Cross Regulation Mechanisms in Multiple-Output Forward and Flyback Converters

Cross Regulation Mechanisms in Multiple-Output Forward and Flyback Converters Bob Erickson and Dragan Maksimovic Colorado Power Electronics Center (CoPEC) University of Colorado, Boulder 80309-0425 http://ece-www.colorado.edu/~pwrelect

Cross Regulation Mechanisms in Multiple-Output Forward and Flyback Converters Design of transformer in development of a multiple-output supply: Typically requires substantial engineering effort Can represent the largest risk to success of the project There is a need for increased understanding of the mechanisms that govern behavior of multiple-output converters An old problem that has never been adequately addressed in the literature Ideal transformers are typically assumed Only conduction losses are modeled Reduced-order magnetics models do not predict observed waveforms Problem is considered intractable

Objectives of this seminar Explain the magnetics-associated mechanisms that govern crossregulation in forward and flyback converters, including peak detection discontinuous conduction mode effects of voltage-clamp snubbers Describe magnetics models suitable for cross-regulation analysis Approximate analytical expressions and computer simulation Predict small-signal dynamics Include laboratory measurement methods and experimental examples

Modeling Multiple-Output Converters Cross regulation, CCM/DCM boundaries, dynamics Flyback converter Forward converter with coupled inductors D 1a v 1 V s W 2 V 2 W 1 v g i 2 W 3 V 3 i 3 1 : 1 : N 1 N 2 D 2a D 1b D 2b v s1 v s2 v 2 C 1 R o1 C 2 R o2 V o1 V o2 W 4 i 4 V 4 V g Q 1 D r D 3a v 3 C 3 R o3 N 3 D 3b v s3 V o3

Some ways to view multiple-output converters With ideal transformer Does not predict effect of leakage inductances on cross regulation Does not predict change of operating mode Does not predict observed converter transfer functions Predicts that isolated converters behave as their non-isolated parent converter topologies With actual transformer and its leakage inductances Analysis previously viewed as intractable, with no hope of gaining physical insight into cross-regulation mechanisms (not so!) Waveforms of isolated converters can differ significantly from their non-isolated parent topologies new phenomena that are not observed in non-isolated versions Some outputs may operate in continuous conduction mode (CCM) while others operate in discontinuous conduction mode (DCM) To correctly predict dynamics and conduction losses: must account for leakage inductances

Cross regulation mechanisms Conduction loss Diode forward-voltage drops Resistances of windings Modeling conduction loss Existing averaged modeling methods only partially apply Effect of magnetics Leakage inductances control shapes of winding current waveforms, especially slopes of winding currents Leakage inductances have a first-order effect on cross regulation, as well as dynamics, operating mode, and conduction losses

What we would like to know and what kinds of answers to expect We would like to express the output voltages as functions of the output currents For example: three-output converter CCM matrix equation, 3x3 Flyback converter example V s W 1 v g W 2 V 2 i 2 V 2 V 3 V 4 = R 22 R 23 R 24 R 23 R 33 R 34 R 42 R 42 R 44 I 2 I 3 I 4 V o2 V o3 V o4 Many outputs many equations Similar comments for CCM/DCM mode boundaries W 3 i 3 W 4 i 4 V 3 V 4

What we would like to know and what kinds of answers to expect, p. 2 1. A correct transformer model That is well-suited to analysis of the cross-regulation problem Whose parameters can be directly measured That correctly predicts observed waveforms 2. Equations Of output voltage regulation Of mode boundaries Might be best evaluated by computer when there are many auxiliary outputs 3. Insight Describe fundamental operation of transformer-isolated multipleoutput converters Explain the physical mechanisms that lead to poor cross-regulation

Outline of Discussion 1. Transformer Modeling in context of cross regulation Discussion of transformer models The extended cantilever model The n-port model 2. Cross Regulation in Flyback Converters Qualitative behavior Analytical results Laboratory example Discussion of strategies for improvement of cross regulation Dynamic response 3. Cross Regulation in Forward Converters Coupled-inductor approaches, and their qualitative behavior Analytical results Laboratory example Discussion of strategies for improvement of cross regulation Dynamic response

1. Multiple-Winding Transformer Modeling in the context of the cross regulation problem Multiple-output converters are more than simple extensions of parent single-output nonisolated converters: Imperfect coupling between windings leads to problems in cross regulation, small-signal dynamics, and multiple operating modes, which have not been fully explored in the literature These phenomena are governed primarily by the transformer leakage inductance parameters Need a suitable multiple-winding transformer model that predicts observed waveforms that yields insight into converter cross regulation, CCM/DCM boundaries, and dynamics that explains how converter performance depends on winding geometry that is useful in computer simulation

Approaches to Multiple-Winding Transformer Modeling Inductance matrix v 1 v 2 v 3 v 4 = L 11 L 12 L 13 L 14 L 12 L 22 L 23 L 24 L 13 L 23 L 33 L 34 L 14 L 24 L 34 L 44 d dt i 1 i 2 i 3 i 4 General Reduces the circuit to matrix equations Numerically ill-conditioned in tightlycoupled case Complete model of four-winding transformer contains ten parameters

Equivalent-Circuit Approaches to Multiple-Winding Transformer Modeling Reduced-order equivalent circuit Physically based Not general Does not predict observed waveforms of flyback converter Difficult to apply to some geometries (for ex., toroidal) Full-order equivalent circuits Allow circuit-oriented analysis of converter General Flyback example

Physical-Based Reduced-Order Model 4 winding transformer example Four-winding transformer example Physical modeling approach: equivalent circuit contains seriesconnected leakage inductances Reduced-order approximation based on winding geometry W 4 W 3 W 2 W 1 core Equivalent circuit proposed in [12]: W 1 W 2 W 3 W 4

An Electrically-Equivalent Form of the reduced-order model l 12 l 23 l 34 1 : n 2 1 : n 3 1 : n 4 W 4 L 11 W 3 W 2 W 1 W 2 W 3 W 4 W 1 Contains seven independent parameters Inductance matrix of four-winding transformer contains ten independent parameters Is this model sufficient?

