Introduction. DL Control Signaling Errors in LTE

Transcription

1 Cntrl Signaling Rbustness in LTE This paper appears in: Vehicular Technlgy Cnference Fall (VTC 2009 Fall), 2009 IEEE 70th, Issue Date: Sept. 2009, Written by: Timner, Ylva; Larm, Anna; Wiemann, Henning SECTION I. Intrductin The LTE radi interface has recently been specified fr 3GPP Release 8 [1] [2]. It uses Orthgnal Frequency Divisin Multiplex (OFDM) fr the dwnlink (DL) and Single-Carrier Frequency Divisin Multiple Access (SC-FDMA) fr the uplink (UL). The Medium Access Cntrl (MAC) [3] layer scheduler in the base statin (enb) is in charge f assigning bth UL and DL radi resurces and a suitable mdulatin and cding scheme. A Hybrid Autmatic Repeat request (HARQ) scheme n the MAC sub-layer perfrms re-transmissins f crrupted transprt blcks and requires fast and frequent HARQ feedback frm the receiving prtcl entity. It is crucial that the HARQ peratin in the LTE netwrk is rbust t signaling errrs n L1/L2. This paper describes the effects f different signaling errrs and evaluates the impact n link level and system perfrmance. Als, different ACK/NACK detectin threshlds fr bth UL and DL are evaluated. I.Intrductin II.DL Cntrl Signaling Errrs in LTE III.UL Cntrl Signaling Errrs in LTE IV.Prtcl Simulatin Results V.System Simulatin Results VI.Cnclusin SECTION II. DL Cntrl Signaling Errrs in LTE In the DL, data is transmitted n the Physical DL Shared Channel (PDSCH) [4], and each transmissin is annunced by a DL Assignment n the Physical DL Cntrl Channel (PDCCH). The PDCCH transmissin can take different frmats and thus the frmat needs t be indicated separately. This is dne n the Physical Cntrl Frmat Indicatr Channel (PCFICH). If the UE crrectly receives the DL Assignment, it will try t decde the data n PDSCH, and reprt the success thrugh an ACK r a NACK n the Physical UL Cntrl Channel (PVCCH). The signal sent n PVCCH is detected thrugh ne r tw detectin threshlds. If tw threshlds are used, ACK, NACK and silence (DTX) can be detected. If any f the cntrl signals are lst r missdetected, there will be an effect n the HARQ transmissin. There are three different effects f the mst cmmn errrs:

2 Extra retransmissins causing additinal verhead HARQ failures resulting in data lss nly recvered n higher layers Reduced efficiency due t Sft-Cmbining instead f Incremental Redundancy A. DL Assignment errr If the UE misses a DL assignment, it will nt send any feedback. If silence (DTX) n the PUCCH is crrectly detected in the enb, this will result in an extra retransmissin with the same redundancy versin (RV) as befre. B. DL Assignment errr + DTX t ACK errr If the enb detects an ACK instead f the DTX, it will nt d a HARQ retransmissin, and thus the data will be lst n the MAC layer. Data lss can be avided by Radi Link Cntrl (RLC) prtcl retransmissins, in case RLC Acknwledged Mde (AM) is used. C. DL Assignment errr + DTX t NACK errr If the enb detects a NACK instead f the DTX, it will retransmit the data, but prbably with anther RV than in the lst transmissin. If the lst transmissin was the first transmissin f the radi blck, the UE will nt be able t decde any ther RV than RV0. The effect f this errr is dependent n the number f previus transmissins, as well as n the enb implementatin regarding chice f RV t transmit. D. ACK t NACK r ACK t DTX errr When the UE decdes the data crrectly it sends an ACK. If the enb miss-detects the ACK as a NACK r DTX, it will retransmit the data. The UE will recgnize the data as a repetitin, and retransmit the ACK. E. NACK t ACK errr When the UE fails t decde the data crrectly, it sends a NACK. If the enb errneusly decdes the NACK as an ACK, n HARQ retransmissin will be perfrmed, and thus the data is lst n the MAC layer. Data lss can be avided by RLC retransmissins, in case RLC AM is used. F. NACK t DTX errr If the enb detects DTX instead f the NACK, it will assume that the assignment was lst, and retransmit the data using the same RV as befre. The UE will therefre have less prbability t decde the retransmissin, since the IR gain is lst, but the data will nt be lst. I.Intrductin

