SensorNoise Class Reference

#include <SensorNoise.h>

Public Member Functions
	SensorNoise (unsigned int seed)
bool	IsDeterministic () const
	True when the generator is seeded zero and callers should add no noise.
double	Uniform (double low, double high)
double	Gaussian (double mean, double stddev)
int	Poisson (double lambda)
double	Gamma (double shape)

Detailed Description

Draws the random numbers the AUV sensors need, reproducibly on any platform.

The distributions are hand-rolled on top of std::mt19937 and std::uniform_real_distribution only. This is the single most important property of this class. std::normal_distribution, std::poisson_distribution and std::gamma_distribution have implementation-defined algorithms: the same engine and the same seed produce different numbers under libstdc++, libc++ and the MSVC standard library, so a recorded regression baseline would pass on one platform and fail on another. std::mt19937 and std::uniform_real_distribution are specified exactly by the standard, and every distribution here is built from them by a published algorithm, so the sequence is fixed everywhere. Do not replace any of these with the standard distributions.

A seed of zero means deterministic operation: the sensor is noise free and the simulation is an exact regression baseline, which is also the most useful configuration when debugging a model. Callers ask IsDeterministic and skip the perturbation entirely rather than drawing zero-width noise. A generator seeded zero that is asked for a draw anyway returns the mean of the distribution, so forgetting to branch degrades the model quietly rather than reintroducing randomness.

Note

This is a helper, not a SimObject, so it has no @group block and appears in no model list.

Constructor & Destructor Documentation

SensorNoise()

SensorNoise::SensorNoise ( unsigned int  seed )

explicit

Creates a generator with the given seed.

Parameters

seed	The random seed; zero means deterministic and noise free.

Member Function Documentation

Gamma()

double SensorNoise::Gamma

(

double

shape

)

Draws a gamma distributed number of unit mean.

The Marsaglia-Tsang method is used with scale 1/shape, so the mean is one and the variance is 1/shape. This is the multiplicative speckle factor the acoustic sensors need.

Parameters

shape

The shape parameter; must be positive.

Returns

One gamma distributed sample with mean one.

Gaussian()

double SensorNoise::Gaussian	(	double	mean,
		double	stddev
	)

Draws a normally distributed number by the Box-Muller transform.

Parameters

mean	The mean of the distribution.
stddev	The standard deviation of the distribution.

Returns

One normally distributed sample.

Poisson()

int SensorNoise::Poisson

(

double

lambda

)

Draws a Poisson distributed count.

Knuth's product method is used below kPoissonNormalLimit, where its expected number of uniform draws is small, and a rounded and clamped normal approximation above it, where the product method becomes slow and the approximation is good.

Parameters

lambda

The mean of the distribution; must not be negative.

Returns

One Poisson distributed count.

Uniform()

double SensorNoise::Uniform	(	double	low,
		double	high
	)

Draws a uniform number.

Parameters

low	The lower bound, inclusive.
high	The upper bound, exclusive.

Returns

A number drawn uniformly from [low, high).

---

The documentation for this class was generated from the following file:

• /update_root/webfhsim/reloadrepos/fhsim_fishery/src/auv/SensorNoise.h