Use a hierarchical buffering scheme: keep the front‑end sampling low‑latency, but push the heavier processing to a local co‑processor that can be idled when idle. Keep the wireless link in a low‑power, short‑burst mode and switch to a direct wired link for the 2 ms budget. Reduce the bit‑depth to 12‑bit if possible and apply a simple delta‑encoding to cut bandwidth. That should shave the 2 ms window without blowing the power envelope.