Coordinated Multipoint
How CoMP Turns Interference Into Signal
In a conventional cellular network, a user near the boundary of two or three cells receives a strong wanted signal from its serving site and nearly-as-strong unwanted signals from the neighbors. Because those neighbors operate on the same frequency (frequency reuse of one in LTE and NR), the user's SINR can fall to 0 dB or below, throttling throughput to a fraction of what cell-center users enjoy. CoMP attacks this by treating the surrounding base stations as members of a cooperating set rather than as adversaries. The cells exchange information over the backhaul (the X2 interface in LTE, Xn in NR, or an ideal fiber/CPRI link) and jointly decide how to serve the edge user so that the formerly interfering energy either adds constructively or is steered into a spatial null.
Downlink CoMP comes in three families that differ in what must be shared. Coordinated scheduling and coordinated beamforming (CS/CB) share only CSI: each cell still serves its own users, but they jointly choose beamforming weights and scheduling so that one cell's beam nulls toward another cell's edge user. Joint transmission shares the actual user data across the cooperating set so multiple points transmit the same symbols, combining coherently (with phase alignment) or non-coherently at the terminal. Dynamic point selection also distributes the user data but lets only the instantaneously best point transmit each subframe. Uplink CoMP is conceptually simpler: a single-antenna user is heard by several base stations whose received signals are jointly combined, a form of macro-diversity that adds receive array gain with no penalty at the handset.
The cost of all this is information exchange and timing discipline. Joint transmission is the most demanding because the precoder it computes from reported CSI ages over the backhaul round trip; if the channel changes faster than the network can react, the coherent gain evaporates and CoMP can underperform a non-cooperative baseline. This is why deployment choices revolve around backhaul latency budgets and user velocity far more than around antenna count.
Governing SINR and Combining Relations
SINR = PsGs / (σ2 + ∑i=1N PiGi)
Coherent joint transmission (M cooperating points):
Prx ≈ (∑m=1M √(PmGm))2 (vs. ∑ PmGm non-coherent)
Uplink combining gain (max-ratio, M sites):
SINRMRC = ∑m=1M SINRm
Channel coherence time (CSI staleness limit):
Tc ≈ 0.423 / fD, fD = v × fc / c
Where P = transmit power, G = path gain, σ2 = noise power, fD = Doppler shift, v = user speed, fc = carrier frequency, c = speed of light. Example: v = 60 km/h at fc = 2 GHz gives fD ≈ 111 Hz and Tc ≈ 3.8 ms, so JT backhaul plus processing must stay well under that window.
CoMP Mode Comparison
| Mode | Shared over backhaul | Backhaul latency budget | Cell-edge gain (typ.) | Key requirement | Best fit |
|---|---|---|---|---|---|
| Joint Transmission (JT) | User data + CSI | < 1 ms (coherent < 0.5 ms) | 20 to 40% | Tight phase/timing sync | Dense fiber/CPRI, low mobility |
| Dynamic Point Selection | User data + CSI | 1 to 5 ms | 15 to 30% | Fast point switching | Bursty edge traffic |
| Coord. Scheduling / Beamforming | CSI only | up to ~5 ms | 10 to 20% | Inter-cell scheduler | Routed X2/Xn backhaul |
| Uplink CoMP reception | Soft bits / IQ samples | 1 to 5 ms | 30 to 50% | Baseband combining | C-RAN, centralized RAN |
| No CoMP (baseline) | None | n/a | 0% (reference) | None | Light-load macro cells |
Frequently Asked Questions
What is the difference between joint transmission and dynamic point selection in CoMP?
In joint transmission (JT), multiple points send the same data on the same time-frequency resource so signals combine at the receiver, adding useful power but requiring tight phase/timing alignment for coherent gain. In dynamic point selection (DPS), the user data is available at every candidate point but only the instantaneously strongest point transmits each subframe; the rest stay silent. DPS avoids JT's synchronization overhead at the cost of less array gain. Both need user data at all cells, unlike coordinated scheduling/beamforming, which shares only CSI.
How much backhaul latency can CoMP joint transmission tolerate?
JT is the most latency-sensitive mode because the CSI used to compute precoders ages over the backhaul round trip. LTE-Advanced JT typically needs one-way latency below ~1 ms (ideally < 0.5 ms), implying dark fiber or a CPRI fronthaul rather than routed IP. At 60 to 120 km/h and 2 GHz the channel coherence time falls to a few ms, so stale CSI collapses the coherent gain and can drop performance below the non-CoMP baseline. Coordinated scheduling and DPS tolerate looser backhaul of several ms.
What cell-edge throughput gains does CoMP actually deliver?
Using 5th-percentile user throughput as the metric, 3GPP trials and simulations show roughly 20 to 40% downlink gain for JT with ideal backhaul and 10 to 20% for coordinated scheduling/beamforming. Uplink CoMP reception can reach 30 to 50% because there is no transmit-power or sync penalty at the handset. Average spectral-efficiency gains are far smaller (often single digits) since cell-center users already see good SINR and benefit little from coordination.