Iterative process for simulating a radiofrequency chain

Abstract

The invention relates to an iterative method for simulating a radiofrequency chain comprising: an initialization (4) of a set of combinations of values ​​of parameters representative of the radiofrequency chain; an iterative calculation implementing the steps corresponding to: a simulation (5) of a radiofrequency chain configured to calculate (6), for each combination of the set, the value of at least one metric of the radiofrequency chain, a metric being a function dependent on at least one parameter of the radiofrequency chain; and an optimization (7) by an Artificial Intelligence algorithm of genetic type, in which: a chromosome corresponds to a combination of the set, and a gene corresponds to a combination parameter of the set; and the calculated cost function (8) depends on a weighting of the at least one simulated metric(s). Figure for the abstract: [Fig.2]

Description

Title of the invention: Iterative method for simulating a radiofrequency chain

[0001] The invention relates to an iterative method for simulating a radiofrequency chain

[0002] The invention relates to the configuration of radiofrequency chains in the upstream engineering phases, a parametric exploration of the different components through simulation tools appears to be an interesting solution. To do this, testing a very important combination is sometimes necessary in order to select the configuration(s) associated with the best performance criteria defined by the architect.

[0003] However, if the exploration range is too large, the calculation time can be a hindrance to pre-sizing.

[0004] A radiofrequency chain comprises all the elements participating in the emission and / or reception of radiofrequency waves, as well as their configuration and operation.

[0005] Methods for simulating a radiofrequency chain are known, as illustrated in [Fig.l] comprising: - a determination 1 of an exhaustive set of combinations of values ​​of parameters representative of the radiofrequency chain; - a radiofrequency chain simulation 2 configured to calculate, for each combination of the set, the value of at least one metric of the radiofrequency chain, a metric being a function dependent on at least one parameter of the radiofrequency chain; and - manual post-processing 3, by experts to choose the best possible combination from the predetermined set of combinations.

[0006] Existing modern simulation tools do not allow for automatic exploration of a large range of parameters. In addition, many such sizing operations are carried out using Excel files, making the exploitation of this data difficult.

[0007] Exploration models would need to be fully developed. In the case of exhaustive exploration, which would consist of going through all possible combinations, the calculation time could then be prohibitive.

[0008] One aim of the invention is to overcome the problems mentioned above.

[0009] According to one aspect of the invention, there is proposed an iterative method for simulating a radiofrequency chain comprising: - an initialization of a set of combinations of parameter values ​​re presentations of the radio frequency chain; - an iterative calculation implementing the steps corresponding to: - a radio frequency chain simulation configured to calculate, for each combination of the set, the value of at least one metric of the radio frequency chain, a metric being a function dependent on at least one parameter of the radio frequency chain; and - optimization by an Artificial Intelligence or AI algorithm of the genetic type, in which: - a chromosome corresponds to a combination of the set, and a gene corresponds to a combination parameter of the set; and - the calculated cost function depends on a weighting of at least one simulated metric(s).

[0010] In one embodiment, the optimization step comprises selecting two chromosomes from the set, generating two replacement chromosomes, and replacing, in the set, two chromosomes from among the unselected chromosomes, with the two calculated chromosomes.

[0011] According to one embodiment, the selection of two chromosomes in the set implements a selection by rank, a selection by tournament, or an elitist selection.

[0012] In one embodiment, the radio frequency chain metrics comprise at least one cumulative noise factor of a component of the radio frequency chain, and / or at least one cumulative gain of a component of the radio frequency chain, and / or at least one cumulative third-order intercept point of a component of the radio frequency chain.

[0013] According to one embodiment, the parameters of the radiofrequency chain comprise at least one noise factor of a component of the radiofrequency chain, and / or at least one gain of a component of the radiofrequency chain, and / or at least one third-order interception point of a component of the radiofrequency chain.

[0014] In one embodiment, the iterative calculation has as a stopping criterion the reaching of a predetermined number of iterations.

[0015] Alternatively, the iterative calculation has as a stopping criterion the achievement of a predetermined number of successive iterative calculations of the cost function of a chromosome of the set all included in an interval of values ​​of width less than a threshold.

