reference signals for lte

IEEE Communications Magazine • February 2012132 0163-6804/12/$25.00 © 2012 IEEE

1 In 3GPP E-UTRA sys-tems, each antenna port ischaracterized by a refer-ence signal, and is notnecessarily a physicalantenna.

INTRODUCTION

The Third Generation Partnership Project(3GPP) Evolved Universal Terrestrial RadioAccess (E-UTRA) Release 10 standards (i.e.,Long Term Evolution [LTE]-Advanced) areevolved from the Release 8 and 9 standards tosupport the International Mobile Telecommuni-cations (IMT)-Advanced peak data rate targetsof 100 Mb/s for high mobility and 1 Gb/s for lowmobility. Furthermore, LTE-Advanced systemstarget support for downlink peak spectral effi-ciency of 30 b/s/Hz and uplink peak spectral effi-ciency of 15 b/s/Hz, and support forapproximately 1.5× improved cell average andcell edge spectral efficiency over Release 8 and 9standards [1]. In order to achieve these peakdata rate and peak spectral efficiency targets,LTE-Advanced introduced carrier aggregation,eight-layer downlink, and four-layer uplink mul-tiple-input multiple-output (MIMO) spatial mul-tiplexing. On the other hand, to achieve thecell-average and cell-edge spectral efficiency tar-gets, enhanced intercell interference coordina-

tion (e-ICIC), enhanced single-user (SU-) andmulti-user (MU-) MIMO, and clustered discreteFourier transform (DFT) spread orthogonal fre-quency-division multiplexing (OFDM) for uplinkare introduced. This article presents technologyadvancements of reference signals to efficientlysupport the enhanced transmission schemesnewly introduced in LTE-Advanced.

LTE systems make use of pilot signals, knownas reference signals (RS) for downlink anduplink channel estimation. Downlink RS areprovided for channel state information (CSI)measurement for feedback and channel estima-tion for demodulation. The Release 8 LTEdesign of downlink RS primarily relied on a setof cell-specific reference signals (CRS) that canbe used for both purposes. The CRS are wide-band pilot signals transmitted across the down-link system bandwidth in every subframe and aredefined for up to four antenna ports1 (APs). Tosupport 8-layer spatial multiplexing in LTERelease 10, low duty-cycle wideband referencesignals, called CSI reference signals (CSI-RS)are introduced for CSI measurement. Further-more, a new set of user equipment (UE)-specificRS (UE-RS) is defined to enable channel esti-mation for demodulation instead of introducing4 additional CRS. This Release-10 design is ben-eficial to efficiently support high rank transmis-sions because UE-RS requiring higher overheadis present only in those physical resource blocksin which higher rank is used. On the other hand,uplink RS are provided for uplink CSI measure-ment (also called uplink channel sounding), linkadaptation, timing estimation, power control,and channel estimation for demodulation. LTERelease 8 defines two types of uplink RS:demodulation RS (DMRS) for demodulationand sounding RS (SRS) for the other purposes.DMRS are sent together with uplink data,whereas SRS are periodically transmitted fromeach UE unit according to a radio resource con-

ABSTRACT

3GPP LTE Release 10 standards (also knownas LTE-Advanced) adopted some of the state-of-the-art radio access technologies that include car-rier aggregation, eight-layer downlink spatialmultiplexing, and four-layer uplink spatial multi-plexing. For facilitating these enhancements, ref-erence signals have significantly evolved inLTE-Advanced. This article examines underlyingdesign principles of the LTE-Advanced refer-ence signals. Specifically, newly introduced dedi-cated demodulation reference signals andchannel state information reference signals fordownlink and improvements of demodulationreference signals and sounding reference signalsin uplink are discussed.

LTE-ADVANCED AND 4G WIRELESSCOMMUNICATIONS

Young-Han Nam, Samsung Telecommunications America

Yosuke Akimoto, Sharp Corporation

Younsun Kim, Samsung Electronics

Moon-il Lee, InterDigital Communications

Kapil Bhattad, Qualcomm Inc.

Anthony Ekpenyong, Texas Instruments

Evolution of Reference Signals for LTE-Advanced Systems

NAM LAYOUT 1/25/12 5:16 PM 32

IEEE Communications Magazine • February 2012 133

trol (RRC) layer configuration. To supportchannel estimation on multiple APs for uplinkspatial multiplexing and to more flexibly supportuplink MU-MIMO, Release-10 LTE-Advancedintroduces further enhancements to the uplinkRS. These enhancements include application oforthogonal cover codes (OCC) that improve theorthogonality of multiple DMRS across APs andmultiple UE units, and introduction of aperiodicSRS transmission so that base stations (oreNodeBs) can more efficiently manage SRSresources across the increased number of UEunits and/or APs in LTE-Advanced systems.

