Skip to content

xwr.rsp.torch

Radar Signal Processing in Pytorch.

Info

This module mirrors the functionality of xwr.rsp.numpy.

Warning

This module is not automatically imported; you will need to explicitly import it:

from xwr.rsp import torch as xwr_torch

Since pytorch is not declared as a required dependency, you will also need to install torch yourself (or install the torch extra with pip install xwr[torch]).

Tip

The RSP implementations in this submodule support automatic differentiation in pytorch.

xwr.rsp.torch.AWR1642Boost

Bases: RSPTorch

Radar Signal Processing for the AWR1642 or AWR1843 with TX2 disabled.

Antenna Array

The TI AWR1642Boost (or AWR1843Boost with TX2 disabled) has a 1x8 linear MIMO array: 

1-1 1-2 1-3 1-4 2-1 2-2 2-3 2-4

Parameters:

Name Type Description Default
window bool | Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], bool]

whether to apply a hanning window. If bool, the same option is applied to all axes. If dict, specify per axis with keys "range", "doppler", "azimuth", and "elevation".

 False
size Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], int]

target size for each axis after zero-padding, specified by axis. If an axis is not spacified, it is not padded.

 {}
Source code in src/xwr/rsp/torch/rsp.py 
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
class AWR1642Boost(RSPTorch):
    """Radar Signal Processing for the AWR1642 or AWR1843 with TX2 disabled.

    !!! info "Antenna Array"

        The TI AWR1642Boost (or AWR1843Boost with TX2 disabled) has a
        1x8 linear MIMO array:
        ```
        1-1 1-2 1-3 1-4 2-1 2-2 2-3 2-4
        ```

    Args:
        window: whether to apply a hanning window. If `bool`, the same option
            is applied to all axes. If `dict`, specify per axis with keys
            "range", "doppler", "azimuth", and "elevation".
        size: target size for each axis after zero-padding, specified by axis.
            If an axis is not spacified, it is not padded.
    """

    def mimo_virtual_array(
        self, rd: Complex64[Tensor, "#batch doppler tx rx range"]
    ) -> Complex64[Tensor, "#batch doppler el az range"]:
        batch, doppler, tx, rx, range = rd.shape

        # 1843Boost cast as 1642Boost
        if tx == 3:
            if rx != 4:
                raise ValueError(
                    f"Expected (tx, rx)=3x4 in 1843Boost -> 1642Boost "
                    f"emulation, got tx={tx} and rx={rx}.")
            rd = rd[:, :, [0, 2], :, :]
        else:
            if tx != 2 or rx != 4:
                raise ValueError(
                    f"Expected (tx, rx)=2x4, got tx={tx} and rx={rx}.")

        return rd.reshape(batch, doppler, 1, -1, range)

xwr.rsp.torch.AWR1843AOP

Bases: RSPTorch

Radar Signal Processing for AWR1843AOP.

Antenna Array

In the TI AWR1843AOP, the MIMO virtual array is arranged in a 2D grid: 

1-1 2-1 3-1   ^
1-2 2-2 3-2   | Up
1-3 2-3 3-3
1-4 2-4 3-4 (TX-RX pairs)

Parameters:

Name Type Description Default
window bool | Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], bool]

whether to apply a hanning window. If bool, the same option is applied to all axes. If dict, specify per axis with keys "range", "doppler", "azimuth", and "elevation".

 False
size Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], int]

target size for each axis after zero-padding, specified by axis. If an axis is not specified, it is not padded.

 {}
Source code in src/xwr/rsp/torch/rsp.py 
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
class AWR1843AOP(RSPTorch):
    """Radar Signal Processing for AWR1843AOP.

