Reference Signals (CRS/DMRS/SRS/CSI-RS) in LTE 4G
The always-on Cell-specific Reference Signal and the DMRS/SRS/CSI-RS that enable channel estimation and sounding.
A radio receiver can only undo what the channel did to a signal if it knows what the channel did — and it learns that from reference signals: known symbols the transmitter inserts at fixed positions so the receiver can measure amplitude and phase and estimate the channel. In LTE these pilots are also what the UE measures for cell selection and mobility, and one of them — the always-on CRS — gives LTE a heartbeat that 5G NR deliberately threw away. This page is grounded in TS 36.211.
Introduction
Reference signals (RS), or pilots, are the known symbols the LTE physical layer inserts into the resource grid so a receiver can figure out what the radio channel did to everything else. They carry no user data; their entire purpose is to be predictable, so that comparing "what I received" against "what I know was sent" yields an estimate of the channel's amplitude and phase on each subcarrier and over time.
They are used at every stage of a UE's life. Before it can demodulate a single bit of PDSCH or PUSCH the receiver needs a channel estimate from a pilot; to rank cells for reselection or handover it measures RSRP/RSRQ from the downlink CRS; to feed MIMO it measures CSI-RS; to sound the uplink band it sends SRS; and to be positioned it is measured on PRS. Different jobs, but all the same underlying idea: a known pattern that lets a receiver measure the unknown.
It matters because the whole design tension of LTE reference signalling — and the single biggest change NR made — is about when a pilot is present. LTE's CRS is always on, everywhere in the band; that makes an idle UE able to measure and a fresh UE able to demodulate instantly, but it also burns power and creates constant interference. Understanding which pilots are always-on and which are on-demand explains almost everything that follows, and why NR looks the way it does.
On this page
Why reference signals are needed
In plain words: think of a photographer shooting under strange, shifting stage lighting. Before trusting any colour in the shot, they photograph a known grey card — because they know its true colour, they can measure exactly how the lighting distorted it and correct every other pixel. Reference signals are the grey card of radio: known symbols the receiver measures the "lighting" (the channel) against, so it can correct the data it actually cares about. Some pilots are like a grey card left in the frame permanently (CRS); others are held up only when a specific shot is taken (DMRS).
Concretely, a wireless link needs pilots to do three separable jobs, and LTE builds a different reference signal for each. First, demodulation: every data or control symbol passes through a channel that rotates its phase and scales its amplitude per subcarrier and drifts over time, so the receiver must estimate that channel before it can equalise — that is what CRS and the DMRS family provide. Second, measurement: to choose cells, ranks and modulation the network needs amplitude and channel-state information — RSRP/RSRQ from CRS, and CQI/PMI/RI from CSI-RS. Third, sounding and positioning: the uplink SRS probes the band the UE is not currently using, and PRS lets the network time-of-arrival distant cells. No single pilot does all three well, which is exactly why LTE has a family of them.
Known pilot symbols at defined resource elements, split by job (demodulation vs measurement) and by owner (cell-specific vs UE-specific), across downlink and uplink.
You cannot coherently demodulate without a channel estimate, cannot compute precoding without channel state, and cannot rank cells without a measurable amplitude — pilots supply all three.
The transmitter maps a pseudo-random or Zadoff–Chu sequence onto scattered REs; the receiver regenerates it, divides it out to get the channel at those REs, and interpolates over the data in between.
What Reference Signals Are
Every LTE data or control symbol travels through an unknown radio channel that rotates its phase and scales its amplitude differently on each subcarrier and drifts over time. Before the receiver can demodulate anything, it has to estimate that channel. Reference signals (RS), also called pilots, are complex symbols whose value the receiver already knows; by comparing what it received against what it expected, it interpolates the channel across the rest of the resource grid and then equalises the data.
Known pilot symbols placed at defined resource elements in the time/frequency grid. LTE defines several kinds: CRS, downlink DMRS, uplink DMRS, SRS, PRS and (from Rel-10) CSI-RS, each serving a different job on downlink or uplink.