This article is organized as follows. Wedescribe the motivation, design choices, andstructure of the new downlink RS covering CSI-RS and UE-RS, and discuss the new uplink RS.Some concluding remarks are provided.

EVOLUTION OF DOWNLINKREFERENCE SIGNALS

CHANNEL-STATE-INFORMATIONREFERENCE SIGNALS

CSI-RS are reference signals transmitted fromeNodeB APs in order for UE to measure down-link CSI. The introduction of CSI-RS in LTE-Advanced systems allows reduced RS overheadand flexible configurations of multicell RS mea-surements.

CSI-RS Patterns — A CSI-RS supports up toeight APs and is transmitted in a wideband man-ner. Figure 1 shows CSI-RS mapping patterns ina pair of physical resource blocks2 (PRBs) fornormal cyclic prefix (CP) subframes.

Depending on the number of CSI-RS APs,there are multiple reuse patterns on differentlocations which share a common base pattern. Inthe case of 1-, 2-, 4-, and 8-AP CSI-RS, thereare 20, 20, 10, and 5 reuse patterns, respectively,allowing different cells to utilize different reusepatterns to avoid mutual CSI-RS collision. Inaddition, the base patterns for different numbersof CSI-RS APs have a nested structure, allowingfor simpler CSI-RS transmitter and receiverimplementation.

Each AP’s CSI-RS are spread over two time-consecutive resource elements3 (REs) such thattwo different CSI-RS are code-division multi-plexed (CDM’ed) in the two REs. One of themajor benefits of CDM as compared to othermultiplexing schemes such as FDM is that CDMcan balance transmission power across APs inthe frequency domain.

The CSI-RS density is closely related to thechannel estimation accuracy. In general, largerCSI-RS density may provide better CSI measure-ment accuracy while reducing downlink resourceutilization. Therefore, adequate CSI-RS density,which occupies minimum downlink resourceswhile providing reasonable CSI measurementaccuracy, is desirable. Since CSI-RS is only usedfor downlink channel measurement and feedbackreporting, granularity is coarse due to feedbackoverhead; the system performance is relativelyinsensitive to the channel measurement accuracyas compared to that for UE-RS. Therefore, alow-overhead CSI-RS transmitted only onceevery 5, 10, 20, 40, or 80 ms is employed in LTE-Advanced. A typical transmission of CSI-RS,which may consist of 4 APs with periodicity of 10ms, would only require an overhead equivalent to0.2 percent of the entire time and frequencyresources. The low overhead is achieved by allo-cating a single RE per PRB pair per CSI-RS APexcept when the CSI-RS has only one AP, inwhich case two REs are allocated per PRB pair.During the evaluation phase of CSI-RS in 3GPP[2, 3], it was observed that better measurementaccuracy by having a higher density than 1 didnot improve the performance of downlink trans-mission in a realistic environment.

Performance Impact of CSI-RS on Legacy UE— When CSI-RS are transmitted in PRBsassigned for a legacy UE, the CSI-RS replacelegacy data symbols on CSI-RS REs. Because alegacy UE has no knowledge of CSI-RS, itsreceiver would consider CSI-RS as valid datasymbols in such a PRB. This leads to demodula-tion performance degradation as the legacy UE isnot only losing coded information but is alsobeing interfered by the CSI-RS in its decodingprocess. The adverse impact on the legacy UEbecomes more serious as the number of CSI-RS

Figure 1. CSI-RS mapping patterns for 1-, 2-, 4-, and 8-port CSI-RS.

Data

1-, 2-port CSI-RS patterns

C17C16C15C14C13C12C8C3C7C2

C19C18

C9C4

C11C10

C5C0

C6C1

4-port CSI-RS patterns

C8C3C7C2C6C1C8C3C7C2

C9C4

C9C4

C5C0

C5C0

C6C1

8-port CSI-RS patterns

C3C3C2C2C1C1C3C3C2C2

C4C4

C4C4

C0C0

C0C0

C1C1

CSI-RSUE-RSCRSOFDM symbols

Subc

arri

ers

Controlchannels

2 A pair of PRBs is theminimum resource alloca-tion size in 3GPP E-UTRA systems. EachPRB consists of 12 con-secutive subcarriers in thefrequency domain and xconsecutive OFDM sym-bols in the time domain,where x = 6 for extendedcyclic prefix subframesand x = 7 for normalcyclic prefix subframes. Inthis article, we consideronly normal cyclic prefixsubframes for simplicity ofpresentation.

3 A resource element (oran RE) is a unit elementof the OFDM time-fre-quency grid in 3GPP E-UTRA systems, whichis uniquely identified by asubcarrier index and anOFDM symbol index.

