5G NR Downlink Reference Signals: DM-RS, CSI-RS, TRS and PT-RS Without the Headache
What each downlink reference signal is for, where it sits, and how NR replaced LTE's always-on CRS with four purpose-built signals: DM-RS (for PBCH, PDCCH and PDSCH), CSI-RS, TRS and PT-RS.
The one-line version
LTE had one reference signal that did everything (the CRS) and was on all the time. NR split that job into four signals, each transmitted only when needed and each answering one question:
| Signal | Question it answers | Sent when |
|---|---|---|
| DM-RS | What does the channel look like for this transmission, with this precoder? | Inside every PBCH, PDCCH and PDSCH |
| CSI-RS | How good is the channel on each beam / antenna port, so the UE can report CQI, PMI, RI, beam quality? | Periodically, semi-persistently, or on demand |
| TRS | Exactly what time and frequency offset should the UE lock to? | Periodically (a special CSI-RS) |
| PT-RS | How is the phase drifting within this PDSCH because of oscillator noise? | Inside PDSCH, mostly FR2, high MCS |
If you keep the question in mind, the placement and density of each signal become obvious.
The mental model: the measuring tape vs the weather station
- DM-RS is a measuring tape you send with the parcel. It goes through exactly the same precoder/beam as the data, so what the UE measures is the effective channel it needs to undo. It is useless for anyone else and for any other transmission — which is precisely why it is cheap: no data, no DM-RS.
- CSI-RS is the weather station. It is not precoded like the data (or is precoded per beam), so it tells the UE about the raw channel or a set of candidate beams, which the UE reports back so the gNB can choose the next precoder/beam.
- TRS is the clock on the wall — a stable, known signal the UE uses to keep its timing and frequency locked.
- PT-RS is a tiny level-check in the corner of the parcel: not for the channel, but for the phase wobble of the oscillators.
DM-RS for PBCH
Covered in detail in the PBCH article; the essentials:
- Every 4th subcarrier in the three PBCH symbols of the SSB (symbols 1, 2, 3), shifted by v = PCI mod 4 to avoid collisions between neighbours.
- Gold sequence seeded by PCI, the half-frame bit and the 2 or 3 LSBs of the SSB index — so decoding the DM-RS reveals the beam index.
- 144 of the 576 PBCH REs (25 %). Dense on purpose: the UE has no other channel knowledge at this point.
DM-RS for PDCCH
- Present in every REG of a PDCCH candidate: subcarriers 1, 5, 9 of each RB in each CORESET symbol (3 of 12 REs, 25 %).
- If the CORESET's
precoderGranularityis allContiguousRBs, the DM-RS is present in every REG of the CORESET (not just the candidate), so the UE can estimate across the whole CORESET bandwidth. - Gold sequence seeded by
pdcch-DMRS-ScramblingID(or the PCI for CORESET#0 and when not configured). Using a configurable ID lets different CORESETs / TRPs use different sequences. - Same precoder is assumed across a REG bundle, which is why bundle size matters for estimation quality.
DM-RS for PDSCH — the one you need to understand well
This is the most configurable reference signal in NR, because PDSCH varies so much: 1 to 8 layers, 2 to 14 symbols, static or fast-moving UEs. Every configuration knob answers one of three questions.
Question 1: How many antenna ports do I need? → Type 1 vs Type 2
DM-RS ports are multiplexed by CDM groups (a set of REs shared by ports that are separated by orthogonal cover codes).
| Type 1 | Type 2 | |
|---|---|---|
| Structure in frequency | Comb: a CDM group uses every 2nd subcarrier (6 REs per RB) | Pairs: a CDM group uses 2 adjacent subcarriers, twice per RB (4 REs per RB) |
| CDM groups per RB | 2 | 3 |
| Ports, single-symbol | 4 (2 per group, FD-OCC) | 6 (2 per group, FD-OCC) |
| Ports, double-symbol | 8 (adds TD-OCC) | 12 (adds TD-OCC) |
| DM-RS REs per RB per port | 6 | 4 |
| Best for | Fewer users, better channel estimation per port (denser) | MU-MIMO with many users/layers |
Rule of thumb: Type 1 for quality, Type 2 for quantity. Type 1 is the default and what DCI 1_0 always assumes.
