Reference Signals (DMRS/PTRS/SRS/CSI-RS) in 5G NR

Introduction

A radio channel does violent things to a signal on its way from transmitter to receiver: it rotates the phase, scales the amplitude, and does both differently on every subcarrier and every fraction of a millisecond. A receiver cannot undo any of that blind — it has to measure the channel first. Reference signals are how it measures: known sequences the transmitter injects at agreed positions, so the receiver can compare "what I know was sent" against "what I received" and solve for what the channel did.

In NR these signals are everywhere in the lifecycle but almost never wasted. When a UE first powers on, PSS/SSS let it find the cell; PBCH-DMRS lets it read the MIB. Once connected, DMRS demodulates each scheduled transmission, CSI-RS and SRS feed the scheduler's link-adaptation and beamforming loops, and PT-RS keeps high-order modulation alive at mmWave. Unlike LTE, none of these are transmitted "just in case" — they appear only when there is a transmission or a measurement that needs them.

This page walks the five families — DM-RS, CSI-RS, PT-RS, SRS, and the synchronization signals PSS/SSS — explaining what each measures, how it is configured, and where it sits in the grid. The demodulation reference signal has its own deeper treatment on the DMRS page.

Why NR killed the always-on CRS

In LTE the Cell-specific Reference Signal (CRS) was transmitted continuously across the whole bandwidth in (almost) every subframe, on up to four antenna ports, regardless of load. It was convenient — every UE could always measure the channel and demodulate — but it was also a permanent tax: constant energy on the air (bad for interference and for gNB sleep), a fixed pilot pattern that could not follow a narrow beam, and overhead that did not scale with actual traffic.

NR throws that model out. There is no cell-wide always-on pilot. Instead the channel is measured with signals that are transmitted only when needed, sit only where a transmission lives, and are precoded with the same beam as the data. The consequences ripple through the whole design: the gNB can go into micro-sleep between bursts, mmWave beams can carry their own pilots, and overhead tracks the schedule rather than the clock.

DM-RS — demodulation reference signal

DM-RS is the pilot bonded to a specific physical channel — there is a distinct DM-RS for PDSCH, PUSCH, PDCCH, PBCH and PUCCH. Its only job is to let the receiver estimate the channel for that exact transmission so the data can be coherently demodulated. Because it is precoded with the same precoder/beam as the data it accompanies, the receiver never needs to know the precoder — it estimates the effective end-to-end channel directly. This "transparent precoding" is precisely what the LTE CRS could not do.

Front-loaded. NR DM-RS is placed early in the allocation. For PDSCH/PUSCH mapping type A the first symbol is governed by dmrs-TypeA-Position (symbol 2 or 3, signalled in the MIB); for mapping type B it sits on the first symbol of the allocation. Front-loading means the receiver can start channel estimation and pipeline decoding without buffering the whole slot — the enabler for low latency.

Additional positions. For fast-changing channels (high Doppler) one or more additional DM-RS symbols are inserted later in the slot via dmrs-AdditionalPosition (0..3, up to 3 extra), so the estimate tracks the channel as it drifts. The scheduler trades pilot overhead against mobility: many pilots for a bullet-train UE, one for a stationary one.

Config type 1 vs type 2. dmrs-Type selects the frequency comb pattern and how many orthogonal ports fit in one DM-RS symbol. Type 1 uses a comb-2 pattern (alternate subcarriers) and supports up to 4 ports on one front-loaded symbol, 8 across two. Type 2 uses grouped subcarriers and supports up to 6 ports on one symbol, 12 across two — more MU-MIMO layers at the cost of denser packing. Ports are separated by CDM groups: within a group, ports are kept orthogonal by Orthogonal Cover Codes (OCC) in frequency and, when a second symbol is used, in time; different CDM groups sit on different subcarriers. The DCI field Antenna Port(s) tells the UE which ports and how many CDM groups are in use — REs in an unused CDM group can be left blank so the UE knows they carry no data. DM-RS antenna ports are numbered from p = 1000 for PDSCH.

CSI-RS — channel state, beam management, and tracking

CSI-RS is the downlink measurement signal. The gNB transmits it so the UE can measure the downlink channel and report what modulation, rank and beam to use. It comes in two flavours and one specialised sub-role.

NZP-CSI-RS (Non-Zero-Power). The workhorse. Configured by NZP-CSI-RS-Resource and grouped into resource sets, it carries actual pilot energy and serves two purposes. (1) CSI acquisition: the UE measures it and returns CQI/PMI/RI/LI so the gNB can pick MCS, precoder and rank (see CSI Feedback). (2) Beam management: multiple NZP-CSI-RS resources, each swept on a different Tx beam, let the UE measure L1-RSRP per beam and report the best one for refinement beyond the coarse SSB beams (see Beam Management). It can be configured with 1–32 ports and flexible density (0.5, 1 or 3 REs/RB/port). Like SRS, it can be periodic, semi-persistent (RRC-configured, MAC-CE activated) or aperiodic (triggered by a DCI field).

ZP-CSI-RS (Zero-Power). A resource that carries no energy — it is a hole. Its job is rate-matching: the gNB tells the UE "PDSCH is not mapped onto these REs," so the UE's decoder skips them. This is how one cell's PDSCH avoids colliding with another cell's (or its own) NZP-CSI-RS, or with any reserved resources. Configured via ZP-CSI-RS-Resource and pointed to by the PDSCH rate-match configuration.