[0016] The invention will be better understood by studying a few embodiments described as non-limiting examples and illustrated by the appended drawings in which:

[0017] [Fig.l] schematically illustrates a method of simulating a radiofrequency chain, according to the state of the art;

[0018] [Fig.2] schematically illustrates an iterative process for simulating a radio chain frequency, according to one aspect of the invention; and

[0019] [Fig.3] schematically illustrates an example of the implementation of an iterative method of si- emulation of a radiofrequency chain, according to one aspect of the invention.

[0020] Throughout the figures, elements having identical references are similar.

[0021] [Fig.2] schematically illustrates the iterative method of simulating a radiofrequency chain according to one aspect of the invention.

[0022] The iterative method of simulating a radiofrequency chain comprises: - an initialization 4 of a set of combinations of parameter values ​​representative of the radiofrequency chain; - an iterative calculation implementing the steps corresponding to: - a radiofrequency chain simulation 5 configured to calculate 6, for each combination of the set, the value of at least one metric of the radiofrequency chain, a metric being a function dependent on at least one parameter of the radiofrequency chain; and - an optimization 7 by a genetic type Artificial Intelligence algorithm, in which: - a chromosome corresponds to a combination of the set, and a gene corresponds to a combination parameter of the set; and - the calculated cost function 8 depends on a weighting of at least one simulated metric(s).

[0023] The optimization step comprises a selection 9 of two chromosomes from the set, a generation 10 of two replacement chromosomes whose cost function value is improved compared to the values ​​of the two selected chromosomes, and a replacement 11, from the set, of two chromosomes among the unselected chromosomes, in this case the two worst ones having the lowest cost values, by the two calculated chromosomes. As a variant of the maximization of the cost function, in the case of minimization of the cost function, the two worst ones would be the highest cost values.

[0024] The iterative calculation has as stopping criterion 12 the achievement of a predetermined number of iterations, or the achievement of a predetermined number of successive iterative calculations of the cost function of a chromosome of the set all included in an interval of values ​​of width less than a threshold.

[0025] In other words, the present invention consists in particular of adding an AI overlay on an already existing simulator.

[0026] If we assume that we have a radio frequency chain simulator (commercial tool, functions developed in a programming language), then a parametric exploration can be carried out as explained previously.

[0027] Indeed, the user can enter exploration ranges exhaustively so that the tool explores all possible combinations. The basic functions of the si emulator are therefore used as many times as there are combinations. This approach also requires manual post-processing because it is then necessary to use all the solutions obtained in order to bring out the best solution(s) according to criteria to be established.

[0028] The present invention makes it possible to avoid a blind browsing of all the combinations with manual post-processing by experts. This is avoided by a symbolic AI algorithm, i.e. based on functions to be optimized and not large quantities of data, of the genetic type.

[0029] More specifically, a genetic type AI algorithm is used to automatically converge towards the best configurations. This algorithm is particularly suited to our particular context of radiofrequency chain simulation by a study of its different parameters such as the probability of mutation, the size of the population or the type of selection.

[0030] [Fig. 3] represents an embodiment in which the Ci represent the combinations or chromosomes of the set, in this case comprising 4 combinations, Cl, C2, C3 and C4.

[0031] Each combination Ci comprises several genes or parameters, in this case 3 genes G1, G2 and G3.

[0032] The two best combinations in the sense of the cost function are Cl and C2. The algorithm calculates two new combinations from these two combinations Cl and C2, by mutation, and replaces two of the combinations in the set which are not Cl and C2, and in this case the two chromosomes C3 and C4 having the worst cost function values, in this case the lowest.

[0033] The terms used are the classic terms of the genetic type AI algorithm.

[0034] This genetic AI algorithm uses a set of initial combinations or initial chromosomes also called "population" in English. Each combination is called a chromosome. All chromosomes contain several parameters called genes.

[0035] In the context of the present simulation method, the parameters may correspond to parameters specific to each radiofrequency component. Each chromosome is evaluated using a cost function sometimes called a "fitness function" in English in the context of the genetic algorithm.

