Model calculation unit and control unit for calculating a data-based function model having data in various number formats

Technical Field

The invention relates to a model calculation unit of the controller, especially hard wiring model calculation unit as in among them or that can be calculated by means of the model based on the data of the function of the hardware unit, in particular for controlling the motor system. The invention also relates to this kind of model calculation unit of the model based on the data of the function of providing and handling of the configuration data.

Background Art

Disclosed in the prior art with the main computing unit and separate used in the calculation of the model based on the data the function of the model calculation unit of the controller. For example, documentation DE   10   2010   028   266   A1 shown with additional logic line as the model calculation unit of the controller. The additional logic circuit in terms of hardware used to calculate the index function and an addition function. In the hardware unit Bayesian regression support, in particular the need for the calculation of the regression states Baye Gaussian process model.

Model calculation unit design as used for implementing mathematical process based on parameter/ultra-parameters and the control point data (Stützstellendaten) or the training data based on data (Trainingsdaten) function of the calculated model. states especially model calculation unit in terms of hardware structure for effectively calculating exponential function, so as to to the main computing unit by appropriate software is compared to the calculated speed of the more high calculation speed to calculate the gauss process model.

Provide configuration data is normally used for model calculating unit and the calculation of the model calculation unit based on the hardware implementation of the calculation of the configuration data, wherein the configuration data includes used for calculating the function of the on the basis of the data and control the parameter of the model.

The control point data usually to floating-point data or given in advance in the form of fixed-point data. However, when implemented in hardware, based on floating-point arithmetic algorithm must be based on the fixed-point arithmetic algorithm realized separately.

Document US   4,675,809 described, by using the switching unit to the application system different in the floating-point data type.

Document US   5,161,117 has described a kind having different basis for use of the method for different floating point value.

Content of the invention

According to the invention provides according to claim 1 wherein the hardware model is used to calculate the unit according to parallel and with the model calculation unit of the controller.

Other advantages of the present invention the design of the programme was illustrated by dependent claims.

According to the 1st aspect, is provided in the controller for calculating a function of the data-based model, especially gaussianity process model model calculation unit, which comprises:

-Computing core, the core structure is used for pure hardware-based based on data for the function of the algorithm of the model, wherein the data is calculated, in particular ultra-parameter data and the control point under the condition of the function of calculating the model based on the data; and

-Pure hardware-based conversion unit, the conversion unit structure used for to the stated computing core in order to pre-given digital format provides at least a part of the calculation data, in particular the provision of control point data.

Described at the beginning of the software control of the controller in addition to the main computing unit implemented in hardware with the model calculation unit, the model calculation unit in addition to the index function computing unit also has at least one cycle (Schleife) for calculating and in a hardware logic unit. The calculation used to calculate the data of the pre-given, in particular parameters and the control point data is used for Gaussian process model, the calculated data is stored in terms of the model calculation unit in a storage area can be visited.

In design the stated model calculation unit, are usually so design the hardware module, i.e. the calculation of the maximum that can be used is to calculate the value of the resolution. This, for example, in a controller of the conventional floating-point digital format equivalent to 32 bit resolution.

Calculation data or floating-point data is generally used in the form of fixed-point data given in advance. However, when implemented in hardware, based on floating-point arithmetic algorithm must be based on the fixed-point arithmetic algorithm realized separately. However in order to limit the computation unit for model of the hardware requirements of the area of the integrated structure, only the calculation processing of the data is in a digital format. If the calculated data is completely or partly in the presence of another digital format, then to the must be provided to the computing data processing, thereby converting it to digital format of desired, that is, the fixed point or floating-point format or digital format.

In order to calculate the transfer function model based on data of a sufficient to provide, for example, 8 bit or 16 bit precision as the calculation of the data or spectral floating point value , usually in the main computing unit in a 32 bit floating-point calculation to floating point value.

