5G NR Antenna Ports and Quasi Co-Location: What the UE Is Really Being Told

Two badly named concepts

Both names mislead. An antenna port has nothing to do with a physical connector, and quasi co-location has nothing to do with where antennas are. Once you replace the names with what they mean, the concepts are simple:

• An antenna port = "a channel you can learn from one reference signal."
• Two ports are QCL = "what you learned about the channel on port 1 also applies to port 2."

Antenna ports: the mental model

Suppose you receive a symbol on some resource element and you want to demodulate it. You need to know the channel it went through. The only way to learn a channel is to receive a known reference signal through the same channel. So the spec defines (TS 38.211 §4.4.1):

An antenna port is defined such that the channel over which a symbol on the antenna port is conveyed can be inferred from the channel over which another symbol on the same antenna port is conveyed.

In practice: a DM-RS port and the data symbols it accompanies share one antenna port — same precoder, same beam, same physical path. Whether that "port" is one antenna element, 64 elements beamformed together, or two panels combined, the UE neither knows nor cares. It just knows: "estimate on this DM-RS, apply to these data REs".

This is why NR ports are numbered by signal type rather than by hardware:

Two consequences that trip people up:

1. Different signals live on different ports by definition, even if they are physically transmitted identically. The SSB is port 4000; a CSI-RS is port 3000+. The spec never assumes you can use the SSB's channel estimate for a CSI-RS unless it tells you so. That "unless" is QCL.
2. The number of PDSCH antenna ports = the number of layers. A 4-layer transmission has 4 DM-RS ports. The gNB's 64 physical antennas are invisible.

Why LTE needed less of this

LTE's CRS was on fixed ports 0–3, always on, unprecoded. A UE could estimate timing, frequency, Doppler and delay spread from CRS and apply it to everything. NR removed the CRS, replaced it with several sparse, beamformed signals — and each of those is on its own port, possibly on its own beam. So the UE needs to be told which measurements carry over to which signals. That is QCL.

Quasi co-location: the mental model

The channel between a gNB port and a UE has large-scale properties — average delay, delay spread, Doppler shift, Doppler spread, and (in FR2) the spatial direction the energy arrives from. These are slow-changing statistics, unlike the instantaneous fading coefficients.

TS 38.214 §5.1.5: two antenna ports are quasi co-located if the large-scale properties of the channel over which a symbol on one port is conveyed can be inferred from the channel over which a symbol on the other port is conveyed.

Read it as a permission slip from the network: "You have been tracking signal X. You may reuse what you learned (timing / frequency / beam) when receiving signal Y."

The four QCL types

Types A–C let the UE reuse its time/frequency/channel-statistics estimates. Type D lets it reuse its receive beam.

Why Type D is the important one

At mmWave the UE has a handful of narrow receive beams (analog beamforming). It must choose one before it can receive anything — you cannot decode first and pick the beam afterwards. So before every PDCCH or PDSCH the UE must be told "this will arrive from the same direction as SSB 5 / CSI-RS 12 that you already measured, so point the beam you used for that". That instruction is QCL-TypeD. Without it, FR2 does not work.

In FR1 with wide beams the UE receives omnidirectionally, so Type D is usually not configured.

How QCL is signalled: TCI states

The network does not signal QCL relationships one at a time. It packages them as Transmission Configuration Indicator (TCI) states. Each TCI state says:

• Source reference signal 1 (an SSB index or a CSI-RS resource) and its QCL type (usually A or C), and
• optionally source reference signal 2 and its QCL type (usually D). The two types must differ.

Then:

1. RRC configures a list of TCI states per BWP (up to 64 or 128 depending on UE capability, for PDSCH).
2. MAC CE activates a subset (up to 8 for PDSCH) and maps them to codepoints; another MAC CE picks one TCI state per CORESET for PDCCH.
3. DCI 1_1 carries a 3-bit TCI field selecting one of the activated states for the scheduled PDSCH. (If the field is absent or the scheduling offset is too short for the UE to switch beams, the PDSCH is assumed to share the TCI of the PDCCH's CORESET.)

Reading a TCI state in words: "TCI state 3: QCL-TypeA with TRS resource 0, QCL-TypeD with CSI-RS resource 12" means "take timing and Doppler from the TRS; point the Rx beam you used for CSI-RS 12".

Rel-17 introduced a unified TCI framework (one indication covering PDCCH, PDSCH, PUCCH, PUSCH together) to reduce signalling — the concept is unchanged.

The QCL chain in a real cell

SSB (port 4000)  ──QCL-C (and D)──▶  TRS (CSI-RS, port 3000)
                                          │
                              QCL-A (and D from a beam CSI-RS)
                                          ▼
                              PDCCH DM-RS (port 2000)  and  PDSCH DM-RS (ports 1000+)

The SSB anchors everything; the TRS refines time/frequency; beam-management CSI-RS refines direction; the data channels inherit from those. Every arrow is a configured QCL relationship — the UE assumes none of them by default.

Common confusions

Spec pointers

Recap

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.