ADAPTIVE CONTROLLER BASED ON TRANSIENT NORMALIZATION

Abstract

A controller is provided for controlling a power stage of a power converter according to a control law, the control law implementing a specific type of compensator and being pre-designed to generate an objective response of a default power converter for a default parameter value of a component of the power stage. The controller is further configured to determine an actual response for an actual parameter value of the component of the power stage and to alter the control law for the actual parameter value of the component of the power stage such that the actual response matches the objective response. The controller determines a degree of matching between the actual response and the objective response by filtering the actual response to generate a filtered actual response and integrating a product of the filtered actual response and a delayed actual response.

Claims

1. A controller for controlling a power stage of a power converter according to a control law, the control law implementing a specific type of compensator and being pre-designed to generate an objective response of a default power converter for a default parameter value of a component of the power stage; the controller being further configured to determine an actual response for an actual parameter value of the component of the power stage; the controller being further configured to alter the control law for the actual parameter value of the component of the power stage such that the actual response matches the objective response by determining a degree of matching between the actual response and the objective response by filtering the actual response by an inverse filter of the objective response to generate a filtered actual response and integrating a product of the filtered actual response and a delayed actual response.

2. The controller according to claim 1, wherein the delayed actual response is delayed by one sample.

3. The controller according to claim 1, wherein a filter used for filtering the actual response comprises an inverse filter of the objective response such that an actual response that exactly matches the objective response results in a zero output from the filter apart from a first sample of the filtered actual response.

4. The controller according to claim 1, wherein the controller is configured to scale a degree of matching between the actual response and the objective response to alter the control law.

5. The controller according to claim 4, wherein the controller is configured to scale the degree of matching by a linear time invariant gain.

6. The controller according to claim 4, wherein the controller is configured to scale the degree of matching by a gain that is dependent on the magnitude of the actual response.

7. The controller according to claim 6, wherein the controller is configured to scale the degree of matching by a constant divided by a 2-norm of the actual response.

8. The controller according to claim 1, further configured to alter the control law by selecting another type of compensator of a plurality of types of compensators and by altering the control law implementing the other type of compensator for an actual parameter value of the component of the power converter such that an objective response of the other type of compensator corresponding to the default power converter matches the actual response.

9. The controller according to claim 8, further configured to optimize for a type of compensator of the plurality of types of compensators and the corresponding control law.

10. A power converter comprising: a controller according to claim 9, and a non-volatile memory for storing the plurality of types of compensators.

11. A method for controlling a power stage of a power converter according to a control law, the control law implementing a specific type of compensator, the method comprising: providing the control law being pre-designed to generate an objective response of a default power converter for a default parameter value of a component of the power stage; determining an actual response for an actual parameter value of the component of the stage; and altering the control law for the actual parameter value of the component of the power stage such that the actual response matches the objective response by determining a degree of matching between the actual response and the objective response by filtering the actual response by an inverse filter of the objective response to generate a filtered actual response and integrating a product of the filtered actual response and a delayed actual response.

12. The method according to claim 11, further comprising: delaying the actual response by one sample.

13. The method according to claim 11, wherein filtering the actual response comprises using an inverse filter of the objective response such that an actual response that exactly matches the objective response results in a zero output from the filter apart from a first sample of the filtered actual response.

14. The method according to claim 11, wherein altering the control law comprises: selecting another type of compensator of a plurality of types of compensators and altering the control law implementing the other type of compensator for an actual parameter value of the component of the power stage such that the actual response matches the objective response of the other type of compensator.

15. The method according to claim 13, further comprising: optimizing for a type of compensator of the plurality of types of compensators and the corresponding control law.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Reference will be made to the accompanying drawings, wherein

[0042] FIG. 1 shows the load step responses; and

[0043] FIG. 2 shows the load step response characterization system; and

[0044] FIG. 3 shows vector u when a=0.5 (objective response); and

[0045] FIG. 4 shows vector y when a=0.5; and

[0046] FIG. 5 shows vector u when a=0.2; and

[0047] FIG. 6 shows vector y when a=0.2; and

[0048] FIG. 7 shows vector u when a=0.8; and

[0049] FIG. 8 shows vector y when a=0.8; and

[0050] FIG. 9 shows vector u when a=0.5; and

[0051] FIG. 10 shows vector y when a=0.5; and

[0052] FIG. 11 shows vector u resulting from a 2.sup.nd order impulse response; and

[0053] FIG. 12 shows vector y resulting from a 2.sup.nd order impulse response; and

[0054] FIG. 13 shows an automatically tunable compensator; and

[0055] FIG. 14 shows the output voltage and inductor current of a buck converter with characterization turned on at 4.0 ms resulting in improved Load-Step response thereafter.

DETAILED DESCRIPTION OF THE INVENTION

[0056] The load-step response is a very important dynamic characteristic of DC-DC converters, but the response is dependent on both the loop gain/phase and the open-loop output impedance of the converter. Although the loop gain/phase alters the closed-loop output impedance, converters with similar loop characteristics may have different load-step responses. Therefore, an approach based on characterizing the shape of the load-step response is advantageous compared to methods that characterize the loop bandwidth/phase-margin.

