Ferkans — Interactive Telecom Tutor

From Multi-Cell CoMP to Multi-TRP

Chapter 11 introduced cell-free massive MIMO as a radical reimagining of multi-cell cooperation: the network as a pool of distributed access points serving every user. 5G NR takes a more incremental path with multi-TRP operation, in which two or more transmission-reception points (TRPs) — sectors of the same gNB, or tightly-connected gNBs sharing a backhaul — coordinate to serve a single UE. Multi-TRP is the standardized descendant of LTE's CoMP (coordinated multipoint) and inherits its central tradeoff: spatial gain versus backhaul latency.

The key design choice is whether the TRPs transmit coherently (sharing phase so that their signals add up constructively) or non-coherently (each TRP chooses its precoder independently and the UE decodes via diversity/multiplexing). Coherent schemes demand tight synchronization and CSI exchange; non-coherent schemes are cheaper but leave up to 3 dB of coherent-combining gain on the table.

Definition:
Multi-TRP Operation in NR

A TRP (transmission-reception point) is a set of co-located antennas serving one or more cells. Two or more TRPs are said to engage in multi-TRP operation when they jointly transmit to (or receive from) the same UE within overlapping time-frequency resources. 5G NR Rel-16 defines four multi-TRP schemes:

Scheme 1a (NCJT / SDM): Each TRP transmits a different MIMO layer on the same time-frequency resource. The UE separates layers via spatial multiplexing.
Scheme 2a (NCJT / FDM): Each TRP transmits on a different frequency subband within the same slot. UE combines across frequency.
Scheme 3 (TDM intra-slot): TRPs transmit on different OFDM symbols of the same slot. Sequential transmission provides diversity with moderate latency.
Scheme 4 (TDM inter-slot): TRPs transmit on different slots. Longest latency but simplest coordination.

A true coherent joint transmission (CJT) scheme, in which multiple TRPs transmit the same layer with phase-aligned precoders, is not part of Rel-16 but is under study in Rel-18 — it requires phase-level synchronization across TRPs, which in turn requires a calibration mechanism akin to a cell-free massive MIMO fronthaul (Chapter 14).

,

Definition:
CJT vs NCJT: The Phase-Coherence Dichotomy

Consider a UE with $N_r = 2$ antennas served by $M_{\text{TRP}}$ TRPs, each with precoder $\mathbf{v}_{m}$ applied to data symbol $s$ . The received signal is $\mathbf{y}_{\text{UE}} = \sum_{m=1}^{M_{\text{TRP}}} \mathbf{H}_{m}\,\mathbf{v}_{m}\,s_m + \mathbf{w}.$ Two regimes arise depending on whether the data symbols $s_m$ are identical or distinct:

CJT (Coherent Joint Transmission): $s_m = s$ for all $m$ , and the precoders are designed jointly so that the effective channel $\sum_m \mathbf{H}_{m} \mathbf{v}_{m}$ constructively combines. Requires phase-coherent synchronization across TRPs and joint CSI knowledge at a central entity.
NCJT (Non-Coherent Joint Transmission): $s_m$ are distinct layers of a multi-layer codeword, each with its own precoder. The UE separates them via MIMO detection. No phase coherence required.

CJT delivers $10\log_{10}(M_{\text{TRP}})$ dB of coherent combining gain on a single layer; NCJT delivers an $M_{\text{TRP}}$ -fold multiplexing gain across layers. The right choice depends on whether the UE is SNR-limited (CJT) or DoF-limited (NCJT).

Theorem: CJT vs NCJT Rate Gap

Let two TRPs with equal-power precoders serve a single UE with per-TRP SNR $\text{SNR}$ . The CJT achievable rate is $R_{\text{CJT}} = \log_2(1 + 2\text{SNR})$ (one layer, 3 dB coherent gain), while the NCJT achievable rate with two spatial layers is $R_{\text{NCJT}} = 2 \log_2(1 + \text{SNR}).$ The two schemes cross over at $\text{SNR} = 1$ (0 dB): CJT wins below 0 dB, NCJT wins above. At high SNR, NCJT's multiplexing advantage dominates; at low SNR, CJT's coherent combining dominates.

At low SNR, receive power is the bottleneck and the $2\times$ gain from coherent addition outweighs the loss of a second layer. At high SNR the channel can support two independent streams and the multiplexing gain of two parallel $\log(1+\text{SNR})$ beats a single $\log(1+2\text{SNR})$ .

Show Hint

Expand $\log_2(1+2x)$ and $2\log_2(1+x)$ at $x \to 0$ and $x \to \infty$ .

At $x \to 0$ : $\log_2(1+2x) \approx 2x/\ln 2$ , $2\log_2(1+x) \approx 2x/\ln 2$ — equal!

Check the second-order term to find which is larger when the linear terms agree.

