Processing of a Signal Representing Radiation

• US 20080215267A1
• Filed: 07/12/2006
• Published: 09/04/2008
• Est. Priority Date: 07/12/2005
• Status: Active Grant

• Current Assignee: Commissariat a L'Energie Atomique (AES Corporation)
• Original Assignee: Commissariat a L'Energie Atomique (AES Corporation)
• Inventors: Barat, Eric, Eglin, Luc
• Granted Patent: US 8,600,694 B2
• Time in Patent Office: Days
• US Class Current: 702/66
• CPC Class Codes: G01R 35/00 Testing or calibrating of apparatus covered by the other groups of this subclass
• CPC Class Codes: G01T 1/17 Circuit arrangements not adapted to a particular type of detector pulse-selection circuits H03K, G01R

Abstract

A method for processing a noisy digital time signal yk of digital pitch k, corresponding to an initial analog signal St after being conditioned by a conditioning chain. The initial analog signal St comprises at least one pulse representing an information concerning at least one radiation (30) from a radiation, the radiation source (30) and the pulse having an energy distribution. The invention is characterized in that it consists in determining from the noisy time signal yk a non-noisy digital estimation signal Sk by means of a state model representing the conditioning imposed by the conditioning chain and in that the state model includes a Markovian variable rk to be estimated whereof at least two values are associated with physical characteristics of at least two typical pulses constituting a possible representation, at least approximate, of the pulse in the signal St.

Claims

1. A method for the treatment of a noisy temporal digital signal (y_k) with digital pitch (k), corresponding to an initial analogue signal (s_t) after it has been conditioned by a conditioning chain, with the said initial analogue signal (s_t) including at least one pulse that is representative of information on at least one radiation pulse (30) coming from a radiation source, with this radiation (30) and this pulse exhibiting an energy distribution, characterised in that one determines from the noisy temporal signal (y_k) a non-noisy estimated digital signal (s_k), by means of a state model that is representative of the said conditioning imposed by the conditioning chain, and in that the state model includes a Markov variable (r_k) to be estimated, of which at least two values are associated with physical characteristics of at least two model pulses, with these model pulses constituting a possible representation, or at least closely approaching it, of the pulses in the signal (S_t).

2. A treatment method according to claim 1, characterised in that it also includes a preliminary stage in which one segments the noisy temporal digital signal (y_k) into successive sections of N samples.

3. A treatment method according claim 1, characterised in that one of the values of the Markov variable corresponds to a pulse mode, known as the “

noise”

mode, for which it is considered that there exists no pulse in the signal with a given pitch (k).

4. A treatment method according to claim 3, characterised in that the state model also includes a vector (F) with constants that depend on the Markov variable, and that define, as physical characteristics, an energy range linked to the model pulse in question.

5. A treatment method according to claim 4, characterised in that the constants of vector F are proportional to a maximum predetermined amplitude (A_min) of a model pulse linked to the lowest energy range, with the “

noise”

mode not being concerned here.

6. A treatment method according to claim 1, characterised in that the state model includes a state vector (X_k) to be estimated, which includes a first component that correspond to the non-noisy digital estimate (s_k), that is to the digitised representation of the initial temporal signal (s_t) at the instant (t) corresponding to the pitch (k) in question, and in that the determination, from the noisy temporal signal (y_k), of the non-noisy estimated digital signal (s_k) by means of the state model includes an intermediate stage of implementation of a Kalman filter for observation of the noisy temporal digital signal (y_k), with this intermediate stage supplying an intermediate estimate (X_k/k) of the state vector (X_k), thus including a non-noisy intermediate digital estimation signal (s_k/k), with the latter corresponding to a first determination of the non-noisy estimated digital signal (s_k) that one is seeking to obtain.

7. A treatment method according to claim 6, characterised in that the determination, from the noisy temporal signal (y_k), of the non-noisy estimated digital signal (s_k) by means of the state model, includes a second intermediate stage of implementation of a smoothing algorithm, with this intermediate stage supplying an improved smoothed estimate (X_k/N) of the state vector (X_k), and therefore an improved smoothed estimate (s_k/N) of the non-noisy estimated digital signal (s_k) that one is seeking to obtain.

8. A treatment method according to claim 1, characterised in that the hidden variable (r_k) is estimated by means of a Viterbi algorithm.

9. A treatment method according to claim 1, characterised in that one uses a criterion for optimisation of the Maximum A Posteriori (MAP).

10. A treatment method according to claim 9, characterised in that the criterion of the Maximum A Posteriori (MAP) is combined with an Expectation-Maximisation (EM) algorithm.

11. A treatment method according to claim 10, characterised in that the criterion of the Maximum A Posteriori combined with the Expectation-Maximisation (EM) algorithm is implemented as appropriate:

for estimating the hidden variable (r_k)

for estimating the state vector (X_k), or

for estimating the hidden variable (r_k) and the state vector (X_k).

12. A treatment method according to claim 1, characterised in that the state is expressed in the following form:

$\{\begin{array}{c}{X}_{k+1}=A<br><br>\left(r_{k+1}\right)<br><br>X_k+G<br><br>\left(r_{k+1}\right)+B<br><br>\left(r_{k+1}\right)<br><br>v_{k+1}\\ y_k={\mathrm{CX}}_k\end{array}\hspace{1em}<br><br>$

Where;

$X_k=\left(\begin{array}{c}{s}_k\\ b_k\\ n_k\end{array}\right),$

in which A is a state matrix, C is an observation matrix, v_k is a Gaussian white-noise vector of unity variance, G is a vector that includes the constants of vector F, B is a dynamic noise matrix, b_k is a bias, and n_k is a noise that is intrinsic to a detector of the radiation used in the conditioning chain.

13. A treatment method according to claim 12, characterised in that it includes an initialisation stage in which the dynamic noise matrix B associated with the weakest energy channel is equal to that associated with the next highest energy channel, namely

B(r_k=0)=B(r_k=1)

and in which, for the values of r_k equal to or greater than 1, the matrix B(r_k) is determined by

B(r_k)−

F(r_k)−

F(r<sub>k−

1</sub>), where r_k≧

1

14. A computer program for the treatment of a noisy temporal digital signal (y_k) with digital pitch (k), corresponding to an initial analogue signal (s_t) after it has been conditioned by a conditioning chain, with the said initial analogue signal (s_t) including at least one pulse that is representative of information on at least one radiation pulse (30) coming from a radiation source, with this radiation (30) and this pulse exhibiting an energy distribution, characterised in that it includes instructions for implementation of the method according to claim 1.

16. A spectrometer that includes means that are designed to implement a method according to claim 1.

17. A bolometer that includes means that are designed to implement a method according to claim 1.

15. A system for the treatment of a noisy temporal digital signal (y_k) of pitch k corresponding to an initial analogue signal (s_t) after it has been conditioned by a conditioning chain, with the said analogue signal (s_t) including at least one pulse that is representative of information on radiation coming from a radiation source, with this radiation and this pulse exhibiting an energy distribution, characterised in that it includes means that are designed to cooperate with a state model that is representative of the said conditioning imposed by the conditioning chain in order to determine, from the noisy temporal signal (y_k), a non-noisy estimated digital signal (s_k), and in that the state model includes a Markov variable (r_k) to be estimated, of which at least two values are used to determine the physical characteristics of at least two model pulses, with these model pulses constituting a possible representation, or at least closely approaching it, of the pulse in the signal (s_t).