detectors¶
lowpassdetector¶
responsivity¶
The detected photo-current \(I_p~[C/s]\) can be described in terms of the optical power \(P_o~[W]\) through the quantum efficiency \(\eta\) and the carrier frequency \(f=c/\lambda~[1/s]\) of the light:
However, it’s often more useful to describe it in terms of a single quantity \(r~[A/W]\):
where we defined \(r~[A/W]\) as the responsivity of the detector. At a wavelength of \(1550nm\) and for a quantum efficiency of \(\eta=1\), we have:
hence a default value of 1 for the responsivity is appropriate.
implementation¶
A simple detector based on low-pass filtering the signal.
The LospassDetector performs the (differentiable) PyTorch equivalent of the
following numpy/scipy function:
from scipy.signal import butter, lfilter
def detect(x, bitrate, samplerate, cutoff_frequency, responsivity, filter_order):
normal_cutoff = cutoff_frequency / ( 0.5 * samplerate)
b, a = butter(N=filter_order, Wn=normal_cutoff, btype='lowpass', analog=False)
return lfilter(b, a, responsivity * x, axis=0)
-
class
photontorch.detectors.lowpassdetector.LowpassDetector(bitrate=40000000000.0, samplerate=160000000000.0, cutoff_frequency=20000000000.0, filter_order=4, responsivity=1.0)¶ Bases:
photontorch.nn.nn.ModuleDetect by lowpass filtering the signal.
The LowpassDetector transforms a raw optical power [W] to a detection current [A].
-
__init__(bitrate=40000000000.0, samplerate=160000000000.0, cutoff_frequency=20000000000.0, filter_order=4, responsivity=1.0)¶ - Parameters
bitrate (float) – [1/s] data rate of the signal to filter
samplerate (float) – [1/s] sample rate of the signal to filter
cutoff_frequency (float) – [1/s] cutoff frequency of the detector
filter_order (int) – filter order of the butter filter
responsivity (float) – [A/W] resonsivity of the detector
-
forward(signal, num_splits=1, split_padding=5, bitrate=None, samplerate=None, cutoff_frequency=None, filter_order=None, responsivity=None)¶ detect a bitstream by low-pass filtering
- Parameters
signal (Tensor) – [W] optical power to detect.
num_splits (int) – number of parallel parts to split the timestream in
split_padding (int) – number of bits padding when splitting the timstream in parts.
bitrate (optional, float) – [1/s] override data rate of the signal to filter
samplerate (optional, float) – [1/s] override sample rate of the signal to filter
cutoff_frequency (optional, float) – [1/s] override cutoff frequency of the detector
filter_order (optional, int) – override filter order of the butter filter
responsivity (optional, float) – [A/W] override resonsivity of the detector
Note
If a bitrate and/or samplerate can be found in the current environment, those values will be regarded as keyword arguments and hence get precedence over the values given during the detector initialization.
Note
Splitting the signal in parts to be processed in parallel can considerably speed up the detection process. However, it remains an approximation, as in theory each detected signal point depends (with an exponentially decreasing factor) on all previous detected signals.
To partly circumvent any large discontinuities when recomposing the detected signal from the detected parts, a split_padding should be defined, which pads a number of bits from the previous / next signal part before and after the current signal part to detect.
-
training¶
-
photodetector¶
responsivity¶
As mentioned above, a responsivity of \(r=1 A/W\) is a valid value, but almost always it will be lower. If the detected signal is not as noisy as expected, you probably used a too high responsivity.
thermal noise¶
On top of the detected current, thermal noise should be added. The thermal noise variance \(\sigma_t^2\) [\(C^2/s^2\)] is given by:
with \(f_c~[1/s]\) the cutoff frequency of the signal and \(N_t~[C/\sqrt s]\) the thermal noise spectral density. This represenation (which is the representation used in VPI for example) requires you to use a quantity with \(N_t~[C/\sqrt s]\) with weird units. We therefore prefer another representation (see for example wikipedia):
To find a good default value for the load resistance \(R_L\), we can equate this equation with the previous one using the default VPI value for \(N_t=10^{-11}C/\sqrt s\):
where we assumed for the room temparture \(T=300K\).
shot noise¶
The shot noise variance \(\sigma_t^2\) [\(C^2/s^2\)] for a PIN photodetector is given by:
with \(q~[C]\) the elementary charge and \(I_d\) the dark current of the photodetector. Notice that this power is dependent on the photocurrent \(I_p~[C/s]\) itself.
In some representations, \(\mu_p=\left< I_p \right>~[C/s]\) is used instead. We choose not to take this average to have a more accurate reprentation. Moreover, low-pass filtering already takes a kind of moving average into account.
total noise¶
The total post-detection noise variance \(\sigma_p^2\) [\(C^2/s^2\)] now depends on the thermal noise variance \(\sigma_t^2\) and the shot noise variance \(\sigma_s^2\):
implementation¶
Photodetector
The photodetector in this module transforms a raw optical power [W] to a (possibly noisy) detection current [A]. It takes thermal noise and shot noise into account.
-
class
photontorch.detectors.photodetector.Photodetector(bitrate=40000000000.0, samplerate=160000000000.0, cutoff_frequency=20000000000.0, responsivity=1.0, dark_current=1e-10, load_resistance=166, filter_order=4, seed=None)¶ Bases:
photontorch.detectors.lowpassdetector.LowpassDetectorRealistic Photodector Model.
The photodetector transforms a raw optical power [W] to a (possibly noisy) detection current [A].
This photodetector takes thermal noise and shot noise into account.
-
__init__(bitrate=40000000000.0, samplerate=160000000000.0, cutoff_frequency=20000000000.0, responsivity=1.0, dark_current=1e-10, load_resistance=166, filter_order=4, seed=None)¶ - Parameters
bitrate (float) – [1/s] data rate of the signal to filter
samplerate (float) – [1/s] sample rate of the signal to filter
cutoff_frequency (float) – [1/s] cutoff frequency of the detector
responsivity (float) – [A/W] responsivity of the photodector
dark_current (float) – [A] dark current adding to the noise
load_resistance (float) – [Ohm] load resistance of the detector
filter_order (int) – filter order of the butter filter
seed (int) – random seed for the detector noise
-
forward(signal, num_splits=1, split_padding=5, bitrate=None, samplerate=None, cutoff_frequency=None, responsivity=None, filter_order=None, dark_current=None, load_resistance=None)¶ detect a bitstream by first adding physical noise and then low-pass filtering.
- Parameters
signal (Tensor) – signal to detect.
num_splits (int) – number of parallel parts to split the timestream in
split_padding (int) – number of bits padding when splitting the timstream in parts.
bitrate (optional, float) – [1/s] override data rate of the signal to filter
samplerate (optional, float) – [1/s] override sample rate of the signal to filter
cutoff_frequency (optional, float) – [1/s] override cutoff frequency of the detector
filter_order (optional, int) – override filter order of the butter filter
responsivity (optional, float) – [A/W] override responsivity of the photodector
dark_current (optional, float) – [A] override dark current adding to the noise
load_resistance (optional, float) – [Ohm] override load resistance of the detector
Note
If a bitrate and/or samplerate can be found in the current environment, those values will be regarded as keyword arguments and hence get precedence over the values given during the detector initialization.
Note
Splitting the signal in parts to be processed in parallel can considerably speed up the detection process. However, it remains an approximation, as in theory each detected signal point depends (with an exponentially decreasing factor) on all previous detected signals.
To partly circumvent any large discontinuities when recomposing the detected signal from the detected parts, a split_padding should be defined, which pads a number of bits from the previous / next signal part before and after the current signal part to detect.
-
training¶
-