[0057] In order to characterize the load-step response it is necessary to have an objective load-step response that represents the desired response. The characterization method identifies the salient features of the load-step response in comparison to the objective load-step response. Bearing in mind, that the magnitude of the response varies with load-step magnitude and edge-rate for example, a method involving some function of the difference, i.e. subtraction, between the response and the objective response would be problematic. Referring to FIG. 1 the objective load step (a) represents the characteristics of the desired response; the under-damped (b) and over-damped (c) responses are shown for comparison.

[0058] In order to characterize the load-step response and quantify how well it matches the objective response the load step response (u), is applied to a filter 21 as shown in FIG. 2.

[0059] The filtered actual load step response is multiplied, see stage 23, by the actual load-step response and integrated by integrator 24 in order to ascertain the degree of matching between the actual load step response and the objective load step response.

[0060] A delay 22 is required to remove the first sample from the filter. The filter may be designed as an inverse filter of the objective load-step response such that an actual load-step response that exactly matches the objective response results in a zero output from the filter, neglecting the first sample, and therefore the integral of the product of the filtered and original actual load step response is zero.

[0061] For example, considering an objective load step response represented by the vector u (FIG. 3), where u=[1, a, a.sup.2, a.sup.3, . . . , a.sup.n], applied to a filter whose impulse response is vector h where h=[1, a]. The resulting signal from the filter is vector y (FIG. 4), where y=u. h and therefore y=[1, a-a, a.sup.2-a.sup.2, . . . , a.sup.n-a.sup.n] which simplifies to y=[1, 0, 0, 0, . . . , 0]. Assuming zero valued signals apriori, delaying u by one sample yields u where u=[0, 1, a, a.sup.2, a.sup.3, . . . , a.sup.n] and the result of the integral of the product is therefore v, where v=u. y=0.

[0062] Now considering u=[1, b, b.sup.2, b.sup.3, . . . , b.sup.n] applied to a filter whose impulse response is vector h where h=[1, a]. The resulting signal from the filter is y=[1, b-a, b.sup.2-ab, b.sup.3-ab.sup.2, . . . , b.sup.n-ab.sup.n-1]. When b>a, the vector y simplifies to a vector of positive values (neglecting the first value), and the result of the integral of the product is therefore positive (FIG. 7, FIG. 8). When b<a, the vector y simplifies to a vector of negative values (neglecting the first value), and the result of the integral of the product is therefore negative (FIG. 5, FIG. 6).

[0063] Negative values of the parameter a model an oscillatory response (FIG. 9), which results in a vector y (FIG. 10), whose integral of the product (neglecting the first value), is negative.

[0064] Therefore, it is clear that the proposed characterization system yields a value whose magnitude and sign is a measure of matching between the actual response and the objective response with a zero result value for an exact match to the objective response, a positive result value when a is greater than the desired value and a negative result when a is less than the desired value or negative.

[0065] A simple two-tap (first order) FIR filter has been considered for clarity of explanation but it is clear that higher order FIR filters or IIR filters may be employed to characterize higher order objective responses. For example the objective response vector equal to the impulse response of a filter whose transfer function is (10.1z.sup.1)/(11.3z.sup.1+0.36z.sup.2) is illustrated in FIG. 11. FIG. 12 shows this is correctly characterized by the 2nd order IIR filter whose transfer function is: (11.3z.sup.1+0.36z.sup.2)/(10.1z.sup.1).

[0066] The output of the characterization system of FIG. 2 may be used to adjust a PID compensator as shown in FIG. 13, where the compensator block 133 is a component of a DC-DC converter and the scaling block interfaces 132 the characterization block 131 to the compensator. The compensator 133 is adjusted by the adjustment value w.

[0067] The scaling block 132 may be suitably [0068] i) a linear time invariant gain;

[0069] ii) a gain that is responsive to the magnitude of the signal being characterized (u) e.g. K|u| where |u| represents the 2-norm of u or another suitable function. The advantage of (ii) is that the resulting signal from the characterization block is amplified more if it is resulting from a small input signal u. Therefore it represents a greater requirement for adjustment in the compensator than if the same signal resulted from a large input signal u.

[0070] FIG. 14 shows the output voltage and inductor current of a buck converter with characterization turned on at 4.0 ms resulting in improved Load-Step response thereafter. The adjustment value w is also shown to characterize the pulse immediately resulting in improved compensator tuning after only one load-step pulse, as required.

[0071] Because of the characterization is carried out on the load-step pulse response as described it is clear that this method may operate with non-linear compensators, for example where different compensators are activated according to the system state at a specific instance in time, and furthermore is compatible with non-linear DPWM restart techniques.

[0072] Following characterization the adjustment value w may be stored in NVM to be applied when the converter is next powered up following power down. Also, the adjustment value (or the like) may be communicated over a communication bus (serial or parallel) to provide information regarding the characterization of the response which would be useful in the design and quality control of the end power system. For example, if it was observed that the value had changed since the previous characterization or was very different from expected then the user may be alerted to act accordingly (on an impending component failure for example).

[0073] It is clear that such combinations and others would be very beneficial.