Consumer demand for speed and capacity is driving today's complex communications products. Best in class EDA simulation tools not only have to work harder during the design process, they also need to facilitate intellectual property flow throughout product development. Development of advanced modeling techniques is critical to this process. This article examines the role of RF component measurement based behavioral modeling for systems that contain RF, and illustrates the designer's modeling options.

Reducing cost is a driving factor for designers. With companies focusing on their core competencies, designers need not develop every component of their system. Make or buy decisions are made when engineers elect to reuse existing designs or purchase off the shelf parts to facilitate system development and lower costs. Measurement based behavioral modeling allows designers access to intellectual property (IP) that can lower development or manufacturing costs. These models can be reused internally within system simulations, allowing evaluation of existing designs for reuse while incorporating real world impairments and physical effects.

IC vendors also can use behavioral modeling to benefit the supplier food chain. For example, chip houses can build behavioral models and supply them to their customer base, allowing the system designer to evaluate the part early in the design phase. The chip designer also benefits from hooking the product into the design process rather than the verification or prototyping process. The system designer takes advantage of the economy of scale achieved by the chip supplier and the predeveloped IP. And, because the model is behavioral and not a circuit schematic representation, the IP is automatically encoded. The model can be shared with colleagues, vendors and customers without worrying about IP protection.

Miniaturization is pushing the need for chip verification before tapeout. Measurement based behavioral modeling facilitates verification, because models built from existing system hardware can be used in simulation, along with the EDA designed chip, to verify the system performance.

Measurement based models ease the burden of system complexity by forming a compact simulation model that provides simulation speed and capacity improvements. The speed gain is nearly exponential vs. increasing complexity. Simulation speed improvements of 1,000x or more have been reported. In addition, because the model is simpler, better simulation convergence is often achieved for large systems.

The complexity of communication systems also limits the value of parametric characterizations such as third order intercept and the noise figure. Today, the emphasis is on functional and system verification, and especially on system performance attributes such as error vector magnitude (EVM) and bit error rate. Measurement based behavioral modeling decreases time to market thanks to IP reuse, faster simulation and a faster verification process.

Verification process

Creating behavioral models for system verification is accomplished with a verification model extractor in an EDA tool or through physical measurement. Verification models are extracted in Agilent's Advanced Design System (ADS) by invoking the Verification Model Extractor. It extracts a verification model for RF circuits such as amplifiers and modulators/demodulators using a circuit design encapsulated in ADS. A similar model is constructed through measurement instrumentation, such as a network analyzer, vector signal generator or vector signal analyzer, and ADS.

In a behavioral modeling system verification process, system designers can choose among behavioral models depending on the application. If a new circuit is developed for use in the system, the designer may choose a Verification Model Extractor that con structs models from circuit schematics (unfabricated hardware, as one example). If hardware exists, the designer can employ a measurement based behavioral model built from hardware that originates from in house or from out of house designs.

The modeling and verification process provides an amplifier behavioral model for which a system can be evaluated and optimized against requirements or specifications. This is a time consuming, manual process, taking up to two days for certain amplifiers. To speed the process, a hardware/software Connected Solution using the Agilent Performance Network Analyzer and the ADS Connection Manager builds a behavioral model in a few seconds.

To illustrate the model building and the capabilities of measurement based behavioral modeling, an example general purpose amplifier suitable for 3GPP/W CDMA systems an Agilent MGA 72543 was characterized.

For power amplifiers, amplitude, phase and time distortions can have a detrimental effect on system performance. In the spectral domain, phase distortion of a complex signal can manifest itself in a requirement such as adjacent channel leakage ratio (ACLR). In a complex vector domain, amplitude and phase distortions affect the complex vector. This distortion is expressed as EVM.

Although all amplifiers contribute distortions such as random noise, phase noise, and AM/AM and AM/PM nonlinearity to communication systems, amplifiers used near the output of a TX chain are dominated by AM/AM and AM/PM distortion. These distortions can be characterized in steady state mode. Therefore, a network analyzer is an ideal instrument to provide data for a measurement based behavioral model.

Traditionally, time based distortions originating from bias circuitry, self heating and secondary effects were characterized by recovery time and "droop" specifications. In today's systems these are also captured in an EVM measurement. For most well designed amplifiers, time distortions are dominated by AM/AM and AM/PM nonlinearity. The time varying nature of time based distortions makes network analyzers unsuitable for providing measurement based behavioral model data, because the instrument does not gather time based information. Therefore, a combination of vector signal generator and vector signal analyzer can provide the required data. This example focuses on measurement based modeling that captures and predicts AM/AM and AM/PM nonlinearity.

ADS contains four amplifier models that can capture AM/AM and AM/PM behavior (AmplifierP2D, AmplifierS2D, AmpSingleCarrier and GainRF). Each can be applied where it best fits the designer's needs. For many power amplifiers, an AmplifierP2D model is best for nonlinear modeling applications. This model can handle small signal, large signal data such as S11, S12, S21 and S22 over multiple frequencies and power levels.

To build an AmplifierP2D model, large signal S parameters extracted from a network analyzer are used. The device under test (DUT) is placed between the ports of the network analyzer, while ADS is connected from PC to network analyzer. It controls the instrument, extracts the data and builds the model. The ADS Connection Manager contains a built in power swept S parameter application, including automatic instrument discovery, and provides nearly plug and play functionality. Using Connection Manager, the process of interrogating the instrument and extracting data from the MGA 72543 amp takes about 2 seconds.

The MGA 72543 has about 13 dB of gain and can be used as a general purpose amplifier for PCS applications. In this case, we are not only interested in the general behavior performance, but also in the ACLR and EVM verification requirements. General performance was obtained by placing three amplifier models into the ADS simulator and extracting AM/AM and AM/PM behavior.

To verify ACLR performance to the 3GPP wideband CDMA standard, the amplifier model was ported into the Agilent Ptolemy system simulation environment, and a Wireless Library testbench was used to simulate the performance.

This result was compared with measured performance that was analyzed using a hardware/software Connected Solution. First, the ADS W CDMA signal was ported to an ESG signal generator. The DUT was then measured with a performance spectrum analyzer (PSA). Consistency in the algorithms between the ADS simulator and the PSA guarantees that measured and simulated results are arrived at equally. Similarly, EVM was verified, with good agreement.

Chris Meuth (chris_meuth@agilent.com) is product marketing manager with Agilent Technologies (Palo Alto, Calif.).

See related chart