In the PDCCH region in DL radio frame, there can be many places where a specific PDCCH is located and UE searches all the possible locations. The possible location for a PDCCH differs depending on whether the PDCCH is UE-Specific or Common, and also depend on what aggregation level is used.  All the possible location for PDCCH is called 'Search Space and each of the possible location is called 'PDCCH Candidates'.

The UE is never told where its PDCCH is. The eNB sends no pointer to it, so the UE has to try each candidate position and check the CRC with its RNTI. A search space keeps that work bounded: it limits the positions a UE must try in each subframe to a small, known set.

Followings are the topics to be covered in this page.

• Common and UE-specific Search Space
• Where the Candidates Sit - Hashing Function
• Blind Decoding Attempts
• Reference

Common and UE-specific Search Space

Why are there two kinds of search space? Some DCIs are meant for every UE in the cell, and others for one UE only. The two kinds of search space follow that split, and they differ in where they start and in which aggregation levels they use.

The search space indicates the set of CCE locations where the UE may find its PDCCHs. Each PDCCH carries one DCI and is identified by RNTI. The RNTI is implicitly encoded in the CRC attachment of the DCI.

There are two types of search space : the common search space and the UE-specific search space. A UE is required to monitor both common and UE-specific search space. There might be overlap between common & UE-specific search spaces for a UE

• The common search space would carry the DCIs that are common for all UEs. For example, system information (using the SI-RNTI), paging (P-RNTI), PRACH responses (RA-RNTI), or UL TPC commands (TPC-PUCCH/PUSCH-RNTI). The UE monitors the common search space using aggregation level 4 and 8. Maximum number of CCEs present in common search space is 16.

• The UE-specific search space can carry DCIs for UE-specific allocations using the UE's assigned C-RNTI, semi-persistent scheduling (SPS C-RNTI),or initial allocation (temporary C-RNTI). The UE monitors the UE-specific search space at all aggregation levels (1, 2, 4, and 8).

The two spaces also carry different DCI formats. The common search space carries formats 0, 1A, 3 and 3A, which share one payload size, and format 1C. The UE-specific search space carries formats 0 and 1A and the format of the configured transmission mode, such as 2A or 2C. The DCI formats that the UE monitors therefore depend on the transmission mode, as 36.213 v19.4.0 clause 9.1.1 states with reference to clause 7.1.

A table from 36.213 shows these relationship as below.

36.213 Table 9.1.1-1. The size of each search space is the aggregation level times the number of candidates, for example 2 x 6 = 12 CCEs at level 2.

The table gives 16 candidates in the UE-specific search space and 6 in the common search space. The UE-specific space has more candidates at levels 1 and 2, because a UE in good radio conditions needs only a few CCEs. The common space starts at level 4, because its DCIs must reach every UE, including those at the cell edge. Table 9.1.1-1 is still the same in 36.213 v19.4.0. Later releases add Table 9.1.1-1A for LAA SCells and a candidate reduction factor, pdcch-candidateReductions, which scales the number of candidates.

• Common search space : levels 4 and 8, 6 candidates, CCE 0 to 15.
• UE-specific search space : levels 1, 2, 4 and 8, 16 candidates.
• Size of a search space : aggregation level times the number of candidates.

Where the Candidates Sit - Hashing Function

The table says how many candidates a UE has, but not where they are. The position of the common search space is fixed, while the position of the UE-specific search space moves with the RNTI and the subframe. This section shows how 36.213 clause 9.1.1 places each candidate.

For candidate m at aggregation level L, the UE reads the CCEs L {(Y_k + m) mod floor(N_CCE,k / L)} + i, with i = 0 to L - 1. Here N_CCE,k is the number of CCEs in the control region of subframe k. For the common search space, Y_k = 0, so the common candidates always start at CCE 0. That is why the common search space never goes beyond CCE 15.

For the UE-specific search space, Y_k = (A x Y_k-1) mod D, with Y_-1 = n_RNTI, A = 39827 and D = 65537. The index k is floor(n_s/2), the subframe number, where n_s is the slot number. So each UE gets its own sequence of start positions, and the sequence changes from subframe to subframe. Two UEs that collide in one subframe therefore do not collide in every subframe.

The table below works through one example. The cell uses 10 MHz, 2 CRS ports, CFI 2 and Ng = 1. Two symbols give 100 + 150 = 250 REGs, the PCFICH takes 4, and 7 PHICH groups take 21. That leaves 225 REGs, which are N_CCE,k = 25 CCEs. The UE has C-RNTI 100.

Here Y_k is 50480 in subframe 0 and 53948 in subframe 1. At level 1, the candidates move from CCE 5 to 10 to CCE 23 to 3, and they wrap around at CCE 25. At levels 2, 4 and 8 they stay in place, only because both values give the same remainder modulo 12, 6 and 3. The example also shows the overlap that the text above mentions: at level 4, the UE-specific candidates at CCE 8 and 12 are the same CCEs as two common candidates.

The mod floor(N_CCE,k / L) term also explains a limit of small carriers. At level 8, this cell has only 3 positions, so the 2 UE-specific candidates cover 16 of the 25 CCEs. When several UEs need level 8 at the same time, their candidates can block each other, and the eNB then has to delay one of them. The PDCCH page describes how the CCEs themselves are built.

• Common search space : Y_k = 0, always from CCE 0.
• UE-specific search space : start moves with the RNTI and the subframe.
• Y_k = (39827 x Y_k-1) mod 65537 : Y_-1 = n_RNTI.
• Candidates wrap at floor(N_CCE,k / L) : few positions at level 8 on small carriers.

Blind Decoding Attempts

How much work does the UE do in each subframe? The UE decodes each candidate once for each DCI payload size it monitors there. The count of these blind decoding attempts follows directly from Table 9.1.1-1 and the DCI formats of each search space.

The common search space has 6 candidates and two payload sizes: one for formats 0, 1A, 3 and 3A, and one for format 1C. That gives 12 attempts. The UE-specific search space has 16 candidates and two sizes, formats 0 and 1A and the transmission-mode format, which gives 32 attempts. A UE therefore makes up to 44 blind decoding attempts per subframe on one carrier.

The number grows with later features. A UE configured for uplink spatial multiplexing also monitors DCI format 4 in the UE-specific space, which adds 16 attempts for 60 in total. With carrier aggregation and cross-carrier scheduling, the UE monitors one UE-specific search space for each scheduled cell, so the count grows with the number of cells. EPDCCH moves the UE-specific search space out of the control region in the subframes where it is configured, as clause 9.1.1 describes.

Each attempt ends in a CRC check, and that check has a cost. 36.212 clause 5.3.3.2 attaches a 16-bit CRC to the DCI and scrambles it with the RNTI. A random candidate therefore passes the check with a probability of about 1 in 65536. With 44 attempts in every subframe, a UE can occasionally accept a DCI that was never sent to it. The small search space keeps this false detection rate low, and the UE also discards a DCI whose fields are not valid.

• Common search space : 6 candidates x 2 sizes = 12 attempts.
• UE-specific search space : 16 candidates x 2 sizes = 32 attempts.
• 44 attempts per subframe : 60 with DCI format 4.

Reference

[1] 3GPP TS 36.213 v19.4.0 - clause 9.1.1, PDCCH assignment procedure, and Table 9.1.1-1

[2] 3GPP TS 36.211 v19.3.0 - clause 6.8, Physical downlink control channel

[3] 3GPP TS 36.212 v19.3.0 - clause 5.3.3.2, CRC attachment for DCI