A preprint proposes using the speed of a movable antenna as an extra way to carry wireless data. At the paper’s baseline 28 GHz setting, the antenna would have to cross a 107-millimetre track at 107 metres per second—385 km/h—within a 1-millisecond slot, a requirement the paper describes as infeasible by roughly five orders of magnitude.
That physical hurdle sits alongside a sharp receiver trade-off. On a track 10 wavelengths long, the study’s local reference used 42 velocity levels and five index bits; an oracle-AoD receiver—one given the signal’s angles of departure—reached that payload from about 0 dB. The AoD-free covariance-matched receiver saturated at three bits from about 10 dB onward, while requiring no AoD, path-gain, transmitted-symbol or instantaneous channel-state information.
Making motion carry a label
The proposed VIM-MA scheme turns deliberate antenna motion into an index, or label, by using movement-induced Doppler—the motion-related frequency shift—as an information-bearing signal. Its velocity codebook is optimized for spacing and size under Doppler-resolvability, finite-track and decorrelation constraints; a convex reformulation yields a closed-form globally optimal solution and identifies the constraint that binds first.
Under the paper’s isotropic-scattering model, the decorrelation spacing is bounded at 0.383 wavelengths or less for every maximum-correlation threshold. When decorrelation is the active limit, the reported codebook is at least 2.61 times finer than the paper’s Rayleigh Doppler reference, reaching 4.13 times finer at a threshold of 0.5.
For the AoD-free receiver, however, reliability—not spatial decorrelation—sets the practical limit: the ambiguity constraint, labelled C1b, is active instead of C3. In plain terms, the receiver’s ability to tell neighboring velocity choices apart becomes the tighter constraint in the reported simulation.
The receiver makes the difference
That distinction carries into effective throughput, measured in bits per channel use. The AoD-free receiver saturated near 4.97 bpcu, while the oracle-AoD reference approached 6.95 bpcu against an AIF-CRB/decorrelation reference of about 7 bpcu. The authors caution that the formula is optimistic because an index error can also corrupt the QAM symbol decision.
A matched comparison with position indexing narrowed the difference. Under the same maximum-correlation threshold of 0.5 on a 10-wavelength aperture, FA-IM supported 42 positions and five index bits, making the velocity and position domains rate-equivalent for that aperture. That comparison did not equalize receiver side information or pilot costs.
At matched reliability, perfect-CSI FA-IM reached five index bits from about 15 dB, but AoD-free VIM-MA did not support a 42-state codebook. With the same pilot budget, VIM-MA delivered 4.97 bpcu from about 10 dB; FA-IM needed about 20 dB for three bits and 25 dB for four.
The physical limits remain
The document’s evidence comes from a modeled system and Monte Carlo simulations. The modeled system was a single-user downlink with movable antennas on one-dimensional tracks and a multipath channel with complex-Gaussian path gains and independent isotropic angles of departure. Default runs used 1,500 Monte Carlo slots per point, and most error, codebook and throughput results used L=32.
Frequency offsets were another sharp vulnerability. Under the reported offset model, index-error probability was 1.5×10−2 without impairment, 8.3×10−2 at 0.05 normalized units, 0.17 at 0.1 and 0.87 at one full level; the estimated tolerance was about 0.02 normalized units.
The AIF-CRB benchmark also needs a qualification. It is an oracle, local bound: with unconstrained unknown angles of departure, velocity is not identifiable from a single slot; the model instead relies on an isotropic-scattering spectrum for AoD-free identifiability.
Taken together, the preprint presents velocity indexing as a CSI-light option whose payload is limited by ambiguity in the no-AoD receiver and whose headline operating point is constrained by mechanics and frequency-offset compensation. It does not demonstrate hardware operation: the 107 m/s estimate is a model-based lower-bound calculation, not a hardware measurement.
The document is an arXiv preprint, version 1, dated 20 Aug 2026.
Paper data and sources
Original title: Velocity Index Modulation for Movable Antenna Systems
Authors: Yan Zhang, Indrakshi Dey, Shuaishuai Han et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text