3GPP TSG-RAN WG1 Meeting #59 Jeju, Korea, 9th 13th November 2009 Agenda item: Source: Title:

: Document for: 7.3.1 Nokia, Nokia Siemens Networks Inter-cell CSI-RS design and performance Discussion and Decision

R1-094648

1 Introduction
In RAN1#58bis and in the e-mail discussion following the meeting, the CSI-RS considerations have been so far mostly focusing on the intra-cell case. In , we are highlighting our views on intra-cell CSI-RS design aspects, also already touching the issue of possible extension of the design to inter-cell case by allowing data blanking in neighbour cells to reduce inter-cell interference on CSI-RS. In , we have presented preliminary results on the benefits of data blanking for multi-cell channel estimation. The results there suggested that in order to have benefits from data blanking, the reuse factor of CSI-RS should be designed to be clearly larger than the maximum size of the CoMP measurement set. Moreover, it was concluded that inter-cell multiplexing scheme for CSI-RS should be FDM or CDM or a combination of them. In this contribution, we provide further results on the impact of data blanking, considering coordinated beamforming throughput performance. We also again discuss the possibility of extending the intra-cell CSI-RS design to allow data blanking properly.

2 CSI-RS design aspects

Let us start by summarizing our views - on the CSI-RS design aspects that were also discussed in the CSI-RS e-mail discussion prior to RAN1#59: -Placement of the CSI-RS should avoid REs used for Rel9-Rel10 DM-RS as well as OFDM symbols used for Rel8 CRS. If needed, one could further consider using the third PDCCH symbol taking into account the reduced PDSCH capacity in the CSI-RS subframe. This might be needed mainly to cover the extended CP case where smaller number of symbols is available. -Required overhead per antenna port in intra-cell case is 1.5 REs 2 REs per PRB based on results shown in and also in numerous earlier contributions. This does not yet factor in the PDSCH blanking overhead, nor is it considering the fact that in inter-cell case the increased path losses are causing the received CSI-RS power to be relatively weak for neighbour cells. However, the power boosting impact due to data blanking helps channel estimation in inter-cell case and hence 1.5 REs 2 REs per PRB per port seem to be enough also in that case, as we will show in this contribution. -Taking into account both CSI-RS overhead and blanking overhead, the total overhead per PRB should be no more than 32 REs. In we have shown that the impact of 32 punctured REs per PRB on Rel8 UEs is not much worse than that of 16 REs. Moreover, distributing the puncturing impact over time, hence allowing inter-cell CSI-RS to be TDM:ed with lower densities per PRB, provides worse performance impact on Rel8 UEs than when sending inter-cell CSI-RS in one subframe . -Considering the previous paragraph, CSI-RS from all cells should be transmitted in one subframe within the CSI-RS period. Furthermore, to properly enable frequency-selective feedback and precoding, the CSI-RS spacing in frequency should be small enough, e.g. multiple PRBs are clearly too much, see e.g. our results for intra-cell case and spacing of 12 REs . -In we concluded that best choices for inter-cell multiplexing of the CSI-RS are FDM, CDM and a combination of them. TDM was deemed not to be efficient in terms of power handling, hence it was ruled out. Also the TDM solution suffers from the frequency drift issues and if done over multiple subframes, causes worse legacy impact . CDM suffers from the loss of orthogonality and resulting near-far effect, however the

impact of that was shown to be fairly minor. However, in we have shown that the performance of CDM for intra-cell case suffers badly compared to FDM. Hence, since we are seeking for a unified design for both intra- and inter-cell cases, FDM would be our main preference for CSI-RS multiplexing. However, accommodating 32 REs in one PRB requires a CDM component to be included; hence in our pattern 5 (see Appendix 1) also a CDM component is utilized.

2.1 Introduction of inter-cell CSI-RS

As mentioned, so far the discussion on CSI-RS has focused on the intra-cell case. We note that allowing data blanking from other cells seems beneficial for proper multi-cell channel estimation for CoMP purpose, see our simulation results later. Introducing data blanking later than in the first release that supports CSI-RS (likely Rel10) would be extremely difficult since it would imply similar legacy issues as we are currently facing with CSI-RS design and Rel8 and Rel9 UEs. Hence, if supported, one should consider introducing the data blanking already in the first release supporting CSIRS to allow scalable and forward compatible design to all releases, including those beyond Rel10 that may support enhanced multi-cell features. Note that this should be the case even if only simple CoMP schemes would be supported in the first release.

