Power limiter

At the low frequencies, to protect the circuits from external large signals, which could tentatively destroy the hardware, the power limiter was employed. Typically the power limiter circuits consist of a reversely biased diode. However, the disadvantage is that such diodes typically introduce large input capacitance, which degrade the input and noise matching. Compensation for such capacitance in a broadband operation becomes impossible, especially because there is a limitation in cut-off frequency of the arbitrary transmission lines (TLs). To minimize the circuit input capacitance introduced by the power limiter, a novel active power limiter architecture was designed (Fig. 2). The circuit consists of a power sensor, comparator, AND gate and a power switch.

Fig. 2. Design topology of the active power limiter. $Z_{1, 2}$ is the resulting characteristic wave impedance, $l_{1, 2}$ is the length of the corresponding TL [Reference Sakalas, Joram and Ellinger7].

All these components are designed by using high-voltage field effect transistor (HV-FET) devices to ensure no breakdown occurs in the power limiter itself. The circuit is implemented under large cross-section TL structures, which are designed for both to match the capacitance introduced by the power limiter and to be able to accommodate large impinging currents. The circuit functionality can be described as follows: the power sensor initially has a high DC potential at the output. In case an input power exceeding a certain threshold is applied, the HV-FET devices at their inputs are switched on and the DC output voltage of the power sensor drops. The comparator circuit is used to create a hysteresis and by this means to protect the power limiter from switching on and off at a single power level. The AND gate is an optional circuit, allowing externally to switch the active power limiter off. If the comparator provides a low signal, it gets then inverted by the AND gate and is further passed to the power switches and the power switches ground the signal. In case of an impinging small signal the power switches remain closed and create an open, letting the signal pass to the LNA. The total parasitic capacitance introduced by the proposed active power limiter topology in off-state can be approximated as follows: $C_{Sensor} + C_{Switch} = ( C_{GD} + C_{GS}) + ( C_{DB} + C_{GD})$. $C_{Sensor}$ and $C_{Switch}$ describe the capacitance introduced by the power sensor and the power switch, $C_{GD},\; C_{GS}$, and $C_{DB}$ are the gate–drain, gate–source, and the drain–bulk capacitances of the HV-FET devices respectively. For small impinging signals the power sensor and the power switch transistors are in the off state. Hence, the major capacitance $C_{GS}$ disappears due to the lack of carriers in the channel. This strongly reduces the overall capacitance introduced by the active power limiter for the low impinging power case. An approximation can be done: $C_{Sensor} + C_{Switch} = 2 \times C_{GD} + C_{DB}$ [Reference Sakalas, Joram and Ellinger7].

Considering a passive diode based power limiter on the other hand, two diodes one forward biased and one reverse biased are needed to absorb the positive and negative half-waves. Hence, in contrast to the proposed topology, the parasitic junction capacitance remains, regardless if the impinging power is high or low. In the particular case, based on components provided within the process development kit, the power switches were modeled for a 10 dBm RF power absorption. The active power limiter introduced a total parasitic capacitance of 52 fF in off-state. On the other hand, the power limiter designed from the diodes with the same power absorption performance at the frequency of interest introduced a total of 780 fF of capacitance Fig. 3 [Reference Sakalas, Joram and Ellinger7].

Fig. 3. Impact by the parasitic capacitance, modeled in the $50\, {\Omega }$ domain. Solid blue: active power limiter, black triangles: equivalent performance diode limiter 1.5–18 GHz [Reference Sakalas, Joram and Ellinger7].

LNA

The design goal for the LNA and mixer components was to maximize the dynamic range and to yield the flat $CG$ response along the 4–40 GHz bandwidth. Addressing the direct path, it is important that the LNA and mixer components remain in the linear region. The large impinging signals of the nearby object (or the direct path) are down-converted to very low frequencies and thus can be easily suppressed by simple off-chip high-pass filters on the IF side. To address the trade-off between the input power ruggedness and the ultra-wide band operation, a distributed traveling wave amplifier topology was chosen for the LNA design (Fig. 4). The topology was optimized to withstand at least 20 dBm of input power level, without any damage being made to the circuit. This was achieved by choosing large input transistor devices, large cross section TLs and wide resistors along the input signal path. With increased number of cascode stages, the linearity is enhanced, where the limitation is given by the cut-off frequency of the resulting overall input/output TL length. The optimum linearity versus noise matching versus bandwidth trade-off was found when employing four input cascode stages (Fig. 4). The matching was achieved by adjusting $l_{1}$, $l_{2}$ (Fig. 4) and the size of the blocking capacitor $C_{b}$ (Fig. 5) for the given biasing condition. The length of output TLs $l_{2}$ was adjusted iteratively with respect to the chosen values of $C_{b}$ and $l_{1}$. Peaking inductors $L_{peak}$ were employed to increase the gain flatness over the bandwidth. Furthermore, with the balanced design the even harmonics could be nearly eliminated, which further enhanced the linearity.

