`b_asic.fft_operations`¶

B-ASIC FFT Operations Module.

Contains selected FFT butterfly operations.

class b_asic.fft_operations.R2Butterfly(src0: SignalSourceProvider | None = None, src1: SignalSourceProvider | None = None, name: str = '', latency: int | None = None, latency_offsets: dict[str, int] | None = None, execution_time: int | None = None)¶

Bases: AbstractOperation

Radix-2 butterfly operation for FFTs.

Gives the result of adding its two inputs, as well as the result of subtracting the second input from the first one. This corresponds to a 2-point DFT.

\[\begin{split}y_0 & = x_0 + x_1\\ y_1 & = x_0 - x_1\end{split}\]

Parameters:

src0, src1SignalSourceProvider, optional: The two signals to compute the 2-point DFT of.
nameName, optional: Operation name.
latencyint, optional: Operation latency (delay from input to output in time units).
latency_offsetsdict[str, int], optional: Used if inputs have different arrival times or if the inputs should arrive after the operator has stared. For example, {"in0": 0, "in1": 1} which corresponds to src1 arriving one time unit later than src0 and one time unit later than the operator starts. If not provided and latency is provided, set to zero. Hence, the previous example can be written as {"in1": 1} only.
execution_timeint, optional: Operation execution time (time units before operator can be reused).

Evaluate the operation and generate a list of output values.

Parameters:

*inputs: List of input values.
data_typeDataType, optional: Data type to use for quantization during evaluation.

is_linear = True¶

classmethod type_name() → TypeName¶: Get the type name of this graph component.

class b_asic.fft_operations.R2DIFButterfly(w: complex, src0: SignalSourceProvider | None = None, src1: SignalSourceProvider | None = None, name: str = '', latency: int | None = None, latency_offsets: dict[str, int] | None = None, execution_time: int | None = None)¶

Bases: AbstractOperation

Radix-2 decimation-in-frequency butterfly operation for FFTs.

Gives the result of adding its two inputs, as well as the result of subtracting the second input from the first one. This corresponds to a 2-point DFT. The second output is multiplied with the provided twiddle factor (TF).

\[\begin{split}y_0 & = x_0 + x_1\\ y_1 & = w(x_0 - x_1)\end{split}\]

Parameters:

wcomplex: Twiddle factor to multiply the second output with.
src0, src1SignalSourceProvider, optional: The two signals to compute the 2-point DFT of.
nameName, optional: Operation name.
latencyint, optional: Operation latency (delay from input to output in time units).
latency_offsetsdict[str, int], optional: Used if inputs have different arrival times or if the inputs should arrive after the operator has stared. For example, {"in0": 0, "in1": 1} which corresponds to src1 arriving one time unit later than src0 and one time unit later than the operator starts. If not provided and latency is provided, set to zero. Hence, the previous example can be written as {"in1": 1} only.
execution_timeint, optional: Operation execution time (time units before operator can be reused).

Evaluate the operation and generate a list of output values.

Parameters:

*inputs: List of input values.
data_typeDataType, optional: Data type to use for quantization during evaluation.

is_linear = True¶

classmethod type_name() → TypeName¶: Get the type name of this graph component.

property w: complex¶: Get the twiddle factor.

class b_asic.fft_operations.R2DITButterfly(w: complex, src0: SignalSourceProvider | None = None, src1: SignalSourceProvider | None = None, name: str = '', latency: int | None = None, latency_offsets: dict[str, int] | None = None, execution_time: int | None = None)¶

Bases: AbstractOperation

Radix-2 decimation-in-time butterfly operation for FFTs.

Gives the result of adding its two inputs, as well as the result of subtracting the second input from the first one. This corresponds to a 2-point DFT. The second input is multiplied with the provided twiddle factor (TF).

\[\begin{split}y_0 & = x_0 + w*x_1\\ y_1 & = x_0 - w*x_1\end{split}\]

Parameters:

wcomplex: Twiddle factor to multiply the second input with.
src0, src1SignalSourceProvider, optional: The two signals to compute the 2-point DFT of.
nameName, optional: Operation name.
latencyint, optional: Operation latency (delay from input to output in time units).
latency_offsetsdict[str, int], optional: Used if inputs have different arrival times or if the inputs should arrive after the operator has stared. For example, {"in0": 0, "in1": 1} which corresponds to src1 arriving one time unit later than src0 and one time unit later than the operator starts. If not provided and latency is provided, set to zero. Hence, the previous example can be written as {"in1": 1} only.
execution_timeint, optional: Operation execution time (time units before operator can be reused).

