5G New Radio Physical Downlink Control Channel Reliability Enhancements for Multiple Transmission-Reception-Point Communications

Non-coherent transmission from multiple transmission-reception-points (TRPs), i.e., base stations, or base station panels to a user equipment (UE) is exploited in 5G New Radio (NR) to improve downlink reliability and cell-edge throughput. Ultra reliable low-latency communications (URLLC) and enhanced Mobile BroadBand (eMBB) are prominent target use-cases for multi-TRP or multi-panel transmissions. In Third-Generation Partnership Project (3GPP) Release 17 specifications, multi-TRP-based transmissions were specified for the physical downlink control channel (PDCCH) specifically to enhance its reliability and robustness. In this work, a comprehensive account of various multi-TRP reliability enhancement schemes applicable for the 5G NR PDCCH, including the ones supported by the 3GPP Release 17 specifications, is provided. The impact of the specifications for each scheme, UE and network complexity and their utility in various use-cases is studied. Their error performances are evaluated via link-level simulations using the evaluation criteria agreed in the 3GPP proceedings. The 3GPP-supported multi-TRP PDCCH repetition schemes, and the additionally proposed PDCCH repetition and diversity schemes are shown to be effective in improving 5G NR PDCCH reliability and combating link blockage in mmWave scenarios. The link-level simulations also provide insights for the implementation of the decoding schemes for the PDCCH enhancements under different channel conditions. Analysis of the performance, complexity and implementation constraints of the proposed PDCCH transmission schemes indicate their suitability to UEs with reduced-capability or stricter memory constraints and flexible network scheduling.


I. INTRODUCTION
The associate editor coordinating the review of this manuscript and approving it for publication was Bilal Khawaja . Radio (NR) standards starting from 3GPP Release (Rel.) 27 16. A user equipment (UE) may receive physical downlink 28 shared channel (PDSCH) transmission(s) that are multiplexed 29 in space, time or frequency using two different reception 30 settings [1], wherein each reception setting may correspond 31 to a different TRP or base station panel. Independent schedul-32 ing of simultaneous PDSCH receptions from different TRPs 33 within a cell, repetition of a PDSCH transmission from 34 different TRPs and PDSCH diversity -a single PDSCH 35 • PDCCH enhancements are considered for two cate- Frequency Range 2 (FR2) (the frequency ranges are 91 specified by 3GPP in [8]) according to the evaluation 92 methodology agreed by 3GPP RAN WG1 in [9]. 93 • An analysis on the trade-offs involved in each scheme 94 with respect to error performance, network complexity, 95 UE complexity and PDCCH overhead provides insights 96 regarding target use-cases and supported UE/network-97 types, which are elaborated towards the end of the paper. 98 The paper is organized as follows. Section II gives a brief 99 introduction to the 5G NR PDCCH. Section III provides 100 a detailed account of various PDCCH reliability enhance-101 ments for multi-TRP scenarios including the 3GPP-supported 102 transmission schemes. Section IV describes the associated 103 receiver processing. The impacts on the 5G NR specifications 104 and various implementation issues concerning the receiver 105 processing methods are provided in Section V. Section VI 106 presents numerical results on the performance of the PDCCH 107 enhancements and decoding complexity. Section VII con-108 cludes the paper by summarizing the key aspects regarding 109 each multi-TRP-based PDCCH enhancement.

112
The hierarchy of components that constitute the physical 113 downlink control channel configuration are as follows: the 114 control resource set (CORESET), the search space set and 115 the PDCCH candidate. PDCCH transmissions are performed 116 on predefined spaces in time and frequency in the NR radio 117 frame called the Control Resource Set (CORESET) [10]. 118 A carrier component or cell that the UE is configured with 119 may comprise multiple bandwidth parts (BWP). Each BWP 120 in a cell can be configured with one or more CORESETs. 121 Each CORESET is associated with one or more search space 122 sets. A search space set is associated with a CORESET and 123 comprises one or more PDCCH candidates. An individual 124 PDCCH transmission is performed in a PDCCH candidate 125 of a search space set [11], [12]. These components of the 126 PDCCH are explained in detail in this section.    A DCI comprising a payload of K bits is attached with a 205 CRC of C = 24 bits scrambled with the applicable RNTI, 206 as mentioned above. The K + C message bits are polar-207 encoded and rate-matched to E bits that are then modulated to 208 the resource elements corresponding to the DCI [3]. The first 209 step in the encoding involves inserting the K +C message bits 210 in a N = 2 n -bit sequence (row vector) x with N ≥ K + C. 211 The value of n is determined based on the DCI payload size 212 K and the number of rate-matched bits E as described in [3]. 213 The positions of the K + C message bits in the N -length 214 sequence and the rest N −(K +C) 'frozen' bits are determined 215 from the universal reliability sequence provided in 3GPP 216 Technical Specification 38.212 [3]. The sequence x is then 217 applied with the polar code generator G n F n B N , where 218 F n = 1 0 is the number of resource elements in the CCEs associated 226 with the PDCCH that are used for DMRS. Depending on 227 the values of N and E, the rate-matching may be performed 228 using repetition, puncturing or shortening [3]. Interleaving 229 is performed before polar-coding and before rate-matching, 230 but they are left out of Fig. 1 and Fig. 3 for compactness of 231 illustration.

