Impact of Flexible RLC PDU Size on HSUPA Performance

Transcription

1 Nash Technologies Your partner for world-class custom software solutions & consulting Enrico Jugl, Michael Link, Jens Mueckenheim* *Hochschule Merseburg, Germany

2 Outline Motivation Flexible RLC PDU Size Feature Packet Data Performance - Single User Performance - Multi-User Performance VoIP Performance - Transmission Technique - Performance Criteria - Simulation Results Conclusions Slide 2

3 Motivation (1) 3GPP introduced an enhanced layer 2 (i.e. flexible RLC PDU size) for the downlink in Release 7 allowing for a more efficient transmission of higher data rates - Required for evolution of HSDPA, e.g. 64 QAM, MIMO and dual-cell HSDPA In Release 8 a similar layer 2 enhancement was added for the uplink by introduction of a MAC-i/is entity handling flexible RLC PDU sizes - Allows for higher data rates given by advanced E-DCH features like 16 QAM and dual-cell HSUPA Maximum achievable RLC throughput: R RLC max RWIN ( N = RTT RLC PDU N + TSP - RWIN: RLC window size in number of RLC PDUs - N RLC PDU : size of the RLC PDU (e.g. 336 bits, 656 bits) - N RLC header : size of the AM RLC PDU header (16 bits) - RTT: round trip time - TSP: timer status prohibit RLC header ), Slide 3

4 Motivation (2) With RWIN = 2047, RTT = 70 ms, TSP = 50 ms - R RLC max = 5.4 Mbps for 336 bits PDU size - R RLC max =10.8 Mbps for 656 bits PDU size By increasing RLC PDU size the maximum RLC data rate can be increased - Problematic at cell edge if the UE is in power limitation, where a large PDU cannot be transmitted at all or with insufficient power only Enhanced layer 2 can alleviate this tradeoff - Large PDUs can be used if allowed by radio conditions - Small PDUs can be used in power limited situations If large PDUs are used - RLC overhead is reduced, as well as the padding in the MAC-i PDUs - Transmission of less PDUs in a TTI allows for reduction of processing load in the terminals and the network equipment Slide 4

5 Flexible RLC PDU Size Feature Example for a single logical channel: TCP/IP header RLC PDU Max RLC PDU TCP/IP Payload MTU: 576 or 1500 RLC SDU RLC PDU Flexible size RLC provides segmentation/concatenation of variable sized RLC SDUs (IP packets) into RLC/MAC-d PDUs E.g. a RLC SDU contains an IP packet of 1500 bytes (MTU=1500) The maximum RLC PDU size is 1505 octets (configurable) The length of the data field is a multiple of 8 bits MAC-i header H RLC PDU RLC PDU Pad. RLC PDU size can vary according to the MAC-is PDU H: MAC-is header amount of data requested by current E-TFCI selection Cf Rel-8 Slide 5

6 Simulation Scenario Parameter # of NodeB (sites)/ sectors Pathloss model Cell radius Shadow fading UL receive diversity Channel Model Mobility UL Target Load Service 1 Service 2 Value Single user: 1 sector Multi-user: 12 sites/ 3 sectors each (wrapped around) COST 231 Okumura Hata urban 1000 m Single cell: no Multi-cell: 7dB standard dev, 50 m correlation length 2 way Single user: AWGN Multi-user: Mixture Single user: no Multi-user: random movement with soft/softer handover 85% ( 8dB noise rise) 2 MByte FTP upload VoIP: 12.2 k AMR speech, 50% activity Slide 6

7 Packet Data Performance Single User (1) Isolated radio cell with good radio conditions (AWGN) and a HARQ retransmission rate of 1% For UE categories 5 & 6 about 5% throughput improvement compared to fixed RLC PDU size of 336 bits due to the reduced RLC overhead UE category 7: throughput significantly drops down to 6.5 Mbps due to RLC window size limitation Maximum RLC PDU size: bits Slide 7

8 Packet Data Performance Single User (2) RLC buffer occupancy limited to available RLC window size Fixed RLC PDU size (336/ 656 bits): drops of available RLC PDUs in the RLC window to zero disrupting the continuous data flow Flexible RLC PDU size: there are always PDUs available for transmission Slide 8

9 Packet Data Performance Multi-User Flexible RLC PDU size provides cell throughput increase of ~8% - Reduced RLC overhead - Finer granularity of the RLC PDU size, allowing for a better exploitation of the uplink resources - Reduced probability of residual MAC-e block errors after HARQ (reduced TCP impact) Only slight impact of the maximum RLC PDU size on throughput (should be chosen > 5000 bits) Slide 9

10 VoIP Transmission Technique/ Performance Criteria MAC-d PDU size of 296 bits for the voice packet and 96 bits for the SID packet Transmission over E-DCH using non-scheduled transmission mode with 2ms TTI RLC UM 8 bit RoHC header 4 bytes RTP 12 bit AMR frame 244bit Non-scheduled grant of 318 bits (transport block size table 0) Maximum number of HARQ transmissions is 4, target average value 2.05 Hdr./ Pad. MAC-d PDU 296 bit Minimum set E-TFCI: 318 for fixed and120 for flexible PDU size MAC-e transport block: 318 bits Performance criteria: - Packet delay <= 90 ms - 95%tile of the VoIP frame loss rate <= 2% - Probability of exceeding 80% uplink cell load <= 2% Slide 10