3 II.DL Cntrl Signaling Errrs in LTE III.UL Cntrl Signaling Errrs in LTE IV.Prtcl Simulatin Results V.System Simulatin Results VI.Cnclusin SECTION III. UL Cntrl Signaling Errrs in LTE In the UL, data is transmitted n the Physical UL Shared Channel (PUSCH) [4], and each transmissin allcatin is signaled by a UL Grant n the PDCCH. Again, the PDCCH frmat is indicated n the PCFICH. If the UE crrectly receives the UL grant, it transmits data in the UL. Assuming nn-adaptive retransmissins, the enb will reply with an ACK r a NACK n the Physical HARQ Indicatr Channel (PHICH), and the UE will perfrm a retransmissin n a predefined resurce if it receives a NACK. If any f the cntrl signals are lst r miss-detected, it will have effect n the HARQ transmissin. The mst cmmn errr cases cause tw different effects: Missed transmissin pprtunities, causing extra delays and underutilizatin f the link. Unscheduled transmissins, causing extra interference, and pssible cllisin with scheduled transmissins. A. Lst UL Grant If the UE cannt decde the Scheduling Grant, it will nt transmit any data (DTX n the PUSCH). If the enb detects that there was n data transmitted, it can retransmit the Grant. The effect will be a lst transmissin pprtunity and an extra delay f ne RTT. If the enb des nt detect the DTX, the enb will send a NACK and expect a new retransmissin. The UE will nt listen t the PHICH, and will thus nt send anything. Again, the enb has the pssibility t detect the errr by means f DTX detectin, and thus send a new Grant. The effect will be anther missed transmissin pprtunity. B. ACK t NACK errr n PHICH When the enb decdes the UL data crrectly, it sends an ACK, and, if the enb expects that the UE has mre data t transmit, a Grant fr new data. An ACK t NACK errr may result in unnecessary retransmissins and, even wrse, in cllisins with UL transmissins by anther UE. This wrst case scenari ccurs when n new Grant is sent t the first UE. When there is an ACK t NACK errr but n new grant, the UE will d a nn-adaptive HARQ retransmissin that is nt scheduled by the enb. After the transmissin, the UE expects feedback n the same PHICH resurce as previusly used. The PHICH resurce might be unused, r used fr anther allcatin t anther UE, s the

4 value that the UE receives is randm. The UE will either detect an ACK r a NACK. If the UE detects a NACK als this time, it will d anther unscheduled retransmissin. The effect f the unscheduled retransmissins is always extra interference, and if the enb happened t schedule at least part f the resurce t anther UE, there will be a cllisin due t tw UEs transmitting n the same resurce. A cllisin will likely cause a transmissin errr fr the scheduled UE, s the scheduled UE will need t retransmit. If als the UE with the ACK t NACK errr retransmits anther time, there will be a secnd cllisin. In wrst case, bth UE's might retransmit until at least ne f them (typically the unscheduled UE) reaches the maximum number f HARQ transmissin attempts. C. NACK t ACK errr n PHICH When the enb fails t decde the data crrectly, it sends a NACK (with r withut a grant in case f Nn-adaptive r adaptive HARQ respectively). If the UE errneusly decdes the NACK as an ACK, it will nt retransmit the data. Assuming that DTX detectin n PUSCH is available in the enb, it will, as in the case f lst UL Grant, be able t detect that there is n signal transmitted n the PUSCH, and will thus retransmit the ld Grant (with the same NDI). The UE will then retransmit the data. The result f the signaling errr is in this case a missed transmissin pprtunity and an extra delay f ne HARQ RTT. I.Intrductin II.DL Cntrl Signaling Errrs in LTE III.UL Cntrl Signaling Errrs in LTE IV.Prtcl Simulatin Results V.System Simulatin Results VI.Cnclusin SECTION IV. Prtcl Simulatin Results In the previus sectin, we listed the mst cmmn errr cases and their effect n the prtcl peratin. In this sectin, we present simulatin results btained frm prtcl level simulatins perfrmed in Matlab. The results aim t quantify the effect f signaling errrs in terms f verhead, packet lss rate, and resurce underutilizatin. A. DL Simulatin mdel and assumptins The BLER f the data transmissins is assumed t be cnstant fr the link, and the cmbining gain f retransmissins is neglected. This is a pessimistic assumptin as it increases the required number f HARQ transmissin attempts and cnsequently the prbability f cntrl channel errrs. As a simplificatin, the number f transmissin attempts is nt cunted, s HARQ failures due t reaching the maximum number f transmissin attempts are neglected.