2.2 Inter-cell CSI-RS patterns

Based on the design aspects listed in the beginning of this section, we have designed five patterns for further simulation studies, see Appendix 1 for the figures. Note that these are the same patterns as studied in for intra-cell case: -Pattern 1: 16 REs per PRB overhead, 2 REs / port / PRB, reuse factor 4 for 2TX, FDM -Pattern 2: 16 REs per PRB overhead, 1 RE / port / PRB, reuse factor 8 for 2TX, FDM -Pattern 3: 24 REs per PRB overhead, 2 REs / port / PRB, reuse factor 6 for 2 TX, FDM -Pattern 4: 24 REs per PRB overhead, 1.5 REs / port / PRB, reuse factor 8 for 2 TX, FDM -Pattern 5: 32 REs per PRB overhead, 2 REs / port / PRB, reuse factor 8 for 2 TX, FDM+CDM The extension from intra-cell case to inter-cell case is done simply by proper allocation of the CSI-RS ports to different cells, and blanking the PDSCH transmission on REs corresponding to CSI-RS transmission from neighboring cells. In order to allow a good reuse factor through blanking, OFDM symbols not containing other Rel9/Rel10 DM-RS should be used for CSI-RS port mapping. Since we are using FDM, the left-over power available from blanked REs can be used to boost the power of the CSI-RS. The same antenna ports are still used as in the intra-cell case, which allows simpler channel estimator implementation from UE point of view. Since the inter-cell CSI-RS are mainly needed for CoMP purpose, if data blanking is supported, one should consider allowing the possibility of switching data blanking on/off depending on whether CoMP is used in the cell or not.

Handling different TX antenna configurations We want to point out that in our examples shown in Appendix 1 we have 2 TX antennas per cell. With reuse factor of four that implies eight orthogonal antenna ports in total, and with reuse factor of eight that implies 16 orthogonal antenna ports. Obviously, once the number of orthogonal antenna ports has been fixed, the reuse factor completely depends on how many antenna ports there are per cell. As an example we have tabulated the reuse factor with different total number of orthogonal CSI-RS ports per antenna configuration, see Table 1. Here we have assumed same antenna configuration in each cell, but also different configurations could be allowed, however it requires different signalling of the antenna ports to the UEs. Table 1. Reuse factors with different antenna configurations and different total numbers of orthogonal antenna ports.
2 TX per cell 4 TX per cell 8 TX per cell 8 orthogonal antenna ports 4 2 1 16 orthogonal antenna ports 8 4 2

As already pointed out, some signalling will be required to inform UE about the exact antenna configurations behind the CSI-RS ports. Note that the number of orthogonal antenna ports should be decoupled from the number of ports that the UE needs to measure. As will be shown in our simulations, a reuse factor of four is not enough to benefit from data blanking. With reuse 8 case one would need a total of 16 CSI-RS ports (see the patterns in Appendix 1), however the UE would be only measuring a subset of them.

3 Inter-cell CSI-RS simulations

The CSI-RS patterns listed in Appendix 1 were simulated in the context of coordinated beamforming (CBF), and also benchmarked against Rel8 CRS, with and without averaging over multiple subframes. We also benchmarked CSI-RS with data blanking to CSI-RS without data blanking.

3.1 Simulation setup

Simulations were run on a multi-cell link simulator that explicitly models the radio links between the UE and each base station. Chosen scenario was 3GPP Case 1. CSI-RS reuse pattern was planned and optimized separately for reuse factors four, six and eight. UEs were dropped in the network such that a sector in the center site of the network is the best quality cell. The CoMP measurement set and also the cooperating set was assumed to be the set of best cells within a 10 dB path loss window. Still the maximum number of cells in the CoMP measurement set was limited to four, i.e. the size of our CoMP measurement set was 2-4 cells. Based on the measured channel, UE was reporting the best companion PMI for the neighboring cells (minimizing interference) and the best rank-1 PMI for the serving cell, i.e. we assumed rank-1 closed-loop SU-MIMO transmission from serving cell. For the neighboring cells, we assumed that the recommended companion PMI was taken into use, i.e. the related scheduling restrictions were not modelled. Feedback delay was modelled explicitly, as well as demodulation with the LTE-Advanced DM-RS. Link adaptation was used to set the MCS, and the cooperation (CBF) was taken into account in the CQI calculation. More detailed simulation parameters are listed in Table 2. We measured channel estimation MSE and precoder selection error rate as in as well as the resulting user throughput. Table 2. Simulation assumptions. Parameter description Transmission bandwidth Transmit antenna configuration Receive antenna configuration Channel model, UE velocity, spatial correlation Number of cells within CoMP measurement set Path loss window for CoMP measurement set PDCCH length Channel estimation for PMI/CQI computation Channel estimation for demodulation Detector Channel coding Modulation, code rates HARQ Precoding Value / Comment 10 MHz 2 Tx per cell 2 Rx ETU 3 km/h, spatially uncorrelated 2 4, depending on how many cells fall within the path loss window 10 dB 2 symbols 2D realistic channel estimation on CSI-RS or Rel8 CRS 2D realistic channel estimation on LTE-Advanced DM-RS MMSE receiver Rel8 turbo coding, CBRM Link adaptation over Rel8 MCS IR used 2-bit i.i.d. codebook

