Will RF Detectors Measure Up To 3G?

Many systems need to measure radio frequency (RF) power. Some examples include communications transceivers, instrumentation, industrial controls, and radar. Sometimes, these RF-power measurements are required to assure compliance with government regulations. In other cases, they help ensure efficient system operation. Over the years, the technology that is used to detect RF-signal levels has improved dramatically. From a primitive-diode beginning, it advanced to multi-function-detector integrated circuits (ICs).

Diode rectifier circuits have been used for signal-level detection for almost a century. This type of detection can be achieved with a very simple half-wave rectifier circuit. Such a circuit includes a rectifying diode, filter capacitor, resistor, and possibly an RF choke and a second capacitor. The two simplest detector circuits are half-wave rectifiers.

Until the early part of the 20th century, solid-state detectors were comprised of a crystal like galena (lead sulfide). Galena has rectifier properties when it's contacted by a metal. An improved detector was the point-contact diode, which consisted of a uniformly doped semiconductor into which a very small, pointed metal whisker was driven. Most frequently, this whisker was made of tungsten. The rectifying junction was formed where the whisker contacted the semiconductor. The metal whisker formed the diode's anode.

Point-contact diodes are still in production today. These diodes produce a very low forward voltage, very small parasitic capacitance, and reasonably high reverse breakdown voltage. All of these characteristics are advantageous for this diode's use as a detector. The point-contact diode also is a majority carrier device. As a result, its ability to rectify high-frequency signals is good. Many radar and communications receivers have utilized point-contact diode detectors.

The point-contact diode does have two major disadvantages, however. First of all, it is quite fragile. It also is difficult to manufacture in a repeatable manner. Vibration or mechanical shock can cause the whisker to be displaced or even temporarily or permanently lose contact with the semiconductor die. Consequently, the reliability of such a device is severely compromised—especially for mobile-equipment applications.

The RF performance of the point-contact diode is affected by the following: the location on the die at which the contact is made; the pressure exerted by the whisker on the die; and the deformation of the whisker due to the force that is required to make contact. Indeed, point-contact diodes are sometimes "tuned" by the manufacturer. In other words, the manufacturer measures their detector performance and then strikes them with a hammer to adjust the contact!

The pn-junction diode solves the mechanical-fragility problem of point-contact diodes. Of course, it also introduces a number of other problems. The pn-junction diode is formed by mating a layer of p-doped semiconductor material with a layer of n-doped material. Depending on the semiconductor material that's used, the forward voltage of a pn junction is much higher than the forward voltage of a point-contact junction. For example, a Germanium (Ge) diode produces a forward voltage of approximately 400 mV, while a Silicon (Si) diode produces about 700 mV. A Gallium-Arsenide (GaAs) pn diode creates a forward voltage of about 1.2 V.

Large forward voltages limit the sensitivity of pn-junction diodes to very small signals. Compared to a point-contact diode, the junction capacitance of pn diodes also can be an order of magnitude or more larger. When it is under forward bias, the pn junction temporarily "stores" minority charge carriers in its depletion region. For the diode to rectify, this charge must either be conducted out of the depletion region or undergo recombination. In the latter case, this process may take many microseconds to complete. The diode's ability to rectify high-frequency signals will therefore be poor. For these reasons, pn-junction diodes are almost never used as RF detectors.

Now, take a look at the Schottky detector diode. This diode has many of the advantages of a point-contact diode without the mechanical fragility. It's formed by depositing a very thin, very small layer of metal on a uniformly doped semiconductor die. Because of their physical contact, the Fermi levels of the metal and the doped semiconductor are forced to be equal. The difference between the metal-work function and the semiconductor material's electron affinity determines the barrier height. It therefore defines the forward voltage of such a junction.