233
The receive-processing for a PDCCH after the collection of 234 the associated CCEs and equalization is shown in Fig. 3. The 235 LLRs of the E transmitted bits obtained from the E/2 equal-236 ized symbols after soft demodulation is fed to the polar 237 decoder after rate dematching. The CRC polynomial is pro-238 vided to the decoder in Fig. 3 considering list-based polar 239 decoding [13].

240
The UE has limited capability for the number of PDCCHs 241 it can decode in a given slot or in a span of symbols which is 242 reported to the network [4], [12]. The network may schedule 243 PDCCH candidates more than the UE's capability to decode, 244 which is called PDCCH overbooking [4]. The 3GPP spec-245 ification instructs the UE to decode the scheduled PDCCH 246 candidates in a slot or span of symbols via an assignment 247 of priority to them. The PDCCH candidates lower in priority 248 and ultimately outside the UE's blind decoding capability are 249 dropped. This understanding is shared by the network due 250 to the UE's reporting of its blind decoding capability. The 251 reported value may depend on the memory available at the 252 UE and waveform numerology, among other parameters.

254
The multi-TRP reliability enhancement schemes for 5G NR Reference Signal (CSI-RS) [1]. The reference RS in each 279 TCI-state may be associated with a different TRP to implicitly 280 configure multi-TRP reception at the UE. No explicit spec-281 ification of configurations or parameters are thus required 282 to identify the TRPs transmitting to the UE. This transmis-283 sion scheme is supported in 3GPP Rel. 17 by enabling the 284 assignment of a CORESET with multiple TCI-states [14]. 285 The reception settings corresponding to all the indicated TCI-286 states are applied by the UE for the demodulation and decod-287 ing of the PDCCHs transmitted on the PDCCH candidates 288 on the CORESET. The PDCCH overhead for SFN-based 289 repetition is identical to the single-TRP transmission as the 290 PDCCH is repeated on the same PDCCH candidate by all the 291 TRPs. However, this scheme poses stringent synchronization 292 requirements among the TRPs as the repetitions should be 293 received on the same resources in time and frequency at 294 the UE. This may be realized only with ideal or near-ideal 295 backhaul across the TRPs. The PDCCH repetitions are multiplexed in time and/or fre-302 quency via multiple TRPs in this scheme. An example is 303 shown in Fig. 5b, where the PDCCHs generated from a 304 given DCI are repeated in full on each of the two different 305 PDCCH candidates P i and P j . By associating the PDCCH 306 candidates with different TCI-states, each corresponding to 307 a different TRP, multi-TRP transmission of the repetitions 308 is enabled. Assigning identical TCI-states to the repetitions 309 leads to single-TRP-based repetition. This scheme does not 310 require stringent time synchronization as in the case of SFN. 311 It can be implemented even with non-ideal backhaul across 312

379
The processing of a DCI at the UE is shown in Fig. 3 for 380 the case of decoding an individual PDCCH candidate. The 381 applicable receiver processing for both SFN-and non-SFN-382 based PDCCH transmission schemes are provided in Table 1. 383 For the SFN-based repetition and the split-PDCCH transmis-384 sion case, the channel estimation on the CCEs corresponding 385 to the PDCCH are performed according to the TCI-states 386 indicated for the CCEs followed by a single PDCCH blind 387 decoding. For non-SFN-based repetition, the combining of 388 the repetitions before blind decoding is an obvious method 389 to achieve SNR or coding gain. Performing multiple PDCCH 390 blind decoding attempts from two or more PDCCH candi-391 dates that carry the same DCI content is another possibility 392 considered in 3GPP and in this work. The various possi-393 ble receiver processing methods for non-SFN-based PDCCH 394 repetition are described in the following.      In the case of non-SFN-based repetition, SD and HD pro-451 vide error performances that are either worse or equal to that 452 of SC as combining with every additional repetition improves 453 the SNR or coding gain [16]. However, the advantages posed 454 by methods involving multiple decoding attempts such as SD 455 and HD are in terms of reducing PDCCH latency, PDCCH 456 scheduling flexibility, memory usage and supporting reduced 457 capability UEs.