TRS (Tracking Reference Signal). Not a separate signal type but a specially configured set of NZP-CSI-RS resources — trs-Info flags the resource set as a tracking set. It is a periodic burst (typically 4 resources across 2 consecutive slots, wideband, low density) that gives the UE a fine time and frequency reference to keep its oscillator and FFT timing locked between the sparse SSB occasions. Think of it as the lightweight replacement for the frequency-tracking role the ever-present LTE CRS used to provide for free.

PT-RS — phase-tracking reference signal

At high carrier frequencies (FR2 / mmWave) the oscillators introduce phase noise that rotates the whole constellation over the slot; at high modulation orders (64/256-QAM) even a small common rotation smears neighbouring points into each other. PT-RS targets exactly this: a set of tones that are dense in time but sparse in frequency, which the receiver uses to measure and undo the Common Phase Error (CPE) symbol by symbol.

It is always associated with a PDSCH or PUSCH and shares that channel's DM-RS for its starting reference. Its time density (present every 1, 2 or 4 symbols) is tied to the scheduled MCS, and its frequency density (every 2nd or 4th RB) is tied to the scheduled bandwidth — both mapped from thresholds configured in PTRS-DownlinkConfig/PTRS-UplinkConfig per TS 38.214. It is enabled only when it pays off, so low-MCS or FR1 transmissions usually skip it entirely.

There is a second, distinct use: for the uplink DFT-s-OFDM (single-carrier) waveform, PT-RS is inserted before the DFT spreading and helps track phase noise on the transform-precoded signal. So PT-RS shows up both for CP-OFDM at FR2 and for DFT-s-OFDM regardless of band.

SRS — sounding reference signal and its usages

SRS is transmitted by the UE on the uplink so the gNB can measure the uplink channel across a wide band — even where the UE currently has no data to send. It is the uplink counterpart to CSI-RS, but it does far more than uplink CSI. Each SRS-Resource lives in a SRS-ResourceSet whose usage field declares what the sounding is for. An SRS resource can span 1, 2 or 4 consecutive OFDM symbols and use 1, 2 or 4 antenna ports.

Reciprocity, the headline use. In TDD uplink and downlink share the same frequency, so the uplink channel the gNB measures from SRS also predicts the downlink channel. That lets the gNB compute massive-MIMO downlink precoders directly, without waiting for a heavy CSI feedback loop — the reason SRS is central to massive-MIMO beamforming.

Time behaviour. Like CSI-RS, SRS can be periodic (fixed period/offset, always on), semi-persistent (configured by RRC, then activated/deactivated by a MAC CE), or aperiodic (a single shot triggered by a field in DCI). Aperiodic sounding is the cheapest — the gNB asks for a sounding exactly when it needs fresh channel knowledge. SRS is placed in the last 6 symbols of a slot and can hop across the band over successive transmissions (frequency hopping controlled by b-SRS/b-hop) to cover more bandwidth than fits in one shot.

PSS / SSS — synchronization signals in the SSB

Before any of the above can be measured, a UE has to find the cell in time, frequency and identity. That is the job of the two synchronization signals, PSS (Primary) and SSS (Secondary), which together with PBCH and its DM-RS form the SSB (Synchronization Signal Block) — 4 OFDM symbols by 240 subcarriers (see SSB & Cell Search).

PSS is an m-sequence carrying one of 3 values; the UE correlates against it to grab OFDM symbol timing and coarse frequency, and to learn N_ID⁽²⁾ (0–2). SSS is a Gold sequence carrying one of 336 values (N_ID⁽¹⁾, 0–335); combining the two yields the Physical Cell ID, PCI = 3 × N_ID⁽¹⁾ + N_ID⁽²⁾, one of 1008. The UE also measures SS-RSRP/SS-RSRQ/SS-SINR on these signals for cell selection and mobility.

Crucially, SSBs are beam-swept — the gNB sends a burst of up to 4 SSBs below 3 GHz, 8 in FR1 above 3 GHz, or 64 in FR2, each on a different beam, so the UE finds not just the cell but the best coarse beam. This "beam a signal at a time" idea is the ancestor of the finer CSI-RS and SRS beam management above, and it exists precisely because there is no always-on omni CRS to lean on. The SSB is periodic (default 20 ms for initial access, configurable 5–160 ms) and its measurements underpin Xn handover decisions too.

Cheat sheet & where each signal sits in the grid

The four data-plane families plus the sync signals, side by side, then a picture of how DM-RS and CSI-RS occupy a slot.

The figure below shows a single BWP over one 14-symbol slot: the front-loaded plus additional DM-RS columns for a PDSCH, and a scattered CSI-RS resource that the UE measures separately.

Summary

NR's reference signals are the whole toolbox that replaced LTE's single always-on CRS, and each is specialised for one job. DM-RS demodulates — a private pilot per channel, precoded with the data, front-loaded for latency, with additional positions for Doppler and config types 1/2 setting the MU-MIMO ceiling. CSI-RS measures the downlink (NZP for CSI/beam/TRS; ZP for rate-matching). PT-RS tracks phase noise at FR2 and high MCS and stabilises DFT-s-OFDM. SRS sounds the uplink and — in TDD — hands the gNB the downlink channel by reciprocity. PSS/SSS in the beam-swept SSB are the entry point: time, frequency, and the 1008-value PCI.

The unifying theme is on-demand, beamformed, minimal-overhead pilots. Nothing is transmitted "just in case"; each signal appears only where and when a transmission or a measurement needs it, which is exactly what lets NR run a lean carrier, sleep between bursts, and beamform every pilot with its data.

Quick Q&A