5 The ACK/NACK signal n PUCCH is detected thrugh tw symmetrical threshlds, s that ACK, NACK, and DTX can be detected. Tw different threshld cnfiguratins were investigated: a lw threshld, which seldm detects DTX, and a high threshld which detects DTX mre ften. With the lw threshld, the prbability f ACK t DTX and NACK t DTX errrs are lwer, but the DTX misdetectin prbability is higher. The effects f the different signaling errrs are mdeled as detailed as pssible. Since RVs are nt mdeled, the effect f errneus RVs can nt be fully mdeled. The effect f DTX t NACK errr is therefre apprximated as tw extra retransmissins. B. DL Prtcl Level Simulatin results Figure 1 shws the fractin f transmissins that result in an extra HARQ retransmissin due t signaling errrs. UEs with a cntrl channel C/I belw 5 db are ut f cverage and have errr rates exceeding the 3GPP requirements [5]. It can be nted that the impact f the chsen ACK/NACK detectin threshld and the data channel BLER is small cmpared t the effect f the cntrl channel C/I. Extra retransmissins are caused either by a DL assignment errr, r by an ACK misdetectin. The DL assignment errrs are significantly mre frequent, and thus the impact f the ACK/NACK detectin threshld and BLER is lw.

6 Figure 1. Link effects f signaling errrs, DL transmissin. View All Next The fractin f HARQ failures due t signaling errrs is shwn in Figure 2. At the cverage limit f 5 db the HARQ failure rate is belw 0.3% even in the wrst case scenari. With VIP, the Frame Errr Rate (FER) requirement is 1% per link, s a value f 0.3 is acceptable. Figure 2. Link effects f signaling errrs, DL transmissin. Previus View All Next At the cverage limit f 5 db, the cntrl channel errrs (mainly lst DL assignments) cause an extra verhead f apprximately 3%, while the verhead is 16 18% at 8 db C/I n the cntrl channels. The impact f the ACK/NACK detectin threshld is large; there are abut 10 times mre HARQ failures with the lw threshld than with the high threshld. HARQ failures are caused by either NACK t ACK errrs, r by missed DL assignments fllwed by a DTX t ACK errr.

7 A high threshld that ften detects DTX will minimize the prbability fr bth NACK t ACK errrs, and DTX t ACK errrs. With the high detectin threshld, the HARQ lss rate is kept arund 0.2% even at the wrst channel investigated, which is a remarkably gd result. C. UL Simulatin mdel and assumptins The BLER f the data transmissin is assumed cnstant fr the link, and the cmbining gain f incremental redundancy is neglected. This is a pessimistic assumptin as it increases the required number f HARQ transmissin attempts and cnsequently the prbability f cntrl channel errrs. The maximum number f transmissin attempts is assumed t be six, and nce that is reached, the transmissin f the radi blck is cnsidered as failed. The number f transmissin attempts is incremented fr each scheduled transmissin, even if n data was transmitted due t signaling errrs. Three different scenaris were investigated fr the PHICH detectin in the UE using different threshlds: The symmetrical threshld scenari crrespnds t a UE that is n the cell edge, while the ther scenaris represent UEs utside f cverage. The latter tw threshlds priritize detectin f the crrespnding feedback signal. The effects f the different signaling errrs are mdeled in detail, using a few assumptins as described belw. The effects f an ACK t NACK errr n PHICH varies depending n traffic cnditins. In this paper a wrst case scenari is assumed where the UE has an empty buffer, and n grant fr new data is sent. In case f a cllisin, it is assumed that the same PHICH resurce is used fr the scheduled UE as was used fr the UE with the ACK t NACK errr. This assumptin maximizes the risk f repeating cllisins. The mdeled behavir is shwn in Figure 3. The prbability fr cllisin is the same as the channel ccupancy, which is assumed t be 80%, and if a cllisin ccurs, the BLER fr the ther UE is assumed t be 80%. The effect fr the UE experiencing the ACK t NACK errr is mdeled as unscheduled retransmissins and an increase in number f transmissin attempts. Fr the scheduled UE, the effect f the cllisin is mdeled as extra retransmissins, but the increase in number f transmissin attempts is nt mdeled.