458
SC and SD or HD require almost the same amount of 459 memory to store the PDCCH symbols or LLRs when all 460 the PDCCH candidates of a PDCCH repetition are obtained 461 during a single monitoring occasion of the search space 462 set or across few OFDM symbols. However, if two repe-463 titions of the PDCCH are obtained on PDCCH candidates 464 whose monitoring occasions are well separated in time, i.e., 465 separated by multiple symbols within a slot or monitored 466 on different slots altogether, then the LLRs from the first 467 repetition have to be retained on memory until the second 468 repetition for soft-combining the LLRs. This memory cannot 469 be used for decoding of another PDCCH until the arrival 470 of the second linked PDCCH repetition. Since the number 471 of blind decoding attempts per slot or span of symbols is 472 limited, as described above, holding this memory for a given 473 duration implies the dropping of other PDCCHs due to mem-474 ory shortage. SC thus requires higher memory when PDCCH 475 repetitions across monitoring occasions are well separated 476 in time. To increase the reliability of the PDCCH and/or to 477 accommodate UEs that do not have sufficient memory to 478 store the LLRs or symbols of each repetition before combin-479 ing them, e.g., reduced-capability UEs, SD can be considered 480 as an alternative. In addition, SD has the advantage of not 481 waiting on the second PDCCH candidate, and therefore it is  Inter-slot repetition is not supported. if a UE implements only SC to process a pair of linked 537 PDCCH candidates, it performs just one blind decoding 538 attempt after soft-combining the PDCCHs. But, it may report 539 a value of 2 or 3 depending on implementation factors such 540 as the memory buffer for LLR storage, number of PDCCH 541 blind decoding attempts withheld due to decoding a pair of 542 linked candidates, etc. With this additional reporting and a 543 renewed set of priority rules for decoding specified in Rel. 544 17 based on the search space set ID associated with the 545 linked PDCCH candidates, the UE and the network have a 546 common understanding of the PDCCH candidates attempted 547 for decoding in a slot or a span of symbols comprising both 548 legacy PDCCH candidates and linked PDCCH candidates 549 for repetition [12]. This enables backward compatibility with 550 legacy 5G NR PDCCH scheduling.

551
To address the memory usage across spans of symbols for 552 inter-span repetition, the UE is enabled to report the number 553 of PDCCH candidates that can be received in a span of 554 symbols, each of which has a linked PDCCH candidate that 555 is yet to be received (in a future slot or span of symbols). 556 This enables the network to gauge the buffer memory of 557 the UE that is blocked for linked PDCCH candidates spread 558 across different spans of symbols, and thereby limit the num-559 ber of scheduled inter-span PDCCH repetitions to the UE's 560 capability.

562
For both SD and HD, the decoding order of the PDCCH 563 candidates determines how soon a valid DCI is detected, i.e., 564 the number of blind decodes required until a valid DCI is 565 detected, thereby improving the PDCCH decoding latency. 566 Such an ordering makes sense for repetitions within a mon-567 itoring occasion or across a series of a few symbols. When 568 the repetitions are within a PDCCH monitoring occasion, the 569 decoding may be performed in a determined order after the 570 reception of all the PDCCH repetitions using a metric related 571 to the PDCCH candidates used for the repetitions. Doing the 572 same for inter-slot or inter-span PDCCH repetition, however, 573 may result in significantly higher decoding latency. Two met-574 rics that can be considered for the decoding criterion are as 575 follows: The pros and cons of choosing a given metric and the scenar-600 ios in which they are advantageous are discussed in the next 601 section.

603
The simulation parameters used for the numerical evaluations 604 are based on the criteria agreed in 3GPP for the evaluation 605 of multi-TRP PDCCH enhancements [9], and are summa-606 rized in Table 2. For FR1, an urban macro cell scenario is  For the non-SFN-based PDCCH repetition scheme, the 646 DCI is transmitted from two different TRPs on two differ-647 ent CORESETs. The CCE-to-REG mapping for both TRPs 648 is interleaved with the interleaving parameters shown in 649 Table 2. When the PDCCH repetitions are obtained on two 650 different PDCCH candidates of the same AL, the transmit-651 processing parameters for the PDCCH repetitions are iden-652 tical. For PDCCH repetitions with different ALs, the CRC-653 attached DCI is rate-matched for a baseline PDCCH with 654 AL L and a PDCCH of AL 2L or L/2 for transmission 655 from a first TRP and second TRP, respectively. The param-656 eters of the encoding and rate-matching schemes are deter-657 mined as described in [3]. Note that in the case of PDCCH 658 repetitions with different ALs, for the DCI size chosen in 659 Table 2, the values of N determined according to the 5G 660 NR specifications [3]

667
In the case of single-TRP transmission, the PDCCH is 668 transmitted by a single TRP. For split-PDCCH transmission, 669 the first half of the PDCCH's CCEs is transmitted on a first 670 PDCCH candidate from the first TRP and the remaining 671 CCEs on a second PDCCH candidate from the second TRP. 672 SFN-based repetition is realized by the superposition of the 673 PDCCHs from two TRPs.

