1. Assignment AMAP Design Document Number: Slides S80216m-09/0953 Date Submitted: 2009-4-27 Source: Yi Hsuan, Hujun Yin Email: {yi.hsuan@intel.com, hujun.yin@intel.com}yi.hsuan@intel.comhujun.yin@intel.com Intel Corporation Venue: IEEE Session #61, Cairo, Egypt. Re: AWD comments/ Area: Chapter 15.3.6 (DL-CTRL), “Comments on AWD 15.3.6 DL-CTRL” Base Contribution: N/A Purpose: For TGm discussion and adoption of 802.16m AWD text. Notice: This document does not represent the agreed views of the IEEE 802.16 Working Group or any of its subgroups. It represents only the views of the participants listed in the “Source(s)” field above. It is offered as a basis for discussion. It is not binding on the contributor(s), who reserve(s) the right to add, amend or withdraw material contained herein. Release: The contributor grants a free, irrevocable license to the IEEE to incorporate material contained in this contribution, and any modifications thereof, in the creation of an IEEE Standards publication; to copyright in the IEEE’s name any IEEE Standards publication even though it may include portions of this contribution; and at the IEEE’s sole discretion to permit others to reproduce in whole or in part the resulting IEEE Standards publication. The contributor also acknowledges and accepts that this contribution may be made public by IEEE 802.16. Patent Policy: The contributor is familiar with the IEEE-SA Patent Policy and Procedures: and.http://standards.ieee.org/guides/bylaws/sect6-7.html#6http://standards.ieee.org/guides/opman/sect6.html#6.3 Further information is located at and.http://standards.ieee.org/board/pat/pat-material.htmlhttp://standards.ieee.org/board/pat

2. AMAP Region Location In the DL subframes where the A-MAP regions can be allocated, each frequency partition may contain an A-MAP region. An A-MAP region, if present, shall occupy the first few distributed LRUs in a frequency partition. Figure: Example on the Location of A-MAP regions in a TDD system with a 4:4 subframe DL:UL split

3. AMAP Region Structure

4. Assignment AMAP Design Decisions in TGm Various AMAPs are separately coded for different users. MLRU is the minimum resource unit for assignment AMAP. MLRU is formed from distributed LRUs in the time first manner. The size of MLRU (48 tones) is subject to change. Two MCS levels (including repetitions) can be used for assignment AMAPs. The exact MCSs are not decided yet. –Our analysis shows QPSK ½ with no repetition and repetition 4 are good choices. TBCC is used to code assignment AMAP Non-user specific information indicates the size of assignment AMAP in the unit of one or multiple MLRU. CRC in Assignment AMAP IE is masked by station ID. –Blind detection using station ID is needed.

5. Remaining Design Issues MLRU numbers per assignment AMAP –Current view is to use 1 MLRU per AMAP for basic AMAP IE and 2 MLRUs per AMAP for extended AMAP IE (group assignment, persistent scheduling, etc.) Assignment AMAP structure –Organization of assignment AMAPs with different MCSs and MLRUs so that decoding complexity, false detection probability, signaling overhead, and resource waste are balanced. Non-user specific information and coding –A tradeoff of blind detection complexity and signaling overhead. –It’s related to assignment AMAP structure.

6. Concerns About Blind Detection Detection complexity –Because MS doesn’t know the location, the MCS, and/or the size of AMAP IE of its assignment AMAP, it needs to perform blind detection in a search space. The detection complexity depends on the size of search space. False positive probability –If a search candidate passes CRC check after decoding, it’s considered a valid assignment AMAP for the MS. If decoding error happens, it’s possible that CRC still passes (false positive). The probability needs to be minimized especially for uplink persistent scheduling. False detection in noiseless scenario –If two or more AMAP IE sizes need to be searched over the same physical resource, false detection could happen even in noiseless and perfect channel condition. This can create testing complexity and cause MS consistently missing its AMAP.

