Uninterruptible power supplies (UPSs) fall into two distinct classes—standby or online. While the relatively low-cost, standby UPS has found favor in the consumer desktop market, the 5- to 10-ms delay these devices introduce when switching from utility to battery cannot be tolerated in critical applications. Here, the online UPS, which avoids this switching action, is preferred. But not all UPSs are created equal; each has its own balance of size, weight, cost, reliability, and performance. So system designers and integrators must understand the internal workings of these devices to ensure that choosing among the various topologies does not result in costly error.
The basic premise of all online UPSs is that the output is glitch free—a sinewave, well regulated, and well protected against overloads and shortcircuits. Also, it should be able to drive both resistive and certain reactive (nonlinear) loads. Because most systems running on UPSs are not purely resistive, their current and voltage consumption are not completely in phase, resulting in a reactive power component known as the "power factor." The rating of a UPS reflects its ability to deal with this reactive load. So, UPS capacity is measured in volts * amperes (VA), rather than watts.
Commonly available online UPSs start at 1 kVA (the size of a computer box), and go as high as several hundred kVA (large cabinets weighing hundreds or thousands of pounds). For backup, sealed, maintenance-free, lead-acid batteries are used for their ruggedness, durability, and ability to supply high inrush currents. They are easy to charge, and they maintain that charge over long periods when not in use. However, the batteries deteriorate at temperatures of 60°C and above, and at low temperatures they lose capacity—in terms of ampere hours (Ahr)—very quickly. The batteries are connected in series to arrive at the desired voltage (12 V * the number of batteries), and in parallel to arrive at the desired capacity.
If the inverter within the online UPS runs on battery at all times, the batteries must be charged continuously from the utility supply. The charger is rather large because it supplies the entire current needs of the inverter, as well as furnishing current to charge the battery. Because the charger is a dc-dc converter (from the rectified utility voltage to battery voltage), and the inverter follows it, this online topology, in effect, completes a double conversion. Hence, the name "double-conversion UPS."
The inverter does not recognize a utility failure because, during the absence of utility power, it will continue running on the battery and delivering its output. But the battery is now in a discharge mode, and will run the inverter for a period proportional to its capacity, temperature, and current drain. The smaller the load, the longer the backup time.
Online UPS manufacturers employ several major design architectures. All of them are in common use in commercial, industrial, and military UPSs. As noted, in online UPS, the battery powers the inverter stage at all times. The utility (ac) drives the high-power battery charger, which charges the battery, and supplies the inverter simultaneously (Fig. 1). The inverter stage—equipped with a stable oscillator, pulse-width-modulator (PWM), and feedback loop—resembles a switch-mode power supply, except that its output is ac rather than dc. The high-frequency PWM stage is modulated by the line frequency (50 or 60 Hz) in a class D fashion, resulting in a high-quality sinewave (after some minimal filtering).
Were it not for the fact that the inverter is limited in capacity and peak current capability, the load would not know the difference between the utility and the UPS output. The UPS proves advantageous however, as it is well regulated (±2%), while the utility voltage fluctuates widely (±15%). This regulation is provided by the feedback loop. A small transformer at the output generates a signal for the feedback loop for error amplification. A ±2% regulation at the output is feasible in this topology from no-load to full-load conditions. The battery-voltage fluctuation is ±15%.
The power transformer at the output of the inverter is a line-frequency transformer, and is, therefore, large and heavy (depending on the UPS capacity). After the battery bank, this transformer is the single most dominant weight and size element within what could already be quite a heavy UPS system.
Notice also, that this transformer, together with the capacitor across its output, acts as a filter inductor (shown by dotted lines)(Fig. 1, again). This bonus feature is accomplished by purposely allowing a high leakage inductance between the primary and secondary of the transformer. The result is a virtual filter inductor, which, together with the capacitor, forms a low-band filter for the high-frequency carrier.
The current-sense circuitry provides overload and short-circuit protection to the output. In addition, the input is bridged to the output of the UPS by a bypass switch. This switch can connect its output to the input utility in case of inverter failure. The transfer switch is a key feature in online UPSs, but it is not standard. Far-East manufacturers make another good use of this switch. At turn on, they connect the load to the utility for 10 s, and only then transfer it to inverter. The first 10 s permits loads with high inrush current—such as switch-mode power supplies (SMPSs), motors, and compressors—to start well on utility, and only then transfer to the inverter under more stable conditions.
Obviously, the line-frequency transformer makes this type of UPS relatively heavy and bulky. On the other hand, it is also the most reliable. The 60/50-Hz transformer acts as a buffer between the inverter and the load, absorbing harmful transients. It makes the bypass switch less sensitive and the entire UPS more rugged and solid. Although this topology reflects somewhat old technology, it's the UPS of choice for many customers who don't mind the extra size and weight, but like the extra reliability.
A variation on the theme of the "heavy UPS" involves driving the inverter from the rectified utility voltage rather than the battery. In this case, a relatively small charger supplies current only to the battery, and not to the inverter. This approach offers advantages, in terms of size, weight, cost, and performance. Also, because there is no real double conversion, the efficiency is significantly higher (80% compared to 65%).