    !!! info "Antenna Array"

        In the TI AWR1843AOP, the MIMO virtual array is arranged in a 2D grid:
            ```
            1-1 2-1 3-1   ^
            1-2 2-2 3-2   | Up
            1-3 2-3 3-3
            1-4 2-4 3-4 (TX-RX pairs)
            ```

    Args:
        window: whether to apply a hanning window. If `bool`, the same option
            is applied to all axes. If `dict`, specify per axis with keys
            "range", "doppler", "azimuth", and "elevation".
        size: target size for each axis after zero-padding, specified by axis.
            If an axis is not specified, it is not padded.
    """

    def mimo_virtual_array(
        self, rd: Complex64[Tensor, "#batch doppler tx rx range"]
    ) -> Complex64[Tensor, "#batch doppler el az range"]:
        _, _, tx, rx, _ = rd.shape
        if tx != 3 or rx != 4:
            raise ValueError(
                f"Expected (tx, rx)=3x4, got tx={tx} and rx={rx}.")

        return torch.swapaxes(rd, 2, 3)

xwr.rsp.torch.AWR1843Boost

Bases: RSPTorch

Radar Signal Processing for AWR1843Boost.

Antenna Array

In the TI AWR1843Boost, the MIMO virtual array has resolution 2x8, with a single 1/2-wavelength elevated middle antenna element: 

TX-RX:  2-1 2-2 2-3 2-4           ^
1-1 1-2 1-3 1-4 3-1 3-2 3-3 3-4   | Up

Parameters:

Name Type Description Default
window bool | Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], bool]

whether to apply a hanning window. If bool, the same option is applied to all axes. If dict, specify per axis with keys "range", "doppler", "azimuth", and "elevation".

 False
size Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], int]

target size for each axis after zero-padding, specified by axis. If an axis is not spacified, it is not padded.

 {}
Source code in src/xwr/rsp/torch/rsp.py 
 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
class AWR1843Boost(RSPTorch):
    """Radar Signal Processing for AWR1843Boost.

    !!! info "Antenna Array"

        In the TI AWR1843Boost, the MIMO virtual array has resolution 2x8, with
        a single 1/2-wavelength elevated middle antenna element:
        ```
        TX-RX:  2-1 2-2 2-3 2-4           ^
        1-1 1-2 1-3 1-4 3-1 3-2 3-3 3-4   | Up
        ```

    Args:
        window: whether to apply a hanning window. If `bool`, the same option
            is applied to all axes. If `dict`, specify per axis with keys
            "range", "doppler", "azimuth", and "elevation".
        size: target size for each axis after zero-padding, specified by axis.
            If an axis is not spacified, it is not padded.
    """

    def mimo_virtual_array(
        self, rd: Complex64[Tensor, "#batch doppler tx rx range"]
    ) -> Complex64[Tensor, "#batch doppler el az range"]:
        batch, doppler, tx, rx, range = rd.shape
        if tx != 3 or rx != 4:
            raise ValueError(
                f"Expected (tx, rx)=3x4, got tx={tx} and rx={rx}.")

        zeros = torch.zeros(
            (batch, doppler, 2, range), dtype=rd.dtype, device=rd.device)
        el_0 = torch.cat([zeros, rd[:, :, 1, :, :], zeros], dim=2)
        el_1 = torch.cat([rd[:, :, 0, :, :], rd[:, :, 2, :, :]], dim=2)
        return torch.stack([el_0, el_1], dim=2)

xwr.rsp.torch.AWR2944EVM

Bases: RSPTorch

Radar Signal Processing for AWR2944EVM.

Antenna Array

The AWR2944EVM has a virtual array on a 2x12 grid: 

        2-1 2-2 2-3 2-4
1-1 1-2 1-3 1-4 3-1 3-2 3-3 3-4 4-1 4-2 4-3 4-4
The horizontal spacing is 1/2 wavelength, and the vertical spacing is 0.8 wavelength.

Parameters:

Name Type Description Default
window bool | Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], bool]

whether to apply a hanning window. If bool, the same option is applied to all axes. If dict, specify per axis with keys "range", "doppler", "azimuth", and "elevation".

 False
size Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], int]

target size for each axis after zero-padding, specified by axis. If an axis is not spacified, it is not padded.