Silicon Schottky diodes are commercially available in four different versions. They offer forward voltages of approximately 600 mV for high barrier, 330 mV for medium barrier, 280 mV for low barrier, and 180 mV for zero-bias detectors. GaAs Schottky diodes produce a forward voltage of approximately 700 mV. Like point-contact diodes, Schottky diodes are majority-carrier devices. As a result, they can switch impedance very rapidly—in most cases well under 1 ns.

Because a Schottky diode's junctions can be made very small, the junction capacitance can be correspondingly small. These two factors make Schottky diodes good candidates for use at the higher microwave and lower to moderate millimeter-wave frequencies. Note that Schottky diodes are extremely sensitive to electrostatic discharge (ESD). They also are easily damaged.

The diode-detector transfer function can be divided into two distinct regions, which are known as "square law" and "linear" (FIG. 1). The square-law region is operative for very small input-signal levels. In this region, the detected output voltage is proportional to the square of the input-signal voltage. For larger input signals, the detector varies linearly with the input-signal voltage.

This phenomenon occurs with all of the diode types that are discussed here. One distinction does exist, though. For each of the junction types, the input-signal level at which this transition occurs is different. This transition from square-law to linear-transfer function does not occur abruptly as the input-signal magnitude approaches the transition region. Rather, it occurs gradually, so that there is little chance for a potential error in determining the input-signal magnitude.

The total voltage gain of a precision log amp, such as the AD8306, can be as high as 120 dB (a factor of one million). Even with no input signal applied to the first amplifier stage, the output stage is near compression. Such compression results from the cascaded amplification of internally generated noise. As the amplitude of the input signal is increased, each of the gain stages goes into compression. This compression starts at the output stage and progresses toward the input stage.

The detectors that are connected to the outputs of these gain stages produce currents that are proportional to the signal voltages at these points. The sum of the output currents is logarithmically related to the input signal's magnitude. The detected output signal has a linear-in-dB variation with respect to the input-signal voltage. The log-amp detector's linear-in-dB response offers two important advantages over square-law detection:

In addition to these advantages, the log amp is typically comprised of several hundred transistors. As a result, the use of a few additional transistors to perform temperature compensation adds virtually no extra cost. Yet it greatly simplifies the task of the circuit designer who uses the log amp.

As the crest factor of the input signal to a log amp increases, the output voltage of a log amp will be shifted (transfer function will shift vertically; mV/dB slope is unchanged) in response to the changing voltage peaks of the input signal. Table 1 shows the correction factor for various input signals. An error in the interpretation of the log amp output can occur if the crest factor of the signal is unknown, as is the case for a multi-carrier W-CDMA base station transmitter with constantly varying call loading and carrier powers. Crest factor affects the performance of diode detectors in the same way, since these detectors are also not rms-responding.

The RF input signal is connected to the input of the differential ladder attenuator. It consists of several cascaded 6-dB sections. Transconductance cells are connected to each node between those sections. They also are connected to the input of the first section and the output of the final section. The attenuation of 6 dB and multiples thereof is achieved by turning one transconductance cell on and turning all of the others off.

Say the required attenuation must fall between multiples of 6 dB. In this case, the transconductance cells that attenuate just greater than and just less than the desired attenuation are turned on. Their outputs are then summed under control of the Gaussian interpolator. That interpolator determines how hard each transconductance cell should conduct in order to produce the correct output-signal amplitude. This interpolation is responsible for the ripple in the law-conformance curves.

Next, the amplifier's output is applied to a squaring cell. This squaring cell's output current, in turn, is applied to an internal summing node. A stable on-chip voltage reference produces a voltage that is applied—via a buffer—to a second, identical squaring cell. The output current from this squaring cell is subtracted from the output current of the signal-path squaring cell. The resulting current is averaged by internal and external capacitors. This voltage is buffered and connected to the set-point circuit's input.

The set-point circuit then applies a scaled version of this voltage to the attenuation control of the variable-ladder attenuator. The ladder's attenuation is automatically adjusted until the signal that's applied to the signal-path squaring cell is at the target value. This condition occurs when the mean current out of the signal-path squaring cell exactly equals the current into the reference-path squaring cell:

When this condition exists, then:

where VATG is the scaled target voltage applied at pin VTGT.