A Thought Experiment using the 4-winding reduced-order model l 12 l 23 l 34 1 : n 2 1 : n 3 1 : n 4 v 1 L 11 i i i 2 3 4 W 1 W 2 W 3 W 4 Apply a voltage to winding 1, short windings 2, 3, and 4. Measure short-circuit currents in each winding. Model predicts that i 3 and i 4 are zero.

A Full-Order Model Extended Cantilever Model l 14 l 13 l 24 l 12 l 23 l 34 1 : n 2 1 : n 3 1 : n 4 L 11 W 1 W 2 W 3 W 4 Include leakage inductances between each winding

Thought ExperimentÑRevisited l 14 l 13 l 24 l 12 l 23 l 34 1 : n 2 1 : n 3 1 : n 4 v 1 L 11 i i i 2 3 4 W 1 W 2 W 3 W 4 Again apply a voltage to winding 1, short windings 2, 3, and 4. Measure short-circuit currents in each winding. Model predicts nonzero i 3 and i 4.

Discussion It is always possible to connect a transformer such that a reduced-order model does not predict the actual waveforms Are such connections actually encountered in multiple-output converters?

Flyback converter circuit V s W 1 v g W 2 V 2 i 2 W 3 V 3 i 3 W 4 V 4 Three outputs Primary-side voltageclamp snubber Cross-regulation is strongly influenced by commutation interval, when transistor turns off and magnetizing current shifts to secondary windings i 4

Commutation Interval Flyback converter exampleñsimilar to thought experiment l 14 Sunbber voltage L 11 l 13 l 24 l 12 l 23 l V V 34 V Magnetizing 1 : n 2 1 : n 3 1 : n 4 current i i 2 3 V 1 i 4 W 1 W 2 W 3 W 4 Reflected output voltages are nearly identical l 23, l 24, l 34 are irrelevant Magnetizing current divides between the output windings according to l 12, l 13, l 14 V 2 Output voltages V 3 V 4

Commutation Interval Flyback converter exampleñsimilar to thought experiment l 14 l 13 Sunbber voltage L 11 l 12 V V V Magnetizing 1 : n 2 1 : n 3 1 : n 4 current i i 2 3 V 1 i 4 W 1 W 2 W 3 W 4 Reflected output voltages are nearly identical l 23, l 24, l 34 are irrelevant Magnetizing current divides between the output windings according to l 12, l 13, l 14 V 2 Output voltages V 3 V 4

Conclusion: Reduced-Order Modeling Approximate reduced-order model derived via physical approach does not correctly predict behavior of multiple-output flyback converter Approximations must not be based solely on winding geometry Application and circuit behavior must be considered before attempting to reduce the order of the model Need a suitable full-order model

Transformer Equivalent Circuit Models Two-winding transformer models T Pi L s1 L s2 i 1 1 : n i 2 v 1 L m v 2 L s i 1 1 : n i 2 v 1 L m1 L m2 v 2 l 12 i 1 1 : n 2 i 2 v 1 L 11 Extension of cantilever model to four-winding case i 1 W 1 i 4 n 4 :1 l 12 l 13 l 14 l 23 l24 l 34 1 : n 2 1 : n 3 v 4 v 3 W 4 W 3 i 2 v 2 W 2 i 3 Cantilever v 1 L 11 v 2

N-winding transformer models used here n-winding transformer Extended cantilever model n-port model v 1 (t) i 1 (t) i 2 (t) v 2 (t) i N (t) v N (t) v 1 i 1 W 1 L 11 l 1N l 12 l2n 1 : n N i N v N 1 : n 2 i 2 W 2 v 2 v 1 i 1 W 1 v T2 L 11 L o2 N Σ k =1 i 2 n k i k W 2 v 2 W N v TN L on i N v N W N

Relationship between inductance matrix and extended cantilever model The inductance matrix: v = sli L = L jk inductance matrix B = L 1 = b jk inverse inductance matrix For an N-winding transformer, contains N(N 1)/2 independent parameters Extended cantilever model also contains N(N 1)/2 independent parameters, related to the inductance matrix as follows: L 11 = L 11 n j = L 1j L 11 l jk = 1 n j n k b jk v 1 v 4 i 1 W 1 i 4 L 11 n 4 :1 l 14 l 12 l 13 l24 l 34 W 4 W 3 l 23 1 : n 2 1 : n 3 i 2 W 2 i 3 v 2 v 3

Measurement of Leakage Inductance Parameters To measure leakage inductance parameter l 34 Current probe i 1 W 1 i 4 n 4 :1 i 2 i 3 l 34 1 : n 3 W 4 W 3 Measurement frequency must be sufficiently high, so that leakage reactance >> winding resistance W 2 v 3 Inject ac voltage Inject ac voltage at winding 3 Short all other windings Measure current in winding 4 l 34 is given by l jk = v j (s) sn j n k i k (s) Must carefully observe polarities, since l jk can be negative

Measurement of Effective Turns Ratios To measure effective turns ratio n 3 Inject ac voltage i 1 v 1 i 2 W 2 Inject ac voltage at winding 1 Open-circuit all other windings Measure voltage in winding 3 W 1 i 4 1 : n 3 i 3 v 3 Measure ac voltage n 3 is given by n k = v k v 1 W 4 W 3

