GS01 — MIMO Wireless Capacity
| Field | Value |
|---|---|
| Domain | Grassmannian Systems |
| System | k transmit beams in n-antenna channel |
| Group | U(n) acting on Gr(k, n) |
| H^k tier | H¹ |
| ISA | Forge (β ≈ β*) |
| Status | Validated |
| Opcodes | ORBIT · TWIST · LABEL |
| Papers | Paper 574 |
Physical system
A MIMO (Multiple-Input Multiple-Output) wireless system uses n_T transmit and n_R receive antennas. The channel is an n_R × n_T complex matrix H drawn from an isotropic distribution. The transmitter chooses k beams (spatial directions) from the n_T-dimensional transmit space — a point V ∈ Gr(k, n_T). Capacity depends only on the singular values of HV, not on the unitary rotation of V within Gr(k, n_T): the channel capacity is a function on the Grassmannian.
Zheng and Tse (2002) established the diversity-multiplexing tradeoff: at SNR ρ, a MIMO channel with n_T transmit, n_R receive antennas achieves
- multiplexing gain r = k (spatial streams = dimension of V ∈ Gr(k, n_T))
- diversity gain d(r) = (n_T − r)(n_R − r) (number of independent fading paths)
The tradeoff d(r) is the Schubert calculus intersection number of two Schubert varieties in Gr(k, n_T + n_R): it counts the dimension of the set of channels H where exactly k beams are supported. Capacity = volume of Gr(k, n_T) in the Fisher-Rao metric of the channel distribution.
Target category
Ch(n_T, n_R) — the category of MIMO channels with n_T × n_R complex matrices as objects and unitary beam-forming transformations as morphisms. The optimal beam-former V* is the point in Gr(k, n_T) that maximises mutual information I(x; y) = log det(I + ρ/k · H V V† H†). The capacity-achieving V* is the leading k-dimensional eigenspace of H†H — the same object as the PCA subspace (GS02).
Interpretation functor
F: C → Ch(n_T, n_R) defined by:
| Opcode | F(opcode) |
|---|---|
| ORBIT | Beam update: V → exp(tX)·V for X ∈ u(n_T); geodesic on Gr(k,n_T) toward capacity-maximising subspace; implements gradient ascent of mutual information |
| TWIST | Phase coherence: the H¹ Berry phase φ = i∮⟨V|dV⟩ accumulated as the beam rotates through the channel; determines coherent vs non-coherent capacity gap |
| LABEL | Singular value λ_i(HV): the eigenvalue of each beam’s effective channel gain; capacity = Σ_i log(1 + ρ λ_i²) |
ISA programme
CHANNEL: LABEL[H | n_R x n_T complex fading matrix] -- observe channel
SVD: ORBIT[H = U Sigma V† | singular value decomp] -- ORBIT to eigenspace
BEAMS: LABEL[V* = top-k columns of V] -- optimal Gr(k,n_T) point
PHASE: TWIST[phi = i oint <V|dV>] -- coherent beam phase (H1)
CAPACITY:LABEL[C = sum_i log(1 + rho * sigma_i^2)] -- Shannon capacity (LABEL eigenvalue)
TRADEOFF:LABEL[d(r) = (n_T - r)(n_R - r)] -- diversity-multiplexing tradeoff
Computable output
- Shannon capacity C = Σ_i log₂(1 + ρ σ_i²(HV)) bits/s/Hz: the ORBIT output. Validated against the Telatar (1999) formula; exact for i.i.d. Rayleigh fading channels.
- Diversity-multiplexing tradeoff d(r) = (n_T − r)(n_R − r): the Schubert intersection number. For a 2×2 MIMO: d(0)=4, d(1)=1, d(2)=0. This is the H¹ TWIST content of the channel: how many independent fading paths support each multiplexing gain level.
- Grassmannian code distance: the minimum chordal distance between codewords in a Grassmannian code is d_c(V_i, V_j) = √(k − ‖V_i†V_j‖_F²) — the ORBIT geodesic distance on Gr(k,n). Optimal Grassmannian codes (packings in Gr(k,n)) maximise this distance; they are ORBIT-maximally-separated points, analogous to sphere packing.
- Coherent vs non-coherent gap: the TWIST phase φ distinguishes coherent reception (channel known, H¹ TWIST computed) from non-coherent reception (channel unknown, only H⁰ ORBIT magnitude available). The coherent capacity advantage is exactly the H¹ Berry phase correction to the H⁰ power gain.
Grassmannian Systems context
MIMO is the engineering entry point to Grassmannian computation: it is the first application where working engineers routinely think in terms of subspaces, SVD, and Grassmannian geometry — often without using those words. The optimal beam-former is a point in Gr(k, n_T); beamforming codebooks (3GPP Type I/II) are Grassmannian codes; massive MIMO (n_T → ∞) is the large-n limit of the Grassmannian. Modern 5G and 6G systems are, in operational practice, Grassmannian computing systems running an implicit OPU (P01) with H⁰ and H¹ opcodes.
Validation
- Telatar (1999): MIMO capacity formula for i.i.d. Rayleigh fading. The foundational result; exact.
- Zheng & Tse (2002): diversity-multiplexing tradeoff d(r) = (n_T−r)(n_R−r). Proved rigorously; matches simulation.
- Grassmannian codes: Love, Heath & Strohmer (2003) — systematic construction of Grassmannian codebooks achieving the Hamming-Rankin bound on chordal distance.
- 3GPP standards: LTE/5G NR codebooks are standardised Grassmannian packings in Gr(k, 32) (32-antenna base station). Deployed globally.