When this condition is achieved, the voltage at the buffer amplifier's output is proportional to the logarithm of the input signal's rms voltage. This value is directly proportional to the input-signal power. The response is linear-in-dB, as is the case for a demodulating log amp. In this system, it is unnecessary to perform the square-root function. The value of the reference-path current is properly selected in the design of this circuit.

In terms of performance, the exponential AGC log amp with large crest-factor signals is an improvement over the rms-to-dc detector. The average error for a W-CDMA signal with 15-call loading is only 0.5 dB compared to performance with a pure sine-wave input signal.

The exponential AGC log amp's performance can be further improved through the use of dithering. Depending on the type of input signal to be measured, a noisy signal is applied either to the VSET or VTGT pin. The goal is to exercise the Gaussian interpolator in such a manner that the ripple's amplitude in the law-conformance curve is reduced.

Wherever the RF input signal is noise-like, such as CDMA or W-CDMA signals, the modified log-amp circuit should be used (FIG. 3). The external filter capacitor, which is connected to the low-pass-corner-frequency (CLPF) pin, is intentionally selected to provide very little filtering of the net current that's applied to the output buffer amp. Yet the filtering is large enough to allow the valid averaging of the input-signal voltage's square.

Consequently, the amplifier's output voltage is noisy. But its average value is centered upon the correct voltage. It corresponds to the rms value of the input-signal voltage. The noisy signal, which has an amplitude that is about 300 mVp-p, is connected directly to the set-point voltage input, VSET. This connection forces the Gaussian interpolator, which controls the input-signal attenuation, to fluctuate about the value that's required to balance the AGC loop. As a result, the attenuation that's produced becomes more of a function of the input-signal voltage's rms value. It is less a function of the attenuation ripple that's inherent in Gaussian interpolation. In simple terms, the ripple in the law-conformance curve is substantially flattened.

If the input signal has a constant envelope, such as a CW sine wave or FM signal, dithering can still be accomplished. But it requires an external noise source to be ac-coupled to the VTGT pin. In this scenario, the current from the reference squarer fluctuates about the dc value that was established by the voltage from the VREF pin. This fluctuation causes noise on the buffer amplifier's output at pin VOUT, which is ultimately connected to the Gaussian interpolator attenuator control. Consequently, the net effect is the same as described above.

A set-point voltage, which corresponds to the desired signal level, is applied to the VSET input. The log amp's output voltage will then be forced (within its limits) to the value that's required to force the external amplitude-controlling element (the VVA in this example) to produce the input signal that balances the squaring-cell currents at their summing node. The application of the set-point voltage to the VSET pin fixes the attenuation of the ladder attenuator. The external system-level control therefore accomplishes the function of balancing the AGC loop.

Note that the set-point voltage that's applied to VSET determines the rms signal voltage that's required to balance the loop. It follows that this set-point voltage would also determine the output power of a high-power amplifier. For example, take the value of the set-point voltage that produces a given signal level at the input under controller mode. It is the same voltage that the log amp would produce in measurement mode when the desired input-signal voltage is present.

The function of the block labeled "Optional Conditioning Circuit" is to shift and amplify the log amp's output voltage. It can range from 0 to 3.6 V up to the control voltage range that's required by the VVA. This block may be omitted if the control-voltage range of the VVA is the same or slightly smaller than the log amp's output-voltage range.

Clearly, the technology that's available for the detection of RF-signal levels has undergone substantial improvement. From primitive crystal detectors, the industry has advanced to demodulating and exponential logarithmic-amplifier architectures as well as direct RF rms-to-dc converters. Such advances enable RF signal levels to be determined with more accuracy and stability. This statement even holds true for the large crest-factor signals that are widely found in modern communications systems.