735
The results for the FR2 scenario using the CDL channel 736 parameters given in Table. 2 are provided in Fig. 8    The performance comparisons between SFN and non-767 SFN-based repetition with identical AL values, with respect 768 to the BLERs in various SNR ranges and the crossing of the 769 BLER curves for different ALs, are similar to that of the FR1 770 scenario.

771
In both FR1 and FR2 scenarios, SFN and non-SFN-based 772 repetition with AL L for both PDCCHs performs better than 773 single-TRP transmission with AL 2L. In addition, any multi-774 TRP based repetition for a baseline PDCCH of AL L per-775 forms better than the transmission of the PDCCH with AL 776 2L from a single TRP in FR2. This means single-TRP-based 777 PDCCH transmissions with a lower coding rate perform 778 worse than PDCCH transmissions from multiple TRPs with 779 higher coding rates. These results lead to the following con-780 clusions. Multi-TRP PDCCH transmissions can be exploited 781 in handover or cell-edge procedures in both FR1 and FR2 782 to improve the SNR. With the prominence of link blockage 783 in FR2, diversifying PDCCH transmission via repetition or 784 split via multiple TRPs is important for the robustness of the 785 transmission.

786
SFN-based repetition provides a good trade-off between 787 error performance and PDCCH overhead, UE complexity and 788 PDCCH latency compared to the other methods, by lever-789 aging network complexity. Tight frame synchronization and 790 near-ideal backhaul across TRPs, at the very least, are crucial 791 for PDCCH repetition with identical scheduling from the 792 TRPs. Minor changes regarding TCI-state application are 793 required at the UE, while existing PDCCH processing rules 794 including overbooking and blind decoding capabilities can be 795 reused for SFN-based repetitions. There is only a marginal 796 increase in PDCCH channel estimation complexity at the UE. 797 The channel estimation has to be performed with respect 798 to the TCI-states of all the involved TRPs instead of just 799 one TCI-state as in the single-TRP case. The impact of the 800 SFN-based repetitions on the 3GPP specifications is also 801 minimal as all the burden is shifted to the network imple-802 mentation. From the perspective of UE implementation, SFN-803 based repetition is better suited among the discussed methods 804 to improve PDCCH reliability.

805
The split-PDCCH scheme provides considerable gains 806 in FR2 which demonstrates the effectiveness of multi-TRP 807 diversity transmissions in the presence of blockage. There is 808 a slight increase in UE complexity compared to SFN-based 809 repetition in terms of processing the CCEs associated with 810 a PDCCH, but it incurs the same overhead as the baseline 811 single-TRP-based PDCCH scheme. When PDCCH split is 812 performed within a slot, span or monitoring occasion, the 813 UE's blind decoding capability and PDCCH overbooking can 814 be reused from existing 3GPP specifications which simplifies 815 UE implementation. With such a PDCCH split, the improved 816 PDCCH robustness is traded for higher network complex-817 ity and a marginal increase in PDCCH latency and/or UE 818 complexity.

819
For non-SFN-based repetitions, the UE incurs higher com-820 plexity than the other PDCCH enhancements, while the 821 tions. The gains are more pronounced than the split-PDCCH   is decoded first followed by the other PDCCH candidate. 877 'RSRP' denotes that the candidate with the higher DMRS-878 RSRP is chosen to decode first.

879
For PDCCH repetition with the same AL L as the baseline 880 PDCCH, decoding using RSRP leads to consistently lower 881 average number of decoding attempts than sequential decod-882 ing for both FR1 and FR2 scenarios (the curve for sequential 883 decoding coincides with the cyan and yellow curves). In the 884 case of repetition with a PDCCH of AL L/2, either the same 885 or a higher number of decoding attempts on average are 886 required compared to other methods in the FR1 scenario. 887 PDCCH repetition with AL 2L performs the best in FR1 when 888 decoding is started with the PDCCH candidate of higher AL 889 or RSRP.

890
A stark contrast is observed between the FR1 and FR2 891 scenarios for the PDCCH repetition methods with AL L/2 892 and 2L. In FR1, choosing to decode the candidate with higher 893 VOLUME 10, 2022