8 Figure 3. Effects f ACK t NACK errr when n new Grant is sent, assuming that the ld PHICH resurce is used fr feedback als fr scheduled UE. Previus View All Next D. UL Prtcl Level Simulatin results Figure 4 shws the number f missed PUSCH transmissins as a fractin f scheduled transmissins. Missed transmissins are caused by lst UL grants r NACK-t-ACK errrs, where the lst UL grants dminate. Nn-adaptive HARQ is simulated, s the number f UL Grants per transmissin decreases when the fractin f retransmissins increases. Therefre, the impact f missed Grants is lwer with high BLER. The figure shws furthermre, that the ACK/NACK threshld has a minr impact as lng as the BLER is at a mderate level. The result with the symmetrical threshld is identical t the NACK threshld. Finally, ne shuld nte that the desired wrst-case perating regime is in the highlighted area, i.e., abut 5dB cntrl channel SIR and 0 30% HARQ BLER. Figure 5, n the ther hand, shws the percentage f unscheduled and thus unwanted UL transmissins. These ccur when the UE interprets an ACK n the PHICH as a NACK, and perfrms a nn-adaptive HARQ retransmissin. It shuld be nted that this evaluatin assumes that the UE des nt receive an UL grant fr new data transmissin either. Such a grant wuld verride the missdetected NACK and thereby avid the unwanted retransmissin. In this respect, Figure 5 shws a wrst case result. The figure shws that the fractin f the unwanted retransmissins in relatin t the actually scheduled transmissins becmes significant if the ACK/NACK threshld is chsen inapprpriately. The NACK Th. leads t ~3% unwanted retransmissins at a reasnable HARQ BLER f 0 30%. The tw ther threshld cnfiguratins give identical results, with 0.25% unwanted retransmissins at the same HARQ BLER level. Please nte that bth the NACK threshld and ACK threshld scenari crrespnd t UEs utside f the target cverage area accrding t [1], while the symmetric threshld scenari assumes UEs just within cverage. If the scheduled transmissin fails due t a cllisin, there will be an extra retransmissin caused by the signaling errr. The number f extra retransmissins is less than the number f unscheduled transmissins (abut 80% in the simulated scenari). These retransmissins might cause extra HARQ failures. Hwever, the number f HARQ failures caused by HARQ signaling errrs was extremely small. (Belw 0.01%)

9 Figure 4. Link Effects f signaling errrs, UL transmissn. The marked area shws the target perating area (Cntrl channer errr rates accrding t 3GPP requirements [5], 0 30% HARQ BLER.). Previus View All Next

10 Figure 5. Link Effects f signaling errrs, UL transmissn. The marked area shws the target perating area (Cntrl channer errr rates accrding t 3GPP requirements [5], 0 30% HARQ BLER.). Previus View All Next I.Intrductin II.DL Cntrl Signaling Errrs in LTE III.UL Cntrl Signaling Errrs in LTE IV.Prtcl Simulatin Results V.System Simulatin Results VI.Cnclusin SECTION V. System Simulatin Results In this sectin, we investigate the impact f cntrl signaling errrs n the verall end-t-end perfrmance, i.e., n the data rates achieved fr file transfers. We d that by means f

