| |
|
October 22, 1998
DC/DC-converter Spice model speeds simulation
A dc/dc-converter macro model provides fast and accurate simulation of
any fixed-frequency converter. By replacing the oscillator with a stable element, the
model provides simulation speeds more than 100 times higher than an equivalent oscillator
in transient-mode simulation.
Martin O'Hara, Newport Components Ltd
DC/DC converters are notoriously difficult to simulate as component models in Spice
because of the inherent instability of oscillator circuits in computer simulation. All too
often, Spice settles into a static mode in which the circuit no longer oscillates, or it
takes so long to converge between time iterations that the simulation is useless.
For a designer of dc/dc converters, these problems are difficult enough. But for a user
who requires the simulated function of an off-the-shelf device as a part of a larger
circuit, these problems are intolerable. Attempting to obtain a load-circuit dc-bias point
further exacerbates the dc/dc converter's oscillator-circuit problems because oscillators
don't produce an output during a dc simulation.
Fortunately, you can avoid the oscillator-related problems simply by not using an
oscillator in the model. Instead, you can model a dc/dc converter's dc-transfer
characteristic by using a voltage-controlled voltage source. Despite the limitations that
are inherent in all macro models, the resultant model closely represents the behavior of a
fixed input dc/dc converter (Figure 1). Listing 1 is the corresponding netlist.
The model provides a fast and accurate simulation in both ac- and dc-analysis modes.
The results for the model and the actual measured results for a 1W board-mounted dc/dc
converter show an exceptionally close correlation between the dc-transfer curves of load
regulation, line regulation, and efficiency, as well as accurate noise spectra at the
correct dc-bias offset during a transient simulation.
Manufacturers of dc/dc converters can use the model to provide their customers with
accurate modeling data without disclosing the entire circuit design of their products.
Likewise, customers using dc/dc converters can produce their own models when such models
are unavailable from the supplier. The user of the dc/dc converter can then model the
characteristic of their target circuits, including the behavior of the dc/dc converter,
without the previously high simulation overhead or the nonconvergence problems of an
oscillator circuit.
To develop the model, you first need to identify the key aspects of the dc/dc converter
that describe its transfer characteristic. An unregulated, 1W, single-output, board-level
dc/dc converter, the NME0505S (Newport Components Ltd, www.newport-comps.com), is an example of a
relatively straightforward device. Operating from a 5V-dc input and producing an isolated
5V-dc output, this device features a fixed switching frequency of 100 kHz, a load
regulation better than 15%, and a line regulation better than 1.1%. Other characteristics
you might include in the model are the efficiency load curve (for the converter in this
example, efficiency is a maximum with 1W load applied) and the ripple noise at the input
and output terminals. All the required data is available from the dc/dc-converter data
sheet.
Construct the macro model
When developing a macro model, your aim is to use stable components and elements to
model the function of the target circuit. For this reason, the Spice model in Figure 1 uses a voltage-controlled voltage source to describe
the dc-transfer characteristic. This approach avoids oscillator and transformer
requirements and maintains isolation from input to output. You can easily model the
no-load power consumption by using a resistor (RNL) across the input terminals to model
the dissipation due to the oscillator operation.
Input and output terminals should appear electrically as they do in the actual
component, which means that the model needs a low-impedance input provided by the no-load
resistor and a dc blocking output. In the real dc/dc converter, the output dc block uses a
diode-rectifier arrangement. Because this arrangement is stable, the model can also use
it. Other input and output features include ripple capacitors. If you use the model with
additional external filtering, these capacitors ensure that the simulation includes the
effects of any potential poles and zeros.
One of the most unique features of this model is the dual nature of the VRPL source,
which is a 0V-dc, ac noise source. This source serves as a current-feedback element and as
a ripple generator (hence, the source's name). Emulating the real converter's efficiency
curve requires feeding back dc output current to a current sink at the input. Without this
feedback, the converter would be more than 100% efficient. This effect results directly
from the voltage-controlled voltage source, which provides the voltage-transfer
characteristic. The VRPL source provides the necessary feedback. The model effectively
connects the voltage-controlled voltage source to the current-controlled current source
back-to-back to emulate the action of a transformer-to-voltage conversion in one direction
and a current conversion in the reverse direction.
VRPL saves elements
VRPL's second role, as a ripple generator, reduces the number of necessary elements in
the model. VRPL's ac source produces ripple at the output so that an explicit input ripple
generator is unnecessary. The ripple voltage is offset to produce a zero-dc level that
doesn't affect the dc-level output from the voltage-controlled voltage source. Also, the
effective feedback of the VRPL-generated ripple current produces an input ripple that is
synchronous with the output-voltage ripple. You can observe this synchronous-ripple effect
in the real device because the same oscillator produces both the input and output ripple.