You cannot coherently demodulate PDSCH/PUSCH without a channel estimate, you cannot compute good precoding without channel state, and you cannot rank cells for handover without a measurable amplitude. Pilots provide all three — a demodulation reference, a measurement anchor, and a sounding probe.
The transmitter maps a pseudo-random sequence (seeded by cell identity or a UE-specific ID) onto scattered resource elements. The receiver regenerates the same sequence, divides it out to get the channel at those REs, and interpolates over the data-bearing REs in between.
The crucial distinction to hold in your head is who a pilot belongs to and when it is present. A cell-specific pilot is common to everyone in the cell and can be transmitted whether or not anyone is scheduled; a UE-specific pilot is precoded for one UE and only appears inside that UE's allocation. A pilot can also be dedicated to demodulation (it must go through the same precoder as the data it serves) or to measurement (it can be sparse and infrequent because nobody demodulates data from it). Every LTE reference signal falls somewhere in this grid, and getting those two axes right explains almost everything that follows.
The big LTE-vs-NR difference: LTE's CRS is always on — it is transmitted across the whole band in every subframe whether or not anyone is scheduled. NR deliberately removed any always-on wideband pilot (it uses SSB plus on-demand DMRS/CSI-RS) to save energy and cut interference. Understanding CRS is the single most useful thing about LTE reference signals.
CRS — Cell-specific Reference Signal
The CRS (Cell-specific Reference Signal) is the workhorse of the LTE downlink and the defining pilot of the whole system. It is transmitted on antenna ports 0–3 (a cell uses 1, 2 or 4 ports depending on configuration), spread evenly across the entire downlink bandwidth, in every single subframe. Because it is common to the cell — not tied to any one UE — every UE in the cell, connected or idle, can use it without any dedicated signalling. That universality is exactly what makes it so useful, and so expensive.
CRS does three distinct jobs at once, and it is worth separating them because NR later split them across different signals:
| Job | What CRS provides |
|---|---|
| Channel estimation | Wideband, every-subframe pilots the UE interpolates to estimate the downlink channel across all resource blocks, in both frequency and time. |
| Demodulation | The demodulation reference for the common channels and the classic transmission modes (TM1–TM6): the UE demodulates PDSCH, PDCCH, PCFICH and PHICH coherently against CRS. |
| Measurement | The signal the UE measures for RSRP, RSRQ and (together with the whole received band) RSSI — the quantities that drive cell selection, reselection and handover. |
Because CRS is always present, it gives LTE its "heartbeat": even an idle UE camped on a cell can keep measuring RSRP from CRS to decide whether a neighbour is stronger, without the network having to schedule anything. A UE arriving on a new carrier can find and measure a cell purely from CRS. And because the reference is common, a UE can start demodulating broadcast and control channels the instant it has synchronised, long before it has any UE-specific configuration. That reliability is also CRS's cost: it burns transmit power on pilots continuously, and every cell's CRS is a permanent source of inter-cell interference for its neighbours — which is exactly why NR dropped it.
The CRS sequence itself is a length-defined pseudo-random (Gold) sequence generated per OFDM symbol and initialised from the physical cell identity PCI, the slot number and the CP length, then QPSK-modulated (one complex reference symbol of ±1±j per pilot RE). Because the sequence depends on PCI, two cells with different identities produce different, effectively uncorrelated pilots — but if their pilots also land on the same subcarriers, they still collide in power. That is where the frequency shift comes in.
Frequency shift by PCI: the CRS resource elements are shifted in frequency by v-shift = PCI mod 6. Neighbouring cells given PCIs with different values of (PCI mod 6) place their CRS on different subcarriers, so their pilots do not sit on top of each other. This is why PCI planning in LTE cares about the PCI-mod-6 (and, for other signals, mod-3) relationship between neighbours: a bad plan puts strong neighbour CRS directly onto your CRS REs and wrecks channel estimation at the cell edge.
