_frames.get(ack_num)
 
-        fut = self._pending_data_frames.get(ack_num)
+            if fut is None or fut.done():
+                continue
 
-        if fut is None or fut.done():
-            return
-
-        # _LOGGER.debug("Resolving frame %d", ack_num)
-        self._pending_data_frames[ack_num].set_result(True)
+            self._pending_data_frames[ack_num].set_result(True)
 
     def frame_received(self, frame: AshFrame) -> None:
         _LOGGER.debug("Received frame %r", frame)
@@ -537,13 +555,16 @@ def error_frame_received(self, frame: ErrorFrame) -> None:
         self._ncp_state = NcpState.FAILED
 
         # Cancel all pending requests
-        exc = NcpFailure(code=self._ncp_reset_code)
+        self._enter_failed_state(self._ncp_reset_code)
+
+    def _enter_failed_state(self, reset_code: t.NcpResetCode) -> None:
+        exc = NcpFailure(code=reset_code)
 
         for fut in self._pending_data_frames.values():
             if not fut.done():
                 fut.set_exception(exc)
 
-        self._ezsp_protocol.reset_received(frame.reset_code)
+        self._ezsp_protocol.reset_received(reset_code)
 
     def _write_frame(
         self,
@@ -582,7 +603,7 @@ async def _send_data_frame(self, frame: AshFrame) -> None:
                 for attempt in range(ACK_TIMEOUTS):
                     if self._ncp_state == NcpState.FAILED:
                         _LOGGER.debug(
-                            "NCP is in a failed state, not re-sending: %r", frame
+                            "NCP is in a failed state, not sending: %r", frame
                         )
                         raise NcpFailure(
                             t.NcpResetCode.ERROR_EXCEEDED_MAXIMUM_ACK_TIMEOUT_COUNT
@@ -618,6 +639,9 @@ async def _send_data_frame(self, frame: AshFrame) -> None:
                         self._change_ack_timeout((7 / 8) * self._t_rx_ack + 0.5 * delta)
 
                         if attempt >= ACK_TIMEOUTS - 1:
+                            self._enter_failed_state(
+                                t.NcpResetCode.ERROR_EXCEEDED_MAXIMUM_ACK_TIMEOUT_COUNT
+                            )
                             raise
                     except NcpFailure:
                         _LOGGER.debug(
@@ -635,6 +659,9 @@ async def _send_data_frame(self, frame: AshFrame) -> None:
                         self._change_ack_timeout(2 * self._t_rx_ack)
 
                         if attempt >= ACK_TIMEOUTS - 1:
+                            self._enter_failed_state(
+                                t.NcpResetCode.ERROR_EXCEEDED_MAXIMUM_ACK_TIMEOUT_COUNT
+                            )
                             raise
                     else:
                         # Whenever an acknowledgement is received, t_rx_ack is set to
@@ -649,9 +676,14 @@ async def _send_data_frame(self, frame: AshFrame) -> None:
                     self._pending_data_frames.pop(frm_num)
 
     async def send_data(self, data: bytes) -> None:
-        await self._send_data_frame(
-            # All of the other fields will be set during transmission/retries
-            DataFrame(frm_num=None, re_tx=None, ack_num=None, ezsp_frame=data)
+        # Sending data is a critical operation and cannot really be cancelled
+        await asyncio.shield(
+            create_eager_task(
+                self._send_data_frame(
+                    # All of the other fields will be set during transmission/retries
+                    DataFrame(frm_num=None, re_tx=None, ack_num=None, ezsp_frame=data)
+                )
+            )
         )
 
     def send_reset(self) -> None: