Solving The Valley Jumping Issue in Quasi-Square Wave Resonant Power Supplies

Quasi-square wave resonant converters, also known as quasi-resonant (QR) converters, allow the design of flyback Switched-Mode Power Supplies (SMPS) with a reduced Electro-Magnetic Interference (EMI) signature and an improved efficiency at full load. However, as the switching frequency increases when the load decreases, the frequency excursion must be limited to avoid additional switching losses. Traditional QR controllers feature a frequency clamp to limit the frequency excursion. When the switching frequency of the system reaches the frequency clamp limit, valley jumping occurs: the controller hesitates between two possible valley choices, resulting in an instable operation and noise in the transformer.

A new technique to overcome this problem is to decrease the switching frequency step by step by changing the valleys as the load decrease. Once the controller selects a valley, it stays locked in this valley until the output power changes significantly: this is the valley lockout technique recently introduced by ON Semiconductor.

We will explain the valley jumping problem and introduce the valley lockout technique as a way to solve this issue. A practical application example will be presented with experimental results backing up the theoretical study.

INTRODUCTION TO QUASI-SQUARE WAVE SIGNALS

Quasi-square wave resonant power supplies generally referred as quasi resonant (QR) power supplies are widely used in notebook adapters or in TV power supplies. The principal feature of this architecture is the zero voltage switching (ZVS) operation which allows the reduction of switching losses and helps to soften the EMI signature. The ZVS operation is achieved by turning on the MOSFET once the demagnetization of the transformer is completed, at the minimum of the free oscillation - hence the term “valley switching” - caused by the resonance of the LC network present on the drain node. This network is actually made of the primary inductance, L^p, and the parasitic capacitance present on the drain node, C_lump.

By nature, a QR power supply exhibits a highly variable switching frequency which depends on the input and output loading conditions. Unfortunately, the switching frequency increases as the load decreases resulting in poor efficiency at light load because of the increased switching losses budget. To improve the light load efficiency, a means has to be found to clamp down on the switching frequency.

TRADITIONAL QR CONVERTERS

Traditional QR controllers include an internal timer that prevents the free-running frequency from exceeding an upper limit. The frequency limit is generally fixed to 125 kHz to stay below the 150-kHz CISPR-22 EMI starting point. Fig. 1 shows a simplified view of the internal architecture of a controller featuring an 8-µs timer to clamp the switching frequency.

In order to turn the MOSFET on, not only a valley must be detected by the Zero-Crossing Detection (ZCD) comparator but the 8-µs timer must have elapsed (Fig. 2). If a valley occurs within the 8-µs window, the MOSFET is not allowed to start. As a result, the off-time of the power MOSFET is only allowed to change through discrete steps of one free-oscillation cycle.

At low line and heavy output load, the demagnetization time is long and exceeds 8 µs: the controller will switch the MOSFET on in the 1^st valley. However, as the power demand decreases, the demagnetization time shrinks and when it passes below 8 µs, the frequency is clamped. In this case, the transformer's core is told to be reset (meaning the secondary current has reached zero and the internal magnetic field has returned to zero) before the 8-µs timer has finished counting. Rather than immediately re-starting, the 8-µs window keeps the MOSFET blocked and some valleys are ignored. If the output power is at a level such that the needed off-time for cycle-by-cycle energy balance falls between two adjacent valleys, the power supply will operate with uneven switching cycles: this is called valley jumping. Longer switching cycle will be compensated by shorter switching cycle and vice versa. In Fig. 3, two or three cycles of first valley switching are followed by one cycle of second valley switching. The valley jumping phenomenon generates big variations in the switching frequency, compensated by large peak current jumps. These current jumps cause audible noise in the transformer.

Clamping the switching frequency alone thus cures instabilities in light output load conditions, but does not improve the efficiency at that particular operating point. Thus, in traditional QR converters, the frequency clamp is associated with either a skip cycle circuit or a frequency foldback circuit.

FREQUENCY FOLDBACK

The frequency foldback circuit is generally a Voltage Controlled Oscillator (VCO) which reduces the switching frequency as the frequency clamp is hit (Fig. 4). By reducing the operating frequency, the switching losses are also decreased and the efficiency at light load improves. However, during the foldback mode, the MOSFET turn-on event is still synchronized with the detection of a valley: valley jumping occurs as the controller hesitates between two adjacent valleys, resulting, again, in an audible noise in the QR power supply.

Another constraint brought by this technique is the choice of the minimum frequency at full load and lower input voltage. Indeed, the frequency clamp imposes to choose a low minimum frequency that must be above the audible frequency range (usually around 30 kHz). Because of this low minimum frequency, the primary inductance value increases to deliver the necessary output power and the transformer size follows accordingly.

