Uninterruptible power supply (UPS) system designs have taken a decided turn in recent years as drivers such as lower utility costs and greener technologies become a greater part of every design discussion. Of course, system reliability and performance remain important decision drivers. Many new products have been introduced to meet these changing imperatives, but two different transformerless approaches seem to be gaining traction in the marketplace.


The earliest reliable large UPS systems were rotary, with massive motors and generators. There were a number of negatives associated with that approach, including poor dynamic response time, poor efficiency at light loads, maintenance issues, high initial cost, and a large footprint. For example, the partial load efficiency, at which most UPS systems operate at today, is routinely well over 90 percent in most units. Early rotary systems rarely (see figure 1) reached 80 percent efficiency because the frictional and windage losses were too large to overcome. Many of the newer designs mitigate these negatives, with large rotary systems now capable of mid 90 percent efficiencies in the half load to full load range. Rotary UPS or continuous power system (CPS) can be cost-effective today at medium voltage, high-megawatt design levels.


Figure 1. Block diagram of early rotary UPS topology.

Almost 50 years ago, the industry moved to a static design approach to address the rotary shortfalls. The static designs incorporated rectifiers and inverters to allow battery use for required energy storage, as well as power quality regulation. Early designs had poor dynamic performance and were not at all reliable compared to modern units, particularly in larger parallel configurations. And though more efficient than rotaries of the day, early static designs were not very efficient, barely breaking 90 percent at partial loads. The industry suffered through many years of module failures as well as occasional cascading parallel failures. 

An IEEE-IAS paper (editor’s note: Dennis DeCoster, author of this article, co-authored the paper) published in September 1987 described a “breakthrough” in its day--the Piller Uniblock UPS--which prospered because of those static UPS reliability, efficiency, and performance issues. Static units slowly improved such that by the turn of the century, large static UPS were generally deemed fairly reliable, and dynamic transient performance for most loads actually surpassed even the best of the rotaries. Early static UPS designs also routinely required isolating transformers, which are expensive, large, and cut efficiency even more, in order to operate reliably. Over time, vendors improved these products, as the use of better, faster-switching inverters and IGBTs led to reliable and relatively efficient UPS by the turn of the century. 



Figure 2. Modern transformerless UPS units (clockwise from upper right): Mitsubishi 9900B, Poweraware 9390, Powerware 9395, and the Toshiba G9000.

Current UPS offerings from the vendor community still include these older design approaches, including output transformers in some cases. These systems have been updated in many ways, although some limitations still exist. Some vendors have radically changed direction, opting for one of two new transformerless designs.

Transformerless UPS architecture depends on state-of-the-art inverter switching and controls that are fast enough to smooth the ac waveform without significant magnetic filtering. At first considered a very dangerous bet due to the theoretical lack of isolation, vendors who got into the game seem to have addressed this possible shortcoming.

The first approach is a modular design capable of running in nearly zero-loss off-line mode, as well as conventional double-conversion mode. Eaton Powerware’s 9395 and 9390 series UPS products exemplify this approach. The second approach uses very high-tech tri-level rectification and inversion in concert with transformerless architecture to deliver very high on line or double-conversion efficiency in a single UPS module. Both Mitsubishi electric and Toshiba use this topology in their latest generation 9900B and G9000 Series UPS products (see figure 2).


Figure 3. ITIC curve depicts the levels of server tolerance.


Only a few years ago, the critical facilities engineering community labeled off-line UPS as a profoundly bad idea. The idea of ramping up an off-line inverter to full load was deemed far too unreliable for any large data center or critical user to consider. But time marches on. In a remarkable turn of events, some of the same vendors who felt that both steady-state voltage regulation and frequency regulation to be required features have developed systems that have neither in a strict sense.

