Three bands, one room.

A Wi-Fi channel is named by a number, but the radio is tuned to a frequency, and the frequency fixes the wavelength. Everything a wave does in a room, how it bends round a shelf, how much of a body it sees, how fast its phase turns when that body moves, is set by the wavelength against the size of the things it meets.

The three carriers are not far apart on a logarithm, but they are far apart on a ruler. A 2.4 GHz wave is about twelve centimetres long; a 5 GHz wave is about six; a 6 GHz wave is five. Halving the wavelength halves every length in the rest of this page.

A person walking towards a receiver shortens the reflected path, and the phase of the reflection turns faster. The rate of that turning is the Doppler shift, and it is the whole of the band argument for motion sensing.

The band changes λ and nothing else in that expression. The same walk therefore writes about twice the Doppler on 5 GHz as on 2.4 GHz, and a little more again on 6 GHz. That is arithmetic, not a trend, and it is the mechanism behind the fleet's choice of 5 GHz for its paced arms. The cost is on the same line: a faster turn needs a faster sampler. The reduction that scores every minute refuses a sub-window with fewer than thirty frames, which is a delivered rate of 120 Hz.

An isotropic transmitter spreads its power over a sphere, and the receiving antenna's effective area shrinks with the wavelength squared. In free space the loss between two isotropic antennas is

so at the same distance 5 GHz arrives about 6.6 dB weaker than 2.4 GHz and 6 GHz about 7.7 dB weaker, before any wall. Indoors the slope is steeper than free space once the line of sight is blocked; Bensky's Table 2.1 gives an exponent of 1.6 to 1.8 for a building with line of sight and 4 to 6 for an obstructed one, and both are empirical, not derived. The fleet's one paired measurement on this point is the empty-office ABBA of EXP-011: RSSI −36 and −51 dBm on 2.4 GHz against −54 and −54 dBm on 5 GHz at one desk-pair geometry, three pairs, one room.

The three bands are not three versions of one thing. The 2.4 GHz band is 83.5 MHz wide with thirteen channels five megahertz apart, so a 20 MHz channel overlaps its neighbours and shares the air with DSSS frames that yield no channel estimate at all. The 5 GHz band is seven hundred megahertz of non-overlapping 20 MHz channels, some of which must vacate when radar appears. The 6 GHz band is twelve hundred megahertz, of which Europe opened the lower 480, with preferred scanning channels every sixteen numbers.

A CSI record is one complex number per used tone per antenna chain. How many tones and how far apart they sit is the numerology, and it is set by the frame's PHY, not by the band: a legacy OFDM frame carries 52 tones at 312.5 kHz whether it was sent on channel 11 or channel 48. The band decides which frames are on the air. On 2.4 GHz the fleet's records are legacy and HT with a thin HE presence; the VHT frames at 40 and 80 MHz exist only on 5 GHz.

Two echoes can be told apart only when their delays differ by more than one over the bandwidth, so a 20 MHz record blurs any two paths within fifteen metres of each other into one tap. Eighty megahertz brings that to under four metres. HE's finer spacing does not change the resolution; it lengthens the delay the grid can see without aliasing.

Put the five numerologies on one axis around one channel and the family resemblance is plain: four of them share one spacing and differ only in how many tones they pack, and HE alone packs four times as many into the same twenty megahertz. What the fleet actually holds of each is a different question, and the answer is lopsided. Almost every decoded record is legacy OFDM at 52 tones, because that is what the paced illuminator sends; the HT, VHT and HE classes are the building's own frames, thin on every band and absent on most.

Here are three real captures, one per band, decoded on this request from the committed sample catalogue: the mean amplitude across the tone grid over the first two hundred full records, per antenna chain, in decibels from each record's own median. Amplitude is normalised per record by the radio's gain control, so this is the shape of the channel and not its level; the level is the RSSI printed under the figure.

The declared layers above say what a band can do. This one says what the fleet has done. Per channel: the measurement sessions, their hours, the yields recomputed from summed counts, and the delivered rate by the local hour a session started. The two-clause rule holds here as everywhere: quote useful_yield, and read a low capture_yield on 2.4 GHz as DSSS/CCK traffic, not as a quiet channel.

Two designs on the fleet held enough fixed to be a contrast. A node that ran two or three of the channel arms on one day gives a within-day, within-node contrast across channels 36, 44 and 48, at the observed rung because the hours differ inside the day. The ABBA pairs of 10 and 11 August held hour, room, nodes and illumination on an empty office, so they are measured, and they measure the instrument: on 2.4 GHz the injector delivered 95.8 % of its commanded rate, on 5 GHz 100.1 %, six pairs of six in the same direction.

The corpus holds no measurement of 6 GHz for sensing and no ranking of channels within a band; both silences are findings. The fleet's own 6 GHz record is one hour on one evening from an arm with two misfires in three attempts, which is a pilot. The physics above says what the wavelength does to a walker's Doppler and to free-space loss; it says nothing about which band counts people better, because that needs the same people on two bands in one hour, and the paired design that buys it (CR-040, ABBA against the ambient arm) has not run.

JSON: physics · census · spectra · Related: Format · Corpus · Occupancy