The N-Port Transformer Model Useful in deriving expressions for current ripples and zero-ripple condition, and for computer simulation Primary winding is represented by its currentcontrolled Norton equivalent Each secondary is modeled by a voltagecontrolled Thevenin equivalent Secondary winding output inductance: v 1 i 1 W 1 v T2 L 11 L o2 N Σ k =1 i 2 n k i k v 2 L ok = n k 2 l 1k l 2k l (k 1)k l (k1)k l Nk W 2 Secondary winding controlled voltage source: v Tk = L ok v n k l 1 L ok L v 1k n k n 2 l 2 ok v 2k n k n k 1 l k 1 (k1)k v TN L on i N v N L ok n k n k 1 l (k1)k v k 1 L ok n k n N l Nk v N W N

Flyback Transformer Example Experimental example: Three-output flyback converter Winding and core geometry air gap Application specifications: Input: 30 V (winding W 1 ) Output: 12 V (winding W 2 ) Output: 12 V (winding W 3 ) Output: 3.3 V (winding W 4 ) W 3 : 15T #22AWG EC41-3C80 core air gap W 4 : 5T #16AWG W 2 : 15T #18AWG 1 layer W 1 : 36T #18AWG 2 layers

Measured Model Flyback transformer extended cantilever model v 1 W 1 i 1 L 11 14 µh 220 l 14 µh 130 l 24 µh l 13 l 12 4.5 µh 13 µh l 23 n 2 v 2 i 4 n 4 1:0.42 34 µh 1:0.42 0.14 :1 l 34 n 3 i 2 W 2 35 µh i 3 v 4 W 4 W 3 v 3

Negative l 34 Directions of induced winding currents, when winding W 3 is driven and windings W 1, W 2, and W 4 are shorted Negative l 34 indicates reversal of polarity of induced current i 4 Side-by-side winding geometry leads to negative leakage parameter W 3 W 4 W 2 W 1

Measured n-port parameter model Flyback transformer example N-port parameters are computed from extended cantilever model parameters as follows: v 1 i 1 L 11 220 µh 0.42i 2 0.42i 3 0.14i 4 Winding output impedance L ok = n k 2 l 1k l 2k l (k 1)k l (k1)k l Nk W 1 0.28v 1 0.09v 3 0.68v 4 L o2 0.52 µh i 2 v 2 Voltage-controlled voltage source v Tk = L ok n k l 1k v 1 L ok n k n 2 l 2k v 2 L ok n k n k 1 l (k1)k v k 1 L ok v n k n k 1 l k 1 (k1)k L ok v n k n N l N Nk Alternatively, these parameters could be directly measured 0.41v 1 0.39v 2 1.13v 4 0.02v 1 0.47v 2 0.18v 3 L o3 2.3 µh L o4 0.36 µh W 2 i 3 W 3 i 4 v 3 v 4 W 4

SummaryÑPart 1 Extended cantilever model, and N-port model, correctly predict observed waveforms of multiple-output converters These models are full-order: the number of independent parameters is the same as in the inductance matrix, and the parameters are directly related to the entries of the inverse inductance matrix Each model parameter can be directly measured, and the model can be checked using several other measurements Reduced-order models generally do not predict the observed phenomena of multiple-output converters Next: the mechanisms of cross-regulation in flyback converters can be explained using the extended cantilever model

2. Cross-Regulation in Flyback Converters Widespread applications, usually at low to medium power levels Multiple-output flyback transformer design is usually based on practical experience, trial and error Operation, steady-state and dynamic properties are strongly affected by transformer leakage inductances Modeling is considered intractable (especially if the number of outputs exceeds two) R s V g V s Q s C s D s Q i 1 v 1 v q W 1 W 2 v 2 W 3 v 3 W 4 i 2 C 2 i 3 C 3 i 4 R o2 R o3 V 2 V 3 Very few analytical results or models are available to aid the designer v 4 C 4 R o4 V 4 Poor cross-regulation is often observed in practice

Flyback Converter Circuit R s V s C s D s i 1 v 1 W 1 W 2 v 2 W 3 i 2 C 2 i 3 R o2 V 2 Example: three outputs Two snubber configurations: Passive voltage-clamp snubber: D s, C s, R s V g Q s Q v q v 3 C 3 R o3 V 3 Active-clamp snubber: Q s /D s, C s Cross-regulation affected by: W 4 v 4 i 4 C 4 R o4 V 4 Conduction losses Transformer leakage inductances

Objectives Explain qualitative behavior using the extended cantilever magnetics model for the transformer Derive general steady-state analytical model capable of predicting static cross-regulation for any number of outputs and arbitrarily complex magnetics configuration Compare model predictions with experimental results Discuss model implications and strategies for improvement of static cross regulation Point to dynamic response considerations

Experimental 3-Output Flyback Converter MUR810 R s 4kΩ V s C s 100µF D s i 1 v 1 W 1 W 2 v 2 W 3 MUR810 i 2 C 2 330µF i 3 R o2 V 2 Application specifications: Input: 30 V (winding W 1 ) Output: 12 V (winding W 2 ) Output: 12 V (winding W 3 ) Output: 3.3 V (winding W 4 ) V g 30V Q s IRF640 v Q q IRF640 v 3 MUR810 C 3 330µF R o3 V 3 W 4 i 4 v 4 C 4 330µF R o4 V 4

Measured Model Flyback transformer extended cantilever model W 3 : 15T #22AWG air gap W 4 : 5T #16AWG W 2 : 15T #18AWG 1 layer v 1 W 1 i 1 L 11 14 µh 220 l 14 µh 130 l 24 µh l 13 l 12 4.5 µh 13 µh l 23 34 µh n 2 1:0.42 i 2 W 2 v 2 W 1 : 36T #18AWG 2 layers i 4 n 4 n 3 l 34 1:0.42 0.14 :1 35 µh i 3 EC41-3C80 core air gap v 4 v 3 W 4 W 3