[0036] Depending on the form of the cost function, we seek to maximize or minimize it. In the first case, a solution will be more qualitative, the higher its cost function. Conversely, the lower the cost function, the more qualitative the cost function. Also, the higher (or lower) the cost function value of the solution, the higher the quality of the latter.

[0037] The genetic algorithm is to renew at each iteration the set of chromosomes by replacing some chromosomes with new solutions. At each iteration, a first selection step allows two chromosomes to be selected from the population.

[0038] There are several selection methods: the elite selection method, the tournament selection method, or the rank selection method.

[0039] The rank selection method gives a selection probability proportional to the quality of the solution. Finally, tournament selection allows a subgroup of elements to be selected from the population and the two best elements to be drawn from this subgroup.

[0040] Preferably, the elitist method consists of selecting the two best elements in the sense of the fitness function.

[0041] Two operations specific to the algorithm are then carried out on the two selected elements to produce two new solutions: a step of selecting two chromosomes or "crossover" in the English language in the set and a step of mutation or generation of two replacement chromosomes. These two operations are associated with execution probabilities

[0042] The crossover step can consist of associating the two combinations in a more or less equal manner. Then, the mutation step consists of randomly modifying certain genes with a certain probability. The two new solutions thus created replace the two worst solutions of the population in the sense of the cost function. If at the current iteration, the algorithm has reached a stopping criterion or if the fixed number of iterations is reached, the iterations are stopped, otherwise we move on to the next iteration.

[0043] This genetic AI algorithm is complemented by the development of radio frequency cost functions in order to guide convergence.

[0044] These cost functions are created from different radio frequency metrics such as the cumulative noise factor of at least one component, the cumulative gain of at least one component, the cumulative third-order intercept point of at least one component, etc.

[0045] Cost functions can have different forms depending on the desired objective. It is possible to minimize a radio frequency metric or tend towards a desired value or target value.

[0046] We can introduce thresholding notions into the cost function using indicators. Weights or weightings can be applied to a metric to give it more importance compared to another. The choices are multiple and allow freedom to the user. The present invention allows a significant saving of time through this more relevant exploration which can thus result in a reduction of the product development cycle. (Less post-processing and limitation of calculation time).

Claims

Claims

1. Iterative method for simulating a radiofrequency chain comprising: - an initialization (4) of a set of combinations of values ​​of parameters representative of the radiofrequency chain; - an iterative calculation implementing the steps corresponding to: - a simulation (5) of a radiofrequency chain configured to calculate (6), for each combination of the set, the value of at least one metric of the radiofrequency chain, a metric being a function dependent on at least one parameter of the radiofrequency chain; and - an optimization (7) by an Artificial Intelligence algorithm of genetic type, in which: - a chromosome corresponds to a combination of the set, and a gene corresponds to a combination parameter of the set; and - the calculated cost function (8) depends on a weighting of the at least one simulated metric(s).

2. A method according to claim 1, wherein the optimization step (7) comprises a selection (9) of two chromosomes from the set, a generation (10) of two replacement chromosomes, and a replacement (11), from the set, of two chromosomes among the unselected chromosomes, by the two calculated chromosomes.

3. Method according to one of the preceding claims, in which the selection (9) of two chromosomes in the set implements a selection by rank, a selection by tournament, or an elitist selection.

4. Method according to one of the preceding claims, wherein the radio frequency chain metrics comprise at least one cumulative noise factor of a component of the radio frequency chain, and / or at least one cumulative gain of a component of the radio frequency chain, and / or at least one cumulative third-order intercept point of a component of the radio frequency chain.

5. Method according to one of the preceding claims, in which the parameters of the radiofrequency chain comprise at least one component noise factor of the radiofrequency chain, and / or at least one radio frequency chain component gain, and / or at least one third-order intercept point of radio frequency chain component.

6. Method according to one of the preceding claims, in which the iterative calculation has as stopping criterion (12) the achievement of a predetermined number of iterations.

7. Method according to one of claims 1 to 6, in which the iterative calculation has as stopping criterion (12) the achievement of a predetermined number of successive iterative calculations of the cost function of a chromosome of the set all included in an interval of values ​​of width less than a threshold.

Images