Verifying RF transceivers

The global market for portable communication products such as pagers, two-way radios, cordless phones, mobile phones, personal GPS receivers, wireless internet browsers and portable video phones is growing at a rapid pace.

These small wireless handsets carry voice, data, image and video and will revolutionise the way people communicate and access information.

Manufacturers of these high-performance communication systems must compete on the basis of power consumption, cost, size, weight and features.

The radio-frequency transceiver is the critical hardware that dictates the performance, as well as the cost, the size and the useful battery life in a portable handheld communication product. However, design and verification of the RF transceiver can be difficult or impossible without the right tools.

Architecture of an RF transceiver

The RF transceiver architecture in a digital mobile phone chipset is shown in Figure 1. Typical RF carrier frequency is in the range of 1-2 GHz with the baseband modulation data rate in the Kbps range in the current second-generation (2G) wireless systems and in the Mbps range in third-generation (3G) systems.

The RF receive path consists of a low-noise amplifier, an image reject filter, and a low-noise mixer that translates RF signals to IF. This is followed by a channel select filter (not shown), an IF amplifier (not shown), and a demodu-lator that down converts IF data to quadrature baseband components.

Figure 1.

The RF transmit path has a modulator that translates baseband I/Q data to RF, followed by a bandpass filter and a high-efficiency power amplifier.

The frequency synthesiser consists of one or more phase locked loops that generate stable periodic signals for the mixers, which must perform channel selection.

It contains circuit blocks such as voltage-controlled oscillators, crystal oscillators, frequency dividers and multipliers, phase detectors, etc.

For optimal performance, the transceiver may also have an automatic power control loop for the RF transmitter, an automatic gain control loop for the RF receiver, and an automatic frequency control loop for the frequency synthesiser. The RF chip can be implemented using a number of process technologies: silicon bipolar, BiCMOS, GaAs, SiGe and CMOS. It has several thousand transistors and some on-chip spiral conductors.

The filters in the RF transceiver are typically implemented using off-chip surface acoustic wave (SAW) filters, and perhaps micromechanical (MEMS) filters in the future. Chip packages and substrates have an impact on transceiver performance as well.

The baseband chip is large and complex. Here, transmit and receive data are processed in the digital domain. The baseband transmitter performs channel encoding, pulse shaping and D/A conversion. The baseband receiver performs complex functions such as A/D conversion, multipath delay estimation, channel estimation, interference cancellation and data decoding.

Baseband processing can also compensate for small distortions in analog filters and signal path nonlinearities.

Accuracy and capacity of RF simulators

Although the RF transceiver is a small part of a complete portable communication system, it is by far the most complex to design and verify. It takes twice as many design iterations as other mixed-signal chips when conventional time-domain simulators are employed.

One reason that the verification of the RF transceiver has been difficult is because the signal path contains thousands of active devices, high-Q components such as crystal oscillators, and distributed elements such as SAW filters, transmission lines and chip packaging.

SPICE (simulated program for integrated circuit emulation) uses first- or second-order polynomials to represent transient voltage and capacitor charge waveforms locally in time, and approximates capacitor current waveforms by polynomial differentiation.

Long transient runs are required to simulate high-Q circuits with widely separated time constants. A time-domain RF simulator will accumulate numerical errors and exhibit false convergence when solving for the steady-state waveforms.

Distributed elements are routinely measured and characterised using tables of S-parameters. It is difficult to model frequency-dependent distributed elements in the time domain. Even when this is possible, the simulation results are inaccurate and are obtained only at a great computational expense.

Frequency-domain solution methods are better suited to simulating high-frequency communication circuits containing high-Q components and distributed elements.

Frequency-domain simulators use Fourier series to capture steady-state voltage and capacitor charge waveforms and obtain accurate steady-state capacitor current waveforms by exact time-differentiation of a Fourier series.

The principle limitation of traditional harmonic balance algorithms is capacity. Run time grows as the number of Fourier components increases, making it impractical to simulate circuits containing tens of transistors except on supercomputers. Recent advances in multi-tone harmonic balance algorithms have overcome this limitation. Modern RF simulators that employ Newton-Krylov solution methods exhibit linear memory complexity and almost linear time complexity.

Some of these new methods can be run in parallel, ideal for today's multiple-processor computers. In a highly optimised simulator, sparse, narrow-band and multi-rate structures of the signal frequency spectrum found in different parts of the RF transceiver and the baseband interface are fully exploited, making it possible to simulate RF integrated circuits containing thousands of transistors in minutes.

Figure 2.

Selectivity and dynamic range

The high selectivity and resistance to interference required of modern wireless communication systems present another challenge.

GSM mobile phones operate within a narrow band near 900 MHz and each channel uses 200 kHz of bandwidth.

The low-noise amplifier in the RF receiver is designed to amplify weak, narrow-band signals without adding excessive noise.

