File size: 8,750 Bytes
570b87b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 | """Fast HOA rotation via Wigner-D matrices (real Ambix ACN / SN3D).
Within each order n the (2n+1) coefficients transform by a real matrix built
from complex Wigner D functions. Normalization (SN3D vs N3D) cancels inside
an order, so the same matrices apply to Ambix SN3D.
Public:
rotation_matrix_zyx — 3×3 active rotation (yaw/pitch/roll)
hoa_rotation_matrix — full ((N+1)²)×((N+1)²) block-diagonal matrix
apply_hoa_rotation — a' = M @ a (also (C,T) streams)
"""
from __future__ import annotations
import math
from functools import lru_cache
from typing import Tuple
import numpy as np
from .basis import MAX_ORDER, N_CHANNELS, acn_index
def rotation_matrix_zyx(
yaw: float, pitch: float, roll: float, *, degrees: bool = True
) -> np.ndarray:
"""R = Rz(yaw) @ Ry(pitch) @ Rx(roll); active rotation of column vectors."""
if degrees:
yaw, pitch, roll = map(math.radians, (yaw, pitch, roll))
cy, sy = math.cos(yaw), math.sin(yaw)
cp, sp = math.cos(pitch), math.sin(pitch)
cr, sr = math.cos(roll), math.sin(roll)
Rz = np.array([[cy, -sy, 0.0], [sy, cy, 0.0], [0.0, 0.0, 1.0]])
Ry = np.array([[cp, 0.0, sp], [0.0, 1.0, 0.0], [-sp, 0.0, cp]])
Rx = np.array([[1.0, 0.0, 0.0], [0.0, cr, -sr], [0.0, sr, cr]])
return Rz @ Ry @ Rx
def _fact(n: int) -> float:
if n < 0:
return 0.0
return float(math.factorial(n))
@lru_cache(maxsize=8192)
def wigner_d(j: int, mp: int, m: int, beta: float) -> float:
"""Wigner small-d matrix element d^j_{mp,m}(beta), real.
Explicit finite sum (safe for j ≤ 7 used here).
"""
if abs(mp) > j or abs(m) > j:
return 0.0
# Numerical stability: beta in [0, pi] preferred but general ok
cb = math.cos(beta * 0.5)
sb = math.sin(beta * 0.5)
# Avoid 0**negative
s_min = max(0, m - mp)
s_max = min(j + m, j - mp)
total = 0.0
for s in range(s_min, s_max + 1):
den = (
_fact(j + m - s)
* _fact(j - mp - s)
* _fact(s)
* _fact(s + mp - m)
)
if den == 0.0:
continue
num = _fact(j + m) * _fact(j - m) * _fact(j + mp) * _fact(j - mp)
pref = ((-1.0) ** (mp - m + s)) * math.sqrt(num) / den
cpow = 2 * j + m - mp - 2 * s
spow = mp - m + 2 * s
# handle base cases
cterm = 1.0 if cpow == 0 else (cb ** cpow if abs(cb) > 1e-15 or cpow > 0 else 0.0)
sterm = 1.0 if spow == 0 else (sb ** spow if abs(sb) > 1e-15 or spow > 0 else 0.0)
if cpow < 0 and abs(cb) < 1e-15:
cterm = 0.0
if spow < 0 and abs(sb) < 1e-15:
sterm = 0.0
total += pref * cterm * sterm
return total
def wigner_D_complex(
j: int, alpha: float, beta: float, gamma: float
) -> np.ndarray:
"""Complex Wigner-D matrix for order j, indices m',m ∈ [-j..j].
Ordering: row/col index = m + j (m from -j to +j).
D_{mp,m} = e^{-i mp α} d_{mp,m}(β) e^{-i m γ}
"""
dim = 2 * j + 1
D = np.zeros((dim, dim), dtype=np.complex128)
for imp, mp in enumerate(range(-j, j + 1)):
for im, m in enumerate(range(-j, j + 1)):
d = wigner_d(j, mp, m, beta)
D[imp, im] = (
math.cos(mp * alpha)
- 1j * math.sin(mp * alpha)
) * d * (
math.cos(m * gamma) - 1j * math.sin(m * gamma)
)
# e^{-i θ} = cosθ - i sinθ
return D
def _real_to_complex_matrix(j: int) -> np.ndarray:
"""Unitary map U: real ACN coeffs (m=-j..j) → complex m=-j..j.
Convention (common in Ambisonics / real SH):
c_0 = r_0
c_{+m} = (-1)^m / √2 * (r_{+m} - i r_{-m})
c_{-m} = 1 / √2 * (r_{+m} + i r_{-m})
so r = U^H c and c = U r with U unitary.