Proof

Low-SNR expansion

At $\text{SNR} \to 0$ , $R_{\text{CJT}} \approx 2\text{SNR}/\ln 2 - 2\text{SNR}^{2}/\ln 2$ , $R_{\text{NCJT}} \approx 2\text{SNR}/\ln 2 - \text{SNR}^{2}/\ln 2$ . The first-order terms are equal; the second-order correction is smaller (more negative) for CJT. But at the linear approximation they cross. We need a more precise argument at small but nonzero SNR.

Difference and monotonicity

$\Delta(\text{SNR}) \triangleq R_{\text{CJT}} - R_{\text{NCJT}} = \log_2(1+2\text{SNR}) - 2\log_2(1+\text{SNR}) = \log_2\frac{1+2\text{SNR}}{(1+\text{SNR})^2}$ . At $\text{SNR} = 0$ : $\Delta = \log_2 1 = 0$ . At $\text{SNR} = 1$ : $\Delta = \log_2(3/4) < 0$ . At large $\text{SNR}$ : $\Delta \to -\infty$ . The derivative is $\Delta'(\text{SNR}) = \frac{2}{(1+2\text{SNR})\ln 2} - \frac{2}{(1+\text{SNR})\ln 2} = \frac{-2\text{SNR}}{(1+\text{SNR})(1+2\text{SNR})\ln 2} \leq 0$ , with equality only at $\text{SNR} = 0$ . So $\Delta$ starts at 0 and strictly decreases.

Interpretation

$\Delta(\text{SNR}) < 0$ for all $\text{SNR} > 0$ : NCJT with two layers always beats CJT with one layer once $\text{SNR} > 0$ , because multiplexing scales with $\log(\text{SNR})$ while coherent combining only shifts the log by $\log 2$ . But this assumes the UE can support two independent layers, which requires $N_r \geq 2$ . For a single-antenna UE, NCJT is infeasible and CJT is the only option. $\blacksquare$

,

CJT vs NCJT Rate Comparison

Sum rate of CJT and NCJT multi-TRP schemes as functions of SNR and TRP count. The crossover SNR is marked; below it CJT wins, above it NCJT wins (when the UE can support multiple layers).

Parameters

TRPs

M_{\text{TRP}}

2

Per-TRP SNR (dB)10

UE antennas

N_r

2

Definition:
PDCCH Reliability Enhancement via Multi-TRP

5G NR Rel-16 introduces PDCCH (physical downlink control channel) repetition across multiple TRPs as a reliability feature for ultra-reliable low-latency communication (URLLC). The same DCI (downlink control information) is transmitted from two TRPs at different spatial precoders and potentially different time-frequency resources. The UE combines the repetitions to improve decoding reliability by roughly a $2\times$ SNR gain.

The key design decision is resource multiplexing: TDM (sequential symbols), FDM (different subbands), or SDM (different spatial layers via different DMRS ports). TDM has the lowest coordination cost but the highest latency; SDM has the lowest latency but requires the UE to support spatial separation. Commercial URLLC deployments typically use TDM inter-symbol repetition as the baseline.

Example: URLLC Reliability with 2-TRP Repetition

A URLLC UE has a target PDCCH BLER of $10^{-5}$ . The single-TRP PDCCH BLER at the cell edge is $10^{-3}$ . Assume independent fading across TRPs and identical per-TRP BLER. How many TRPs (via repetition) are needed to meet the target, and what is the airlink latency?

Solution

Repetition model

With $M_{\text{TRP}}$ independent TRPs and equal BLER $p$ , the post-combining BLER (maximal-ratio combining) is approximately $p^{M_{\text{TRP}}}$ for independent errors and moderate SNR.

Required $M_{ ext{TRP}}$

Setting $(10^{-3})^{M_{\text{TRP}}} = 10^{-5}$ , we get $M_{\text{TRP}} = \lceil 5/3 \rceil = 2$ . So two-TRP PDCCH repetition meets the target.

Latency

With TDM repetition on consecutive symbols at $\mu = 1$ ( $T_{\text{OFDM}} \approx 33\,\mu s$ ), the extra delay is one symbol $\approx 33\,\mu s$ . This is well within URLLC's 1-ms latency budget.

Practical note

The independence assumption is the weak link: if both TRPs see the same blockage (e.g., a pedestrian between the UE and both TRPs), the BLERs are correlated and the combining gain is reduced. Commercial deployments place TRPs on geographically separated towers to maintain independence. $\blacksquare$

Multi-TRP Scheme Comparison

Scheme	Resource	Coherence	Gain	Latency	Rel-16 support
SDM (NCJT)	Same time-freq, different layers	None needed	Multiplexing $M_{\text{TRP}}$	Zero extra	Yes
FDM (NCJT)	Different subbands	None needed	Diversity + moderate multiplexing	Zero extra	Yes
TDM intra-slot	Different symbols, same slot	None needed	Diversity $M_{\text{TRP}}$	$\leq$ slot	Yes
TDM inter-slot	Different slots	None needed	Max diversity	Multi-slot	Yes
CJT	Same time-freq, same layer	Full phase	$M_{\text{TRP}}\times$ coherent SNR	Zero extra	Rel-18 study

