This month I need to fulfill the promise I made about explaining what all we did with electronic ballast technology in India, thousands of years ago (or so it seems).
The most common (and commercially viable) fluorescent ballasts still use bipolar transistors (BJTs), not MOSFETs. They are also self-oscillating, and therefore need no PWM control IC. This is actually an advantage. Engineers who have worked with self-oscillating topologies (like the well-known Royer oscillator) know not to underestimate them. They are inherently self-protecting and tend to be very rugged. Short their outputs and the frequency automatically adjusts itself to maintain critical conduction mode, so there is no staircasing of the current or magnetic flux in the inductor.
In the electronic ballast too, there is a ferrite inductor in series with the tube. Its basic purpose is to limit the current, as in conventional copper ballasts. The difference being that copper ballast needs a large line-frequency choke made of iron/steel, whereas the electronic ballast is much smaller, lighter, and made of ferrite material. The size advantage is exactly what we expect and invariably achieve by creating switching action at a very high frequency. However, one basic difference as compared to a conventional half-bridge switching power supply (which the electronic ballast shown in Figure 1 resembles), is that the ballast is actually a resonant topology. The L in series with the tube forms a series-resonant circuit with the two C's of the half-bridge (which are effectively in parallel from the AC point of view). So the current sloshes back and forth, and it makes perfect sense to therefore make the circuit self-oscillate, with the help of base-drive transformers as shown in the oval circle.
Click to Enlarge
Note that when the tube has not fired, i.e. when we first apply AC input, the small unnamed capacitor across the tube is the effective series capacitance of the resonant tank circuit. It is the starter cap. The oscillations are in fact at even a higher frequency than in normal operation on startup. But what's more, since there is almost no damping resistance during startup, a high voltage is created across the tube, to get it to fire. We must remember from our high-school EM course that a series LC presents a very low input impedance at its resonant frequency, and at that moment, the voltages across the L and the C can both be very high, though opposite in phase, thereby effectively canceling out as far as the input is concerned.
This is just the opposite as compared to a textbook parallel LC tank circuit, which presents a very high impedance at its resonant frequency, and in which the currents in each component can be very high though, opposite in phase, thus effectively canceling out as far as the input is concerned. We also remember, that if we drive such a low-impedance series LC tank circuit with the driving frequency equal to its natural resonant frequency (which is what we really do by using a self-oscillatory scheme), the oscillations build up every cycle, and so, though the input voltage remains the same, the currents and the voltages across each reactive component keep building up every cycle. Finally, the tube 'fires', thus effectively bypassing the small starter capacitance. Thereafter, the circuit lapses into a more stable, damped, and lower frequency oscillation based on the resonant frequency created by the C's of the half-bridge.
We can clearly see one major problem already. That is, what if the tube does not fire? This is a real-world possibility, since the seals at the ends of the tube may leak, thus affecting the 'vacuum' inside the tube over a period of time. In this situation, we are expecting to replace the tube, not the ballast! But in a virtually undamped LC circuit, the oscillations will build up every cycle, and eventually the transistors, which see the same current when they turn on, will be destroyed. This is what leads to the 'deactivated tube' test. The tube does not fire and the filaments at the end of the tube are typically of such low resistance, that they really can't damp out the steadily escalating oscillations. Some engineers therefore try to place an additional resistor in series with the small starter capacitance, but this certainly affects the ability to start the tube, especially at lower mains input voltages.
A PTC (a thermistor with a positive temperature coefficient) can be used, but it is an expensive solution and also has response-time limitations. In the case of the existing ballast design (just before we set to work on it), the previous engineers had try to circumvent the deactivated tube failures by using more expensive and hefty 'horizontal deflection' transistors (the well-known BU508A). But these have low gain, and they run inefficiently and get hot. So now heatsinks had to be added. In addition, there was still a need to turn off the ballast after several such unsuccessful attempts to fire the tube. So in came an expensive mechanical thermal overload relay, fastened to the heatsinks. But then they found out that it just failed to act fast enough to protect the transistors, given the high heatsink thermal capacity involved.
Our contribution was to use the principle of the forward converter to recover energy back from the inductor. See Figure 2. So an additional winding (with turns ratio and polarity carefully worked out) recovered the excess energy and delivered it back to the main input bulk capacitor. But to eventually cease the switching operation, a sense resistor (or diode in our case) was added that would charge up a capacitor and eventually trigger a small NPN-PNP latch sitting on the base of the lower transistor. Now mains resetting would be required to make the ballast try again. Fail-safe really.
Click to Enlarge
Another critical area of improvement was in the base drive. Historically, many vendors use a single-inductor approach (see Figure 3). However, we should remember that the key to turning on a BJT efficiently is a little different from the way we want to turn it off. In particular, the turn-on must be a little slower (delayed) and it has been shown that the actual crossover duration is significantly reduced if in fact we don't do a hard turn-on. On the other hand, for turn-off we do want to create a hard turn-off, yanking the base momentarily, several volts below the emitter voltage.
Click to Enlarge
The problem with the single-inductor drive is that the turn-on waveform of one transistor is the exact inverse of the turn-off waveform of the other. So there is no possibility of driving them appropriately and differently, and thereby efficiently. The transistors can thus run very hot with the single-inductor base drive. In Figure 1 and Figure 3 we have 'hi' for the high-side transistor, 'lo' for the low-side' and 'sw' stands for the primary winding i.e. the loop of wire passing through the base-drive current transformer from the switching node. To conquer the limitations of the single-inductor drive, engineers often use the double-toroid approach. However, if the permeabilities and dimensions of the two toroids are not well-matched, there are again discrepancies during the crossover, resulting in losses. But realizing that the actual permeabilities of the two toroids are not really important, but their relative permeability is, we started using an innovative 'balun' core to drive the transistors. The advantage is that 'both halves' of the balun are created in the same batch, so though the permeability may be have a lot of process tolerance, the two halves are still very well-matched. Besides, having the two halves still with some uncoupled inductance, allows us to create the appropriate turn-on and turn-off waveforms by a little 'wave-shaping' circuit as shown in Figure 4. Baluns, usually made for RF suppression, and on Ni-Zn ferrite, can be made to order with the more preferred Mn-Zn material. Then they require lesser turns and run very cool themselves too.
Click to Enlarge
With these improvements, we could do away with the heatsinks altogether. The transistors would now run cool-to-the-touch, even while free-standing. No thermal trip was required. The transistors could also now be the cheaper and higher gain MJE13005. As for the input surge test, we replaced the filter with a smaller differential mode toroidal filter deliberately made of lossier Ni-Zn ferrite material, and simultaneously increased the input bulk capacitance slightly, for that is the only way to really pass the lightening surge test without MOVs etc.
Lastly I will pass along an interesting E-Mail I received from Pat Rossiter in Denver, describing that such reliability problems exist not just in India, but right here in US too. He writes:
Dear Sanjaya / Mr.Maniktala [please circle correct choice]:
I read with great interest your article in today's Planet Analog Newsletter. I'm not as technically versed as I'd like to be, but it looks like the ballast that you invented (way to GO!) would be just the ticket for our lights here at Yellow Cab. We are a stone's throw from a sub station and our power fluctuates a bit and in the past month I have replaced 5 ballast packs.
The point of this email--and it is certainly about time that I got to it--is where can I get the ballast packs you described? Eagerly awaiting your illumination on the matter.
With that I put my pen down again for this month. Hope to hear from you at email@example.com or firstname.lastname@example.org. But do copy the person who started this column off: Steve Ohr at email@example.com. Thanks a lot!!