Thus, the VRPL source effectively provides current feedback, output-ripple generation, and
input-ripple generation.
Finally, the model includes an element for the isolation capacitance (CISOL), which
allows you to include common-mode effects if necessary.
Choose values for the elements
Once you decide on the elements, you're only halfway to completing the model. The next
step is to determine values for all of the elements and modeling terms. You can choose the
input and output capacitors from the known input and output capacitor values (for this
example, COUT=CIN=1 µF). The manufacturer does not always explicitly state these values,
but you can determine them from application notes that discuss appropriate filter values.
In reality, isolation capacitance is a meas-ure of the capacitive coupling across the
transformer and is a much lower value than that of most discrete capacitors. For this
case, CISOL=24 pF.
You can calculate the no-load resistor value (RNL) using the no-load power consumption
(PinNL) or input current (IinNL) at the nominal input voltage (VNOM),
as follows:
For the NME0505S, PinNL=100 mW, and VNOM=5; hence, RNL=250 Ohm.
You can determine the voltage- and current-transfer characteristics from the transfer
ratio of the internal transformer; use the direct ratio for the voltage-transfer
characteristic, and use its reciprocal for the current. If the transformer ratio is
unknown, you can obtain an approximation from the transfer ratio of the dc/dc converter.
For a 5V-to-5V converter, the ratio is 1-to-1.
The chosen ac ripple voltage gives a zero dc-bias effect so that the model doesn't
offset the output-voltage bias levels during ac simulations. Thus, the model produces a
small positive and negative excursion. The ripple frequency equals twice the nominal
switching frequency because the NME0505S uses full-wave rectification. The oscillator's
switching waveform is a square wave. To produce this signal, the model uses a pulse
waveform with fast rise and fall times.
Determining values for the diode is the most difficult decision. In Spice, a diode is a
model in itself, and the diode model for DOUT includes an estimation of all of the
dc-resistance effects due to tracking, wire, and the active components (Listing 1). This model is similar to the actual diode
manufacturer's model but includes a better estimation of the resistive contribution. The
model in Listing 1 also adjusts the values of the
parameters that govern resistance (RS) and saturation current (IS) to produce a load curve
that matches that of the data sheet. If you don't have a model of the actual diode, you
can create an adequate model by using an ideal diode and manually adjusting the RS and IS
parameters.
Measurements of the output voltage and efficiency from an NME0505S device were
graphically compared to the simulated load and efficiency curves (Figure 2 and Figure 3). The
results illustrate that the model is an extremely close representation of the real
circuit's dc operating characteristic. Additional tests of the line regulation at a fixed
output load exhibit similar levels of agreement.
Ripple was measured on a digital sampling oscilloscope and compared with the simulated
response (Figure 4). The measured result has much more
noise than the simulation, and the switching frequency is not an exact match. Despite
these small differences, the general shape of the ripple is not drastically different,
although the measured result is slightly more rounded than the simulation. The magnitudes
of the measured and simulated ripples are approximately 60 and 75 mV, respectively. The
simulated dc offset differs from the measured results by only 40 mV. Hence, the ac
simulation is a reasonably close representation of the measured result.
Take note of the model's limitations
A macro model is always a compromise; it usually sacrifices accuracy under all
operating conditions for convenience and ease of use. This converter model is no
exception, and, although the ripple, dc-load, line-regulation curves match well with the
device's actual performance, this situation is true only over the device's normal
operating conditions.
Outside the maximum recommended load condition, the real oscillator circuit produces a
relatively rapid fall in output voltage as the ripple capacitor fails to hold up the
output. The model continues to hold a linear regulation curve. At zero load, the actual
output voltage rises to a high value as the ripple voltage charges up the capacitor until
the output capacitor leakage and ripple charge are in equilibrium. Simulations using this
model indicate a much lower zero-load output voltage.
The output ripple voltage is constant in the model, regardless of load conditions, but
the real converter's ripple is much lower at light loads. The oscillation frequency also
changes with input voltage. The model doesn't include these features, and simulation
results using the model always indicate a constant ripple magnitude and frequency.
References
- Pressman, AI, Switching Power Supply Design, McGraw-Hill, 1991.
- IsSpice User's Guide, Intusoft, San Pedro, CA, 1994.
- Power Supply Design Seminar, SEM-1000, Unitrode Integrated Circuits, Merrimack, NH,
1994.
|
![[Download PDF version]](images/pdf.gif){kind=small}