In order to use for computing data storage needs of and still provided with a minimized for high-precision calculation data of the programme of calculation model calculation unit, therefore, can set up a kind of model calculation unit, its as the input stage has a conversion unit. Through the conversion unit can be used for calculating the model calculation unit to directly supply the other digital format to calculate the data of, for example, 16-bit floating point number format values or 16 bit fixed point digital format and where the value of the use of these data, the model is computed without the need of separate hardware is set up in the unit.

Furthermore, the conversion unit realizes, does not need to be in the main computing unit in the implementation of the data required for the calculation of the conversion, the power of the main computing unit usually limited in the control device. Usually as a spectral sensor data provided by the converting unit can also be provided in the by means of a common index parameter (on-the-fly) rapid write of nailing, it is not by means of a main computing unit, for converting in an appropriate manner.

Furthermore, conversion unit can be constructed and used in accordance with a selection signal of the pre-given different digital format digital format, the calculation of the data into pre-given digital format.

According to one embodiment, the conversion unit can have at least one digital format for the 1st pre-given data into digital format and the data of the conversion block of the multiplier, according to the selection signal provided by either the calculation of the data either through the at least one conversion block into a predetermined one of the calculation data in digital format of the core is transmitted to the calculating.

Is able to specify, the pre-given equivalent to the digital format of the 32 bit floating-point format.

1st conversion block can, in particular, from the data structure is used for the pre-given digital format with less bits compared with the floating-point format into a pre-given digital format.

2nd conversion block can be constructed and used in the data from fixed-point format into a pre-given digital format.

Furthermore, conversion block 2nd can be structure is used for in the data from fixed-point format into a digital format of pre-set to take into account pre-given value of the exponent.

According to another aspect of the controller, in particular, for example, with the integrated structure of the form of the chip structure. The controller includes:

-For the software control implementation of the function or functions of the main computing unit (Funktion);

-Used to store at least one digital format to calculate the data storage unit; and

-The above-mentioned model calculation unit.

Furthermore, the main computing unit can be on the basis of the calculated data stored in the storage unit of the selected digital format signal is supplied to the model calculation unit, according to the selection signal to the calculation of the data into pre-given digital format.

Description of drawings

According to the Figure a detailed explanation of the preferred embodiment of this invention. In the attached drawing:

Figure 1 show and has the main computing unit and a model calculation unit a schematic diagram of the controller;

Figure 2 has a conversion unit is shown, Figure 1 a schematic diagram of the model calculation unit;

Figure 3 shown fig. 2 a detailed diagram of a conversion unit; and

Figure 4 is shown with a common index of from 16 bit fixed-point value into the conversion of the value of the floating-point format illustrated.

Mode of execution

Figure 1 is a schematic diagram of the controller used for integrating 1 of the hardware structure of the, in an integrated manner in the controller (such as the monolithic to) mode the microcontroller such as in the form of the main computing unit 2 and, in particular, used for hardware-based function is calculated based on data of the model (Funktionsmodell) model calculation unit (Modellberechnungseinheit) 3. The main computing unit 2 and model calculation unit 3 through the system bus 6 are in communication with each other. Model calculation unit 3 configured to only hardware (hard wired) and therefore cannot execute the software code. Therefore, in the model calculation unit 3 is not arranged in the processor or similar device. This can resources optimally realize this kind of model calculation unit 3.

Furthermore, controller 1 has an internal memory unit 4 and DMA unit 5 (  Access   Memory DMA=Direct), via the internal communication connecting their 6, for example, via the internal data bus and the main computing unit 2 and model calculation unit 3 in connecting.

In the Figure 2 shows in detail the model calculation unit 3. Can see that, in order to calculate the transfer function model based on data, computing core 31 is provided with index unit 32, in some cases can be combined to are set up in the common FMA unit in (  Add   Multiply FMA=Fused) the addition and multiplication unit states the unit the use of the front 33 and 32, 33 for the computation process of the provides for the determination of the logic unit 34.