7. Proposed Assignment AMAP Structure

8. Properties of the Proposed Structure Signaling overhead: non-user specific AMAP needs to indicate the number of channels of each MLRU group. Each MLRU is decoded once only. The number of detections required, given M MLRUs, is no more than M. For the example on the previous page, M=16 and the number of detection is only 5. No resource is wasted if the exact number of channels in each group is signaled. No false detection in noiseless case if only one MAP IE size needs to be detected in each MLRU group. –The MAP IE design target is to have only one MAP IE size for 1 MLRU and one MAP IE size for 2 MLRUs.

9. Signaling Overhead For 10 MHz system bandwidth and AMAP region in every DL subframe, assume that 17 MLRUs are used for assignment AMAPs. To allow full flexibility to signal the size of each group, non-user specific AMAP should have 5, 4, 3, and 2 bits for each group respectively. If table lookup is applied, the number of bits is less by just considering possible combinations of the 4 groups of AMAPs. For 17 MLRUs, there are 201 combinations so 8 bits are enough. TBCC can be used to code the 8 bits non-user specific information. We can use an effective code rate of 1/8 like assignment AMAP. Power boosting can be applied to help robustness. The required resource is therefore 1/3 of a 96-tone LRU. Higher code rate is possible for scenarios like frequency reuse 3. SFH can signal the exact code rate of non-user specific AMAP. For 5 Mhz bandwidth, AMAP can be allocated every 2 DL subframes to reduce the signaling overhead. For other bandwidth and AMAP periodicity, 8-bit non-user specific AMAP can still be used by scaling the number of channels each table entry accordingly. For example, for 20MHz band and AMAP region every two subframes, the number of assignment AMAPs in each table entry should be multiplied by 4.

10. More Analysis on Number of Detection Structure 1 (tree structure) requires the largest number of detection trials. With 18.75% of resources for assignment channel (9 out of 48 RU) each subframe, MS needs to do 33 detections for 10 MHz and 67 detections for 20 MHz. A variation of structure 1 is the LTE design which requires a maximal 44 detections every 1ms independent of system bandwidth. However this approach causes significant control channel resource waste. See backup slides for more details. The detection number of Structure 2 depends on the total number of users to be scheduled and therefore depend on the bandwidth and traffic model too. The worst case would be 20 MHz and VoIP users. –Suppose that AMAP uses no more than 22% of total resource. Each 20MHz subframe has about 21-1(non-user specific)-2(Ack/Nack, PC)=18 LRUs or 36 MLRUs for assignment AMAP. For VoIP, most users should be scheduled through extended AMAP like group or persistent scheduling, so many assignment AMAP should be in 2MLRU group or 8 MLRU group. –Suppose 1MLRU group and 2 MLRU group each occupies 18 MLRUs. The total detection is 18+9=27 per subframe or 0.625ms. Compared to LTE, it’s 27/0.625=43 detections/ms, still less than 44.

11. False Positive Detection False positive detection happens when CRC check passes by mistake for a detection trial. This is caused by certain decoding errors such that error bits in the AMAP IE and CRC are aligned to allow CRC check to pass. Such decoding errors can be caused by –Channel impairment or RF impairment. This factor cannot be avoided in any blind detection scheme. –Rate matching or repetition mismatch. This cause of false positive can be eliminated in structure 2 if each MLRU group has only 1 AMAP IE size. AMAP false positive detection can interrupt both DL an UL traffic, especially for persistent scheduling. LTE has done significant work to analyze the false activation frequency of semi-persistent scheduling (SPS). [Ref: R1-082766, R1-082280, R1- 082297, R1-082542, R1-083757, etc.] The following analysis is based on the methodology used in LTE.

12. LTE False Positive Probability VoIP-only MS can on average decode a false SPS UL activation every 5 * 2 16 *2*2/(22*1000)= 60 second. –The factor of 5 is for DRX operation, meaning UE wakes up only 1 out 5 subframes. –2 16 is from 16-bit CRC. –2 is from the bit in IE to distinguish UL and DL. –2 is from voice activity factor of ½. –22 is the decoding trials for UL DCI format. –1000 is the number of subframes per second. Some bits in the control IE have known values for SPS. These bits can be used as “virtual CRC” to reduce false detection –9 or 10 bits in format 0 can be used as virtual CRC. The resulting false SPS activation duration becomes 60s * 2 [9-10] /3600=8.5 to 17.1 hours.