 {}
Source code in src/xwr/rsp/torch/rsp.py 
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
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
class AWR2944EVM(RSPTorch):
    """Radar Signal Processing for AWR2944EVM.

    !!! info "Antenna Array"

        The AWR2944EVM has a virtual array on a 2x12 grid:
        ```
                2-1 2-2 2-3 2-4
        1-1 1-2 1-3 1-4 3-1 3-2 3-3 3-4 4-1 4-2 4-3 4-4
        ```
        The horizontal spacing is 1/2 wavelength, and the vertical spacing is
        0.8 wavelength.

    Args:
        window: whether to apply a hanning window. If `bool`, the same option
            is applied to all axes. If `dict`, specify per axis with keys
            "range", "doppler", "azimuth", and "elevation".
        size: target size for each axis after zero-padding, specified by axis.
            If an axis is not spacified, it is not padded.
    """

    SAMPLE_TYPE = "I"

    def mimo_virtual_array(
        self, rd: Complex64[Tensor, "#batch doppler tx rx range"]
    ) -> Complex64[Tensor, "#batch doppler el az range"]:
        batch, doppler, tx, rx, range = rd.shape
        zeros_2 = torch.zeros(
            (batch, doppler, 1, 2, range),
            dtype=torch.complex64, device=rd.device)
        zeros_6 = torch.zeros(
            (batch, doppler, 1, 6, range),
            dtype=torch.complex64, device=rd.device)
        row0 = torch.cat([zeros_2, rd[:, :, 1:2, :, :], zeros_6], dim=3)
        row1 = torch.cat(
            [rd[:, :, 0:1, :, :], rd[:, :, 2:3, :, :], rd[:, :, 3:4, :, :]],
            dim=3)
        return torch.cat([row0, row1], dim=2)

    def elevation_azimuth(
        self, rd: Complex64[Tensor, "#batch doppler tx rx range"]
    ) -> Complex64[Tensor, "#batch doppler el az range"]:
        """Calculate elevation-azimuth spectrum from range-doppler spectrum.

        !!! warning

            Special treatment is needed for the AWR2944EVM since the two
            rows of virtual elements are 0.8 wavelength apart instead of
            0.5. We compute the DTFT along the elevation axis with the
            steering matrix corresponding to the 0.8 lambda spacing.

        Args:
            rd: range-doppler spectrum.

        Returns:
            Computed elevation-azimuth spectrum, with windowing and padding if
                specified.
        """
        mimo = self.mimo_virtual_array(rd)

        if self.window.get("elevation", self._default_window):
            mimo = self.hann(mimo, 2)
        if self.window.get("azimuth", self._default_window):
            mimo = self.hann(mimo, 3)

        az_size = self.size.get("azimuth", mimo.shape[3])
        spectrum = self.fft(mimo, axes=(3,), shift=(3,), size=(az_size,))

        el_size = self.size.get("elevation", mimo.shape[2])
        sin_theta = torch.linspace(-1, 1, el_size, device=rd.device)
        el_elements = torch.arange(
            mimo.shape[2], dtype=torch.float32, device=rd.device)
        phases = (-2j * torch.pi * 0.8
                  * torch.outer(sin_theta, el_elements).to(torch.complex64))
        steering_matrix = torch.exp(phases)

        el_az_spectrum = torch.einsum(
            'bdear,ke->bdkar',
            spectrum, steering_matrix
        )

        return el_az_spectrum

elevation_azimuth 

elevation_azimuth(
    rd: Complex64[Tensor, "#batch doppler tx rx range"],
) -> Complex64[Tensor, "#batch doppler el az range"]

Calculate elevation-azimuth spectrum from range-doppler spectrum.

Warning

Special treatment is needed for the AWR2944EVM since the two rows of virtual elements are 0.8 wavelength apart instead of 0.5. We compute the DTFT along the elevation axis with the steering matrix corresponding to the 0.8 lambda spacing.