Qualitative Behavior Case 1: Passive Voltage-Clamp Snubber 1 : n 2 D 2 i 2 V g V s D s i 1 W 1 v 1 L 11 v q i m Q l 12 l 23 l 13 l 34 l 14 l 24 W 2 1 : n 3 W 3 1 : n 4 v 2 v 3 v 4 D 3 D 4 i 4 i 3 V 2 V 3 V 4 Analysis: Use the extended cantilever model for the transformer Neglect losses (other than snubber losses due to leakage inductances) Neglect capacitor voltage ripples W 4

Start at the Time When the Main Switch is Turned ON: (1) Diode Turn-Off Interval t d 1 : n 2 D 2 i 2 V g i 1 W 1 V s v 1 L 11 D off s Q on i m l 12 l 23 l 13 l 34 l 14 l 24 W 2 1 : n 3 W 3 1 : n 4 v 2 v 3 v 4 D 3 D 4 i 4 i 3 V 2 V 3 V 4 Currents through the leakage inductances decrease to zero after the main switch Q turns on Secondary diodes turn off Diode turn-off times are not equal W 4

(2) Main switch conduction interval 1 : n 2 D 2 off i 2 v 2 V 2 V s D s off i 1 W 1 v 1 L 11 i m l 12 l 13 l 23 l 24 W 2 1 : n 3 W 3 v 3 D 3 off i 3 V 3 The main switch Q is on, all diodes are off The magnetizing current increases All leakage inductor currents are zero V g Q on l 14 l 34 1 : n 4 v 4 D 4 off i 4 V 4 W 4

(3) Commutation interval t c 1 : n 2 v 2 D 2 on i 2 V 2 The main switch Q turns off, the snubber diode D s turns on i 1 W 1 V s v 1 L 11 D s on i m l 12 l 23 l 13 l 24 W 2 1 : n 3 W 3 v 3 D 3 on i 3 V 3 The secondary diodes turn on, and the secondary winding current start increasing The interval ends when the reflected secondary currents add to i m V g Q off l 14 l 34 1 : n 4 W 4 v 4 D 4 on i 4 V 4 The magnetizing current divides between the widnings according to l 12, l 13, l 14

(4) Diode conduction interval 1 : n 2 D 2 on i 2 V g i 1 W 1 V s v 1 L 11 D on s Q off i m l 12 l 23 l 13 l 34 l 14 l 24 W 2 1 : n 3 W 3 1 : n 4 v 2 v 3 v 4 D 3 on D 4 on i 3 i 4 V 2 V 3 V 4 The main switch Q is off, the snubber diode D s is off The secondary diodes are on, and the secondary winding currents increase or decrease at the rates that depend on the initial values and the loads W 4

Qualitative Behavior Case 1: Passive Voltage-Clamp Snubber l 12 n 2 i 2 v 1 (t) V g i m l 13 V s L 11 l 14 l 24 l 23 V 2 /n 2 -V x n 4 i 4 n 3 i 3 i m (t) -V s l 34 V 4 /n 4 L 11 V 3 /n 3 Q off, Q s /D s on Commutation interval V x i m l 14 l 24 n 4 i 4 l 12 l 13 l 23 n 2 i 2 V 2 /n 2 n 3 i 3 i 2 (t) i 3 (t) i 4 (t) t t t All outputs in CCM l 34 V 4 /n 4 V 3 /n 3 Q off, Q s /D s off Diode conduction interval t d diode turn-off interval Q on t c Q off D s on commutation interval Q off D s off Diode conduction interval (1-D)T

Measured and predicted waveforms Flyback converter example with passive snubber Measured waveforms Simulated waveforms using extended cantilever model W 2 current 1A/div W 3 current 1A/div W 4 current 1A/div pri transistor voltage 50V/div Winding 2: CCM with negative ripple Winding 3: DCM Winding 4: CCM with positive ripple

Secondary current waveforms Commutation interval l 12 n 2 i 2 W 2 current 1A/div W 3 current 1A/div V s L 11 i m l 14 l 24 n 4 i 4 l 13 l 23 V 2 /n 2 n 3 i 3 W 4 current 1A/div l 34 V 4 /n 4 pri transistor voltage 50V/div Q off, Q s /D s on Magnetizing current commutes from primary winding to secondary windings Reflected output winding voltages are nearly equal Essentially zero voltage is applied across l 23, l 23, and l 23 Large voltage is applied across l 12, l 13, and l 14 Magnetizing current divides between secondary windings according to relative values of l 12, l 13, and l 14 V 3 /n 3 Circuit model referred to the primary side

Commutation interval analysis Secondary winding current (W2 example): l 12 n 2 i 2 i m l 13 V s L 11 l 14 l 24 l 23 V 2 /n 2 Reflected output voltages are nearly equal: n 4 i 4 n 3 i 3 l 34 V 4 /n 4 V 3 /n 3 Q off, Q s /D s on At the end of the commutation interval the reflected secondary currents add up to the magnetizing current: The magnetizing current divides according to relative values of l 12, l 13, and l 14, not directly related to the loads:

Secondary Current Waveforms Case 1: Passive Voltage-Clamp Snubber i m l 13 l 12 n 2 i 2 i 2 (t) i m n 2 L o1 l 12 V s L 11 l 14 l 24 l 23 V 2 /n 2 n 2 V s - V 2 n 4 i 4 n 3 i 3 2 n 2 l 12 l 34 V 4 /n 4 V 3 /n 3 Q off, Q s /D s on i 3 (t) i m n3 L o1 l 13 t n 3 V s - V 3 V x i m l 12 l 13 n 2 i 2 i 4 (t) 2 n 3 l 13 t L 11 l 14 l 24 l 23 V 2 /n 2 i m L o1 l 14 n4 n 4 V s - V 4 n 4 i 4 n 3 i 3 2 n 4 l 14 l 34 V 4 /n 4 V 3 /n 3 Q off, Q s /D s off Q on Q off D s on t c Diode conduction interval (1-D)T Q off D s off t

