Click here for the enhanced PDF version of this article

The RF portion of a transceiver designed by Dialog Seminconductor was designed using the EDA software package, Advanced Design System (ADS). The most sensitive part of this chip, the receiver (Rx), was extensively simulated and a demo board was tested to verify the design. The chip was fabricated using TSMC 0.25 µm RF CMOS, customized by Dialog.

The Bluetooth transceiver was put on a FR4 four-layer printed circuit board (PCB) for testing and verification. The objective was to check the circuit behavior in the full Bluetooth bandwidth for accuracy without considering every channel point. Three Bluetooth channels were selected to be investigated over three temperature values. They were: channel 0, channel 39 and channel 79, at the temperatures of -30 °C, 27 °C (environmental temperature) and +80 °C. Bluetooth channel 0 (2.402 GHz), channel 39 (2.441 GHz) and channel 79 (2.481 GHz) were representative of the full Bluetooth bandwidth.

All the measurements were made using a measurement chain including a signal generator, the DUT (with its PCB) inside a climatic chamber, and a spectrum analyzer to evaluate the IF power value. The spectrum analyzer data was imported into an Excel spreadsheet for future use. The climatic chamber setup settled for at least 10 to 15 minutes before performing the measurements to get homogeneous temperature for the DUT. The collected data were the measured channel and the operating temperature, as set from the climatic chamber control.

Before physical measurements were made, ADS EDA software was used to mimic the measurement through harmonic balance simulation, with the best possible accuracy, to have a forecast of the measurements. To limit the simulation size and to speed the simulation, some data-sets were built in advance to model the voltage-controlled oscillator (VCO) circuit behavior at different frequencies and temperature values, retaining the VCO characteristics. Hence, the VCO was not simulated with the other Rx chain, but any performance impairments stemming from the VCO behavior could be modeled thanks to the stored VCO datasets. These VCO datasets were used in the subsequent simulations to feed the mixer and check its linearity.

The ADS suite performed the modeling for the PCB and its lumped components (capacitors, inductors in the matching circuitry), and a balun. Additionally, the use of a spectrum analyzer forced the modeling of an additional PCB to match the differential IF output impedance (approximately 4,000 Ω) of the DUT to the spectrum analyzer input impedance (50 Ω). This impedance-matching stage employed a buffer (modeled using its spice model) with some lumped components, which had their respective vendor models.

The DUT input matching optimization involved the creation of a custom library from vendor-supplied data, since the components were not shared and included in the standard ADS vendor libraries at the time the measurements were performed. Tx and Rx from the DUT board were split on the two top and bottom layers of the PCB, to minimize cross couplings. This arrangement allowed modeling of the vias between top and bottom layers and the traces on the two layers for the Rx path.

The final step was to divide the path between the RF connector and the chip in three parts:

1. the RF connector to the balun and the matching networks;
2. the vias between the bottom layer and the top layer;
3. the path from the vias to the IC leads.

Agilent's planar EM simulator, Momentum, modeled parts 1 and 2. Part 1 exhibits different sizes for the various traces, and Momentum's primitive meshing feature was important to achieve accurate results limiting the machine burdening.

Following the matching networks and the paths to the IC leads, next was the package model (LQFP64), which was created by a quasi-static 3-D EM extraction tool. The bond wires are modeled according to Delfts TU models, which are available in ADS. Then there is the EM model (simulated by Momentum) of the IC pads and IC internal traces up to the first block of the LNA plus mixer ensemble. The IC pads model includes the differential input signals and the supplies (2.5 V and ground). The ground reference for the simulation is the PCB ground, as it is in a real-life situation. The LNA and the mixer used some coils that were EM modeled by Momentum.

The LNA plus mixer composite block sports the full RC parasitic back annotation for every item of the ensemble (that is LNA, mixer, polyphase filter and mixer buffer stages to decouple VCO from the mixer) to take into account high-frequency parasitic effects. The simulation results are shown in the table. For simplicity, the table shows the 27 °C simulation with reference to one Bluetooth channel, but all the other data sets are available for the other measurement points and they show the same error magnitude (worst case, about 1.2 dB).

The measurements exhibit good agreement with the two ADS simulation measurements (total power and power at selected frequencies) for the power. The error is close to 1 dB and for high frequencies it is even lower. The measurement was made using a spectrum analyzer with a 100 Hz bandwidth, which helped reduce the noise, and the measurements were ac based. ADS should mimic this situation, so that the pt function that computes the dc components was forced to compute IC and RF components only and the frequency components span up to 5 GHz, as the FR4 board model could be considered reliable to that frequency only, while higher-frequency components are interpolated. Some measurements were repeated under the same environmental conditions with an oscilloscope, but the instrument bandwidth did not allow input levels lower than -64 dBm, as the noise was too high to detect any IF signal. This led us to perform the measurements with a spectrum analyzer.

Emanuele Stavagna graduated from Trieste University in Italy in 1990. He is an RFIC design engineer at Dialog Semiconductor.

RSS    Save to Del.icio.us  Digg This