SOLVING THE VALLEY JUMPING ISSUE

A new solution to avoid valley jumping is to change from one valley location to the next/previous one as the output load varies and lock the controller inside the selected place. This is called the “valley lockout” technique. Once the controller has chosen a valley to operate in, it stays locked in this valley until the output power changes significantly. Practically, the observation of the output power variation can be achieved by monitoring the feedback voltage, VFB. A counter is needed to count the valleys. The valley locking is realized by allowing the power supply to have two possible operating points for a given output load. Thus, when the output load is at a value such that the needed off-time for the cycle-by-cycle energy balance is between two adjacent valleys, the peak current is allowed to increase high enough to find a stable operating point in the next valley. For each output load, there is a corresponding operating point in 2 adjacent valleys (Fig. 5).

Thanks to this technique, there is no valley jumping instabilities anymore, and no audible noise in the transformer can be heard.

Another feature of this technique is the natural switching frequency limitation that it provides. Indeed, each time the controller increment the valleys, the frequency decreases by discrete steps as shown in Fig. 6. The switching frequency drop depends of the free-oscillation period:

Where:

L_p = Primary inductance

C_lump encloses all the parasitic capacitances present at the drain of the power MOSFET (C_OSS, transformer capacitance …)

Fig. 6 portrays the switching frequency evolution of an adapter using a controller with valley lockout such as the NCP1380 from ON Semiconductor. For an input voltage of 115 V rms, the switching frequency excursion is contained between 65 kHz and 95 kHz without using any frequency clamp.

Another benefit of this technique lies in the optimization of the efficiency over the load / input voltage range, particularly at high input voltage. At a high input voltage, the zero voltage switching operation is lost: the switching losses increase. Thus, it could be more advantageous to operate in the 2nd valley instead of the 1st valley or in the 3rd valley instead of the 2nd valley for example, in order to allow the power supply to switch with a lower frequency. This case is well described by Fig. 7. that shows the efficiency variation for output power ranging from 24 W to 34 W depending if the controller operates in the 3rd valley or in the 4th valley. Turning-on the MOSFET in the 4th valley instead of the 3rd valley offers a 0.3% gain in efficiency. The switching frequency in the 4th valley is 15 kHz lower compared to the 3rd valley switching frequency.

IMPLEMENTATION IN AN INTEGRATED CIRCUIT

The valley lockout technique has been implemented inside the NCP1379 and NCP1380 which are QR controllers manufactured by ON Semiconductor. Practically, a gang of comparators monitor the voltage on the feedback pin and feed the information to a counter. The hysteresis on each comparator locks the operating valley. Thus, for a given output power, there are two possible operating points: this ensures a stable operation without valley jumping. In order to further improve the efficiency at light load, a frequency fold-back circuit based on a VCO reduces the switching frequency as the output power decreases. The schematic in Fig. 8 shows a 19-V, 60-W QR adapter controlled by the NCP1380.

Thanks to the valley lockout, the controller changes valley (from the 1st to the 4th valley) as the load decreases without any instability. This helps to extend the quasi-resonance operation range down to 20 W at 230 V rms. The scope shots in Fig 9, Fig 10, Fig 11 and Fig 12 show the first, second, third and fourth valley operations at 60 W, 45 W, 30 W, and 24 W, respectively, for an input voltage of 230 V rms. No valley jumping is observed.

IMPROVED EFFICIENCY

The valley lockout optimizes the efficiency over the line/loading range and improves the overall efficiency, as shown in Table 3 and Table 4.

At V_in = 115 V rms, the average efficiency measured is 87.9%.

At V_in = 230 V rms, the average efficiency is 87.7% which is above the 87% limit defined by the Energy Star norm EPA 2.0.

At a light output load, the efficiency is further improved by the frequency foldback circuit. In this mode, the switching frequency is variable and decreases as the load decreases, thus minimizing the switching losses. When the output power is very low, the peak current is fixed to 17.5% of its maximum value and allows the switching frequency to decrease toward very low values. For example, in the 60-W adapter board, the switching frequency is around 31 kHz at Pout = 10 W and drops to 6 kHz for an output power of 1 W as shown in figures 12 and 13. When no output load is connected to the board, the switching frequency is around 340 Hz. For an output power of 0.7 W, the power drawn from the main by the adapter is well below 1 W. Table 1 summarizes the efficiency at light load, while Table 2 shows power consumption at no-load.

By reducing the switching frequency, the frequency foldback also decreases the power consumed by the adapter in standby (meaning when no output load is connected to the adapter). At 230 V rms, the full adapter consumes 85 mW from the mains (including the discharge resistors of the X2 capacitor), which is an outstanding result for a controller without the high voltage startup circuit.

SUMMARY

Traditional QR controllers are subject to the so-called valley jumping which creates uneven switching cycles and generates acoustic noise in the transformer. Valley jumping occurs under some line/load conditions, when the needed off-time for cycle-by-cycle energy balance falls between two adjacent valleys. In order to solve this problem, the valley lockout technique is introduced. This technique consists in allowing the power supply to choose two stable possible operating points for a given output load. Not only the instability is gone, but combined with a VCO, the overall efficiency figure clearly gains from this implementation. Practical results carried over a NCP1380-based controller have confirmed the validity of the approach.