Because the transmission medium is shared by many users, there exists interferers within the RF receive band. When two or more interferers are near the desired channel, nonlinearity in the LNA creates intermodulation products within the channel that may block reception.

Subsequent filter, mixer and amplifier stages must select the desired channel and process the weak signals despite the presence of interferers.

The GSM air interface calls for phone operation in the presence of strong interferers (blockers) that are 56-99 dB above the desired signal. An RF simulator must faithfully simulate these extreme operating conditions.

Dynamic range, the ability to accurately resolve weak signals in the presence of strong blockers, and spectral resolution, the ability to resolve signals that are closely spaced in frequency, are key considerations.

The time-domain integration methods used in SPICE are optimised to compute large-signal waveforms. SPICE gains speed by ignoring small variations and adjusting the size of the time-step accordingly. It has a numerical noise floor of 60 dBc, and is not designed with dynamic range in mind.

Figure 3.

Accuracy can be improved by using a tight simulation tolerance and very small time-steps, but run times will still be inversely proportional to spectral resolution.

For example, a standard two-tone inter-modulation test of the RF front-end requires pure sinusoidal tones applied at two closely spaced frequencies - say 900 and 900.2 MHz.

Precise measurement will be needed to determine the small third-order intermodula-tion products at 899.8 and 900.4 MHz, and fifth-order intermodulation products at 899.6 and 900.6 MHz.

Transient integration takes a minimum of Q = fc/BW = 900 MHz/200 kHz = 4500 cycles of the RF frequency.

Millions of cycles would be necessary if we were to include the mixer in the simulation. Each cycle requires many time points.

It is fair to conclude that SPICE transient simulations will necessarily be both slow and inaccurate! It should be emphasised that any solution method which uses the SPICE transient algorithm (including shooting methods) will have dynamic range limited by the numerical noise floor of SPICE.

Frequency-domain methods are much more appropriate for these circumstances. While a large number of time-points are necessary to capture the steady-state waveform accurately, the frequency spectrum is very sparse, so frequency-domain simulations are inherently efficient for high-frequency circuits where the majority of the signals are band-limited.

Frequency-domain solution methods solve the dynamic range problem too - an accurate implementation of multi-tone harmonic balance has a numerical dynamic range that exceeds 200 dBc.

Receive path challenges

The RF receive path must be designed to withstand interference and amplify very weak signals without introducing excessive noise and distortion. The mixer in the RF receive path performs the translation of RF signals to a lower intermediate-frequency band.

Conversion gain, image rejection, 1 dB compression point (P1dB), three-order intercept point and noise figure are commonly used metrics that can readily be extracted from multi-tone harmonic balance simulations.

The design issues are quite complex. For example, both noise figure and conversion gain are degraded in the presence of blockers, strong interferers that can be 70 dB bigger than the desired signals.

The graph shows the noise figure of a mixer as a function of local oscillator power, with and without blocking interference.

It illustrates that noise performance worsens when a blocker is present and underlines the need to analyse noise in RF integrated circuits under realistic multi-tone, large-signal excitations.

In a homodyne (zero-IF) receiver, poor port-to-port isolation due to coupling through layout parasitic, substrate and chip packaging can cause a strong LO signal to find its way into the receive chain, distorting the weak RF signals and saturating the subsequent stages.

The LO signal can also reach the antenna causing in-band interference to nearby receivers. A strong interferer can leak to the LO port and mix with itself causing a DC offset.

In-band blockers result in even-order intermodulation products near DC, and can appear as spurious responses due to a direct feedthrough from the RF port to the IF port.

Another concern is system stability, because the total signal gain from the antenna to the baseband analog interface is more than 100 dB and oscillations can arise due to parasitic feedback paths.

RF integrated circuits must function over a wide range of frequency and temperature, despite manufacturing variations and power supply voltage variations. So nominal simulations are rarely sufficient.

RF and analog simulators must be designed to run long sequences of simulations under many operating conditions. Parametric sweep of input power to extract 1 dB compression point and frequency sweep to compute the large-signal conversion gain and image rejection in an image-reject receiver are two basic examples.

In a time-domain simulator, the initial conditions used to start a transient analysis must satisfy the circuit differential equations. Hence time-domain simulation is inherently slow, as every point in a parametric sweep is an independent simulation run.

In a frequency-domain simulator, the frequency-domain solution at one point in a parametric sweep is often an excellent initial guess for the next point, so harmonic balance simulation is both fast and reliable.

Detailed multi-tone distortion, noise and stability analyses of an RFIC chip under extreme operating conditions are essential for detecting unforeseen problems before a design is committed to silicon.

Time-domain simulation approaches cannot handle distributed elements and high-Q components accurately and efficiently. Only frequency domain methods can take advantage of the sparse, narrow-band and multi-rate structures inherent in the signal frequency spectrum encountered in RF transceivers.

Recent algorithmic advances in multi-tone harmonic balance simulation offer the speed, the capacity, the spectral resolution and the dynamic range required to verify today's high-performance radio-frequency transceivers.