Secondary current waveforms Diode conduction interval W 2 current 1A/div V x i m l 13 l 12 n 2 i 2 W 3 current 1A/div L 11 l 14 l 24 l 23 V 2 /n 2 W 4 current 1A/div pri transistor voltage 50V/div n 4 i 4 n 3 i 3 Circuit model referred to the primary side l 34 V 4 /n 4 V 3 /n 3 Q off, Q s /D s off The sum of the reflected secondary currents equals the magnetizing current Current slopes can be positive or negative, depending on the initial values and the loads Increased output voltage reduces slope of winding current waveform, leading to reduced average output current Decreased output voltage increases slope of winding current waveform, leading to increased average output current

Diode conduction interval analysis Secondary winding current (W2 example): Winding Thevenin output inductances (ref.to primary): V x l 12 n 2 i 2 All secondary winding currents: i m l 13 L 11 l 14 l 24 l 23 V 2 /n 2 n 4 i 4 n 3 i 3 l 34 V 4 /n 4 Q off, Q s /D s off V 3 /n 3

General Steady-State Solution Case 1: Passive voltage-clamp snubber, all outputs in CCM Averaging of the secondary winding currents gives: where V x is found from the volt-second balance on L 11 :

Thevenin Output Resistance Matrix Passive voltage-clamp snubber, all outputs in CCM Referred to the primary side: Referred to the secondaries:

Predicted and measured cross-regulation (with passive voltage-clamp snubber) 0.4 V 2 [V] 0.2 I 4 [A] 0.5 1 1.5 2-0.2-0.4 V 3 [V] Operating conditions: D=0.52, I 2 =I 3 =0.4A, I 4 =0.1A to 2.0A Changing load on the main output has opposite effects on the two auxiliary outputs!

Predicted and measured output resistance matrix Operating point: D=0.52, I 2 =I 3 =0.4A, I 4 =1A Predicted: Measured variations in the output voltages corresponding to load changes:

Experimental verification of the model Waveforms obtained by simulation match experimentally observed waveforms: the extended cantilever magnetics model can be used to predict complex operation of multiple-output flyback converters The extended cantilever model allows easy qualitative explanation of the converter operation, as well as derivation of a general steady-state cross-regulation model Good agreement between predictions of the steady-state crossregulation model and measured results Off-diagonal terms ( mutual resistances ) in the Thevenin output resistance matrix are determined mainly by the transformer leakage inductances. Prediction of these terms is very good Diagonal terms ( self resistances ) in the Thevenin output resistance matrix are affected by conduction losses. Hence the measured values are greater than predicted by the model.

Discontinuous conduction modes Initial secondary winding currents at the end of the commutation interval are proportional to the magnetizing current, i.e. to the total load, not to the load on the individual output The magnetizing current is divided according to the relative values of l 12, l 13, and l 14, not according to individual loads

Discontinuous conduction modes A lightly loaded output well coupled to the primary (l 12 small) Starts with large initial value i k i k (t) i k Large negative slope in the diode conduction interval => DCM with much increased output voltage T t Increasing load on another output increases i k further Deeper DCM, even larger output voltage i k (t) i k Very poor cross-regulation (often referred to as peak detection ) T t

CCM boundaries General result for operation of all outputs in CCM: The winding best coupled to the primary is most likely to operate in DCM Increasing load on one winding eventually drives all other outputs into DCM Improper relative primary-to-secondary couplings and the resulting DCM modes are the major cause of poor cross-regulation General rule: the winding with the widest load range should be best coupled to the primary

CCM boundaries: experimental example Experimental example: Predicted CCM conditions: Good agreement with experiment

Qualitative Behavior Case 2: Active-Clamp Snubber 1 : n 2 D 2 i 2 V g V s D s Q s i 1 W 1 v 1 v q i m L 11 Q l 12 l 23 l 13 l 34 l 14 l 24 W 2 1 : n 3 W 3 1 : n 4 v 2 v 3 v 4 D 3 D 4 i 4 i 3 V 2 V 3 V 4 Analysis: Use the extended cantilever model for the transformer Neglect losses (other than losses due to leakage inductances) Neglect capacitor voltage ripples W 4

Diode turn-off interval t d (same as in Case 1) 1 : n 2 D 2 i 2 V s D s off i 1 W 1 v 1 L 11 i m l 12 l 23 l 13 l 24 W 2 1 : n 3 W 3 v 2 v 3 D 3 i 3 V 2 V 3 Currents through leakage inductances decrease to zero after the main switch Q turns on; secondary diodes turn off Diode turn-off times are not equal V g Q on l 14 l 34 1 : n 4 v 4 D 4 i 4 V 4 W 4

Main switch conduction interval (same as in Case 1) 1 : n 2 D 2 off i 2 v 2 V 2 V s D s off i 1 W 1 v 1 L 11 i m l 12 l 13 l 23 l 24 W 2 1 : n 3 W 3 v 3 D 3 off i 3 V 3 The main switch Q is on, all diodes are off The magnetizing current increases All leakage inductor currents are zero V g Q on l 14 l 34 1 : n 4 v 4 D 4 off i 4 V 4 W 4

