Computer software tools have been employed for years at both the so-called “circuit” and “system” levels. The use of such tools has become increasingly important as product lifecycles have become shorter and shorter. Electronic products are now often obsolete within months of delivery. This fact has provided great impetus for the combination and streamlining of all phases of product research, design, production and testing. Modern electronic design automation (EDA) tools are beginning to support a more efficient, integrated approach to development, although this trend is quite recent. This article describes the different types of EDA that have traditionally been applied to electronic system development. It also describes the need to integrate these tools to harmonize the work of various design teams from different disciplines. Finally, it illustrate the state-of-the-art of the tools with some examples of typical design processes.

In this article, we will show how system and circuit engineers have used various types of computer simulations to assist in the development process. We also describe recent trends in EDA tool design that enable the work of system and circuit engineers to be better harmonized.

First, let's define some terms. The terms “system-level” and “system-level design” are somewhat vague. These terms take on different meanings within different segments of the electronic design community. Instead, we prefer the term “electronic system level” design or ESL design. This term has gained popularity lately within certain forums. When employing this term, we refer to the process of conceptual or functional design of an overall system, irrespective of any ultimate hardware implementation. We don't mean, however, to imply that the system-level design process takes place in a vacuum. To the contrary, the sharing of goals and constraints between system designers and hardware designers is a major subject of this article.

The complete design of a modern electronic system spans both ESL design and design of the actual signal processing hardware. In 2003, connectivity has become the main market driver for electronic systems. This has further underscored the importance of the communications function that has become crucial within these devices. Due to the proliferation of wireless communication systems, radio frequency (RF)/microwave subsystems have become almost ubiquitous. They are increasing in sophistication, with complex microwave circuits now embodied in tiny microchips. In this article, we describe some techniques that are helping system engineers work with RF/microwave hardware engineers to achieve better designs, faster.

Even though today's design tools are based on the current state-of-the-art software, ESL design is far from an automatic process. The notion of behavioral synthesis as “push a button and out comes a design” is misleading. Whereas some progress has been made over the years, for example, in the area of “high-level synthesis” for digital circuit design, the situation for ESL design is quite different. The conceptual design of an electronic system takes place at a much higher level of abstraction, which is generally concerned with functionality and algorithms, rather than specific hardware choices. To synthesize an ESL design, a great deal of peripheral information is needed before a viable hardware architecture can even be suggested. This information spans everything from cost and weight down to “tricks of the trade,” learned only from years of industry experience.

All is not lost, however, since ESL design has been effectively assisted for many years by computer-based tools. System modeling and simulation has traditionally been used as an adjunct to the ESL design process, providing verification of prospective designs and enabling inadequate designs to be eliminated. This narrows down the number of design choices that need to be taken to further, more costly, stages of laboratory testing and prototyping. Viewed alternatively, these tools enable a broader range of designs to be explored in a fixed time. With ESL software tools, some of these explorations can be performed in a virtual environment. This is usually a much more timely and cost-effective process than traditional hardware evaluation. This is not to say that computer-based simulation should be used as a replacement to laboratory prototyping. Computer models are sometimes not accurate enough to provide predictions of absolute performance. By showing trends associated with different types of designs, these tools can reveal critical behavior and help system developers to wade through an overwhelming number of possible design choices.

To make ESL simulations and modeling most useful, accurate models of hardware must be available and easily accessible within a design environment. Without sufficiently accurate models, results gained from such tools can be misleading. With accurate hardware models, ESL design tools can be more effective in producing better designs, faster. Even the design of seemingly simple subsystems can benefit from access to robust hardware models, as the following example shows.

The solid curve in figure 1 shows the amplitude versus frequency response of an ideal eighth order Chebyshev filter. This filter is a classical S-plane pole-zero design. System engineers typically model this type of filter with an ideal rational transfer function, which is mathematically defined. In a practical implementation, however, circuit components implementing such a filter will have losses. This means that filter sections will now have a finite “Q” value, which changes the placement of poles and/or zeroes. This can have a profound effect on the frequency selectivity of the filter. The dotted lines overlaid on figure 1 correspond to a filter with the topology shown in figure 2. This is a filter with four second sections, each having a Q of 50. Parasitic resistances are automatically installed in the circuit, based on the user's specification of inductive and/or capacitive Q factor. Note that the difference in the frequency response is pronounced around the passband edges and may cause the transmitted signal not to meet a given emission specification. The amount of ripple in the passband also exceeds that of the ideal filter by a large margin. It is important to have accurate models of filter hardware at the system design stage in order to avoid surprises during the actual implementation. For details of the actual models used here, see endnote [1].

One barrier that is often not recognized is that system architects and hardware designers often have very different world views. Nomenclature and parameterization is often starkly different between the two design communities. One example is a difference in normalization and presentation of graphs as a function of frequency. Hardware engineers work with components designed to work within specific frequency bands. An example would be a frequency characteristic that is defined over, say, 5160 MHz to 5240 MHz. ESL engineers of digital communication systems are more concerned with bandwidths normalized to the bit or symbol rate, with frequencies expressed relative to the carrier center frequency. A software tool useful to both communities of designers must be capable of supporting different “world views.” Two different frequency normalization are shown in figure 3. This is a plot of an estimate of the power spectrum of an 802.11a signal. Software tools, which are usable by different design communities, can save a tremendous amount of time in the development process.