Evaluate the operation and generate a list of output values.

Parameters:

*inputs: List of input values.
data_typeDataType, optional: Data type to use for quantization during evaluation.

is_linear = True¶

classmethod type_name() → TypeName¶: Get the type name of this graph component.

property w: complex¶: Get the twiddle factor.

class b_asic.fft_operations.R2TFButterfly(w0: complex, w1: complex, src0: SignalSourceProvider | None = None, src1: SignalSourceProvider | None = None, name: str = '', latency: int | None = None, latency_offsets: dict[str, int] | None = None, execution_time: int | None = None)¶

Bases: AbstractOperation

Radix-2 butterfly operation for FFTs with twiddle factor (TF) multiplications at both outputs.

Gives the result of adding its two inputs, as well as the result of subtracting the second input from the first one. This corresponds to a 2-point DFT. Both outputs are multiplied with the provided TFs.

\[\begin{split}y_0 & = w_0(x_0 + x_1)\\ y_1 & = w_1(x_0 - x_1)\end{split}\]

Parameters:

w0complex: Twiddle factor to multiply the first output with.
w1complex: Twiddle factor to multiply the second output with.
src0, src1SignalSourceProvider, optional: The two signals to compute the 2-point DFT of.
nameName, optional: Operation name.
latencyint, optional: Operation latency (delay from input to output in time units).
latency_offsetsdict[str, int], optional: Used if inputs have different arrival times or if the inputs should arrive after the operator has stared. For example, {"in0": 0, "in1": 1} which corresponds to src1 arriving one time unit later than src0 and one time unit later than the operator starts. If not provided and latency is provided, set to zero. Hence, the previous example can be written as {"in1": 1} only.
execution_timeint, optional: Operation execution time (time units before operator can be reused).

Evaluate the operation and generate a list of output values.

Parameters:

*inputs: List of input values.
data_typeDataType, optional: Data type to use for quantization during evaluation.

is_linear = True¶

classmethod type_name() → TypeName¶: Get the type name of this graph component.

property w0: complex¶: Get the twiddle factor that the first output is multiplied with.

property w1: complex¶: Get the twiddle factor that the second output is multiplied with.

class b_asic.fft_operations.R3Winograd(src0: SignalSourceProvider | None = None, src1: SignalSourceProvider | None = None, src2: SignalSourceProvider | None = None, name: str = '', latency: int | None = None, latency_offsets: dict[str, int] | None = None, execution_time: int | None = None)¶

Bases: AbstractOperation

Three-point Winograd DFT.

\[\begin{split}u & = -\frac{2 \pi}{3} \\ c_{30} & = \cos(u) - 1 \\ c_{31} & = j \sin(u) \\ s_0 & = x_1 + x_2 \\ s_1 & = x_1 - x_2 \\ s_2 & = s_0 + x_0 \\ m_0 & = c_{30} s_0 \\ m_1 & = c_{31} s_1 \\ s_3 & = s_2 + m_0 \\ s_4 & = s_3 + m_1 \\ s_5 & = s_3 - m_1 \\ y_0 & = s_2 \\ y_1 & = s_4 \\ y_2 & = s_5\end{split}\]

Parameters:

src0, src1, src2SignalSourceProvider, optional: The three signals to compute the 3-point DFT of.
nameName, optional: Operation name.
latencyint, optional: Operation latency (delay from input to output in time units).
latency_offsetsdict[str, int], optional: Used if inputs have different arrival times or if the inputs should arrive after the operator has stared. For example, {"in0": 0, "in1": 1, "in2": 0} which corresponds to src1 arriving one time unit later than the other inputs and one time unit later than the operator starts. If not provided and latency is provided, set to zero. Hence, the previous example can be written as {"in1": 1} only.
execution_timeint, optional: Operation execution time (time units before operator can be reused).

Evaluate the operation and generate a list of output values.

Parameters:

*inputs: List of input values.
data_typeDataType, optional: Data type to use for quantization during evaluation.

is_linear = True¶

classmethod type_name() → TypeName¶: Get the type name of this graph component.

class b_asic.fft_operations.R4Butterfly(src0: SignalSourceProvider | None = None, src1: SignalSourceProvider | None = None, src2: SignalSourceProvider | None = None, src3: SignalSourceProvider | None = None, name: str = '', latency: int | None = None, latency_offsets: dict[str, int] | None = None, execution_time: int | None = None)¶

Bases: AbstractOperation

Radix-4 butterfly operation for FFTs.