Two kinds of generally known IEEE-754 floating-point standard as the data format. The single-precision in the floating-point standard equivalent to 32 bit resolution, that is, 1 symbol bit, 8 and the indicators 23 and one of the semi-precision mantissa bits is equivalent to in 16 bit resolution, that is, 1 symbol bit, 5 plurality of index bits and 10 a mantissa bits. Furthermore, the 16-bit fixed-point format used in 16 bit value of the data representative of, without setting up the index. In addition to the 16 bit fixed point outside the digital format, can be also a series of the data value is provided with a common index, the common order of magnitude for the data distribution.

In order to make the model calculation unit 3 can use the data having different data format, in the model calculation unit 3 is set in the conversion unit 35, such as the combined Figure 3, as shown in detail. The conversion unit 35 the calculation of the data, such as control point data into a predetermined format, such as into a 32 bit floating-point format and the V as processing data provided to the model calculation unit 3 described in the hardware of the computation algorithm. To calculate the data of the data elements can also exist in different digital format, so as to make the calculation data according to the data element for converting digital format.

The conversion unit 35 comprises a multiplier 36 for the model calculation unit 3 on the basis of data in the calculation of the transfer function model of the processing data to provide a switching of the V. In this case, the control point data can be used as a form of a 32-bit floating point format (with single-precision floating-point format) data D32F, in the form of 16-bit floating point format (with a semi-precision floating-point format) data D16F or form of a 16 bit fixed-point format data D16 combined pre-given common index CE is supplied to the model calculation unit 3.

16-bit floating point format data D16F and 16 bit fixed-point format data D16 in the corresponding 1st and 2nd conversion block 37, 38 in the known manner in the converted into a 32 bit floating-point digital format without conversion and 32-bit floating point format data D32F common input multiplier 36. In a multiplier 36 in, by means of a can, for example, by the main computing unit 2 provides a selection signal to select data format corresponding S D16F, D32F, D16 of one.

In 1st conversion block 37 by a 16 bit floating-point format into a 32-bit floating point format is a simple bit operation. 5 bit index (translation is provided with a symbol (Excess-15) ± 15) and using the 8 bit index (translation with symbol ± 127 (excessive -127)) of the 32-bit floating point number format through 112 (that is, 127-15) adds of the formed index conversion, and ordinarily the integrated structure forms to advantageously as area with two input terminals of the multiplier, the multiplier through the highest index bit control, wherein null value, the data of the or-∞ +∞ and NaN (Not-a-Number, nonnumeric) as a special case, the conversion results to the same value. From the 10-bit extensions to the 23 position of the mode of use of the zero as simply be added to the lowest value bit, this can, for example, through the left 13-bit left-shift operation or the 10 a and the mantissa bits 13 are connected in series to a zero position.

16 bit fixed-point format into a 32-bit floating point number format can, for example, with the aid of Figure 4 as shown in the conversion block 2nd 38 to realize. At the same time able to take into account pre-given common index CE. Figure 4 of the conversion block 2nd 38 can be realized with a symbol of the 16 bit fixed-point format generating 32 bit floating point value.

In this, the symbol extraction block 41 extracting the input and is used as the sign bit of the sign bit of the output value. Furthermore, in the value extraction block 42 and then extracting from the input value in the standardized block 43 is to the left of the moving mobile computing, until the peak value is equal to the bits of " the 1 [...] , wherein the input value of the zero value as a special case, the same conversion result to zero. Standardized block 43 on the mobile of bits needed to provide data of the number of index addition unit 44, the index-addition unit in the common index bits minus the value of the number of moving and as a result of the conversion by providing the as 32 bit floating point value value of an index. By combining the standardized block 43 mantissa value obtained in, symbol extraction block 41 and extracting the sign bit of index addition unit 44 to produce the value of the exponent in a result of the conversion. The conversion result is then provided to the model calculation unit 3 V as processing data, the processing data includes the conversion of the control point data.