Parameters:

Name Type Description Default
rd Complex64[Tensor, '#batch doppler tx rx range']

range-doppler spectrum.

 required

Returns:

Type Description
Complex64[Tensor, '#batch doppler el az range']

Computed elevation-azimuth spectrum, with windowing and padding if specified.
Source code in src/xwr/rsp/torch/rsp.py 
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
def elevation_azimuth(
    self, rd: Complex64[Tensor, "#batch doppler tx rx range"]
) -> Complex64[Tensor, "#batch doppler el az range"]:
    """Calculate elevation-azimuth spectrum from range-doppler spectrum.

    !!! warning

        Special treatment is needed for the AWR2944EVM since the two
        rows of virtual elements are 0.8 wavelength apart instead of
        0.5. We compute the DTFT along the elevation axis with the
        steering matrix corresponding to the 0.8 lambda spacing.

    Args:
        rd: range-doppler spectrum.

    Returns:
        Computed elevation-azimuth spectrum, with windowing and padding if
            specified.
    """
    mimo = self.mimo_virtual_array(rd)

    if self.window.get("elevation", self._default_window):
        mimo = self.hann(mimo, 2)
    if self.window.get("azimuth", self._default_window):
        mimo = self.hann(mimo, 3)

    az_size = self.size.get("azimuth", mimo.shape[3])
    spectrum = self.fft(mimo, axes=(3,), shift=(3,), size=(az_size,))

    el_size = self.size.get("elevation", mimo.shape[2])
    sin_theta = torch.linspace(-1, 1, el_size, device=rd.device)
    el_elements = torch.arange(
        mimo.shape[2], dtype=torch.float32, device=rd.device)
    phases = (-2j * torch.pi * 0.8
              * torch.outer(sin_theta, el_elements).to(torch.complex64))
    steering_matrix = torch.exp(phases)

    el_az_spectrum = torch.einsum(
        'bdear,ke->bdkar',
        spectrum, steering_matrix
    )

    return el_az_spectrum

xwr.rsp.torch.AWRL6844EVM

Bases: RSPTorch

Radar Signal Processing for AWRL6844.

Antenna Array

The AWRL6844 has a 4x4 MIMO virtual array with λ/2 spacing: 

2-1 2-4 1-1 1-4   ^
2-2 2-3 1-2 1-3   | Up
3-1 3-4 4-1 4-4
3-2 3-3 4-2 4-3 (TX-RX pairs)

TX Phase Relationship

TX1 and TX3 are in phase with each other. TX2 and TX4 are also in phase with each other, but are 180° out of phase with TX1 and TX3. Their contributions to the virtual array are negated accordingly.

Source: Table 3-1, EVM User's Guide: AWRL6844EVM IWRL6844EVM.

Parameters:

Name Type Description Default
window bool | Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], bool]

whether to apply a hanning window. If bool, the same option is applied to all axes. If dict, specify per axis with keys "range", "doppler", "azimuth", and "elevation".

 False
size Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], int]

target size for each axis after zero-padding, specified by axis. If an axis is not specified, it is not padded.

 {}
Source code in src/xwr/rsp/torch/rsp.py 
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
class AWRL6844EVM(RSPTorch):
    """Radar Signal Processing for AWRL6844.

    !!! info "Antenna Array"

        The AWRL6844 has a 4x4 MIMO virtual array with λ/2 spacing:
        ```
        2-1 2-4 1-1 1-4   ^
        2-2 2-3 1-2 1-3   | Up
        3-1 3-4 4-1 4-4
        3-2 3-3 4-2 4-3 (TX-RX pairs)
        ```

    !!! info "TX Phase Relationship"

        TX1 and TX3 are in phase with each other. TX2 and TX4 are also in
        phase with each other, but are 180° out of phase with TX1 and TX3.
        Their contributions to the virtual array are negated accordingly.

        Source: Table 3-1, *EVM User's Guide: AWRL6844EVM IWRL6844EVM*.