NAM LAYOUT 1/25/12 5:16 PM Page 133

IEEE Communications Magazine • February 2012134

APs grows larger [2, 3]. To mitigate system per-formance loss arising from this side effect, aneNodeB may allocate additional PRBs in trans-mitting a transport block, or assign a reducedmodulation and coding rate to such legacy UE.The eNodeB may also choose to avoid schedulinglegacy UE in the subframe containing CSI-RS.

Data Resource Element Muting for IntercellCSI Measurement — Although different CSI-RS patterns are likely to be used in neighboringcells to minimize mutual CSI-RS collision, neigh-bor cells’ downlink data signals may still signifi-cantly interfere with CSI-RS received at cell edgeUE in a heavily loaded network in interference-limited scenarios. In addition, accurate CSI mea-surement for neighbor cells is also required forbetter support of coordinated multipoint (CoMP)transmissions in future releases. To cope with theinterference problems and to facilitate CSI mea-surement of neighbor cells for CoMP, LTE-Advanced Release 10 supports data RE muting,whereby data REs colliding with CSI-RS in aneighbor cell are transmitted with zero transmis-sion power (i.e., muted REs). For LTE-AdvancedUE, the eNodeB applies code rate matchingaround the muted REs so that LTE-AdvancedUE is able to decode data accordingly.

UE-SPECIFIC REFERENCE SIGNALSUE-RS are precoded pilots used for data demod-ulation. They are transmitted only on the PRBsallocated for each UE’s data, and are precodedwith the same precoder used for the data.

There are several advantages of UE-RS-basedoperations over CRS-based operations. The UE-RS pilot overhead depends only on the transmis-sion rank (or the number of assignedtransmission layers or streams) and is decoupledfrom the maximum transmission rank supportedby the system. In addition, channel estimation fordemodulation can be performed per layer, whichis desirable for CoMP and heterogeneous net-works. For example, in CoMP joint transmission,where multiple transmission points jointly beam-

form to transmit to UE, the UE does not need toknow which transmission points are involved inthe joint transmissions; hence, the related infor-mation does not need to be signaled. Further-more, UE-RS-based operation facilitatesnon-codebook-based frequency-selective precod-ing, which can improve downlink throughput,especially in MU-MIMO and CoMP transmis-sions. For CRS-based transmissions, a selectedprecoding matrix (or matrices) has to be signaledto UE so that the UE can calculate the precodedchannels utilizing the channel estimates obtainedfrom CRS. However, for UE-RS-based transmis-sions, frequency-selective precoders can be select-ed across the allocated PRBs not beingconstrained by a fixed codebook, since UE canestimate precoded channels per PRB directlyfrom UE-RS. Finally, with UE-RS, UE canobtain more accurate interference estimates inpartial loading scenarios. UE-RS always collidewith either data signals or UE-RS of neighboringtransmission points, which are present only whendata signals are transmitted; however, in CRS-based operations, CRS may collide with CRS ofneighboring transmission points, which makesinterference estimates obtained from CRS incor-rect when data signals are not transmitted.

UE-RS Pattern and Spreading Sequences —The UE-RS pattern is shown in Fig. 2. It con-sists of two CDM groups each of which can mul-tiplex up to 4 UE-RS (for 4 APs). The secondCDM group is used only for rank > 2; hence,the pilot density is 12 REs per PRB pair forranks 1 and 2, and 24 REs per PRB pair forrank > 2. The pattern provides three looks ofeach layer’s channels in frequency in order tosupport reliable channel estimation with delayspread up to 5 μs.

For r-layer data transmissions APs 7 to 6 + rare used, while a single-layer allocation usingAP 8 to UE is also allowed to facilitate MU-MIMO operation. The mapping of APs to CDMgroups and the spreading codes is shown on theright side of Fig. 2. The spreading codes are

Figure 2. UE-RS pattern and spreading sequences.

Antenna port

7

8

9

10

11

12

13

14

CDM group

1

1

2

2

1

2

1

2

Spreading code 1

[+1+1+1+1]

[+1–1+1–1]

[+1+1+1+1]

[+1–1+1–1]

[+1+1–1–1]

[–1–1+1+1]

[+1–1–1+1]

[–1+1+1–1]

Spreading code 2

[+1+1+1+1]

[–1+1–1+1]

[+1+1+1+1]

[–1+1–1+1]

[–1–1+1+1]

[+1+1–1–1]

[–1+1+1–1]

[+1–1–1+1]

CDM group

UE-RS CDM group 2(APs 9,10,12,14)