The CRS Resource-Element Pattern and Overhead
CRS lands on a regular, staggered grid. Within one resource block (12 subcarriers) and one slot (7 OFDM symbols, normal cyclic prefix), the pilots for a given port sit every 6th subcarrier, and the two OFDM symbols carrying that port's pilots are offset by 3 subcarriers from each other — producing the classic diagonal-looking pattern. Ports 0 and 1 use OFDM symbols 0 and 4 of each slot; ports 2 and 3 (only present in 4-port cells) use symbol 1. Critically, wherever one port transmits a pilot, the other ports are muted (DTX) on that resource element, so a UE estimating port 0 does not see port 1's energy leaking onto it.
A UE only ever needs a fraction of the grid for pilots, yet the estimate is dense enough to interpolate reliably because fading is correlated across nearby subcarriers and symbols. In a 1-port cell only R0 is present; a 2-port cell adds R1 (Figure 1); a 4-port cell adds R2 and R3 in symbol 1 of each slot, at half the time-density of ports 0/1 — the assumption being that the extra ports change more slowly or matter less for interpolation.
All of this pilot placement is pure overhead: every RE spent on CRS is an RE that cannot carry data. Because the pattern is fixed per port, the overhead scales cleanly with the number of ports. Counting CRS REs per resource block per subframe (a subframe has 14 OFDM symbols under normal CP, so 168 REs per RB), the cost is:
| CRS ports | CRS REs per RB per subframe | REs available per RB (normal CP) | Approx. overhead |
|---|---|---|---|
| 1 port (R0) | 8 | 168 | ~4.8% |
| 2 ports (R0, R1) | 16 | 168 | ~9.5% |
| 4 ports (R0–R3) | 24 | 168 | ~14.3% |
Two ports double the pilot cost of one; four ports do not quite double it again only because ports 2 and 3 are placed at lower time-density. Roughly one in seven resource elements in a 4-port cell is gone before any user data is sent — and that cost is paid in every subframe, everywhere in the band, forever. This single table is the strongest possible argument for why NR abandoned the always-on wideband pilot.
Downlink DMRS — UE-specific Demodulation Reference Signal
CRS is common to the cell, so it travels through the cell's generic transmit path — it cannot be beamformed toward one particular UE. When the eNB wants to apply UE-specific precoding or beamforming (transmission modes TM7 through TM10) it needs a pilot that goes through the same precoder as the data. That pilot is the downlink DMRS (Demodulation Reference Signal), also called the UE-specific reference signal.
A reference signal transmitted only within a UE's own scheduled PDSCH resource blocks and precoded identically to the data. LTE uses antenna port 5 for single-layer beamforming (TM7) and ports 7–14 for the later single-/dual-/multi-layer MIMO modes (TM8, TM9, TM10) — up to 8 layers.
Because DMRS shares the UE's precoder, the UE estimates the effective channel (physical channel × precoder) and does not need to know the precoding matrix at all. This is what makes beamforming and higher-order / "transparent" MIMO possible — the network can change its precoder freely and the UE just follows the pilot.
The eNB inserts DMRS REs inside the UE's allocation; the sequence is scrambled by a UE-specific identity (an nSCID scrambling ID) so co-scheduled MU-MIMO UEs and multiple layers stay separable via orthogonal cover codes. It is present only when data is scheduled — never always-on like CRS.
The evolution of the ports tells the story. Port 5 arrived in Rel-8 for TM7, a single beamformed layer aimed at TDD deployments with antenna arrays. Rel-9 added ports 7 and 8 for dual-layer beamforming (TM8). Rel-10 extended this to ports 7–14 for up to eight layers under TM9, and Rel-11's TM10 reused the same DMRS framework for Coordinated Multi-Point (CoMP), where several transmission points can serve one UE. In every case the principle is identical: the demodulation pilot is buried inside the data allocation and precoded with it, so the UE never has to be told the spatial processing that was applied. Layers sharing the same DMRS REs are kept apart by length-2 (or length-4) orthogonal cover codes plus the scrambling ID, which is what allows several MU-MIMO UEs to be co-scheduled on the same resource blocks.
