LDMOS transistors are used for RF power amplification in numerous communications applications, with the most popular being cellular base stations. These RF power amplifiers (RFPAs) provide from 5 W to more than 200 W of output power per channel and require extremely good linearity to maximize the data throughput in a given channel. The main consideration in achieving linearity is the DC biasing of the LDMOS transistor. The drain current bias (Idd) needs to be held constant over temperature and time for optimal performance. Typically, the target accuracy for Idd bias current over temperature is ±5%, but ±3% is more desirable for a high-performance, wideband design. Idd bias drift in a typical class AB amplifier will result in reduced power output, increased distortion products and reduced phase linearity, all of which impair the performance in digital communications systems.
The DC bias on these amplifiers is set by applying a DC voltage to the gate (Vgs) and monitoring the Idd current. Ideally, this Idd will be constant over temperature, but since the Vgs of LDMOS amplifier devices varies with temperature, some type of temperature compensation is required. One method of setting this DC bias involves using an adjustable reference, DAC, or digital potentiometer combined with a temperature compensation source, such as a transistor Vbe multiplier. This solution can work well, but getting tight temperature compensation can be problematic because the Vbe junction temperature characteristic for discrete transistors will vary. Also, the Vgs tempco for LDMOS amplifiers will vary with Idd. The result is that there are variations in Vbe junction characteristics as well as the LDMOS characteristics. For optimal temperature compensation, in-circuit adjustments need to be made for the temperature compensation, as well as the Vgs bias.
An alternative solution is available to provide an accurate DC bias design. It involves measuring and digitally converting temperature information, then adjusting the DC bias using stored values in look-up table (LUT) memory. The memory contents are loaded into a DAC to provide a gate bias voltage. The memory is programmed at final test using measured parameters from the amplifier circuit being tested, resulting in highly accurate, in-circuit compensation.
The X96011 device is particularly suited to this application because it has a temperature sensor, 8-bit ADC, a single non-volatile LUT and an 8-bit current output DAC. The LUT and configuration registers in the device are programmed using a serial 2-wire interface. Communication can be via an on-board system microcontroller or a PC in a production environment.
The figure shows the device used in an RFPA biasing application. The LDMOS device is the MRF9080 from Motorola SPS, a 50 W device optimized for GSM applications. The RF portion is available from Motorola. The complete temperature-compensated bias control circuit consists of the X96011, an op- amp and some discretes.
The MRF9080 device requires from 3.25 V to 3.80 V of bias voltage over a -20 to +100°C temp range for an Idd of 600 mA. Although a 50 W RFPA will quickly warm up, even at -20°C, the bias voltage should be set at startup for optimum amplifier operation. The full-scale output current of the X96011 is programmable with ranges of 400 uA, 800 uA and 1.6 mA maximum. The 800 uA range is chosen here to save on power dissipation yet keep op amp bias currents below the control range. A rail-to-rail input and output op amp was chosen to ensure the voltage ranges of the circuit are met using a +5 V supply.
The maximum Vgs control voltage is set to about 4.28 V to allow for device and temperature variation. A low-pass filter is also included in the bias line to block RF energy from entering the bias control circuit. Since the filter presents a capacitive load to the op amp, R4 and C5 are included to isolate the load capacitance and ensure stability.
The X96011 requires a +5 V supply, so U1 is added to regulate the Vdd supply to +5 V.
The X96011 provides a -40 to +100°C temperature measurement range and uses six bits of LUT addressing (64 steps), giving a resolution of 2.2°C/bit, or compensation that is adjusted every 2.2°C. Because Idd can change between 1.0% to 2.0% per °C, this compensation should keep drift within the target of ±3% if the LUT is constructed well.
The other factor in the control of the bias current is the output bias control voltage quantization. There is 8-bit control with the X96011, and with the full range current and the gm of the MRF9080, the calculated Idd step size is about 24 mA/step. Again, if the calibration is done well, this should result in about ±12 mA variation around the target Idd or ±2.0%.
Various methods can compute the lookup table values. An effective way to construct the lookup table is to make measurements at two temperatures that represent the target range for the product, and then interpolate values for the other temperatures with a linear regression. For example, the ADC value (temperature) is recorded for one setting, and then the DAC is adjusted to place the amplifier at a bias point closest to the correct Idd. The amplifier is heated or cooled to the second temperature and allowed to settle, then the corresponding ADC value is recorded while the DAC is readjusted for optimum bias setting. The table is then constructed using all of the values of ADC outputs (64 entries) and the corresponding DAC values interpolated from the measured values.
The RFPA built with this circuit used the two-temperature calibration method and resulted in an accuracy of ±5% over 0 to +90°C. Since Vgs drift is not perfectly linear with temperature, the error in this method increases at temperature extremes. A more accurate method includes more temperature points and then interpolation between those points.