UE-RS CDM group 1(APs 7,8,11,13)DataCRS

OFDM symbols

Subc

arri

ers

Controlchannels



selected such that for up to four-layer transmis-sions, the CDM-multiplexed pilots can beorthogonalized by despreading over two con-tiguous REs in time. Thus, this can provide twolooks at each layer’s channels in time, which isbeneficial to estimate channels at moderate tohigh Doppler. For more than four-layer trans-missions, the pilots can only be orthogonalizedby using all four REs in time on the same sub-carrier where some of the REs are not contigu-ous, and hence the pilots remain orthogonalonly for low speeds. However, this is not a sig-nificant restriction as MIMO with more thanfour layers is likely to be used only at very lowDoppler. The spreading code alternates in fre-quency between spreading codes 1 and 2. Thisalternating pattern helps reduce the peak-to-average-power ratio, for example, when thesame precoding matrix is used across all thePRBs in the system bandwidth. It also helpsmitigate the impact of residual interlayer inter-ference remaining after CDM despreading athigh Doppler on the channel estimate.

Furthermore, a cell-ID-based scramblingsequence is applied to UE-RS to ensure that aUE unit is able to distinguish its UE-RS fromthe UE-RS of neighboring base stations. ForAPs 7 and 8, two scrambling sequences aredefined for each cell ID. This allows MU-MIMOmultiplexing of up to four UE units where UE-RS with the same scrambling sequence areorthogonally multiplexed, while UE-RS with dif-ferent scrambling sequences are non-orthogonal-ly multiplexed.

Control Signaling for UE-RS Allocation —For each UE unit’s downlink receptions, thenumber of transmission layers (i.e., the transmis-sion rank) and corresponding UE-RS AP num-bers are dynamically signaled. When rank > 2 issignaled, a UE unit’s demodulator can safelyassume that there are no co-scheduled UE unitsin the UE’s allocated PRBs as the LTE-Advanced system does not allow more than twoorthogonal UE-RS multiplexing for MU-MIMO.On the other hand, when either rank 1 or rank 2is signaled, it is possible that the UE’s signal ismultiplexed with other UE units’ signals in someof the allocated PRBs. While for optimizing per-formance of an MU-MIMO/SU-MIMO receiverinformation about co-scheduled UEs in each ofthe allocated PRB is essential, LTE-Advancedhas not defined signaling for such information,partly because the signaling overhead is signifi-cantly high when different sets of UE are multi-plexed in different PRBs. Nevertheless,MU-MIMO multiplexed UE can blindly detectco-scheduled UE in each allocated PRB by mea-suring power on UE-RS of the APs and scram-bling sequences not allocated to the MU-MIMOUE itself.

Traffic to Pilot Power Ratio — The traffic topilot power ratio for UE-RS is fixed to 0 dB forranks 1 and 2, and 3 dB for rank > 2, in whichcase the average UE-RS power per RE is thesame as the average data signal power per RE.For MU-MIMO operation, such a conditionhelps to ensure that interference measured onUE-RS REs is same as that seen on data REs.

PRB Bundling — When an eNodeB can esti-mate the downlink channel per PRB (e.g., fromuplink sounding using channel reciprocity), theeNodeB may want to choose frequency-selectiveprecoders in different PRBs to improve thethroughput. However, when an eNodeB’s CSIfor a UE is obtained only from precoder matrixinformation (PMI) and rank information (RI)feedback, the eNodeB is forced to use the sameprecoder across multiple consecutive PRBs inthe frequency domain as feedback granularity isnot a single PRB. To facilitate achieving the fre-quency-selective precoding gains when per-PRBCSI is available, and at the same time improvechannel estimates at the UE when only PMI/RIfeedback is available, the LTE-Advanced speci-fies that UE configured with PMI/RI feedbackmay assume that the same precoder is applied toeach of the PRBs belonging to a precodingresource block group (PRG), whereas UE notconfigured with PMI/RI feedback should assumethat different precoders are applied in differentPRBs for UE-RS channel estimation.

EVOLUTION OF UPLINKREFERENCE SIGNALS

In this section, we describe the basic structure ofuplink DMRS and SRS. The limitations of theRelease 8 design are analyzed, and the designchoices for Release 10 enhancements are pre-sented.

EVOLUTION OFDEMODULATION REFERENCE SIGNALS

Required Property for LTE-Advanced DM RS— For Release 8 LTE uplink DMRS, extendedZadoff-Chu (ZC) sequences are adopted [4]because the ZC sequences ensure low peak-to-average-power ratio (PAPR) and at the sametime provide mechanisms to orthogonalize multi-ple DMRS transmitted in MIMO spatial multi-plexing schemes. Using different cyclic time shifts(CS) of a root ZC sequence, multiple users’DMRS transmitting to the same eNodeB can beorthogonally CDM’ed as long as the multipleDMRS are transmitted in the same set of PRBs.