13. False Positive Probability of the Proposal Assume the worst case scenarios: 20 MHz bandwidth, 18 LRU or AMAP, all users are scheduled with persistent scheduling (extended IE used), no repetition is applied (more detection trials needed). VoIP-only MS can on average decode a false SPS UL activation every 5 * 2 16 *2*2/(18*1600)= 45.5 second. –The factor of 5 is for DRX operation, meaning MS wakes up only 1 out 5 frames. –2 16 is from 16-bit CRC. –2 is from the bit in IE to distinguish UL and DL. –2 is from voice activity factor of ½. –18 is the decoding trials for assignment AMAP. –1600 is the number of subframes per second.

14. Comparison of the Proposed Structure and LTE False Positive Interval From the previous slide, false positive interval of the proposal is a little worse than LTE (45s vs 60s) in an extreme case. However the false positive interval of the proposal can become much larger for the following reasons. LTE’s number doesn’t change with these factors. –Smaller bandwidth. –Repetition used in some assignment AMAPs. –Less resources used for control because of less users. –Existence of some basic AMAP IEs in the AMAP region so detection trials for extended AMAP IE are less. For instance, if 2 AMAPs are transmitted with repetition 4 and 10 are transmitted with no repetition, there are 12 detections per subframe and the false positive interval becomes 68.25s. 16m AMAP IE design needs to maximize the number of bits for virtual CRC, which can increase the false positive interval significantly.

15. Backup

16. LTE Blind Search Algorithm Given a CID, subframe index, and aggregation level, there is a particular search space start Z k L, whereis the total number of CCE (or MLRU in 16m terms). For each aggregation level, MS Rx tries to decode a fixed number of continuous candidates (6, 6, 6, and 4 respectively) starting from Z k L. Therefore the number of detection candidates are fixed to 22, given a DCI format (or MAP IE types in 16m terms).

17. LTE Blind Detection Candidates Search spaceNumber of PDCCH candidates DCI formats TypeAggregation level LSize [in CCEs] UE- specific 1 6 6 0, 1, 1A,1B, 2 2 12 6 4 8 2 8 16 2 Common 4 16 4 0, 1A, 1C, 3/3A 8 16 2

18. LTE DL Control Allocation

19. Cause of Resource Waste Random search space generation –As shown in the previous slide, the search space of each user and each MCS level is randomly generated based on RNTI. Therefore unused control resource may not be used by any user if their search spaces don’t overlap with the unused resources. TDM of control and data –Because LTE uses TDM to separate control and data, the resource granularity for control region is very rough. The DL control region can be 1, 2, or 3 symbols out of 14 symbols. For example, if the required control resource is just a little more than 2 symbols, 3 symbols have to be allocated and most of the resource in symbol 3 is wasted.

20. LTE DL Control Channels vs. Total Resources Simulation Setup: Assume 15 CCEs are for resource allocation in each 10 MHz symbol. Assume 15 CCEs are for resource allocation in each 10 MHz symbol. Assume each PDCCH takes 2 CCEs and has a different n RNTI. Assume each PDCCH takes 2 CCEs and has a different n RNTI. 8 to 15 PDCCHs are considered. 8 to 15 PDCCHs are considered.Observation Although 8 to 15 PDCCHs should all fit in 30 CCEs (2 symbols), 3 symbols are needed sometimes and cause resource waste. Although 8 to 15 PDCCHs should all fit in 30 CCEs (2 symbols), 3 symbols are needed sometimes and cause resource waste.

21. Summary Detection complexity: –LTE blind detection has fixed detection complexity for MS given a DCI format (6+6+2+2+4+2=22 trials). –MS needs to detect two DCI formats at the same time so 44 trials are needed for each 1 ms subframe. Resource waste: –LTE blind detection yields significant amount of resource waste. From a simulated case, it needs more than 40% of the minimum resource required.