Non-linear modeling method

• US 7,091,779 B2
• Filed: 10/29/2002
• Issued: 08/15/2006
• Est. Priority Date: 11/12/2001
• Status: Expired due to Term

• Current Assignee: Telefonaktiebolaget LM Ericsson
• Original Assignee: Telefonaktiebolaget LM Ericsson
• Inventors: Sahlman, Karl-Gosta
• Primary Examiner(s): Mottola, Steven J.
• Application Number:
  US10/494,662
• Publication Number:
  US 20040257157A1
• Time in Patent Office: 1,386 Days
• Field of Search: 330/2, 375/297
• US Class Current: 330/149
• CPC Class Codes: H03F 1/3247 using feedback acting on predistortion circuits H03F1/3264 takes precedence
• CPC Class Codes: H03F 1/3258 based on polynomial terms

Abstract

In the present invention a completely different approach for modeling of non-linear devices or processes is used. A cascade of blocks each depending on a non-linearity parameter are utilized both in non-linear modeling and a pre-distorter design. The new non-linear model description can be used for characterization of a non-linear system or for linearizing a non-linear system. Application to a Communications System Multi-carrier Amplifier is shown as an application example. Also for other application areas of the new mathematical method is feasible echo-canceling, non-linear communications channels etc. Parameter extractions and tables for the new model do not give multiple dimensional mathematical solutions for parameter extractions as in prior art.

Claims

1. A method of modeling a non-linear digital signal processing system for reduction of distortion in the system, the method comprising:

expressing the non-linear digital processing system as a number of processes depending on a set of determined process parameters P1 to PN and non-linear processes designated as NLS(P1,P2, . . . PN), which are described by multiplying blocks or a cascade of non-linear sub-processes each depending on one of the non-linear process parameters P1 to PN, whereby each sub-process is described by a generalized Wiener or Hammerstein process comprising both a linear and a non-linear process and a description of the process or system is made by a model describing first a transfer function of a main non-linear sub-system (MNLS) depending on a process P1, and consecutive multiplying non-linear system processes are each described by a transfer function 1+dNLS(dPi), wherein a process dNLS(dPi) is defined as a difference sub-process to the preceding cascaded processes and a process parameter dPi is defined as equal to a difference between an actual process parameter value Pi and a mean value Pi_mean of the process value Pi used in preceding processes.

2. The method according to claim 1, further comprising evaluating a first main system process MNLS(P1) by a model parameter adaptation which minimizes errors between an observed output from the actual process and a calculated output from the modeled main non-linear process MNLS(P1) depending on parameter P1, whereby both model and actual measured non-linear process are having the same input stimuli, and a next cascaded process Pi is further evaluated and the process dPi value is adapted by further calculations using errors between an actual output from this actual process and an output of so far modeled process and having the same input stimuli.

3. The method according to claim 2, further comprising:

using, when linearizing an actual non-linear process or system this new cascadable non-liner process description being made by connection of cascaded sub-processes each depending on the process parameter Pi in front of the actual non-linear process or system, the connection of the processes being such that a fast cascadable block is a non-linear sub-process depending on a difference process dPN and a process closest to the actual non-linear system is dependent of process parameter P1 related to the main non-linear system model MNLS(P1), whereby each sub-process has an inverse process functionality of the corresponding non-linear model sub process, and the inverse sub-processes is either a direct inversion of the non-linear model sub-process 1/(1+dNLS(Pi) or expressed as 1+dNLSm(Pi) where dNLSm(Pi) is a modified process depending on the same process parameter Pi as in the non-linear model and an inverse functionality of the main process MNLS(P1) depending on parameter P1 functionality is either a direct inverse 1/MNLS(P1) of the model process or a modified non-linear process MNLSm(P1) having the inverse functionality of the main non-linear model process.

4. The method according to claim 3, further comprising performing adaptation and optimization of the linearization process by minimizing error contributions between the input stimuli and the observed output from the total system containing the cascaded linearization system blocks and the actual non-liner system, whereby error minimizing is first done versus the main process P1 in an inverse process block 1/MNLS(P1) or MNLSm(P1) of the cascade, and when the adaptation has reached a minimum and no further error reduction is available by the adaptation process, the linearization system goes on by error minimizing of the next block depending on process P2 and so forth until the last parameter of the cascade linearization system is minimized, and the linearization then may make a second loop or continuous looping of linearization adaptation by doing the same error minimizing process starting with the parameter P1 and ending with the parameter PN.

5. The method according to claim 4, further comprising adaptation linearization of sub-blocks in the linearization cascade either by calculation of a virtual model in a system processor controlling the linearization process and minimizing the errors between the calculated virtual model output with the total system output, whereby the linearization sub-block then can be updated by the direct inverse of the calculated model block, or a mix of adaptation techniques is used by updating some sub-blocks in the linearization circuits by values derived from the virtual non-linear system model, while other blocks are direct updated by direct inverse signal transfer characteristic calculation values minimizing the errors between the observed output and the input stimuli to the linearizing system.

6. The method according to claim 1, further comprising rearranging procedures such that the main process MNLS can depend of more than one process, i.e. MNLS(P1,P2, . . . PM) and sub-processes can also be set as a dNLS (PM,PM+1 . . . ), whereby a computational effort in those cases will be multiple kernel solutions for each sub-process and a mix of single process blocks and multiple process blocks can be used.