It is only in the recent past that electronic product companies have been compelled to create multi-disciplinary design teams. Many modern electronic products have lifecycles that would have been thought absurdly short 20 years ago, or even 10 years ago. Life-cycles of cellular telephones and laptop computers, for example, are currently averaging from six to nine months. Manufacturers must strive for ways to achieve ever-decreasing time-to-market if they are to be competitive. Some progressive electronics companies have, in the past, attempted to harmonize the efforts of ESL and hardware designers in limited ways. It is now becoming absolutely necessary to fully integrate the various parts and stages of electronic system development to avoid time lost in needless design iterations.

This section shows excerpts from a real design that demonstrate how potential problems are identified early in the design phase, long before the product is bread-boarded. Many dollars and much time can be saved through this approach. It is critically important to have the ability of easily importing characteristics of real hardware devices, such as amplifiers. A major focus of this section is showing how important the ability to automatically calibrate drive and power levels of devices, like amplifiers, for an end-to-end link simulation. Without this capability, users are forced to manually estimate gains at various points in the link topology and compensate for them. This limits the number of design choices users are able to explore within a given time, since much time is spent with manual calibration instead of performing and analyzing simulations.

The quest for higher and higher bandwidth efficiencies has produced communication signal designs that are increasingly complex. Many of these signals are stressing conventional hardware designs. Adjustments to current hardware may be required, with completely new designs called for in some cases. EDA tools are being called upon to assess the impact of new signals on conventional hardware subsystems. It is important to be able to import accurate models of devices, such as amplifiers, and set up evaluations of standard measures of performance with a minimum amount of effort.

Figure 4 shows a simulation of certain aspects of the global system for mobile communications (GSM) system. In this simulation, both conventional gaussian minimum shift keying (GMSK) and enhanced data rates for GSM evolution (EDGE) signals are simulated to show the differences in their characteristics. The simulation employs a pseudorandom sequence to drive both GMSK and EDGE modulators. The I/Q envelope trajectory for both signals is overlaid on the I/Q plot. The conventional GMSK signal possesses a constant envelope, as shown in the black trace. The EDGE signal, on the other hand, possesses a high dynamic range. This is evidenced by the high-level amplitude excursions of the signal in red. EDGE is known to possess a high peak-to-average ratio and can affect amplifier operation in a profound way.

The major effect of signals with high peak-to-average ratio is to cause amplifier saturation during extreme amplitude excursions of the signal. It is very useful to be able to record and observe signal excursions, and to compare these with an amplifier's saturation characteristics. The plot in figure 5 shows the amplitude excursions of GSMK and EDGE signals overlaid on a power amplifier's AM-to-AM characteristic. The AM-to-AM characteristic is a standard description used by hardware engineers to characterize the nonlinear behavior of an amplifier. As the input signal level increases, the amplifier is eventually driven into saturation, distorting the signal. The black curve represents the AM-to-AM characteristic of the amplifier. The black dot represents the (constant) projection of the GMSK signal on the AM-to-AM characteristic. The red dots represent the various levels of the EDGE envelope as they are projected on the characteristic. One thousand bits were generated to produce this overlay plot. The plot clearly shows that the EDGE signal is driving the amplifier into saturation. One advantage of a constant-envelope signal, such as GMSK, is that nonlinear amplifiers can actually be operated at, or near, saturation without incurring the deleterious effects of non-linear distortion. The migration to EDGE promises increased throughput, but also requires more sophisticated linear amplifier designs.

When a given signal is applied to a specific amplifier, accurate power calibrations must be performed at various points in the simulation to give correct measures of performance. Adjacent channel power ratio (ACPR) is an example of a standard measure of performance that must be measured accurately. Amplifier bias and output power must be carefully normalized to produce curves, such as those shown in figure 6. This figure shows spectral estimates of both GMSK and EDGE signals after passing through the nonlinear amplifier. GMSK is shown in black, while EDGE is shown in red. The GSM spectral mask is shown overlaid on the graph. To be truly useful in saving design time, EDA environments should support automated calibration of simulations and output results. Without this support, users can spend more time setting up a simulation than performing design trades and analyzing results. In this experiment, power spectra are automatically normalized to provide “dB Down from Carrier” scaling. With this calibration, it is easy to see that whereas the amplified GMSK signal meets the GSM spectral emission requirement, the amplified EDGE signal possesses excessive spectral regrowth.

Integrated EDA tools for electronic system design can be used to harmonize the work of different design teams working in different disciplines. It is important to enable each group of designers to communicate with the tool, in a language with which they are comfortable. Furthermore, such tools can provide real time savings if problem setup is sufficiently automated. It has become increasingly important to utilize design engineers' time efficiently. Properly designed EDA environments are making great strides in this regard.

[1] “Visual System Simulator Model Reference Guide,” Applied Wave Research Inc. (El Segundo, Calif.).

Kurt Matis is the director of systems research at Applied Wave Research Inc. (www.mwoffice.com). He holds a BS in mathematics from Empire State College as well as an MS in electrical engineering and Ph.D. from Rensselaer Polytechnic Institute. Matis can be reached at kurt@mwoffice.com.

Joel Kirshman is a systems engineer at Applied Wave Research. He has a BA in mathematics from Occidental College and an MS in electrical engineering from California State University, Northridge. Kirshman can be reached at joel@mwoffice.com.

Albert Santos is a development engineer at Applied Wave Research. He has a BS and an ME in mechanical engineering from Rensselaer Polytechnic Institute. He can be reached at albert@mwoffice.com.