In practical context, the average UPS user does not need the precise power regulation provided by standard double-conversion UPS the majority of the time. Servers can tolerate up to a +10 percent swing in steady-state voltage, and so even a +3 percent steady-state regulating UPS is not problematic, though all modern systems are much better than that. These loads can also will also tolerate even a total loss of power for up to 20 ms (see figure 3). Some UPS manufacturers have developed systems that idle in off-line mode (labeled “Energy Saver” or “ECO” or similar), then instantly ramp up to full double-conversion when power problems occur. The breakthrough came in designing a bypass power loop that eliminates mechanical switching time, resulting in mode switching fast enough that it appears transparent to the


Figure 4. This is a trace of an Eaton UPS in energy saver/offline mode recovering from a power fail on input. Loading for this particular module (a 160 kVa) was 93 percent. Output V stays inside ITIC guidelines.

vast majority of loads. Mission Critical West witnessed transient testing at eight different sites that confirmed that these systems stay inside the ITIC for transient as well as steady-state performance conditions.

Operation of hundreds of these off-line-capable systems over the past two years or so also demonstrates that this approach actually works quite well. There is a short inverter ramp-up interval under certain conditions when output drops sharply as one would expect, but actual ramp-up curves show that the power quality stays within ITIC recommended limits during the recovery. Inverter reliability appears to be on par with double-conversion mode operation, perhaps due to sharp reduction in thermal stresses (see figures 4 and 5).

Off-line UPS operating modes also do not allow for steady-state frequency regulation. This is not a practical issue in the U.S. as the utility rarely moves more than a tenth of a hertz off the nominal 60-Hz baseline. Generator power, of course, is different, but in the case of the Eaton system, off-line operating mode is disabled and automatically returned when the site is transferred back and power stabilizes.


Until a year or two ago, every major UPS sold in recent years used a conventional two-level IGBT power


Figure 5. Momentary voltage dip/recovery of an off-line UPS

semiconductor design. Then Mitsubishi Electric, which manufactures semiconductors as well as UPS systems, made a breakthrough. They found that a third level of custom-tailored semiconductor arrays can sharply cut switching losses, reduce filtering, and make it easier to optimize sizing. Complexity certainly increases, but efficiency rises dramatically at all load levels. Three-level, or carrier stored trench transistors (CSTBT), when used in both rectifier and inverter, almost double online efficiency. Deployed in other industries, this technology was never attempted in UPS. To date, Mission Critical West has seen no significant field reliability impact from the increased parts count relative to competitors. The technology was also picked up and marketed by Toshiba under a joint agreement (see figure 6).


Paralleling UPS modules was very tricky in early days, due to the intricacies of synchronizing, load sharing, and variations from harmonics, noise, faults, and distance, etc. But vendors have developed solutions to these problems over the past decade.

In fact, many possible paralleling approaches now exist, including conventional paralleling of up to 6 or 8 single-rating UPS modules. Feedback and paralleling control can be achieved by old-school star or other redundantly wired systems or “wirelessly.” The wireless breakthrough eliminated the need for master or slave modules. Identical UPS subsystems all monitor and compare independently, eliminating single points of failure and much of the complexity.


Figure 6. Sharp reduction in harmonics generated and filtering with CSTBT. Courtesy of Toshiba

This approach works very well with transformerless UPS. For some applications using smaller size sub-modules relative to design load can translate into interesting and often cost-effective ways of achieving higher redundancy. For instance, one vendor may use three 750-kilowatt (kW) UPS modules to get to 1,500 kW at N+1 design. But if another vendor’s 750-kW units is really three 250 KW submodules, each capable of independent parallel operation, they can get to 1500 kW at statistically far superior N+3 redundancy using the same three by 750-kW layout. This is because the latter system is actually a six-module system with three extra for redundancy.

Not as practically significant is the bypass function in parallel systems. Bypass is required for load fault clearing and isolating faulted modules within the UPS itself. Static transfer (STS) switches can provide momentary switches with mechanical contactor or full duty-rated switches. Individual STS switches can bypass at the module level, and one large STS can control the whole system level. There is lots of debate on which is better, and many large vendors provide a choice.


UPS systems operate at a variety of load ranges and in a variety of configurations. All these affect efficiency, so advertised “typical” efficiency values have little meaning. A mid-range capacity of 35 to 40 percent of rating, for example, fits with typical load levels of 2N and 2N+1 UPS systems, as well as many UPS that run far less than maximum design rating. The tri-level or CSTBT (Mitsubishi or Toshiba) transformerless UPS operating at that loading run close to 96 percent efficient in full double-conversion mode (see table 1). Eaton’s transformerless UPS is much less efficient at that power module loading, perhaps 92 to 93+ percent depending on model, but since it is modular in construction, a two to four module system may allow shut-down of one or more modules upping efficiency closer to 94 percent or even higher. This is because the practical effect of the partial shutdown is to load the remaining modules to a more efficiency-favorable 50 to 75 percent of capacity. And if that same system is running in off-line mode, efficiency exceeds 98 percent.