Commutation & diode conduction interval merge into one i 1 W 1 V s v 1 L 11 D s /Q s on i m l 12 l 23 l 13 l 24 1 : n 2 W 2 1 : n 3 W 3 v 2 v 3 D 2 on D 3 on i 2 i 3 V 2 V 3 The main switch Q turns off, the snubber diode D s turns on; Q s is turned on The secondary diodes turn on, and the secondary winding currents start increasing at the rates that depend on the leakage inductances and loads V g Q off l 14 l 34 1 : n 4 v 4 D 4 on i 4 V 4 All outputs operate in CCM always W 4

Qualitative Behavior Case 2: Active-Clamp Snubber l 12 n 2 i 2 v 1 (t) V g i m l 13 V s L 11 l 14 l 24 l 23 V 2 /n 2 -V s n 4 i 4 l 34 n 3 i 3 i m (t) V 4 /n 4 Q off, Q s /D s on V 3 /n 3 i 2 (t) Diode conduction interval All secondary winding currents have positive slope There is only one operating mode: CCM i 3 (t) i 4 (t) t t t Current slopes can be calculated as in the passive snubber case t d diode turn-off interval Q on Q off Q s /D s on Diode conduction interval (1-D)T

Measured and predicted waveforms Flyback converter example: Active-Clamp Snubber Measured waveforms Simulated waveforms using extended cantilever model i 2 [1A/div] i 2 [1A/div] i 3 [1A/div] i 3 [1A/div] i 4 [1A/div] i 4 [1A/div] v q [50V/div] v q [50V/div] t [2µs/div] Simulation model included resistive conduction losses

Secondary Current Waveforms Case 2: Active-Clamp Snubber l 12 n 2 i 2 V s L 11 i m l 14 l 24 l 13 l 23 V 2 /n 2 i 2 (t) 2I 2 1-D n 4 i 4 n 3 i 3 l 34 V 4 /n 4 i 3 (t) t Q off, Q s /D s on V 3 /n 3 2I 3 1-D Secondary currents increase at the rates determined by the leakage inductances and the loads i 4 (t) 2I 4 1-D t Current slopes can be calculated as in the passive snubber case Q on Q off, Q s /D s on (1-D)T t

General Steady-State Solution (Active-clamp snubber) Averaging of the secondary winding currents gives: where V s is found from the volt-second balance on L 11 :

Thevenin Output Resistance Matrix (Active-clamp snubber) Referred to the primary side: Referred to the secondaries:

Predicted and measured output resistance matrix Operating point: D=0.52, I 2 =I 3 =0.4A, I 4 =1A Predicted: Measured variations in the output voltages corresponding to load changes: Compare to the passive-snubber case: resistance values can be significantly different

Discussion Cross-regulation (open loop) is in general better if the terms in the Thevenin equivalent output resistance matrix are smaller Output resistances are directly proportional to the leakage inductances between the secondaries. Therefore, tighter coupling between the secondaries improves cross-regulation On lightly loaded outputs, tight primary-to-secondary coupling may lead to DCM operation and significantly worse cross regulation Relative values of primary-to-secondary leakage inductances are important for good cross regulation: the output with the widest load range should have the best coupling to the primary Perfect closed-loop cross-regulation can be obtained even if leakage inductances are not small: attempt to match rows of the resistance matrix R referred to the primary side

Closed-loop cross-regulation Open-loop cross-regulation model: Closed-loop operation: Increasing load on the regulated output increases duty ratio D to compensate for the load-induced drop All other output voltages increase with D If an auxiliary output has the same open-loop dependence on load currents, i.e., the same rows of the resistance matrix R referred to the primary side, then this auxiliary output will have perfect cross-regulation

Experimental example 0.4 V 2 [V] 0.2 I 4 [A] 0.5 1 1.5 2-0.2-0.4 V 3 [V] Winding W4 (the last row) is closed-loop regulated Winding W2 has better match of the resistance terms with W4 than winding W3 Winding W3 has the resistance terms of opposite sign! Expect better closed-loop cross-regulation on the W2 output (as long as the outputs operate in CCM)

Experimental Closed-loop Cross-regulation 17 V 3 [V] 16 15 V 2 [V] 14 13 12 Possible improvements: I 4 [A] 0.5 1 1.5 2 Avoid opposite-sign terms (which in this example came from the negative leakage of the side-by-side winding W3 next to W4) Improve matching of resistance terms by better primary-to-regulated output coupling Avoid DCM operation by relating primary-to-secondary couplings to loads

Prediction of small-signal dynamics The extended cantilever model can also be used to predict the converter control-to-output transfer function Depends on operating modes of auxiliary windings Significant changes observed when auxiliary winding changes from DCM to CCM Computer modeling method is described in reference [18] Small-signal frequency response is generated by Mathematica, based on converter impulse responses generated by PSPICE or PETS Approach automatically accounts for changes in operating mode Transformer was simulated using N-port model Simulations converged quickly and easily, even though system contained eight states

Measured N-port parameter model Flyback transformer example v 1 W 1 i 1 L 11 220 µh 0.28v 1 0.09v 3 0.68v 4 0.42i 2 0.42i 3 0.14i 4 L o2 0.52 µh i 2 v 2 The n-port model is well-suited for simulation in PSpice or PETS The number of inductances is equal to the number of states in a multiple-winding transformer W 2 0.41v 1 0.39v 2 1.13v 4 L o3 2.3 µh i 3 v 3 W 3 0.02v 1 0.47v 2 0.18v 3 L o4 0.36 µh i 4 v 4 W 4

Measured and predicted transfer functions Flyback converter example Small-signal CCM duty-cycle to W 4 output transfer functions Predictions of model (a) with W 2 and W 3 outputs operating in DCM 40 db (b) with W 2 and W 3 outputs operating in CCM 20 db (b) (a) Measurements 0 db 20 db Magnitude 40 db 40 db 20 db (b) 10 Hz 100 Hz 1 khz 10 khz 0 db (a) (a) 0 90 20 db (b) 180 0 (a) 40 db 270 90 180 Phase (b) 10 Hz 100 Hz 1 khz 10 khz