11 VoIP Performance Simulation Results The VoIP packet delay increases with higher path loss caused by - Higher number of HARQ transmissions in case of fixed RLC PDU size - Allocation of several HARQ processes for transmission of the whole MAC-d PDU in case of flexible RLC PDU size A delay higher than 90 ms is considered to be a packet loss About 2 db coverage gain for flexible RLC PDU size In multi-ue scenario, the VoIP capacity is slightly improved by 6% for flexible RLC PDU size compared to fixed PDU size - SID frames can now be transmitted with a smaller RLC PDU size - In case of power limitation the RLC PDU can be segmented by MAC-is at the UE Slide 11

12 Conclusions Flexible RLC PDU size feature in uplink was investigated by dynamic system simulations for packet data services in single- and multi-user scenarios, and for VoIP over E-DCH For UE categories 5 and 6 the single user throughput improves by about 5% compared to fixed RLC PDU size of 336 bits due to the reduced RLC overhead In case of multi-users, a maximum gain of about 8% was detected for UE category 6 - Reduced RLC overhead - Finer granularity of the RLC PDU size allowing for better exploitation of the available uplink load - Reduction of call drops caused by TCP timeouts by improvements of the behavior at cell edge No significant impact of the maximum RLC PDU size on the performance, as long as this parameter is chosen larger than 5000 bits RLC window size limitations are resolved enabling for about 11.3 Mbps RLC throughput with UE category 7 (16 QAM) compared to 6.5 Mbps for fixed RLC PDU size of 336 bits Performance in power limitation at cell edge for VoIP over E-DCH users can be improved too - Using smaller packet sizes in power limitation packet loss can be prevented at cost of an increased transmission delay Improved coverage, about 2 db gain - Capacity gain of about 6% in multi-user scenarios Slide 12

13 Thank you! Nash Technologies GmbH Thurn-und-Taxis-Str. 10 D Nuremberg

14 Backup

15 RLC Rate Limit WS Optimum for Peak Rate Tx window state variable VT(...) Parameters: RLC RTT TimerStatusProhibit TSP > RTT Available MAC-is peak rate r RLC window size WS is optimum for RTT, TSP and r Result: Mean RLC rate R: R = WS / (TSP + RTT) = r One SR arrives per TSP. SR acknowledges PDUs up to the situation one RTT earlier. RLC window jumps by a fraction of WS. Note: TSP > RTT step size > WS/2 TSP = RTT step size = WS/2 TSP < RTT step size < WS/2 SR status report Exactly when Tx window is full, the next SR arrives. MS (upper edge) RTT WS RTT S (actually submitted) A (lower edge) TSP With TSP > RTT: R < WS / (2 * RTT) Time Slide 15

16 UTRAN Architecture Evolution from Rel-7 Enhanced layer 2 which is already available for HSPDA is also supported for E-DCH E-DCH in Rel-8 Additions in RRC to choose between MACe/es and MAC-i/is RLC now supports flexible PDU size (UM & AM) SRNC Logical Channels CRNC w/o MAC-c/sh MAC-es/ MAC-is MAC-d flows MAC-c w/o MAC-c/sh MAC-d flows RRC RLC MAC-d DCH Upper phy MAC-c/sh DCCH DTCH PDCP BCCH New MAC-is entity with link to MAC-d and MAC-c New MAC-i entity located in the Node B NodeB MAC-e/ MAC-i Transport Channels EDCH MAC-hs/ MAC-ehs HS-DSCH DSCH FACH MAC-b BCH MAC-i entities from multiple NodeB may serve one UE (soft HO) Slide 16

17 Data Flow through Layer 2 UTRAN Side RLC DCCH DTCH DTCH RLC PDU: Header DATA TSN: Transmission Sequence Number (6 bits) MAC-d MAC-d PDU: DATA SS: Segmentation Status (2 bits) Disassembly & Reassembly MAC-is Disassembly & Reassembly Reordering Reordering Reordering Reordering queue distribution MAC-d Flows Disassembly & Reassembly Reordering queue distribution Mac-is PDU: TSN SS Mac-is SDU DATA DATA LCH-ID: Logical Channel Identifier (4 bits) - Maps to MAC-d flow ID L: Length of MAC-is SDU in octets (11 bits) MAC-i Demultiplexing Read UE id (FDD only) HARQ MAC-i PDU: LCH-ID L F MAC-i header DATA DATA Padding (Opt) F: Flag indicating if more fields are present in MAC-i header or not (1 bit) - 0: Flag is followed by additional set of LCH-ID, L, F field L1 Transport block: DATA - 1: Flag is followed by MAC-is PDU Mapping info signaled to Node B LCH-ID => MAC-d flow ID Cf Rel-8 Slide 17