A transformerless UPS module may or may not equate to a transformerless system, depending upon voltage service, distribution, and use levels. Finally, as a general statement, redundancy and efficiency have an inverse relationship. More redundancy (reliability) generally means less efficiency.


Table 1. Efficiency comparison of transformerless UPS products at different loads. 


Transformerless UPS always cost less than equivalent traditional UPS with transformers. Initial cost savings result from elimination of copper magnetics and possibly filtering reductions as well. Lower cost may or may not result in lower initial price to the customer, depending on sales pressures. But cost reductions also result from decreased space/footprint, probable utility rebates, and, of course, lower utility operating costs


UPS system space varies greatly with many factors. These include the deployed voltage, input filtering or transformation, type of UPS, ultimate design expansion, the rating of modules (if modular), and other factors. Regardless of application variables, in every case a transformerless UPS will be smaller than a transformer UPS, all else being equal. The difference can be dramatic. Mission Critical West recently had a recent case where a client replaced a 15-year-old 500-kW 480-V legacy UPS with a new transformerless type of same rating and voltage having a footprint more than 50 percent smaller. The older paint lines on the floor made the difference obvious. The older system in this case was negatively impacted by both output transformer and input transformer, along with higher capacitor counts for filtering the slower inverter and rectifier assemblies. No two applications are the same, but all the transformerless UPS products examined by Mission Critical West offer a footprint of 20 square foot or less for a 500-kW product series, to pick one example.


Mission Critical West sees no significant performance defect in any of the transformerless systems discussed for typical data center or server application. This does not mean that testing did not reveal any performance differences. The magnitude of those differences depends on how modular systems are built out, but in rated-module to rated-module testing, the CSTBT/tri-level UPS is unmatched in dynamic performance. For example, 100 percent load step voltage transients are at or under a phenomenal 1 percent (see figure 7). Conventional inverters, including transformerless off-line capable types, can be as much as five times that upset. But at a practical level, 5 percent is fine since data centers rarely experience heavy load steps. For some critical manufacturing loads (chip fabs, some pharmaceuticals, etc), where high inrush motor loads or magnetic loads are routine, this performance difference could be a deciding factor.

Overload capability, gen-set compatibility, harmonic load handling, and fault clearing were excellent for all above units tested, though CSTBT had a technical if not practical edge in many of these tests as well. As always, specific UPS configurations can dramatically effect results and so these comments are guideline only. Redundant UPS sections or UPS modules add reliability as well as performance due to available apparent capacity increase, but penalties exist in cost, footprint, and efficiency.

Figure 7. Excellent block load transient performance. Courtesy of Mitsubishi Electric


If the serving utility is very reliable and voltage is stable, off-line-capable UPS may prove a good choice. Similarly, if utility is very expensive, running at off-line super high efficiencies is clearly attractive. In 2N UPS architectures, it is possible and even desirable to consider running one system in off-line mode, and the other in double-conversion mode, until enough operating time passes (perhaps a year) to confirm which way to operate both systems in a given utility region. Another driver toward this system could lie in flexibility or expandability. In the 9395 format, a single 275 KVA can be expanded up to 4X, or 1100 KVA. Such a four-module system can be operated at 550 N+2, 825 N+1, 1100 at N. But it can also operate with one or modules shut down so to speak, allowing more efficient 550 N+1 operation, with at that point built-in expansion.

The Tri-Level or CSTBT UPS may win out if utility problems are more frequent, problem loads exist, and/or if initial as well as projected loads stay 50 to 90 percent of that module rating. If significant steady-state voltage drop (brownout) conditions occur regularly, say from large industrial neighbor effects, this is a great option. For production or manufacturing applications, where a majority of loads are motors, transformers, high THD drives, etc., this can also tilt things in favor of this technology.

Although very different from each other, both of the UPS approaches discussed here have significant edges over traditional transformer-equipped UPS designs.