Summary - Part 2 Two flyback converter cases considered: (1) passive voltage-clamp snubber and (2) active-clamp snubber Multiple-output flyback converter has complex operation strongly affected by the transfomer leakage inductances Extended cantilever model offers easy qualitative explanation of all details of operation General analytical models derived to predict static cross-regulation in converters with arbitrary number of outputs Cross-regulation mechanisms: DCM operation and output resistances due to leakage inductances are now well understood The models can be used to evaluate and compare magnetics designs, and to test various approaches to improve cross-regulation The extended cantilever model can also be used to correctly predict the converter control-to-output transfer function, and to investigate frequency-responses at various load conditions

3. Cross Regulation in Forward Converters with Coupled Inductors v 1 N 1 D 1a D 1b v s1 C 1 R o1 V o1 Conduction losses 1 : 1 : D 2a v 2 C 2 R o2 diodes windings esr V g Q 1 D r N 2 N 3 D 2b D 3a D 3b v s2 v s3 v 3 C 3 R o3 V o2 V o3 Unequal diode conduction times due to the transformer leakage inductances Discontinuous conduction modes

Objectives Explain effects of inductor coupling on cross-regulation using the extended cantilever magnetics model Find general solution for the discontinuous-mode boundaries and steady-state conversion ratio Discuss approaches to coupled-inductor design: Near-ideal coupling Moderate coupling Zero-ripple approach Compare predictions with experimental results Point to dynamic response considerations

DCM Analysis Using the Extended Cantilever Model D 1a v 1 C 1 R o1 v s1 (t) v 1 1 : 1 : N 1 D 2a D 1b v s1 v 2 C 2 R o2 V o1 V o1 v s2 (t) v s1 v 2 V o1 N 2 D 2b v s2 V o2 v s2 V o2 V g Q 1 D r D 3a v 3 C 3 R o3 V o2 v s3 (t) v 3 N 3 D 3b v s3 V o3 V o3 v s3 V o3 Step 1: Voltage waveforms across coupled-inductor windings (assume all outputs operate in continuous conduction mode)

DCM Analysis Using the Extended Cantilever Model v 1 v s1 (t) v 1 = v s1 - V o1 V o1 V o2 v s2 (t) v s3 (t) v s1 v s2 v 2 v 3 V o1 V o2 l 13 l 12 l 23 n 2 :1 i 2 n 3 :1 L 11 v 2 = v s2 - V o2 V o3 v s3 V o3 v 3 = v s3 - V o3 Step 2: Use the extended cantilever model for the coupled inductor

DCM Analysis Using the Extended Cantilever Model v 1 = v s1 - V o1 v 1 l 13 l 12 l 23 n 2 :1 L 11 i 2 v 2 v 2 = v s2 - V o2 l 12 l 23 n 2 :1 i 2 i 2 Thevenin equivalent of the 2 nd winding l 2 v 2 e 2 n 3 :1 v 3 /n 3 v 3 = v s3 - V o3 Step 3: Find Thevenin equivalent for one of the windings

Thevenin Equivalent for the j th Winding v 1 v 2 /n 2 l 1j l 2j v j - i j v n /n n l jn (a) 1:nj i j v j - (b) l j e j Thevenin equivalent of the j th winding

CCM Condition for the j th Winding i j l j e j v Thevenin equivalent sj of the j th winding C j V oj R j - i j V sj t v sj e j V oj -α j V sj D α j V sj (1-D) DT s D 2 T s D 3 T s t t -α j V sj D

DC Conversion Ratio In CCM: Output j in DCM, other outputs in CCM:

DC Conversion Ratio V oj /V sj 1 α j = 0: No coupling α j = 1: Zero-ripple approach, CCM at any load Smaller l j : DCM operation more likely; Worse cross-regulation in DCM 0.8 0.6 0.4 0.2 0 α j = 0 0.5 0.9 α j = 1 k j = 0.2 0 0.2 0.4 0.6 0.8 1 D

Experimental Example MUR810 MUR810 v s1 i 1 v 1 C 1 100 µf I o1 R 1 v o1 V g 20 V IRF540 MUR810 MUR810 v s2 i 2 v 2 C 2 100 µf I o2 R 2 v o2 MUR810 MUR810 v s3 i 3 v 3 C 3 100 µf I o3 R 3 v o3 3-output converter, V o1 is the main regulated output at 5.1V

Coupled-Inductor Design Examples Design #1: W1 turns: 24 W 3 wound on top of W 1, W 2 W 1, W 2 Bifilar W2 turns: 24, tightly coupled to W1 W3 turns: 24, moderately coupled to W1 W 3 Design #2: W1 turns: 24 W2 turns: 24, tightly coupled to W1 W3 turns: 28, poorly coupled to W1, # of turns selected for zero-ripple W 1, W 2 Bifilar Magnetics Inc. 58254 powdered iron core

Model parameters Design #1: n 2 =1.004, n 3 =0.919, l 12 =0.36uH, l 13 =21.3uH, l 23 =16.4uH l 2 =0.36uH, α 2 =1.006 (W2 tightly coupled to W1) l 3 =7.81uH, α 3 =0.919 (W3 moderately coupled to W1) Design #2 n 2 =0.997, n 3 =0.994, l 12 =0.36uH, l 13 =21.3uH, l 23 =16.4uH l 2 =0.38uH, α 2 =0.99 (W2 tightly coupled to W1) l 3 =44.3uH, α 3 =0.994 ( zero-ripple approach in W3-to-W1 coupling)