"""
dim = 2 * j + 1
U = np.zeros((dim, dim), dtype=np.complex128)
# index helper: m -> i = m + j
def idx(m: int) -> int:
return m + j
U[idx(0), idx(0)] = 1.0 + 0.0j
s2 = 1.0 / math.sqrt(2.0)
for m in range(1, j + 1):
sign = (-1.0) ** m
# c_{+m} from r_{+m}, r_{-m}
U[idx(m), idx(m)] = sign * s2
U[idx(m), idx(-m)] = -1j * sign * s2
# c_{-m} from r_{+m}, r_{-m}
U[idx(-m), idx(m)] = s2
U[idx(-m), idx(-m)] = 1j * s2
return U
def real_sh_rotation_block(
j: int, alpha: float, beta: float, gamma: float
) -> np.ndarray:
"""Real (2j+1)×(2j+1) rotation matrix for ACN order-j block (m=-j..j)."""
if j == 0:
return np.array([[1.0]], dtype=np.float64)
U = _real_to_complex_matrix(j)
D = wigner_D_complex(j, alpha, beta, gamma)
# real coeffs: r' = U^H D U r
R_c = U.conj().T @ D @ U
# Should be real symmetric orthogonal (numerically tiny imag)
return np.real(R_c)
def rotation_matrix_to_zyz(R: np.ndarray) -> Tuple[float, float, float]:
"""Extract ZYZ Euler angles (α, β, γ) from a right-handed rotation matrix.
R = Rz(α) @ Ry(β) @ Rz(γ) (active, column vectors).
"""
R = np.asarray(R, dtype=np.float64)
# β ∈ [0, π]
# R[2,2] = cos β
cbeta = float(np.clip(R[2, 2], -1.0, 1.0))
beta = math.acos(cbeta)
sb = math.sin(beta)
if abs(sb) > 1e-10:
# α = atan2(R[1,2]/sinβ, R[0,2]/sinβ)
alpha = math.atan2(R[1, 2] / sb, R[0, 2] / sb)
# γ = atan2(R[2,1]/sinβ, -R[2,0]/sinβ)
gamma = math.atan2(R[2, 1] / sb, -R[2, 0] / sb)
else:
# Gimbal: β ≈ 0 or π — only α+γ or α-γ determined
alpha = math.atan2(-R[0, 1], R[0, 0])
gamma = 0.0
if cbeta < 0:
# β = π
alpha = math.atan2(R[0, 1], -R[0, 0])
return alpha, beta, gamma
def hoa_rotation_matrix(
R3: np.ndarray,
*,
max_order: int = MAX_ORDER,
) -> np.ndarray:
"""Full Ambix ACN rotation matrix for orders 0..max_order.
Applies the *same* geometric rotation as R3 does to Cartesian vectors.
"""
max_order = int(max_order)
nch = (max_order + 1) ** 2
M = np.zeros((nch, nch), dtype=np.float64)
alpha, beta, gamma = rotation_matrix_to_zyz(R3)
# Order 0
M[0, 0] = 1.0
# Order 1: direct Cartesian for numerical exactness / convention lock.
# ACN: [Y, Z, X] = indices 1,2,3 ; cart = [X,Y,Z]
# v' = R3 @ v ⇒ [X',Y',Z'] = R3 @ [X,Y,Z]
if max_order >= 1:
# Build 3×3 block mapping [Y,Z,X] -> [Y',Z',X']
# [X'] [R00 R01 R02] [X]
# [Y'] = [R10 R11 R12] [Y]
# [Z'] [R20 R21 R22] [Z]
# Y' = R10 X + R11 Y + R12 Z
# Z' = R20 X + R21 Y + R22 Z
# X' = R00 X + R01 Y + R02 Z
# Coefficients of (Y,Z,X):
# Y' = R11 Y + R12 Z + R10 X
# Z' = R21 Y + R22 Z + R20 X
# X' = R01 Y + R02 Z + R00 X
B = np.array(
[
[R3[1, 1], R3[1, 2], R3[1, 0]],
[R3[2, 1], R3[2, 2], R3[2, 0]],
[R3[0, 1], R3[0, 2], R3[0, 0]],
],
dtype=np.float64,
)
M[1:4, 1:4] = B
for n in range(2, max_order + 1):
block = real_sh_rotation_block(n, alpha, beta, gamma)
# Verify det ~ 1; if complex conversion convention is flipped we may
# need transpose — tests catch this against plane-wave re-encode.
i0 = n * n # ACN start for order n is n^2 (m=-n → n*(n+1)+(-n)=n^2)
# Wait: n*(n+1)+(-n) = n^2 + n - n = n^2. Yes.
dim = 2 * n + 1
M[i0 : i0 + dim, i0 : i0 + dim] = block
return M
def apply_hoa_rotation(
hoa: np.ndarray,
M: np.ndarray,
) -> np.ndarray:
"""Apply precomputed rotation matrix to (C,) or (C,T) coefficients."""
a = np.asarray(hoa, dtype=np.float64)
nch = M.shape[0]
if a.ndim == 1:
aa = np.zeros(nch, dtype=np.float64)
n = min(nch, a.shape[0])
aa[:n] = a[:n]
out = M @ aa
if a.shape[0] > nch:
full = np.zeros_like(a)
full[:nch] = out
return full
if a.shape[0] < N_CHANNELS:
full = np.zeros(N_CHANNELS, dtype=np.float64)
full[:nch] = out
return full
return out
if a.ndim == 2:
aa = np.zeros((nch, a.shape[1]), dtype=np.float64)
n = min(nch, a.shape[0])
aa[:n] = a[:n]
out = M @ aa
if a.shape[0] >= N_CHANNELS:
full = np.zeros((max(a.shape[0], N_CHANNELS), a.shape[1]), dtype=np.float64)
full[:nch] = out
return full[: a.shape[0]]
full = np.zeros((N_CHANNELS, a.shape[1]), dtype=np.float64)
full[:nch] = out
return full
raise ValueError("hoa must be (C,) or (C,T)")
def clear_wigner_cache() -> None:
wigner_d.cache_clear()
|