Massive MIMO vs Beamforming: Two Different Gains from One Array

The question gets asked constantly and it is badly formed. Massive MIMO and beamforming are not two options you pick between — one is a way of building a radio, the other is something that radio does, and a 64T64R unit does both in the same slot.

The useful question is the one underneath: which gain am I actually getting, and what does it buy me? Because there are two gains, they behave completely differently, and confusing them is how a site ends up with an expensive radio that changed nothing.

---

Two gains, two shapes of return

Array gain is about where the energy goes. Feed the same signal to many elements with the right phase offsets and the wavefronts add constructively in one direction. Double the element count and you add roughly 3 dB of directivity. The receiver sees more signal, its SINR rises, the scheduler can pick a higher modulation and coding scheme.

Multiplexing gain is about how many things you send at once. With enough antennas at both ends and a channel that supports it, you can transmit independent data streams on the same resource blocks and separate them at the receiver.

The difference that matters is the shape of the return. Capacity per stream grows with the logarithm of SINR, so array gain runs into diminishing returns quickly — the step from 6 dB to 9 dB buys far less than the step from 0 to 3. Capacity across streams grows linearly with the number of them. Two streams is twice the data. Four is four times.

That asymmetry is the entire commercial argument for massive MIMO. Nobody spends money on a 64-element array for 18 dB of directivity. They spend it because a linear term beats a logarithmic one.

They are the same operation

The reason these cannot be separated at the hardware level is that they are produced by the same act: applying a complex weight to each element before transmission.

Choose weights that steer all the energy toward one receiver and you have beamforming. Choose a weight matrix that creates several spatial channels at once and you have spatial multiplexing. In the specification both are just precoding, and beamforming is precoding at rank one.

Which is why "does this radio do beamforming or MIMO?" has no answer. It applies weights. What those weights are is decided per transmission, and the decision is made by the scheduler using channel knowledge — from SRS in TDD, from a reported codebook index in FDD.

The channel decides, not the radio

Here is the part that gets skipped. Rank is a property of the propagation environment. You can buy as many elements as you like; if the channel has one dominant path, there is one usable spatial dimension and the extra ports have nothing to multiplex.

Rich scattering supports high rank. Dense urban, indoors, mid-band — many independent paths, so several streams remain separable. This is where multiplexing gain is real.

Strong line of sight collapses rank. Rural macro, rooftop backhaul, fixed wireless, and most of FR2. One path dominates, rank falls toward one, and the array's contribution is almost entirely pointing. At millimetre wave that is not a disappointment — the link does not close without it — but it does mean the gain you are buying there is coverage, not capacity.

And the device is the third constraint. A phone carries two or four receive antennas. Its own rank is capped there regardless of what the base station has, so a single UE can never use sixty-four ports.

Which is why MU-MIMO is the point

Given that cap, what does an operator do with the other sixty ports? Serve other people with them.

Multi-user MIMO schedules several devices on the same time-frequency resources and separates them spatially. The array is not creating more streams to a user; it is creating a beam per user, shaped so that each one lands on its intended device and nulls toward the others.

This is the gain that justifies the radio, and it has three preconditions that are easy to miss:

Enough RF chains. Simultaneous users on the same resources cannot exceed the number of digital chains. An array with many elements behind few chains — the hybrid architecture that FR2 forces — can point sharply but can only separate a handful of users.

Users that are actually separable. Two devices in nearly the same direction cannot be paired; their channels are too correlated for the nulls to work. Pairing rate depends on how the traffic is distributed in space, which is a property of the site, not the equipment.

Accurate channel knowledge. MU-MIMO nulls are only as good as the channel estimate they were computed from. In TDD, reciprocity gives the gNB a full estimate from the device's sounding reference signals — which is why TDD bands got massive MIMO first. In FDD there is no reciprocity, so the network works from a quantised codebook report, and the nulls are correspondingly blunter.

How to tell which one you are getting

The answer is in counters you already have.

Rank distribution. Reported rank indicator, aggregated per cell. If it sits at one, your array is doing beamforming and nothing else, whatever the marketing said.

MU pairing rate. How often multiple UEs are scheduled on the same resources. Low pairing in a loaded cell means the multiplexing gain is not being realised — usually correlation or channel-estimate quality, occasionally a conservative pairing threshold.

Where the throughput moved. Beamforming raises SINR, so it shows up in the fifth-percentile — cell-edge — throughput. MU-MIMO raises the total served, so it shows up in cell throughput under load and barely at all when the cell is quiet.

That last one is the practical test. If your new radio improved cell-edge throughput but not cell capacity, you bought array gain. If it improved capacity only when busy, you bought multiplexing gain. If neither moved, look at the rank distribution before you look at the vendor.

Where this goes wrong

Expecting capacity from a rank-1 environment. A rural macro with clean line of sight will not produce MU-MIMO gain no matter how many elements it has. It will produce coverage gain, which may well be worth it — but it should be bought deliberately.

Assuming FDD behaves like TDD. Same radio, same element count, materially less gain, because the channel knowledge is a codebook index rather than a measurement.

Comparing element count to element count. 64T64R names sixty-four transceiver chains, not sixty-four elements; the physical element count is typically several times higher, with elements grouped behind each chain. Two products quoting different numbers are often quoting different quantities.

Treating beam count as capacity. More beams is finer pointing. Capacity comes from how many are used simultaneously on the same resources, which is bounded by the digital chains, not by the beam book.

---

The mental model

Two gains, not two technologies. Array gain concentrates energy; multiplexing gain sends parallel streams. The same weights produce both.

Beamforming is precoding at rank one. There is no hardware boundary between them.

Logarithmic versus linear. SINR gains saturate. Stream count does not — which is the whole reason massive MIMO is worth its cost.

Rank belongs to the channel. Line of sight collapses it; scattering supports it; the device's own antenna count caps it.

MU-MIMO is where the money is, and it needs digital chains, spatially separable users, and a good channel estimate. Remove any one and the array degrades to expensive beamforming.

Cell edge moved? Array gain. Loaded-cell capacity moved? Multiplexing gain.

Further reading

• 3GPP TS 38.214 §5.2 — CSI reporting, rank indication and codebook types
• 3GPP TS 38.211 §6.3.1.5 — precoding for uplink, and §7.3.1.4 for downlink
• 3GPP TS 38.331 — CSI-MeasConfig and the SRS configuration behind reciprocity
• Massive MIMO in detail and beamforming architectures
• ShareTechnote's 5G handbook, for codebook structures and CSI report formats