Question 2: Where in time does the first DM-RS go? → Mapping type A vs B
- Mapping type A (slot-based): the first DM-RS symbol is at symbol 2 or 3 of the slot (chosen by
dmrs-TypeA-Positionin the MIB), regardless of where the PDSCH starts. Designed for full-slot allocations that follow a PDCCH. - Mapping type B (mini-slot): the first DM-RS symbol is the first symbol of the PDSCH allocation ("front-loaded"). Designed for short, low-latency transmissions anywhere in the slot: the UE can start channel estimation immediately.
The DM-RS is front-loaded in both cases in the sense that it comes as early as possible — the UE needs the estimate before it can demodulate the data symbols. LTE's CRS was spread through the subframe; NR wants low latency, so the estimate must be available early.
Question 3: How fast is the channel changing? → Additional positions and double symbols
dmrs-AdditionalPosition∈{0, 1, 2, 3}adds up to three more DM-RS symbols spread across the allocation (positions from TS 38.211 Table 7.4.1.1.2-3/4, depending on allocation length). A static UE needs only the front-loaded symbol; a high-speed UE needs several so the UE can track the changing channel across the slot.maxLength= 2 allows a double-symbol DM-RS (two adjacent symbols) — this is what enables the 8 / 12 port counts, and also improves estimation SNR.
The sequence and the DCI
- Sequence: a Gold sequence seeded by one of two configurable scrambling IDs (
scramblingID0/1), selected by the n_SCID bit in DCI 1_1 (DM-RS sequence initialisation). Two IDs let neighbouring cells or TRPs use different sequences for the same UE. - The antenna ports field in DCI 1_1 tells the UE which DM-RS ports (hence which layers) are its own, and how many CDM groups are "without data" — i.e. reserved for other UEs' DM-RS in MU-MIMO. The UE rate-matches its data around those groups even though it does not use them. This is the mechanism that lets two UEs share the same PRBs.
Why NR does it this way (vs LTE)
LTE's CRS was cell-specific, unprecoded and always on. That has three costs NR wanted to eliminate: constant power and interference even in empty cells, no way to beamform the reference signal with the data, and dense overhead regardless of need. NR's DM-RS is UE-specific and precoded exactly like the data, so the gNB may use any precoder (non-codebook, reciprocity-based, anything) and the UE never needs to know it. This "transparent precoding" is the reason NR can beamform every channel.
CSI-RS — the flexible measurement signal
Purpose. A known, unprecoded (or per-beam precoded) signal on 1 to 32 antenna ports, so the UE can measure the channel and report CSI: CQI, PMI, RI, LI (for link adaptation and precoder choice), CRI / L1-RSRP (for beam management), and CSI-RSRP/RSRQ/SINR for L3 mobility. It also serves as the base for TRS (below) and, in zero-power form, for interference measurement and rate matching.
Configuration (TS 38.211 §7.4.1.5). A CSI-RS resource is defined by:
- Number of ports: 1, 2, 4, 8, 12, 16, 24 or 32 — Table 7.4.1.5.3-1 lists 18 "rows", each a template for how the ports' REs are placed within one RB and one slot.
- CDM type: noCDM, FD-CDM2, CDM4 (FD2-TD2), CDM8 (FD2-TD4) — how many ports share each RE group.
- Density: 1 (one RE per port per RB), 0.5 (every other RB — even or odd), or 3 (three REs per RB, only for the single-port row used by TRS).
- Frequency range: start RB and number of RBs.
- Time: which symbol(s) in the slot, and — for periodic / semi-persistent resources — a periodicity (4 to 640 slots) and slot offset.
Three transmission types:
| Type | Activated by | Typical use |
|---|---|---|
| Periodic | RRC | L3 mobility measurements, TRS, always-available beam measurement |
| Semi-persistent | MAC CE | Beam management for a connected UE, switched on/off cheaply |
| Aperiodic | DCI (CSI request in 0_1) | One-shot precise CSI before scheduling a big transmission |
Zero-power CSI-RS (ZP-CSI-RS) is a resource where the gNB transmits nothing. The UE uses the empty REs to measure interference (CSI-IM), and the PDSCH is rate matched around them — which is also how a gNB keeps its own PDSCH away from a neighbour's CSI-RS.
CSI-RS in beam management. A resource set can be marked repetition = on (same beam on every resource: lets the UE sweep its receive beams) or off (different beam per resource: lets the UE compare transmit beams and report the best CRI + L1-RSRP). The SSB does coarse beam selection; CSI-RS does the fine refinement.