Feedback Transmission rank

1 PRB granularity, 5 ms delay Rank 1 always

3.2 Results
The resulting average throughput for CoMP UEs (roughly ~50% of the UEs) is shown in Figure 1. For comparison we have plotted also the single-cell performance for the same UEs, i.e. the case of rank-1 transmission and no coordination of the PMIs. Furthermore, we have also plotted the performance of pattern 4 (FDM 24 REs, reuse 8) without data blanking, as well as the CBF performance when Rel8 CRS are used for CSI estimation. The best CSI-RS patterns seem to be those with 24 REs, both reuse factor 6 and reuse factor 8 perform equally well. Especially it is to be noted that the 32 RE pattern requires a CDM component which seems to perform worse than the 24 RE pure FDM patterns. What can be immediately concluded is also that the reuse factor indeed has to be designed large enough so that one benefits from CSI-RS and data blanking, e.g. reuse factor four is not enough as it is clearly visible. This is basically confirming our conclusion from . Based on the results, there is 12.1% gain over Rel8 CRS when there is no interpolation of the channel estimate over multiple subframes, and 8.4% gain when channel estimate is obtained by interpolating over two consecutive subframes. In our view CRS are configured independently from CSI-RS hence what needs to be pointed out is that when CSI-RS are configured to the cell, one will most likely want to reduce the number of Rel8 CRS ports and it may not be possible to measure CSI from them. In that case one would have to rely on CSI-RS. Our result shows also that data blanking of CSI-RS is indeed beneficial if one wants to use CSI-RS for the CoMP CSI measurements there is 18.8% gain in that case. In summary, the conclusions from this simulation can be listed as follows: -Data blanking is beneficial if CSI-RS are used for CoMP CSI measurements. -Large reuse factor needs to be enabled if one wants to really benefit from data blanking. -Total of 24 REs seems to be enough to cover even larger reuse factor in case of 2 TX antennas per cell.

Figure 1. Average throughput for the CoMP UEs with each CSI-RS pattern. For comparison, we have also plotted the same result without data blanking. As reference, we have plotted the single-cell performance with rank-1 closed-loop SU-MIMO for the same UEs (no coordination). It is acknowledged that coordinated beamforming is clearly not a CoMP scheme that requires the best channel knowledge. True MU techniques such as joint processing might require even better CSI knowledge at eNB, hence

similarly to we have again plotted also the channel estimation MSE for each cell in the CoMP measurement set as well as the precoder selection error rate, shown in Figure 2 and Figure 3 respectively. These are showing that actually without data blanking the CSI-RS based precoder selection is almost random due to very bad CSI (high MSE). This explains the bad performance of coordinated beamforming, and this impact can be assumed to be more severe with any multi-user techniques such as joint processing.

Figure 2. Channel estimation MSE for each cell for the CoMP UEs. For comparison we have again plotted the CRS cases with and without averaging as well as the case of CSI-RS without data blanking.

Figure 3. Precoder selection error rate for each cell for the CoMP UEs. For comparison we have again plotted the CRS cases with and without averaging as well as the case of CSI-RS without data blanking. Finally, Figure 4 shows the throughput CDF of CoMP UEs for coordinated beamforming using CSI-RS, data blanking and pattern 4 as well as the same throughput CDF for the case of using CRS for CSI estimation. Also the case of no

coordination (single-cell rank-1 transmission) is shown. Clear gains from CSI-RS together with data blanking are observed.

Figure 4. Normalized user throughput CDF for the CoMP UEs using CBF with CRS or 24 REs CSI-RS, as well as for the same UEs without coordination.