Near-Ideal Coupling vs. Moderate Coupling Coupled-Inductor Design #1 i [0.5A/div] 3 0 W3 lightly loaded, operates in DCM 0 i 2 [1A/div] i [2A/div] 1 0 v [20V/div] s3 0 V o1 =5.1V, I o1 =2A, V o2 =5.3V, I o2 =0.97A, V o3 =5.6V, I o3 =0.2A, V g =20V, f s =100kHz

Near-Ideal Coupling vs. Moderate Coupling Coupled-Inductor Design #1 Measured closed-loop cross-regulation, V o1 regulated at 5.1V, I o1 =2A 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 V o2 -V o1 [V] V o3 -V o1 [V] Design #1 Worse cross-regulation on the moderately-coupled W3 because of DCM operation -0.2 0.01 0.1 I o2, I o3 [A] 1 10

Near-Ideal Coupling vs. Zero-Ripple Approach Coupled-Inductor Design #2 0 0 0 i 3 [100mA/div] i 2 [100mA/div] i 1 [2A/div] Current spikes and DCM operation on the tightly coupled W2 output CCM operation and small current ripple on the W3 output 0 v s1 [20V/div] V o1 =5.1V, I o1 =2A, V o2 =5.7V, I o2 =0.2A, V o3 =5.7V, I o3 =0.2A

Near-Ideal Coupling vs. Zero-Ripple Approach Coupled-Inductor Design #2 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0-0.1 V o2 -V o1 [V] V o3 -V o1 [V] Design #2-0.2 0.01 0.1 1 10 I o2, I o3 [A] Measured closed-loop cross-regulation, V o1 regulated at 5.1V, I o1 =2A The tightly-coupled output operates in DCM and has large current spikes Tight coupling may not be easy to achieve with larger number of outputs The zero-ripple approach gives slightly better cross-regulation than tight coupling, and can in practice be achieved easily for any number of outputs In the zero-ripple approach, one output operates with non-zero ripple determined by L 11

Zero-Ripple Approach: DCM on Other Outputs 0 0 0 i 3 [0.2A/div] i 2 [0.2A/div] i 1 [0.2A/div] v s1 [20V/div] 0 W1 and W2 in DCM, so: Voltage waveforms across W1 and W3 are no longer propotional Cross-regulation between the regulated W1 output and the W3 output is affected because of relatively poor coupling between W1 and W3 W3 current has increased ripple

Zero-Ripple Approach: DCM on Other Outputs 0.6 0.4 Since W3 is poorly coupled to W1, DCM on W1 causes poor crossregulation on W3 0.2 0-0.2-0.4-0.6-0.8 0.01 0.1 I o1 [A] 1 V o2 -V o1 [V] V o3 -V o1 [V] Design #2 10 Zero-ripple approach works best if the non-zero-ripple output always operates in CCM Possible approach: Use a heavily loaded output as the non-zero-ripple output Adjust the number of turns on the other windings to scale effective turns ratios and achieve the zero-ripple condition of equal induced and applied voltages to the winding

Coupled-Inductor Design Approaches Near-ideal coupling: (very small l j, α j 1) Good cross-regulation even in DCM Exact matching of turns ratios in necessary Significant current spikes, larger ripples May be difficult to achieve in practice Moderate coupling (moderate l j, α j 1) Degraded cross-regulation in DCM Zero-ripple approach (moderate to large l j, α j 1) Effective turns ratios match the ratios of imposed voltages Effective turns ratios differ from the turns ratios of physical windings CCM operation down to almost zero load Very small ripples Best static cross-regulation if non-zero-ripple output always operates in CCM

Frequency-Response Considerations Experimental 5-output forward converter with current-mode programming used in the feedback loop Coupled-inductor had moderately coupled auxiliary windings with physical turns ratios matching the physical turns ratios on the transformer secondaries Load variations on the auxiliary outputs produced very large changes in the experimental control-to-output response and cross-over frequency of the feedback loop Observed behavior can be explained once discontinuous conduction modes on the auxiliary outputs are taken into account

2-output Circuit Model i s v s1 i 1 v 1 C 1 100 µf I o1 R 1 v o1 V g v c R s i s v s2 i 2 v 2 C 2 100 µf I o2 R 2 v o2 V o1 is the main regulated output Coupled-inductor design is using windings W1, W3 of the design #1 with W3 moderately coupled to W1 Control-to-output frequency response was found using the method of reference [18] Load on the auxiliary output changed to move from CCM to DCM

Control-to-output responses 10 G c [db] 0-10 -20-30 0 Output 2 in CCM Output 2 in DCM The model correctly predicts a significant change in the magnitude and phase responses A factor of two change in the cross-over frequency f c would result if f c is between 5% and 20% of the switching frequency -50 G c Output 2 in DCM -100 Output 2 in CCM -150-200 0.001 0.005 0.01 0.05 0.1 0.5 1 f /f s

Explanation Simple model for current-mode programming predicts a single-pole response with pole frequency inversely proportional to the output filter capacitance When the auxiliary output operates in CCM, the effective filter capacitance referred to the main output is C 1 C 2 When the auxiliary output operates in DCM, the effective filter capacitance referred to the main output reduces to C 1 ; C 2 is disconnected from the main feedback loop because the auxiliary output operates in DCM i s v s1 i 1 v 1 C 1 100 µf I o1 R 1 v o1 V g v c R s i s v s2 i 2 v 2 C 2 100 µf I o2 R 2 v o2

Summary - Part 3 Extended cantilever model used to explain operation of the coupledinductor in a multiple-output forward-type converter General analytical solution found for DCM/CCM boundaries, and for conversion ratio when one of the outputs operates in DCM Three coupled-inductor design approaches evaluated and compared: near-ideal coupling moderate coupling zero-ripple approach Effects of possible DCM operation in the auxiliary outputs on the main contol loop were pointed out and explained