CRS vs downlink DMRS in one line: CRS is a cell-wide, always-on, generic pilot that any UE can use for anything; downlink DMRS is a private, on-demand, precoded pilot that exists only inside one UE's grant and only serves that grant's demodulation.
CSI-RS and Zero-Power CSI-RS
For its first releases LTE derived downlink channel-state feedback (CQI/PMI/RI) from CRS, which capped the useful antenna-port count at four and forced measurement to ride on the same expensive always-on pilot. To support more antennas and cleaner measurement, Rel-10 introduced the CSI-RS (Channel State Information Reference Signal).
A sparse, configurable downlink reference signal dedicated to measurement. The UE uses it to compute its CSI report (CQI, PMI, RI). Rel-10 supports up to 8 CSI-RS antenna ports; later releases (full-dimension / massive MIMO enhancements, e.g. Rel-13/14) push this to 16, 32 and beyond.
CRS could not scale beyond 4 ports and is expensive to transmit continuously. CSI-RS decouples measurement from demodulation: it can be low-density and infrequent (configured periodicities of several to tens of milliseconds — e.g. 5, 10, 20, 40, 80 ms), freeing the design to add antenna ports for higher-order MIMO without paying always-on overhead.
The eNB configures CSI-RS resources per UE by RRC: which REs, how many ports, and what periodicity/offset. The UE measures and feeds back CSI; it does not demodulate data from CSI-RS — that remains DMRS's job in TM9/TM10.
A vital companion is Zero-Power CSI-RS (ZP-CSI-RS). Instead of transmitting a pilot, the eNB mutes a set of REs — it sends nothing there. This does two things. First, it lets one cell protect another cell's (non-zero-power) CSI-RS from interference: if cell A mutes the REs where cell B places its CSI-RS, cell B's UEs measure a clean channel. Second, those muted REs form a CSI Interference Measurement (CSI-IM) resource: because the eNB transmits nothing, whatever the UE receives there is pure interference-plus-noise, which is exactly what it needs to compute a realistic CQI. Non-zero-power CSI-RS measures the wanted signal; zero-power CSI-RS / CSI-IM measures what is fighting it.
This measurement/demodulation split is the design idea NR inherited and generalised: NR has no CRS at all and leans entirely on CSI-RS for measurement (including beam management and mobility) and DMRS for demodulation. LTE's Rel-10 CSI-RS is the prototype for that whole philosophy.
PRS — Positioning Reference Signal
Estimating a channel and measuring a cell's power are not the only things pilots are good for — a carefully designed reference signal can also let a UE work out where it is. That is the job of the PRS (Positioning Reference Signal), added in Rel-9 to support OTDOA (Observed Time Difference Of Arrival).
In OTDOA the UE measures the arrival time of downlink signals from several cells and reports the time differences between them (the Reference Signal Time Difference, RSTD). Each time difference constrains the UE to a hyperbola; intersecting several hyperbolas fixes its position. For this to work the UE must be able to hear distant cells, not just its strong serving cell — and against a strong serving cell those far pilots would normally be buried. PRS solves that with three features: a special diagonal pattern spread across the band that gives good time-of-arrival resolution; transmission in configurable positioning subframes grouped into occasions; and, crucially, the option for neighbours to mute their PRS in a coordinated pattern so a UE gets quiet windows in which to detect faint cells. PRS is a measurement-only pilot — no data is ever demodulated from it — and it is transmitted only when positioning is configured, not continuously.
Why not just reuse CRS? Early positioning did use CRS, but CRS from far cells is drowned by the serving cell and its pattern is not ideal for timing. PRS adds bandwidth-spanning coverage, higher effective SINR through muting, and longer coherent measurement occasions — all aimed at hearing weak, distant cells accurately.
Uplink Reference Signals — DMRS and SRS
Everything so far has been downlink except in passing. The uplink has its own two pilots, and both are built from Zadoff–Chu-based sequences chosen for their low peak-to-average power ratio, which matches the single-carrier (SC-FDMA / DFT-spread-OFDM) uplink waveform and keeps the UE's power amplifier efficient.