11 simulatins in a system simulatr laded with web traffic. The system simulatr used mdels the physical layer transmissins n the data channels and includes detailed mdels f the L2 and applicatin level prtcls. The cntrl channels are mdeled with fixed errr prbability. The web traffic simulatins are dne in a single user scenari. We simulate a large number f file transfers f 10 MB files twards and f 5 MB files frm a UE cnnected t an LTE cell in a 10 MHz deplyment with lw lad. The simulatin settings are accrding t the 3GPP case 1 scenari [6], and a simplified 2 2 MIMO scheme is used. The users start frm a randm psitin, and mve with a speed f 3 km/h twards a randm directin. The HARQ BLER target is set t 10%. HARQ failures appear if the here used maximum number f 5 transmissin attempts are exceeded, r due t NACK-t-ACK errrs. The residual lss rate abve the MAC layer is in the rder f The MAC errrs prpagate t higher layers and are recvered by the RLC AM prtcl. The bject bitrates reflect the perfrmance f the entire file transfer including the TCP slw-start phase. Due t the chsen file size, the dwnlad peak rates are limited by TCP, and nt by the radi interface rate. Als, it shuld be nted that the simulatin has nt been tuned fr reaching peak rates, but rather t shw the perfrmance lss caused by the L1/2 signaling errrs. Figure 6 and 7 shw CDFs f the bject bitrates fr file dwnlads and uplads, respectively. As can be seen frm the graphs, the used wrst-case errr levels recmmended by 3GPP, nly cause minr lsses in perfrmance.

12 Figure 6. Dwnlad bject bitrates fr a large number f 10MB files. Previus View All Next

13 Figure 7. Uplad bject bitrates fr a large number f 5 MB files. Previus View All In the DL, the perfrmance is degraded by 2% n average. The difference is mainly caused by the packet lsses n L1 that are recvered by RLC AM, but delay the data transfer. In the UL, the perfrmance is degraded by nly 0.7%. The main reasn fr better UL perfrmance is that signaling errrs cause less HARQ failures and packet lss than in the DL. I.Intrductin II.DL Cntrl Signaling Errrs in LTE III.UL Cntrl Signaling Errrs in LTE IV.Prtcl Simulatin Results V.System Simulatin Results VI.Cnclusin SECTION VI. Cnclusin

14 The prpsed wrst case errr rates accrding t [1] are reasnably lw s that cntrl channel errrs n PCFICH, PDCCH, PHICH, and PUCCH d nt have any significant impact n the system perfrmance, especially when RLC AM is used t recver the lwer layer errrs. The resurce waste caused by redundant retransmissins and missed UL transmissins can be large fr UEs with significantly wrse cntrl channel perfrmance than required by the 3GPP targets. Hwever, even then the mre severe cnsequences like HARQ failures r cllisins can be kept at an acceptable level when apprpriate detectin threshlds are used. T avid cllisins n the PUSCH, it is imprtant that UEs apply a detectin threshld causing few ACK t NACK errrs Ftntes "N Data Available" References 1. E. Dahlman, A. Furuskär, Y. Jading, M. Lindström, S. Parkvall, "LTE: The Evlutin f Mbile Bradband", IEEE Cmmunicatins Magazine, April, 2009, vl. 47, n.4 Shw Cntext 2. A. Larm, M. Lindström, M. Meyer, G. Pelletier, J. Trsner, H. Wieman, "The LTE Link Layer Design", IEEE Cmmunicatins Magazine, April, 2009, vl. 47, n.4 Shw Cntext 3. E-UTRA Medium Access Cntrl (MAC) prtcl specificatin Shw Cntext 4. Physical channels and mdulatin Shw Cntext 5. Overall descriptin; Stage 2 Shw Cntext 6. Physical layer aspects fr evlved Universal Terrestrial Radi Access (UTRA) Shw Cntext

15 Authrs Ylva Timner N Bi Available Anna Larm N Bi Available Henning Wiemann N Bi Available Cited By