4 Conclusions
We have discussed the design and performance of multi-cell CSI-RS, especially considering the possibility to extend intra-cell CSI-RS to support data blanking and hence CoMP features properly. Key observations from our previous contributions on the topic have been emphasized, i.e. that CSI-RS should be confined in one subframe and have dense enough spacing in frequency. Also 1.5 REs or 2 REs per PRB per port seems enough for the overhead and FDM seems to be the best multiplexing scheme. Maximum overhead considering both data blanking and CSI-RS should be kept below 32 REs per PRB; based on results in this contribution, 24 REs would be a good number if data blanking is supported. We have shown simulation results on data blanking performance with coordinated beamforming setup. The conclusions are as follows: -Data blanking is beneficial if CSI-RS are used for CoMP CSI measurements. -Large reuse factor is needed if one wants to really benefit from data blanking. Especially, this means that 16 REs per PRB are not enough if one also factors in the blanked REs. -Total of 24 REs brings a good compromise, providing large reuse factor in case of 2 TX antennas per cell. We note that CQI measurements should be considered as well. Single-cell CQI can be measured if CSI-RS have frequency shifts and data blanking is off, which can be accomplished easily with the shown patterns and the measured CQI will reflect the actual PDSCH CQI accurately enough. However, data blanking would make this more complicated. Anyway, the real problem is in CoMP CQI measurements where neither CSI-RS with data blanking nor without data blanking will reflect CoMP CQI correctly. We propose to continue the studies on data blanking, considering feedback accuracy, time of CoMP introduction, and CQI issues. Our results would already hint that from feedback accuracy perspective data blanking is needed.

References
[1] R1-094647, Intra-cell CSI-RS design aspects, Nokia, Nokia Siemens Networks [2] R1-093909, Multi-cell CSI-RS design aspects, Nokia, Nokia Siemens Networks

[3] R1-094649, Multi-cell CSI-RS transmission and related impact to LTE Rel8, Nokia, Nokia Siemens Networks

Appendix 1 CSI-RS patterns

The considered CSI-RS patterns are listed here. Note that exactly the same patterns have been studied for intra-cell case in . Pattern 1: 16 REs per PRB overhead, 2 REs / port / PRB, reuse factor 4 for 2TX, FDM
CoMP, cell#0
R1 R0 R1 R3 R5 R7

Single-cell 8TX
R0 R2 R4 R6

R0

R1

R0 R2 R4 R6

R1 R3 R5 R7

Figure 5. Pattern 1 with 16 REs per PRB, 2 REs/port/PRB, FDM multiplexing and reuse factor of four, both for inter-cell case with 2TX antennas per cell and for the 8TX single-cell case. Note that one could also use OFDM symbol #10 instead of symbol #4 for CSI-RS in case of normal CP.

Pattern 2: 16 REs per PRB overhead, 1 RE / port / PRB, reuse factor 8 for 2TX, FDM

CoMP, cell#0
R1 R0 R1 R3 R5 R7

Single-cell 8TX
R0 R2 R4 R6

Figure 6. Pattern 2 with 16 REs per PRB, 1 RE/port/PRB, FDM multiplexing and reuse factor of eight, both for inter-cell case with 2TX antennas per cell and for the 8TX single-cell case. Note that one could also use OFDM symbol #10 instead of symbol #4 for CSI-RS in case of normal CP.

Pattern 3: 24 REs per PRB overhead, 2 REs / port / PRB, reuse factor 6 for 2 TX, FDM

CoMP, cell #0
R1 R0

Single-cell 8TX
R1 R0 R3 R2 R5 R4 R7 R6

R0 R1

R0 R1 R3 R2 R5 R4 R7 R6

Figure 7. Pattern 3 with 24 REs per PRB, 2 REs/port/PRB, FDM multiplexing and reuse factor of six, both for inter-cell case with 2TX antennas per cell and for the 8TX single-cell case.

Pattern 4: 24 REs per PRB overhead, 1.5 REs / port / PRB, reuse factor 8 for 2 TX, FDM

CoMP, cell #0
R1 R0

Single-cell 8TX
R1 R0 R3 R2 R5 R4 R7 R6

R0 R1

R0 R1 R3 R2 R5 R4 R7 R6

R1 R0

R1 R0 R3 R2 R5 R4 R7 R6

Figure 8. Pattern 4 with 24 REs per PRB, 1.5 REs/port/PRB, FDM multiplexing and reuse factor of eight, both for inter-cell case with 2TX antennas per cell and for the 8TX single-cell case. Note that to preserve PA balance, one has to flip the two symbols for the next two PRBs. This should be fine since CSI-RS are transmitted over full band.

Pattern 5: 32 REs per PRB overhead, 2 REs / port / PRB, reuse factor 8 for 2 TX, FDM+CDM

CoMP, cell#0

Single-cell 8TX

R0 R0

R0 R0

R1 R1

R1 R1

R03 R03

R03 R03

R47 R47

R47 R47

R1 R1

R1 R1

R0 R0

R0 R0

R47 R47

R47 R47

R03 R03

R03 R03

Figure 9. Pattern 5 with 32 REs per PRB, 2 REs/port/PRB (in power equivalent), FDM+CDM multiplexing and reuse factor of eight, both for inter-cell case with 2TX antennas per cell and for the 8TX single-cell case. Note that we are utilizing CDM with code length 4 here. In CoMP case the 2 TX ports of one cell are FDM multiplexed here.