    Args:
        window: whether to apply a hanning window. If `bool`, the same option
            is applied to all axes. If `dict`, specify per axis with keys
            "range", "doppler", "azimuth", and "elevation".
        size: target size for each axis after zero-padding, specified by axis.
            If an axis is not specified, it is not padded.
    """

    SAMPLE_TYPE = "I"

    def mimo_virtual_array(
        self, rd: Complex64[Tensor, "#batch doppler tx rx range"]
    ) -> Complex64[Tensor, "#batch doppler el az range"]:
        _, _, tx, rx, _ = rd.shape
        if tx != 4 or rx != 4:
            raise ValueError(
                f"Expected (tx, rx)=4x4, got tx={tx} and rx={rx}.")

        tx_idx = torch.tensor(
            [[1, 1, 0, 0], [1, 1, 0, 0], [2, 2, 3, 3], [2, 2, 3, 3]],
            device=rd.device)
        rx_idx = torch.tensor(
            [[0, 3, 0, 3], [1, 2, 1, 2], [0, 3, 0, 3], [1, 2, 1, 2]],
            device=rd.device)
        phase = torch.tensor(
            [[-1, -1, 1, 1], [-1, -1, 1, 1], [1, 1, -1, -1], [1, 1, -1, -1]],
            dtype=torch.float32, device=rd.device)
        return rd[:, :, tx_idx, rx_idx, :] * phase[None, None, :, :, None]

xwr.rsp.torch.RSPTorch

Bases: RSP[Tensor], ABC

Base Radar Signal Processing with common functionality.

Parameters:

Name Type Description Default
window bool | Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], bool]

whether to apply a hanning window. If bool, the same option is applied to all axes. If dict, specify per axis with keys "range", "doppler", "azimuth", and "elevation".

 False
size Mapping[Literal['range', 'doppler', 'azimuth', 'elevation'], int]

target size for each axis after zero-padding, specified by axis. If an axis is not spacified, it is not padded.

 {}
Source code in src/xwr/rsp/torch/rsp.py 
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
class RSPTorch(RSP[Tensor], ABC):
    """Base Radar Signal Processing with common functionality.

    Args:
        window: whether to apply a hanning window. If `bool`, the same option
            is applied to all axes. If `dict`, specify per axis with keys
            "range", "doppler", "azimuth", and "elevation".
        size: target size for each axis after zero-padding, specified by axis.
            If an axis is not spacified, it is not padded.
    """

    def fft(
        self, array: Complex64[Tensor, "..."] | Float32[Tensor, "..."],
        axes: tuple[int, ...],
        size: tuple[int, ...] | None = None,
        shift: tuple[int, ...] | None = None
    ) -> Complex64[Tensor, "..."]:
        if array.dtype == torch.float32:
            fftd = torch.fft.rfftn(array, s=size, dim=axes)
        else:
            fftd = torch.fft.fftn(array, s=size, dim=axes)
        if shift is None:
            return fftd
        else:
            return torch.fft.fftshift(fftd, dim=shift)

    @staticmethod
    def pad(
        x: Shaped[Tensor, "..."], axis: int, size: int
    ) -> Shaped[Tensor, "..."]:
        if size <= x.shape[axis]:
            raise ValueError(
                f"Cannot zero-pad axis {axis} to target size {size}, which is "
                f"less than or equal the current size {x.shape[axis]}.")

        shape = list(x.shape)
        shape[axis] = size - x.shape[axis]
        zeros = torch.zeros(shape, dtype=x.dtype, device=x.device)

        return torch.concatenate([x, zeros], dim=axis)

    @staticmethod
    def hann(
        x: Complex64[Tensor, "..."] | Float32[Tensor, "..."], axis: int
    ) -> Complex64[Tensor, "..."] | Float32[Tensor, "..."]:
        hann = np.hanning(x.shape[axis] + 2).astype(np.float32)[1:-1]
        broadcast: list[None | slice] = [None] * x.ndim
        broadcast[axis] = slice(None)
        window = torch.from_numpy(
            (hann / np.mean(hann))[tuple(broadcast)]).to(x.device)
        return x * window