🎓CommIT Contribution(2024)

Cell-Free Massive MIMO as the Ultimate Multi-TRP

H. Q. Ngo, G. Caire, E. Björnson, E. G. Larsson — IEEE Communications Magazine (invited overview)

The CommIT group's invited 2024 overview on cell-free massive MIMO for 6G makes the multi-TRP connection explicit: a fully-coordinated cell-free system is the asymptotic limit of multi-TRP operation as (i) $M_{\text{TRP}} \to \infty$ , (ii) inter-TRP phase synchronization becomes perfect, and (iii) the joint processing moves from a per-TRP local MMSE (Rel-16 style) to a central LSFD-based combining (Chapter 13). The paper argues that each Rel-16/17/18 multi-TRP feature is a partial step toward this limit: PDCCH repetition is a diversity primitive, NCJT is a multiplexing primitive, and the Rel-18 CJT study is the first explicit introduction of phase-level joint transmission. The 5G NR standard will not fully close the gap to cell-free — but 6G is expected to.

cell-freemulti-trp6gcoordination

NCJT vs CJT Topologies — Left: NCJT — two TRPs send distinct spatial layers to the UE, which separates them via MIMO detection. Right: CJT — two TRPs send the same symbol via jointly-designed precoders, requiring phase synchronization. The CJT path diagram shows the synchronization signal distribution network at the center.

Common Mistake: Coherent Joint Transmission Is Not Free

Mistake:

A plausible-sounding assumption is that CJT's 3-dB gain is essentially free because the TRPs can simply share the data and transmit at the same time.

Correction:

CJT requires three things beyond what NCJT needs: (i) a tight frequency/phase synchronization link between TRPs (typically below 1 ns timing error and 1 degree phase error), usually implemented with a dedicated reference-signal distribution over fiber or CPRI fronthaul; (ii) joint CSI at a central processing entity, which means the CSI from each TRP must be shipped over the backhaul within one coherence time; (iii) calibration of the per-TRP RF chains so that the joint precoder operates on a self-consistent channel matrix. The aggregate fronthaul and synchronization overhead is the reason CJT was excluded from Rel-16 and is only being standardized in Rel-18.

⚠️Engineering Note

Backhaul Latency Budget for Multi-TRP

The coordination quality of any multi-TRP scheme is bounded by the backhaul latency between TRPs. For a 1-ms coherence time, the backhaul must deliver CSI between TRPs in $< 0.5$ ms; fiber backhauls typically meet this (propagation + switching < 100 µs), while microwave backhauls (10-20 ms) cannot. The practical implication is that multi-TRP coordination is only feasible between TRPs served by the same DU (distributed unit) or between DUs connected by direct fiber. Commercial deployments therefore restrict multi-TRP to intra-DU coordination in dense urban scenarios.

Practical Constraints

•
Intra-DU fiber backhaul: < 100 µs (fine for CJT)
•
Inter-DU Ethernet: 1-5 ms (OK for NCJT diversity)
•
Microwave backhaul: 10-50 ms (no multi-TRP possible)

📋 Ref: 3GPP TR 38.801 Section 8.1.2

Historical Note: From LTE CoMP to NR Multi-TRP

2012-2020

Multi-TRP in NR is the direct descendant of LTE's Coordinated Multi-Point (CoMP) feature, introduced in LTE Rel-11 (2012). CoMP included both joint transmission (equivalent to CJT) and coordinated scheduling (akin to NCJT). However, LTE CoMP was widely regarded as under-utilized: commercial deployments reported gains of 5-15% sum rate — below the 30-50% predicted by theory — because backhaul latency and calibration were harder than expected. NR's Rel-16 multi-TRP cleanly separates the easy parts (NCJT, TDM repetition) from the hard parts (CJT, deferred to Rel-18), based on the LTE CoMP lesson that coordination quality determines almost everything about real-world performance.

TRP (Transmission-Reception Point)

A set of co-located antennas at a fixed location with their own baseband processing chain, treated as an independent transmitter/receiver by the MAC scheduler. A cell can contain multiple TRPs when different antenna panels are deployed; multiple cells can contain a single TRP when it is shared between operators.

NCJT (Non-Coherent Joint Transmission)

Multi-TRP scheme in which distinct spatial layers or frequency subbands are transmitted from different TRPs to the same UE. Does not require phase coherence between TRPs. Used in 5G NR Rel-16 as the default multi-TRP mode.

Quick Check

A UE with $N_r = 2$ antennas is served by $M_{\text{TRP}} = 2$ TRPs at per-TRP SNR $\text{SNR} = 10$ dB. Which scheme achieves higher rate: CJT on one layer or NCJT on two layers?

CJT: $\log_2(1 + 2 \cdot 10) \approx 4.4$ bits/s/Hz

NCJT: $2\log_2(1 + 10) \approx 6.9$ bits/s/Hz

Both equal

Depends on TRP spacing