Why decode blindly?
There is no header that says "UE 42, your DCI is here". Instead the DCI's CRC is masked with the RNTI. The UE tries each candidate location and DCI size, unmasks with its own RNTI, and checks the CRC. Pass = the message was for it. This keeps control overhead tiny — but the UE must bound how many attempts it makes.
RNTI-masked CRC
The 24-bit CRC is XORed with the RNTI. Only the intended UE's unmasking yields a valid CRC.
Candidates × sizes
Each search space lists candidates per aggregation level; the UE tries each for every monitored DCI size.
Hard budgets
Per slot the UE can only run so many decodes and estimate so many CCEs — capped by sub-carrier spacing.
Blind decoding in action
Each tile is a (candidate, size) attempt. The UE sweeps them, descrambling by C-RNTI and checking the CRC. Most fail; one — if scheduled — passes. Press play.
Decode & CCE budgets
Two per-slot ceilings (TS 38.213 Tables 10.1-2/3), both set by sub-carrier spacing: the maximum monitored PDCCH candidates (blind decodes) and the maximum CCEs the UE will channel-estimate. Add candidates and watch both.