TRS — the tracking reference signal
Purpose. NR removed the CRS that LTE UEs used to keep time and frequency synchronised while connected. The SSB is too sparse (tens of ms apart) for fine tracking. So NR defines a special CSI-RS configuration for time/frequency tracking, delay-spread and Doppler estimation — the source of QCL Type A properties.
Configuration (TS 38.214 §5.1.6.1.1). An NZP-CSI-RS resource set with trs-Info = true:
- 1 port, density 3 (three REs per RB, spaced 4 subcarriers apart) — dense in frequency for accurate delay-spread and timing estimation.
- Two symbols per slot, 4 symbols apart (e.g. symbols 4 & 8, 5 & 9 or 6 & 10), in two consecutive slots (FR1) — the 4-symbol spacing gives an unambiguous frequency-offset estimate; the slot-to-slot spacing gives fine Doppler.
- Bandwidth: the whole BWP, or 52 RBs if the BWP is wider.
- Periodicity: 10, 20, 40 or 80 ms.
A UE must be configured with a TRS before it can rely on QCL Type A for PDSCH/PDCCH — in practice every connected UE has one.
PT-RS — phase tracking
The problem. At mmWave, the oscillators in both gNB and UE have significant phase noise: the phase of the received constellation rotates within a slot, differently from symbol to symbol. DM-RS is estimated once (or a few times) per slot, so it cannot follow this rotation. For QPSK you might not care; for 64QAM/256QAM a few degrees of rotation causes errors.
The solution (TS 38.211 §7.4.1.2). A sparse reference signal on one subcarrier every K RBs, transmitted every L symbols from the first DM-RS symbol onward, precoded identically to one DM-RS port so the UE can compare it against that port's estimate and track the common phase error.
- Time density L_PT-RS ∈
{1, 2, 4}— chosen by the scheduled MCS against three RRC thresholds (ptrs-MCS1/2/3): higher MCS → denser in time. - Frequency density K_PT-RS ∈
{2, 4}— chosen by the scheduled bandwidth against two thresholds (N_RB0,N_RB1): wider allocation → sparser in frequency (one sample per 4 RBs is enough). - Not present at all if the MCS is below
ptrs-MCS1or the bandwidth is belowN_RB0, or if not configured — so in FR1 with modest MCS you typically never see it. - Associated with the lowest-indexed scheduled DM-RS port; downlink uses at most one PT-RS port.
Why not just denser DM-RS? DM-RS estimates the full channel per subcarrier — expensive. Phase noise is a common rotation across all subcarriers, so a single subcarrier per few RBs, sampled every symbol, is enough to track it. PT-RS is the cheapest possible way to measure that one number per symbol.
Putting the four together for one PDSCH in FR2
- The UE has been tracking time/frequency on the TRS (periodic, 1 port, density 3).
- It measured beams on CSI-RS and reported the best CRI; the gNB set a TCI state pointing to that CSI-RS.
- DCI 1_1 schedules a 2-layer, 256QAM PDSCH with mapping type A: DM-RS Type 1, double-symbol, one additional position because the UE is moving.
- Because MCS is high and the allocation is wide, PT-RS with L = 1, K = 4 rides along on one subcarrier per 4 RBs, every symbol.
- The UE: locks timing (TRS) → points the Rx beam (QCL-D from the CSI-RS) → estimates the effective channel (DM-RS) → corrects per-symbol phase (PT-RS) → demodulates.
Common confusions
- "DM-RS is cell-specific." In NR it is UE-specific and precoded with the data; only the PBCH DM-RS is cell-specific.
- "Type 1/2 is the same thing as mapping type A/B." No. Type 1/2 = frequency structure and port count. Mapping type A/B = where the first symbol sits in time. Any combination is allowed.
- "TRS is a separate signal." It is a CSI-RS resource set with a flag; there is no TRS in TS 38.211's signal list.
- "PT-RS is present in every PDSCH." Only when configured and MCS and bandwidth exceed the thresholds; rarely in FR1.
- "CSI-RS is only for CQI." It is also the beam-management, tracking, interference-measurement and L3-mobility signal.
- "Where is the CRS?" There is none; that is the whole point.