However, the rectified line voltage produces significantly higher voltage (150 V dc for rectified 115 V) than the battery voltage (84 V dc for a 115-V UPS). Thus, the inverter is subjected to a severe "bump" in the form of a voltage step function every time the utility falls and comes back. The bump may be reflected on the output somewhat, even in UPSs with a quick response time. Worse, it may cause a latchup of the UPS, especially when the utility comes back, causing the inverter dc rail to jump by up to 70 V.
The latchup (and subsequent shutdown) occurs when the overload and short-circuit protection circuit senses a severe hike in inverter current. This may be caused by a momentary saturation of the power transformer, or by a sporadic noise pulse that propagates through the system. As a result, circuitry must be added to mitigate the impact of the bump, and slow it down enough, so that it looks more like a ramp than a steep front. This circuitry adds a bit to the circuit's complexity.
High Frequency, Lightweight
To get around the hefty size and weight of the line-frequency transformer, a high-frequency (100-kHz) dc-dc converter can be used (Fig. 2). Dramatic savings in weight and size are accomplished in this way, though the dc-dc converter can add a significant cost of its own.
The absence of the massive transformer at the output means that the load is connected to the inverter transistors (H-bridge topology) directly, without any buffering. This tends to diminish reliability in two ways. Firstly, the reliability of the dc-dc converter (high power) is substantially lower than that of an iron and copper transformer. Secondly, the energy pumped back from the load (due to parasitics or back EMF) may cause transients that can damage the inverter's power transistor. Despite these negatives, there are airborne, portable, or military applications in which the lightweight approach is the most appropriate choice.
All of the design approaches discussed above incorporate galvanic isolation. This is important for the attenuation of high-frequency noise and the elimination of transients riding on the input line. Even so, it's best to tie the input and output neutral lines for safety reasons. It's also worth mentioning that some of the lightweight, online UPSs coming from the Far East are not equipped with galvanic isolation—a move to reduce cost.
UPS designers can essentially eliminate the inverter stage by modulating the dc-dc converter to obtain a wave shape resembling a rectified sinewave (Fig. 3). The change involves a reference voltage (for the dc-dc converter PWM) that looks like a rectified sinewave instead of dc. Because the frequency of the dc-dc converter is relatively high (60 to 100 kHz), the reference frequency of 60 Hz will be treated as a varying dc signal by the PWM.
After rectification, the resulting signal looks like a rectified sinewave with a period of 8.33 ms (for 60 Hz modulation), and a peak amplitude of 160 V. Now it is necessary to invert every second pulse to create an ac voltage of 60 Hz. This is done simply by an H-bridge circuit synchronized to the 60-Hz reference voltage.
The integrated approach is obviously more efficient and less expensive (one conversion instead of two), but it has its problems. For one, the wave shape at the output is not as pure as in the previous approaches (due to crossover distortions). Also, it cannot handle a transfer switch. Our attempts to incorporate a solid-state transfer switch in this topology ended repeatedly with the destruction of the output H bridge.
In newer designs, many of the functions within the UPS are delegated to one microprocessor. These include sinewave generation, PWM, monitoring, and control. It's also possible to incorporate the RS-232 communication into the same microprocessor, though it's not recommended. The microprocessor provides a good approach to circuit simplification, leading to labor and cost reductions.
The majority of online UPSs in the market lack power-factor correction (PFC) at their input. As a result, these units behave exactly as SMPSs due to their input rectification and large filter capacitor. This capacitor causes high-current impulses at the input, causing a substantial amount of harmonic power to be pumped back to the line. The new European Safety and EMC regulations (CE) do not permit such harmonics to be injected into power lines, and as a result, create pressure on exporters to include PFC in their UPSs. The PFC circuit entails conversion (by booster topology), and therefore causes a reduction in the overall efficiency of the UPS—not to mention a 10% to 15% cost increase over non-PFC systems.
Adding to the power-factor problem is the fact that online UPSs range in power into the hundreds of KVA. It is not easy to design-in PFC for such power levels. Above a certain power level, it is likely the European regulations will exempt such apparatus from meeting the recent PF limits set by the EEC.
Many users are oblivious to the fact that some loads (like SMPSs) are extremely reactive, with a poor power factor at their input and, consequently, a very high peak-current demand. Therefore, users tend to purchase a UPS with capacity equal to or slightly higher than their power-system needs, resulting in a useless, undersized UPS.
It should be realized that a UPS rated for 1000 VA is good, at best, for 700 to 750 W. This is how UPS manufacturers specify their units, and it implies that a user cannot expect 1000 W from a 1000-VA UPS. However, what's more important to remember is that high-inrush-load impulses (as in the case of compressors and switch-mode power supplies), are not compatible with UPS capabilities. The end result is that the UPS shuts down to protect itself. SMPS loads are particularly painstaking, and because SMPSs are very common (they are in almost all computer systems), the likelihood is very high that the UPS will be driving an SMPS at its output. However, the UPS and the SMPS can be made more compatible by equipping the SMPS with PFC at its input from the start.
Not all UPSs are the same though, as some are designed for highly reactive loads and will perform better with SMPSs. Some are marginal. The moral of the story is that the user cannot go blindly into this. The best route is to provide the manufacturer of the UPS with an actual (measured) profile of the system's input current to make sure the UPS can handle the expected load. Experience has shown that the user should purchase a UPS with a capacity of at least three times the system's need in watts. This rule of thumb may be challenged by various manufacturers, but for the sake of generalization, we found it to be true time and again.