In LTE-Advanced, SU-MIMO with up tofour-layer spatial multiplexing is supported.Although the DMRS corresponding to differentlayers can theoretically be orthogonalized by theRelease-8 LTE mechanism of assigning differentCS to each layer, the CS orthogonality can bebroken in case of high delay spread. In addition,the CS orthogonality is guaranteed only whenthe root ZC sequences are identical in waveformand length, which implies that the time-frequen-cy allocations of MU-MIMO-multiplexed UEshould be identical. Uplink DMRS in LTE-Advanced are designed to overcome these draw-backs while maintaining backward compatibility.

Structure of LTE-Advanced Uplink DMRS — InLTE and LTE-Advanced, two single-carrier fre-quency-division multiple access (SC-FDMA) sym-bols are configured for DMRS in a subframe withinthe allocated time-frequency resources, as shown inFig. 3a. Each uplink RS sequence is given by

In LTE-Advanced,

SU-MIMO with up to

4-layer spatial multi-

plexing is supported.

Although the DMRS

corresponding to

different layers can

theoretically be

orthogonalized by

the Release-8 LTE

mechanism of

assigning different

CS to each layer, the

CS orthogonality can

be broken in case of

high delay spread.



r(α)(n) = wejαn/12–r(n) (1)

in the frequency domain, where n is the subcar-rier index, –r(n) is the root ZC sequence generat-ed by an extended ZC sequence, and α is the CSvalue [4]. LTE Release-10 employs codebook-based precoding for uplink SU-MIMO datatransmissions, and the same precoding mecha-nism is applied for uplink DMRS transmissions.As such, a receiver can directly estimate the spa-tially combined precoded channel. Note thatDMRS are split in the same manner as datawhen clustered DFT spread OFDM is used.

An eNodeB can choose root ZC sequencesfor the RS by considering the auto- and cross-correlations among neighboring eNodeBs. How-ever, such deployment planning is not alwaysstraightforward for cellular operators. Hence, aroot ZC sequence randomization mechanismcalled sequence hopping and sequence grouphopping (SGH) [5] is available depending oneach operator’s choice, where the root ZCsequence varies in every slot if SGH is enabled.The scalar value of w (i.e., +1 or –1) in Eq. 1 isa newly introduced feature called OCC, which isused to spread the RS sequence into two OFDMsymbols. The motivation to introduce OCC is toefficiently support SU- and MU-MIMO.

Enhancements Aimed at Improving SU-MIMO Performance — In SU-MIMO opera-tion in which a UE transmits multiple layerssimultaneously, multiple DMRS can be orthogo-nally multiplexed by assigning different CS forthe multiple layers. However, the CS orthogo-nality may be broken when the delay spread ishigh, because the impulse response of each layerleaks into the correlation window of anotherlayer in the time domain as the CS separationbecomes smaller. To further separate the multi-ple DMRS, OCCs are introduced in LTE-Advanced. As shown in Fig. 3b, different OCCsare applied for higher-numbered layers (i.e., λ ≥3) and lower-numbered layers (i.e., λ ≥ 2), so themultiple RS for the higher rank transmission(i.e., rank ≥ 3) are still well orthogonalized evenif CS separation is small. Although the orthogo-nality by the OCCs may also be broken by high

Doppler shift, the effect is minimal becausehigher rank transmission is mainly used for low-mobility UE.

Enhancements Aimed at Improving MU-MIMO Performance — From the viewpoint ofDMRS orthogonality, MU-MIMO schemes arecategorized into two types, identical-band MU-MIMO and non-identical-band MU-MIMO, andOCC is used to realize non-identical-band MU-MIMO. Identical-band MU-MIMO is a schemein which the frequency band allocations for MU-MIMO UE are identical, and DMRS orthogo-nality can be achieved by assigning different CS.Although this scheme is Release 8 compatibleand hence can be used for multiplexing Release8 and Release 10 UE, the scheduling restrictionmay negatively impact on the uplink systemthroughput performance. On the other hand,non-identical-band MU-MIMO is a scheme inwhich the frequency band allocations for MU-MIMO UE are not identical. Although thisimproves frequency scheduling flexibility andhence may improve the uplink system through-put [6], it implies that the DMRS root ZCsequences for the two slots of a subframe shouldbe the same to utilize OCC [7].

Signaling Design — In LTE-Advanced, CSand OCC are assigned to a UE unit via physicallayer signaling. To reduce downlink signalingoverhead, a 3-bit field is used to signal a combi-nation of the CS and OCC, as shown in Fig. 3b.This 3-bit field is designed based on the follow-ing aspects.