This corresponds to a 4-point DFT.

\[\begin{split}s0 & = x_0 + x_2\\ s1 & = x_1 + x_3\\ s2 & = x_0 - x_2\\ s3 & = -jx_1 + jx_3\\ y0 & = s0 + s1\\ y1 & = s2 + s3\\ y2 & = s0 - s1\\ y3 & = s2 - s3\end{split}\]

Parameters:

src0, src1, src2, src3SignalSourceProvider, optional: The four signals to compute the 4-point DFT of.
nameName, optional: Operation name.
latencyint, optional: Operation latency (delay from input to output in time units).
latency_offsetsdict[str, int], optional: Used if inputs have different arrival times or if the inputs should arrive after the operator has stared. For example, {"in0": 0, "in1": 1, "in2": 0, "in3": 0} which corresponds to src1 arriving one time unit later than the other inputs and one time unit later than the operator starts. If not provided and latency is provided, set to zero. Hence, the previous example can be written as {"in1": 1} only.
execution_timeint, optional: Operation execution time (time units before operator can be reused).

Evaluate the operation and generate a list of output values.

Parameters:

*inputs: List of input values.
data_typeDataType, optional: Data type to use for quantization during evaluation.

is_linear = True¶

classmethod type_name() → TypeName¶: Get the type name of this graph component.

class b_asic.fft_operations.R5Winograd(src0: SignalSourceProvider | None = None, src1: SignalSourceProvider | None = None, src2: SignalSourceProvider | None = None, src3: SignalSourceProvider | None = None, src4: SignalSourceProvider | None = None, name: str = '', latency: int | None = None, latency_offsets: dict[str, int] | None = None, execution_time: int | None = None)¶

Bases: AbstractOperation

Five-point Winograd DFT.

Parameters:

src0, src1, src2, src3, src4SignalSourceProvider, optional: The five signals to compute the 5-point DFT of.
nameName, optional: Operation name.
latencyint, optional: Operation latency (delay from input to output in time units).
latency_offsetsdict[str, int], optional: Used if inputs have different arrival times or if the inputs should arrive after the operator has stared. For example, {"in0": 0, "in1": 1, "in2": 0, "in3": 0, "in4": 0,} which corresponds to src1 arriving one time unit later than the other inputs and one time unit later than the operator starts. If not provided and latency is provided, set to zero. Hence, the previous example can be written as {"in1": 1} only.
execution_timeint, optional: Operation execution time (time units before operator can be reused).

Evaluate the operation and generate a list of output values.

Parameters:

*inputs: List of input values.
data_typeDataType, optional: Data type to use for quantization during evaluation.

is_linear = True¶

classmethod type_name() → TypeName¶: Get the type name of this graph component.

class b_asic.fft_operations.R7Winograd(src0: SignalSourceProvider | None = None, src1: SignalSourceProvider | None = None, src2: SignalSourceProvider | None = None, src3: SignalSourceProvider | None = None, src4: SignalSourceProvider | None = None, src5: SignalSourceProvider | None = None, src6: SignalSourceProvider | None = None, name: str = '', latency: int | None = None, latency_offsets: dict[str, int] | None = None, execution_time: int | None = None)¶

Bases: AbstractOperation

Seven-point Winograd DFT.

Parameters:

src0, src1, src2, src3, src4, src5, src6SignalSourceProvider, optional: The seven signals to compute the 7-point DFT of.
nameName, optional: Operation name.
latencyint, optional: Operation latency (delay from input to output in time units).
latency_offsetsdict[str, int], optional: Used if inputs have different arrival times or if the inputs should arrive after the operator has stared. For example, {"in0": 0, "in1": 1, "in2": 0, "in3": 0, "in4": 0, "in5": 0, "in6": 0,} which corresponds to src1 arriving one time unit later than the other inputs and one time unit later than the operator starts. If not provided and latency is provided, set to zero. Hence, the previous example can be written as {"in1": 1} only.
execution_timeint, optional: Operation execution time (time units before operator can be reused).

Evaluate the operation and generate a list of output values.

Parameters:

*inputs: List of input values.
data_typeDataType, optional: Data type to use for quantization during evaluation.

is_linear = True¶

classmethod type_name() → TypeName¶: Get the type name of this graph component.

b_asic.fft_operations¶

`b_asic.fft_operations`¶