Spec pointers
- TS 38.211 §7.4.1.1 — PDSCH DM-RS (Types 1/2, mapping A/B, positions, sequences)
- TS 38.211 §7.4.1.2 — PT-RS; §7.4.1.3 — PDCCH DM-RS; §7.4.1.4 — PBCH DM-RS; §7.4.1.5 — CSI-RS
- TS 38.214 §5.1.6 — DM-RS/PT-RS/CSI-RS reception procedures, TRS configuration; §5.2 — CSI reporting
- TS 38.331 — DMRS-DownlinkConfig, PTRS-DownlinkConfig, NZP-CSI-RS-Resource, CSI-ResourceConfig
- TS 38.215 — CSI-RSRP/RSRQ/SINR definitions
Recap
Four signals, four questions. DM-RS travels with each channel, precoded like the data, so the UE sees the effective channel (Type 1/2 for ports, mapping A/B for timing, additional positions for speed). CSI-RS is the flexible, up-to-32-port measurement signal for CSI, beams and mobility. TRS is a dense single-port CSI-RS that replaces the CRS for time/frequency tracking. PT-RS is a sparse per-symbol phase check for high-order modulation at mmWave. Together they do what LTE's CRS did — but only where and when they are needed, and beamformed.
Interview questions
The questions below are drawn from this topic and phrased the way they tend to come up. Try each one out loud before revealing the answer.
18 questions
- DM-RS for demodulating a specific transmission, CSI-RS for channel state and beam measurement, TRS for time and frequency tracking, and PT-RS for phase noise compensation.
- DM-RS is tied to a transmission and is precoded exactly like it, so it reveals the effective channel. CSI-RS is transmitted independently to measure the channel before any decision is made.
- CRS was transmitted continuously across the whole band whether or not there was traffic, which wastes energy and creates constant inter-cell interference. NR made every reference signal on-demand.
- Far lower idle-cell emissions and interference, at the cost that the UE cannot measure the channel whenever it likes — it needs configured CSI-RS.
- Type 1 supports up to 8 ports with a comb structure; type 2 supports up to 12 ports with a different pattern. Type 2 trades density for more ports, which matters for multi-user MIMO.
- Type A places the first DM-RS symbol near the start of the slot, matching slot-based scheduling. Type B places it at the start of the allocation, which is what mini-slots need.
- Tracking a fast-changing channel. A single DM-RS symbol at the start becomes stale by the end of the slot at high speed, so up to three more can be added.
- Every DM-RS symbol displaces data. It is a direct exchange of throughput for channel-estimation accuracy under mobility.
- Doubling the ports available, which supports higher-order MIMO, at the cost of another symbol of overhead.
- A tracking reference signal — in the specification a specific CSI-RS resource set configured for tracking. It gives the UE a periodic, predictable signal for correcting time and frequency drift.
- Phase-tracking reference signal, compensating for phase noise. It matters at mmWave and at high modulation orders, where phase noise otherwise closes the constellation.
- The scheduled MCS sets the time density and the scheduled bandwidth sets the frequency density, so it scales with how vulnerable the transmission actually is.
- Multiple CSI-RS resources are transmitted on different beams, and the UE reports which is strongest — the same role the SSB plays during initial access, at finer granularity.
- Marking resource elements the PDSCH must rate match around, typically because a neighbouring cell is transmitting its own CSI-RS there.
- So the channel the UE estimates is the effective channel including the precoder. It never needs to know what precoding was applied, which is the whole point of the antenna port abstraction.
- From the DCI, which indicates the antenna port set alongside the rest of the scheduling assignment.
- DM-RS additional positions. A single front-loaded symbol goes stale over a slot at speed, so the estimate is fine at the start and wrong by the end.
- DM-RS travels with a transmission so it can be demodulated, CSI-RS is transmitted so a decision can be made, TRS keeps the clock honest, and PT-RS keeps the constellation from rotating.
Related Articles
5G NR Waveforms: CP-OFDM and DFT-s-OFDM, and When the Uplink Should Use Which
Why NR uses CP-OFDM everywhere but keeps DFT-s-OFDM as an uplink option, what PAPR really costs, and how the network chooses per UE.
5G NR Transmitter and Receiver Chain: Following the Bits From MAC to Antenna and Back
A block-by-block walk through the NR physical-layer transmit chain and its mirror-image receiver, explaining what each stage does and why it sits where it does.
5G NR Synchronisation Signals (PSS and SSS): How a Phone Finds a Cell From Nothing
PSS and SSS explained from the UE's point of view: what each signal gives the receiver, how the PCI is built from them, and why NR has 1008 cell identities.