Uplink DMRS is the demodulation pilot for the uplink channels. PUSCH carries DMRS in the middle symbol of each slot (symbol 3 under normal CP) so the eNB can estimate the channel and demodulate the data; PUCCH carries its own DMRS whose placement depends on the PUCCH format. Different UEs and layers are separated by cyclic shifts of the base Zadoff–Chu sequence (and, for MU-MIMO, orthogonal cover codes). Like downlink DMRS it is present only where the UE is actually transmitting, and it covers only those resource blocks.
That last point is the limitation the SRS (Sounding Reference Signal) exists to fix. Uplink DMRS only covers the RBs a UE is currently using, so it tells the eNB nothing about the parts of the band the UE is not using. To schedule the uplink intelligently the eNB wants a wideband view of each UE's uplink channel — and that is what SRS provides.
Mental model: uplink DMRS is a pilot for "the data I'm sending right now"; SRS is a pilot for "the channel across the band, so you can decide where to schedule me next."
The UE transmits SRS (periodically or aperiodically, as configured) over a configured bandwidth, normally in the last OFDM symbol of a subframe, and the eNB uses it for two big purposes:
- Frequency-selective uplink scheduling: knowing which subcarriers are good for this UE, the eNB allocates
PUSCHon the best resource blocks and picks a suitable MCS — the uplink equivalent of frequency-selective downlink scheduling. - TDD downlink via reciprocity: in TDD the uplink and downlink share the same carrier frequency, so the channel measured from SRS on the uplink also describes the downlink. The eNB can compute downlink beamforming/precoding weights directly from SRS — no downlink CSI feedback needed. This is a major reason TDD LTE and TDD massive-MIMO systems lean so heavily on SRS.
Putting Them All Together
Each reference signal answers a different question: demodulate the cell's common channels, demodulate a beamformed stream, measure downlink CSI, position the UE, demodulate the uplink, or sound the uplink band. Laid side by side, the whole family fits on two axes — direction, and demodulation versus measurement.
| Signal | Direction | Antenna ports | Purpose | Always on? |
|---|---|---|---|---|
CRS | Downlink | 0–3 | Channel estimation, demodulation (TM1–TM6), RSRP/RSRQ measurement | Yes — every subframe, whole band |
DL DMRS | Downlink | 5, 7–14 | Demodulation of beamformed/precoded PDSCH (TM7–TM10) | No — only inside a UE's grant |
CSI-RS | Downlink | up to 8 (Rel-10) | CSI measurement (CQI/PMI/RI); ZP-CSI-RS / CSI-IM for interference | No — configured, sparse |
PRS | Downlink | positioning | OTDOA positioning (RSTD), with neighbour muting | No — positioning occasions only |
UL DMRS | Uplink | UE transmit | Demodulation of PUSCH and PUCCH | No — only when transmitting |
SRS | Uplink | UE transmit | Sound UL channel → frequency-selective UL scheduling; TDD DL via reciprocity | No — periodic/aperiodic, as configured |
Notice that only one row says "yes" in the last column. That single always-on entry, CRS, is the whole personality of LTE reference signalling: it is what makes an idle UE able to measure, a fresh UE able to demodulate immediately, and a network able to run TM1–TM6 without any per-UE pilot configuration — and it is also what makes LTE spend power and generate interference even in an empty cell. NR looked at this table, kept the ideas in every "no" row (on-demand DMRS for demodulation, sparse CSI-RS for measurement, SRS for uplink sounding), and simply deleted the "yes" row, replacing an idle UE's measurement anchor with the periodic SSB. That one design choice — no always-on wideband pilot — is the cleanest one-sentence summary of how NR's physical layer differs from LTE's.
