Extract B-HF model for LNA device

How to extract Bilateral High Frequency memory (B-HF) model for Low Noise Amplifier (LNA). The LNA-B-HF model has the same properties as the HPA-U-HF model for frequency dispersion. It therefore takes into account only the dispersive effects present in the amplifier band and ignores other memory effects such as those due to polarization or heating. On the other hand, it makes it possible to take into account the changes in load and source impedance.

1. Create a new LNA device
   In an opened project, you can create a device from Applications window or Workspace window.

   • From Applications window, right-click on Device modeler and click on Create device. You can also right-click on LNA and click on Create LNA device.
   • From Workspace window, click on Device modeler button, select LNA and click on Open button then New button.
   The Create a new device dialog box is displayed.

   
   

   

   
   

   1. In Type field, select LNA.
   2. In Model field, select LNA-B-HF.
   3. In Name field, edit the name of your device. Here, we will name it "LNA_example5".
   Click on Create button to display the new device in the tree of Applications window and the settings of the extraction in Workspace window.

   
   

   

   
   

2. Choose your data file
   
   

   

   
   In the Extraction Settings section, fill in the Data file field with the absolute or relative path of your measurement or simulation file with the extension .dat. Click on Browser button to open the file browser and select your file in the local file system. The file browser opens directly to the data directory specified when creating the project.
3. Tune power and frequency approximation order parameters
   
   

   

   
   In Bilateral model order drop-down list, choose the order corresponding to the measurement data.
   In Power approximation order and Frequency approximation order fields, start to put low orders and checks results graphically after extraction.
   You have the option to add a noise factor file by choosing Yes in the Noise factor field.
   Nota Bene:

   The LNA-B-HF model is identified from the CW load-pull measurement. This roughly corresponds to applying the principle of extracting the HPA-U-HF model on several load impedances. The HPA-B-HF model is declined in several bilateral order, ranging from 0 to 3. The greater the order, the more the model is able to take into account strong load mismatches in the nonlinear operating zone. It takes a minimum of 3 load impedances to be able to identify the simplest bilateral model (Bilateral order 1), 6 impedances for bilateral order 2 and 10 impedances for bilateral order 3.

   The power approximation order can not be greater than the number of power points included in the data file.

   The frequency approximation can be carried out either by polynomial function or by poles-residues decomposition. The polynomial approximation is more adapted to weakly varying characteristics according to the frequency. Otherwise, it is recommended to use poles-residues approximation.

   The frequency approximation order can not be greater than the number of frequency points included in the data file. If exceeded, VISION will send a message in the Output Console window and automatically truncate the order of approximation to the maximum number allowed.

   It is recommended to consider a frequency approximation in poles-residues if you are interested in the transient response of the amplifier. In this particular case, you must also take care to consider the order of poles-residues approximation as low as possible (do not seek a perfect fit of the frequency characteristics by pushing the order of approximation to the maximum).

   The Technological dispersions option allows to specify a distribution law of the gain (module) and phase shift characteristics of the amplifier. Two laws of dispersion are possible (Uniform or Gaussian law). The dispersion is characterized by two parameters: the standard deviation Module, given in % of the nominal value for the gain, and the standard deviation Phase in degrees for the phase shift.

4. Extract behavioral model and check with output graphs
   Click on Extract button to start the extraction process of the model. The output console is displayed:
   

   

   
   
   The message Model Fit Error is showing the normalized mean square error (NMSE) between data and model. Close the window to see in the Applications window the number of the newly created extraction, here, 001. The results are saved and can visualized at any time by designating in the tree the associated extraction. Click on the Output graphs tab to see comparisons between data and model.

   
   

   

   
   

   Various graphs are available to check the quality of the model according to two dimensions: power and frequency. To examine the quality of the approximation of the nonlinear pseudo parameter S21, select Volterra Model LNA-B-HF [Output reflected power wave: B2] in Figures section and choose graphs you want to display in Graphs section:

   • Tick dB[S21 = B2/A1] [par=Pin, Load] to display, for different input power and load impedances, the modulus of the nonlinear pseudo parameter S21 in dB as a function of dFreq, the offset between the central frequency of the device characterization band and the frequency of the CW signal.
   • Tick phase[S21 = B2/A1] [par=Pin, Load] to display, for different input power and load impedances, the phase of the nonlinear pseudo parameter S21 in dB as a function of dFreq, the offset between the center frequency of the device characterization band and the frequency of the CW signal.
   • Tick dB[S21 = B2/A1] [par=Freq, Load] to display, for different carrier frequencies and load impedances, the modulus of the nonlinear pseudo parameter S21 in dB as a function of Pin, the power of the CW input signal.
   • Tick phase[S21 = B2/A1] [par=Freq, Load] to display, for different carrier frequencies and load impedances, the phase of the nonlinear pseudo parameter S21 in dB as a function of Pin, the power of the CW input signal.
   The graphs show the curves of data (from measurement or simulation) in red lines and the extracted model in blue lines. The legend recalls the error NMSE between model and data. If the number of curves makes the graphs unreadable, click on Configure button to reduce the density of curves and/or limit the input power range and frequency band. Select Volterra Model LNA-B-HF [Input reflected power wave: B1] in Figures section to display the nonlinear pseudo parameter S11 depending on input power, frequency and load impedances. Select Volterra Model LNA-B-HF [DC consumption] in Figures section to display power consumption of the device depending on input power, frequency and load impedances.

   
   

   

   
   

5. Check measurement aberration and noise measurement
   The first extraction is an opportunity to verify the data, especially if there is measurement noise or measurement error. This type of phenomena can make it difficult to extract a model and can lead to aberrant results when the model is subjected to more complex signals. VISION provides some tools to limit or eliminate these phenomena in order to avoid doing measurements again. In Applications window, click on your device, here LNA_example5, to show up Settings tab in the Workspace window. Click on Extraction Options section to reveal some options:

   • Measurement aberration and noise polish filters: for the U-HF model, the option CW power gain aberrations is available to filter the noise that can be encountered on the measurement data. The user must choose the filter order appropriately.
   • Extraction power and frequency range tune: depending on the needs or observations on the data, one can modify the range of input power and/or the frequency band of the data with which the extraction of the model can be performed.

   
   

   

   
   The option Frequency grid oversampling increases the frequency step in order to improve the approximation of the phase, especially if this one presents a variation greater than $2\mathrm{\pi }$ in radian according to the frequency.
6. Tune power and frequency range
   If the first extraction is not satisfactory, it is necessary to increase the order of approximation power and/or frequency.

   1. Start by increasing the order of approximation power as long as the error NMSE decreases significantly. Check graphically the comparison between the data and the model.
   2. Then, increase the order of approximation frequency as long as the error NMSE decreases significantly. Check graphically the comparison between the data and the model.
   3. If the error is not small enough, restart in step a from the current settings
   The user can find in the following table an example of the extraction process. Here, the parameters of extraction 007 allow to have the smallest error between the data and the model.

   Table 1. Extraction settings LNA_example5

   Extraction	Power approximation order	Frequency approximation order	NMSE-S11 (dB)	NMSE-S21 (dB)
   001	1	1	-34.56	-30.79
   002	2	1	-35.21	-31.51
   003	1	2	-34.56	-30.79
   004	1	3	-34.56	-30.79
   005	2	3	-35.21	-31.51
   006	2	4	-35.21	-31.51
   007	2	5	-35.21	-31.51

   
   

   

   
   

   You can label an extraction as a reference to differentiate it from others for use in System Architect. To do so, select the appropriate extraction of your device in the Applications , right-click on it, and subsequently select the add to favourites option.