Spatial Reuse, OBSS Interference, and BSS Coloring

Two access points serve adjacent rooms on the same channel. Their coverage overlaps in a corridor. A station in the overlap can hear both. When the first AP transmits, the station hears it, sets its network allocation vector, and defers. The second AP’s own transmissions are blocked while the first is talking, and the stations associated to the second AP receive nothing. The two cells are not sharing a channel; they are taking turns on it, and the aggregate throughput of the pair is close to that of a single cell.

The deferral is correct in the sense that the station could not have decoded two simultaneous transmissions. It is wasteful in the sense that the first AP’s signal at the second AP’s own stations may be far too weak to disturb them. The rule that decides whether a transmission is harmful is a threshold, and for the 20 MHz OFDM PHYs that threshold has stayed at the same level through the amendments that followed: if the energy is above it, defer.

Why the threshold is conservative

The clear channel assessment threshold for a preamble the station can decode is -82 dBm in a 20 MHz channel. For energy the station cannot identify as a valid preamble, the threshold is -62 dBm, twenty decibels higher, because a receiver that cannot decode a preamble cannot tell whether the energy would have damaged a frame. Numbers in this range are not arbitrary: they are set so that a station defers to any signal it could conceivably decode, which is the safest rule and the one that maximises the chance that a reception in progress succeeds.

The cost is that a station defers to signals it could not decode and that would not have collided with anything it cares about. In a dense deployment — an office floor with an AP every ten metres, an apartment building where every unit has its own router — the aggregate of these conservative decisions is that the channel is busy almost all the time even when most of the transmissions are physically separable.

The problem has a name: overlapping basic service set, or OBSS, interference. It is not interference in the sense of a corrupted frame. It is interference in the sense of a shared medium being reserved by a neighbour. The physical layer could carry both transmissions simultaneously at a lower rate each; the MAC layer’s deference rule prevents it.

The colour in the preamble

The high-efficiency amendment attacks the problem by adding information to the physical header that tells a receiver whose transmission it is hearing.

The legacy preamble begins with a field that a station uses to detect energy and to synchronise. The high-efficiency amendment prefixes its frames with a legacy-compatible preamble so that older stations can still defer correctly, then follows it with a new set of fields. One of those fields is 6 bits long and carries a colour: a value from 0 to 63 chosen by the AP, announced in its management frames, and repeated in every frame the AP or its stations transmit.

A station that decodes a frame and finds a colour different from its own knows the frame belongs to another basic service set. If the received signal is also below a threshold that the AP advertises, the station may treat the medium as available and transmit anyway, at a reduced power and with its own colour in the preamble. The receiving station on the other side does the same test, so both cells can transmit concurrently and each can still decode its own traffic.

The mechanism converts a binary deference rule into a two-dimensional one: energy and identity. Energy alone says whether a signal is present; identity says whether it matters.

Two constants carry the design. The colour is 6 bits, so there are 64 values and a network with more than 64 APs on one channel must reuse colours; two APs with the same colour in range of each other will be treated as the same cell and will defer, which is the safe failure. The threshold is advertised in a field in the high-efficiency operation element, and a station that uses a value more aggressive than the AP’s is not conforming; the AP sets it as a policy for its whole cell.

The OBSS packet detection threshold

The threshold is the crux, and its units are dBm. It defines the received power above which a frame from another colour must still be treated as busy. Setting it very low reproduces the old behaviour; setting it high allows more concurrent transmissions and more collisions.

The trade is not symmetric. A station that transmits into an OBSS frame it could have deferred to risks corrupting that frame at its receiver, and the corruption is invisible to the station that caused it. The other cell’s retransmission then consumes airtime that the aggressor also pays for, since the medium is shared. The gain from concurrency and the loss from collision both land on the same channel.

The rule that makes the trade tractable is that the threshold applies to the frame the station hears, while the harm depends on the power that frame’s receiver sees. Those are different quantities, and a station cannot measure the second. The approximation is that the two are similar because the topology is roughly symmetric, which is the same reciprocity assumption the reservation handshake relies on and the same one that fails with hidden terminals.

Colours are chosen, not negotiated

Nothing in the standard assigns colours. Each AP picks one, typically at startup, and the choice is a coordination problem between APs that may belong to different operators and may never exchange a frame.

The naive choice is random, which gives a birthday-problem collision rate: with 64 colours and a handful of neighbouring APs, the chance that two close neighbours share a colour is small but not negligible, and with a dozen neighbours it becomes likely. A collision is not catastrophic — the two cells simply behave as one and defer to each other — but it silently removes the benefit for that pair.

Better implementations listen before choosing. An AP that scans the channel, decodes beacons and high-efficiency frames from neighbours, and notes their colours can pick a value that none of its audible neighbours use, and can change its colour later if a conflict appears. The change is announced in the beacon and takes effect for subsequent frames, which means stations must accept a colour that differs from the one they associated with. A station that caches the colour at association and never updates will misclassify every frame after the change.

There is a second coordination problem that colouring does not solve: transmit power. Two cells may be distinguishable by colour and still interfere if the aggressor’s power at the victim’s receiver is high. The amendment therefore also defines a transmit power envelope, a maximum power an AP advertises for its cell, and stations are expected to stay within it. Lower power means smaller cells and more concurrent transmissions; higher power means better coverage per AP and more overlap.

What concurrency actually buys

The gain from spatial reuse is not proportional to the number of concurrent transmissions. Two simultaneous transmissions each see the other as noise, so each operates at a lower signal-to-interference-plus-noise ratio, which supports a lower modulation. The capacity of the pair is the sum of two reduced rates rather than twice one rate, and if the interference is strong enough, both fail.

The condition for a net gain is that each transmission’s power at the other’s receiver is below the level that would force a large drop in modulation. In a well-planned deployment with directional antennas and low power, that holds over much of the overlap. In a deployment where APs are placed for coverage at maximum power, it does not, and colouring changes little.

A second limit is the hidden node. A station that decides to transmit concurrently because it cannot hear an OBSS frame may still be audible at that frame’s receiver, and the collision it causes is the hidden-terminal problem in a new form. The reservation handshake cannot help, because the station never heard the frame it should have deferred to. This is why the aggressive thresholds are risky in exactly the topologies where spatial reuse would pay most: sparse, oddly shaped coverage with weak mutual hearability.

Measuring whether it worked

The quantities that indicate whether reuse is helping are not the ones a link-rate display shows. A cell that transmits concurrently will show a lower rate per frame and a higher frame count; the aggregate is the product, and only the aggregate matters.

The counters that matter are the ones counting deferrals, the ones counting failed receptions with a valid header, and the ones attributing failures to a colour. Not every driver exposes them, and those that do name them differently. The measurement approach that works regardless is to compare the sum of throughput across the overlapping cells before and after a change in colour or threshold, holding the traffic constant.

The mechanism sits in a chain of decisions that all reduce to the same question: which transmissions can safely share the channel. The reservation handshake described in The hidden terminal problem and why 802.11 reserves the medium answers it for one cell by announcing a reservation; colouring answers it across cells by labelling the announcement. Neither answers it for a station that hears neither. And the way a cell divides its airtime among stations once it has decided it may transmit is the subject of OFDMA scheduling and multi-user transmission.