Maximum separation of CS space for SU-MIMO: The CS values are chosen to maximizethe minimum CS separation considering bothSU-/MU-MIMO. The separation value is deter-mined considering maximum four-layer multi-plexing irrespective of transmission rank. Thus,the CS indication for the first layer implicitlydetermines the CS for all other layers.

OCC to achieve further orthogonality for SU-MIMO: OCCs for SU-MIMO are used toachieve further orthogonality to maximize theRS separation. Along with the CS assignmentrule above, different OCCs can be assigned forlayer {0, 1} and layer {2, 3}. This can be turned

Figure 3. Uplink subframe structure and mapping table of CS and OCC: a) uplink subframe structure; b) mapping table of CS and OCC.

† w denotes OCCs applies for first and second slot

‡ λ denotes layer index for SU-MMO. And CS value(s) for λ=[0..x-1] is used when rank-X data channel transmission is indicated

Cyclic shift (α)λ=0 λ=1 λ=2 λ=3

Cyclic shift (α)λ=0 λ=1 λ=2 λ=3

0 6 9 3 [+1+1] [+1+1][+1–1] [+1–1]6 0 3 9 [+1–1] [+1–1][+1+1] [+1+1]3 9 6 0 [+1–1] [+1–1][+1+1] [+1+1]4 10 7 1 [+1+1] [+1+1][+1+1] [+1+1]2 8 5 1 [+1+1] [+1+1][+1+1] [+1+1]8 2 11 5 [+1–1] [+1–1][+1–1] [+1–1]

10 4 1 7 [+1–1] [+1–1][+1–1] [+1–1]9

000001010011100101110111 3 0 6 [+1+1] [+1+1][+1–1] [+1–1]

Signalingbit

OFDMsymbols

CSOCC

Root sequence

Subc

arrie

rs

Slot 0 Slot 1

CS :UplinkDMRS:Uplinkdata

OCC

(a) (b)



off for non-identical-band MU-MIMO.Different OCC for UE: For the sake of non-

identical-band MU-MIMO, different OCC val-ues of [+1 +1] or [+1 -1] are defined fordifferent UE.

SGH Disabling — Assuming the use of non-identical-band MU-MIMO, OCCs can orthogo-nalize DMRS only when SGH is disabled, thatis, the root ZC sequence is the same betweenthe first and second slots. Therefore, LTE-Advanced allows disabling SGH in UE-specificmanners, in which case the same root ZCsequence is used among slots/subframes.

EVOLUTION OF SOUNDING REFERENCE SIGNALSBackground — LTE Release 8 SRS is based onthe extended ZC sequence [4], which is periodi-cally transmitted in the last SC-FDMA symbol ofa subframe. UE units are multiplexed in the timeand frequency domains, for which sounding sub-frames of periodicity TSRS, frequency comb, andbandwidth are UE-specifically configured. Simi-lar to DMRS, up to eight orthogonal SRS trans-missions can be CDM’ed in the same bandwidthusing different CS of the root ZC sequence.

Achieving the spectral efficiency gainspromised by four-layer uplink spatial multiplexingis dependent on accurate CSI estimation, which,in turn, dictates up to a fourfold increase insounding resources. Furthermore, the LTE-Advanced requirements [1] specified the supportof at least 300 active users, without discontinuousreception, in a 5 MHz bandwidth, which is a 50percent increase over the Release 8 requirements.Hence, it has been observed that the periodicnature of Release 8 SRS transmission is not flexi-ble enough to support sounding from an increasednumber of antennas and/or UE units; nor is itnimble in sounding the channel in response todynamic fluctuations in traffic and channel condi-tions. This is partly because Release 8 UE is con-figured for sounding by semi-static RRC signaling,which incurs a larger signaling latency comparedto physical layer dynamic signaling.

As a result of these limitations in Release 8sounding, aperiodic SRS (A-SRS) transmissionis introduced to complement periodic SRS (P-SRS) transmission, wherein an eNodeB dynami-cally schedules UE for one-shot SRS

transmission on demand. In this manner SRSresources are not tied to a single UE unit untilRRC reconfiguration by the eNB. This mecha-nism allows for efficient management of a fixedset of time/frequency/code SRS resources by alarger pool of UE/APs.