LTE ↔ NR: NR carries over the demodulation/measurement split but removes CRS entirely. NR demodulates PDSCH/PDCCH/PBCH with front-loaded DMRS, measures the channel and manages beams with periodic/semi-persistent/aperiodic CSI-RS (plus a Tracking Reference Signal, TRS, for fine time/frequency tracking), gives idle-mode UEs the periodic SSB as their measurement anchor (SS-RSRP), and keeps SRS uplink sounding (now also usable for antenna switching and codebook/non-codebook precoding). LTE's PRS concept reappears as NR's own DL-PRS and UL SRS-for-positioning.
Summary
LTE reference signals sort cleanly onto two axes: who owns the pilot (cell-specific vs UE-specific) and what it is for (demodulation vs measurement). CRS (ports 0–3, PCI-keyed, shifted by PCI mod 6) is the cell-wide, always-on pilot that does channel estimation, demodulation for TM1–TM6, and RSRP/RSRQ measurement — costing roughly 5/10/14% overhead for 1/2/4 ports. The DMRS family (DL ports 5 and 7–14; UL with PUSCH/PUCCH) is the on-demand, precoded demodulation pilot that makes transparent beamforming and higher-order MIMO possible. CSI-RS (Rel-10, up to 8 ports and beyond) is the sparse measurement pilot, with Zero-Power CSI-RS / CSI-IM measuring interference. PRS serves OTDOA positioning, and uplink SRS sounds the wider band for scheduling and TDD reciprocity.
The single fact that ties the whole page together is the "always on?" column: only CRS answers yes, and that one property — a cell-wide wideband pilot present in every subframe — is both LTE's greatest convenience and its permanent power/interference tax. NR kept every on-demand idea and deleted the always-on one, leaning on SSB for idle measurement and DMRS/CSI-RS for the rest. If you remember one thing, remember why that "yes" row exists and why NR removed it.
Quick Q&A
Q. Why is CRS transmitted even when no UE is scheduled, and why did NR drop it?
A. CRS is cell-common and does triple duty — channel estimation, demodulation of the common channels, and RSRP/RSRQ measurement — so it must be present continuously across the whole band for any UE, including idle ones measuring for mobility. That always-on transmission wastes energy and creates constant inter-cell interference, so NR removed it and instead uses on-demand DMRS for demodulation and CSI-RS/SSB for measurement.
Q. What determines where CRS lands in frequency, and why does it matter for planning?
A. The CRS pattern is shifted by v-shift = PCI mod 6. Neighbour cells assigned PCIs that differ in (PCI mod 6) place their pilots on different subcarriers, avoiding CRS-on-CRS collisions — which is why PCI planning considers the mod-6 and mod-3 relationships between adjacent cells.
Q. Why introduce CSI-RS when CRS already existed, and what is Zero-Power CSI-RS for?
A. CRS caps the antenna-port count at four and is expensive because it is always on. CSI-RS separates measurement from demodulation, allowing up to 8 ports (more later) with sparse, configurable transmission. Zero-Power CSI-RS mutes REs — either to protect a neighbour's CSI-RS or, as CSI-IM, to let the UE measure pure interference-plus-noise for an accurate CQI.
Q. What is the difference between uplink DMRS and SRS?
A. Uplink DMRS is the demodulation pilot inside the RBs the UE is actually transmitting on, so the eNB can decode PUSCH/PUCCH. SRS is a wideband sounding pilot covering parts of the band the UE is not currently using, so the eNB can do frequency-selective uplink scheduling and, in TDD, compute downlink precoding by reciprocity.
Q. How is DMRS able to demodulate a beamformed signal the UE was never told about?
A. DMRS is precoded with exactly the same weights as the data, so the UE estimates the effective channel (physical channel × precoder) and equalises against it. It never needs to know the precoding matrix — the pilot already carries its effect. That is why LTE's beamforming modes (TM7–TM10) use DMRS rather than CRS.
Where reference signals connect
Reference signals are the foundation under demodulation, spatial multiplexing, positioning and the measurements that steer the network. From here, follow the threads into the channels and features they enable — and into how NR reworked all of this once it removed the always-on pilot.