SRS Structure — Periodic and aperiodic SRStransmissions share a common set of cell-specificSRS resources (a subframe configuration periodand a subframe offset, and a maximum SRSbandwidth). Different sets of UE-specific sound-ing parameters are independently allocated forP-SRS and A-SRS transmission including SRSbandwidth, periodicity, a frequency comb, and acyclic shift [6]. UE is configured with a fixed setof subframes — A-SRS opportunities — onwhich it may be scheduled for A-SRS transmis-sion. This reduces the eNB scheduling complexitybecause in any subframe only a subset of UE canbe triggered for A-SRS transmission. A compari-son of the timing relationship between P-SRSand A-SRS is shown in Fig. 4 where P-SRS andA-SRS have the same subframe offset, P-SRShas a transmission periodicity TSRS = 5 ms, andA-SRS transmission opportunities occur at aperiodicity of TSRS,1 = 5 ms. A-SRS transmissionoccurs in subframe n + k if downlink controlinformation (DCI) conveying a positive A-SRStrigger is detected in subframe n and k ≥ 4. Theconstraint of 4 ms is to maintain the same mini-mum timing response to a downlink assignmentor uplink grant contained in a detected DCI.

Signaling Design — The desire to minimizedownlink signaling overhead guided the choiceof maintaining a semi-static configuration of A-SRS parameters similar to P-SRS configuration.Furthermore, A-SRS control signaling bits arepiggybacked in either a downlink assignmentscheduling downlink data transmission or anuplink grant scheduling uplink data transmissionin order to minimize signaling overhead. Sincethe primary goal of A-SRS is to support multi-antenna sounding, some configuration flexibilityis achieved by configuring three independentsets of parameters (sets 1, 2 and 3). A 2-bit SRSfield is added to an uplink grant schedulinguplink SU-MIMO data transmission, where thevalue 00 indicates a negative SRS trigger and the

Figure 4. Comparison of P-SRS and A-SRS transmission.

n n+5 n+10

(a)

(b)

n+15 n+20

n

A-SRS trigger A-SRS triggerA-SRS transmission A-SRS transmission

n+5 n+10

10 ms frame

n+15 n+20

A-SRS control

signaling bits are

piggybacked either

in a downlink

assignment

scheduling downlink

data transmission or

in an uplink grant

scheduling uplink

data transmission in

order to minimize

signaling overhead.



values 01, 10, and 11 schedule an A-SRS trans-mission utilizing the parameters indicated in sets1, 2, and 3, respectively. On the other hand, a 1-bit A-SRS field is added to an uplink grantscheduling uplink single-antenna data transmis-sion, and a separate set of parameters is indicat-ed when a positive trigger (value 1) is detectedin the uplink grant.

It can be seen in Fig. 4 that to satisfy the A-SRS transmission timing for a specific A-SRSopportunity, there are a limited number of sub-frames in which the eNB can send a positive A-SRS trigger. This scheduling constraint is evenmore severe for some time-division duplex (TDD)system configurations where only a few uplinksubframes occur in a radio frame. Hence, toincrease the scheduling opportunities for A-SRStransmission, Release 10 specifies a 1-bit A-SRScontrol field in some downlink assignmentsscheduling downlink data transmission. Anotheradvantage for TDD of triggering A-SRS in down-link assignments is that uplink sounding can beused to support downlink beamforming by exploit-ing channel reciprocity. As such, the need foraperiodic sounding is decoupled from the direc-tion of data transmission (uplink or downlink).

SRS Multiplexing for SU-MIMO — To ensurethat there is no timing mismatch between CSIestimates from multiple antennas, UE is config-ured to simultaneously transmit SRS from allconfigured APs. The orthogonality of the CSbreaks down as the frequency selectivity of thechannel increases, and this degradation is morepronounced as the number of APs increases.Therefore, for both P-SRS and A-SRS, a combi-nation of the frequency comb and CS can beused to multiplex four APs’ SRS in large delayspread channels, while for low delay spreadchannels or UE equipped with two APs, onlydifferent CS values are used to multiplex theSRS. The CS and comb values for all APs areimplicitly derived from the values in each of thesemi-statically configured parameter sets.

CONCLUSIONIn this article, technology advancements of refer-ence signals to efficiently support enhancedtransmission schemes newly introduced in LTE-Advanced were overviewed. For the downlink,design principles for the newly introduced CSI-RS and additional sets of UE-RS were reviewed.In particular, CSI-RS patterns, impacts on thelegacy UE, and data resource element mutingwere examined for the CSI-RS, and benefits ofUE-RS, UE-RS patterns and spreadingsequences, control signaling, traffic-to-pilot ratio,and physical resource block bundling were dis-cussed for UE-RS. For the uplink, enhance-ments of uplink DMRS and SRS were describedin detail. Specifically, benefits of introducingorthogonal cover codes for uplink DMRS andaperiodic SRS were presented.

REFERENCES[1] 3GPP TR 36.913, “Technical Specification Group Radio

Access Network; Requirements for Further Advance-ments for Evolved Universal Terrestrial Radio Access (E-UTRA).”

[2] Samsung, “Performance Evaluation for CSI-RS Design,’’R1-100106, 3GPP TSG RAN1#59bis, Jan. 2010.

[3] LG Electronics, “Investigation on the Number of RE forCSI-RS,’’ R1-100647, 3GPP TSG RAN1#59bis, Jan. 2010.

[4] 3GPP, TS 36.211 v10.1.0, Physical Channels and Modu-lation, Mar. 2011.

[5] 3GPP, TS 36.213 v10.1.0, Physical Layer Procedures,Mar. 2011.

[6] Sharp, “Orthogonal Cover Codes for Uplink DMRS,’’ R1-102465, 3GPP TSG RAN1#60bis, Apr. 2010.

[7] Huawei, “Analysis and Evaluation of UL DM RS Designfor LTE-A Scenarios,” R1-100262, 3GPP TSGRAN1#59bis, Jan. 2010.

BIOGRAPHIESYOUNG-HAN NAM ([email protected]) received his B.S.and M.S degrees from Seoul National University, Korea, in1998 and 2002, respectively. He received a Ph.D. degree inelectrical engineering from the Ohio State University,Columbus, in 2008. Since February 2008 he has beenworking at Samsung Telecommunications America, Richard-son, Texas. He has been engaged in designing and analyz-ing wireless communication schemes and protocols for the3GPP LTE and LTE-Advanced standards. His research inter-ests include MIMO/multi-user/cooperative wireless commu-nications, cross-layer design, and information theory.

YOSUKE AKIMOTO ([email protected]) received hisB.E. degree from Kyoto University, Japan, in 2000, and hisM.I. degree from the Graduate School of Informatics, KyotoUniversity, in 2002. In 2007 he joined Sharp Corporation, andworked on radio access technologies for LTE and LTE-Advanced. His current research interests include system designfor uplink MU-MIMO and cooperative communications.

YOUNSUN KIM ([email protected]) received B.S. andM.S. degrees in electronic engineering from Yonsei Univer-sity, Korea, and a Ph.D. degree in electrical engineeringfrom the University of Washington in 1996, 1999, and2009, respectively. He joined Samsung Electronics in 1999,and has since been working on the standardization ofwireless communication systems such as cdma2000, HRPD,and recently LTE/LTE-A. His research interests include multi-ple access schemes, coordination schemes, multiple anten-na techniques for next generation systems.

MOON-IL LEE ([email protected]) received his M.S.degree in electrical engineering from Ajou University,Korea, in 2005. He joined InterDigital Communications towork in mobile communication technology research in2011. Prior to this he worked for LG Electronics, Korea. Hehas been actively participating in 3GPP standardization forLTE radio access technologies, primarily within the area ofadvanced antenna techniques since 2005. He is active inconcept development and 3GPP standardization of LTE-Advanced and future wireless technologies.

KAPIL BHATTAD ([email protected]) received hisB.Tech. degree in electrical engineering from the Indian Insti-tute of Technology-Madras in 2002 and his Ph.D. degree inelectrical engineering from Texas A&M University, CollegeStation, in 2008. He was a recipient of the Texas Telecommu-nications Engineering Consortium fellowship from 2002 to2003 and was a research assistant at Texas A&M Universityfrom 2003 to 2007. He has been working at Qualcomm Inc.,San Diego, California, since December 2007. During his Ph.D.he worked in the areas of iterative decoding, joint sourcechannel coding, network coding, and MIMO. At Qualcommhe has contributed to LTE prototype and modem develop-ment, and to LTE-A standardization activities, particularly inthe areas of downlink MIMO and heterogeneous networks.

ANTHONY E. EKPENYONG ([email protected]) received hisB.S. from Obafemi Awolowo University, Nigeria, and theM.S. and Ph.D. degrees from the University of Notre Dame,Indiana, all in electrical engineering. From 2006 to 2008 hewas a systems engineer with Texas Instruments San Diego,where he worked on receiver algorithms for HSPA base-band modems. Since October 2008 he has been with theCommunications Infrastructure group of Texas Instruments,Dallas, as a systems/standards engineer actively involved inthe standardization of LTE and LTE-Advanced systems, andin the development of base station receivers for LTE andHSUPA systems. His research interests include signal pro-cessing for wireless communications, multi-user communi-cations, and heterogeneous cellular networks.

For both P-SRS and

A-SRS a combination

of the frequency

comb and CS can be

used to multiplex

four APs’ SRS in

large delay spread

channels, while for

low delay spread

channels or for UEs

equipped with two

APs, only different

CS values are used

to multiplex the SRS.


reference signals for lte

Documents