Fig. 4. Top level view of the chosen four-stage distributed differential LNA topology. $Z_{1, 2}$ is the resulting characteristic wave impedance, $l_{1, 2}$ is the length of the corresponding TL and $R_{B,\; C}$ denote the base, or collector loadline resistor [Reference Sakalas, Joram and Ellinger7].

Fig. 5. Differential cascode stage with controlled feedback and inductive peaking [Reference Sakalas, Joram and Ellinger7].

Two topologies were designed and tested. The first one used adjusting the tail current for implementing the gain control [Reference Sakalas, Joram and Ellinger10]. However, the issue that this topology was facing was that reduced tail current reduces the maximum output current swings and hence linearity of the corresponding gain stage itself. To overcome this issue, a second version of the LNA was designed, where the gain control was implemented by controlling the amount of signal passing through the feedback path. Compared to the previous and other reported gain control topologies [Reference So and Hong11, Reference Elkholy, Shakib, Dunworth, Aparin and Entesari12], this approach had the advantage that when reducing the gain, the bias conditions of the LNA circuit itself remain unchanged and hence the linear operation versus the input power is strongly increased. The same LNA topology was chosen for the LB and HB channels with some difference of scaling to better address the corresponding bandwidth and expected maximum input power levels.

Mixer

Two different mixer topologies were tested. Initially a multi-Tanh input-based Gilbert cell topology [Reference Sakalas, Joram and Ellinger13] was chosen. The core advantage of such topology is its unique capability of spreading the $g_{m}$ curve versus the input voltage. This feature is a very promising design approach for highly linear circuit designs. However, the multi-Tanh topology still suffers from a number of drawbacks. Complex circuitry is penalized by high DC power consumption. Increasing the number of input pairs does spread the $g_{m}$ curve versus input voltage, but it also complicates and makes symmetrical layout impossible, causing the circuit to lose the balanced setup advantages. Broadband matching is difficult, since the input connects to high impedance and highly capacitive transistor gates or bases. The attenuation, introduced by the input source/emitter follower stage further deteriorates the noise performance. The second designed mixer topology (Fig. 6) was implemented to overcome the abovementioned drawbacks and to keep the advantages of a Gilbert cell. This topology uses a simple common base input, which has a low-ohmic impedance that can be approximated as follows (equation (1)):

(1)$$Z_{IN \rm{ + /-}} = {1\over g_{m,\; CB \rm{ + /-}}}\Vert jwL_{IN \rm{ + /-}}$$

$g_{m,\; CB}$ describes the transconductance for a common base configuration and $L_{IN}$ in stands for the input shunt inductance. As seen in equation (1), $Z_{IN}$ is a function of both $g_{m,\; CB}$ and $L_{IN}$. Hence, $Z_{IN}$ can be adjusted by scaling the devices ($Q_{IN +}$ and $Q_{IN-}$), sizing the shunt inductances ($L_{IN +}$ and $L_{IN-}$) or actively by changing the bias points of $Q_{IN +}$ and $Q_{IN-}$ [Reference Sakalas, Sakalas and Ellinger14]. For yielding the optimum broadband noise matching, the input impedance was adjusted with respect to the simulated $Z_{OPT}$ [Reference Sakalas, Joram and Ellinger8].

Fig. 6. Schematic diagram of the designed mixer circuit [Reference Sakalas, Joram and Ellinger7].

Furthermore, it can be demonstrated that in a hetero-junction bipolar transistor (HBT) most of the harmonics are generated in the base–emitter junction [Reference Schroeter15], therefore injecting the input signal into the emitter gives a more linear behavior. In addition to that the shunt input inductors $L_{IN} +$ and $L_{IN}-$ are modeled to create a short (by reaching the resonance frequency) for harmonic waves above the fundamental frequencies in this way adding to the linearity of the circuit. The output is shunted with capacitors $C_{SHUNT}$, which serve for grounding the RF and LO leakages, whereas the charge injection resistors $R_{Injection}$ serve for higher gain. The simplistic topology allows for an easy, symmetrical layout and no buffers are needed for the impedance matching. Some major measured performance figures of the stand-alone LNA and mixer combination ICs are presented below (Figs 7 and 8) [Reference Sakalas, Joram and Ellinger8].

Fig. 7. Combined HB LNA and mixer. $CG$ (triangles) and $NF_{DSB}$ (crosses), $S_{DD11}$ (squares). Measured (solid) versus simulated (dashed). IF freq. = 100 MHz, LO = 0 dBm, RF = −30, dBm, high gain setting [Reference Sakalas, Joram and Ellinger8].

Fig. 8. Combined LB active limiter, LNA, and mixer. $CG$ (triangles) and $NF_{DSB}$ (crosses), $S_{DD11}$ (squares). Measured (solid) versus simulated (dashed). IF freq. = 100 MHz, LO = 0 dBm, RF = $-30\, {\rm dBm}$, high gain setting.

The LB LNA plus mixer circuit was fabricated combined with the active power limiter. Its performance was assessed by increasing and then decreasing the power level of the input signal (Fig. 9). More details about the active power limiter can be found in [Reference Sakalas, Joram and Ellinger7].

Fig. 9. Combined active limiter, LNA, and mixer. Measured $P_{OUT}$ versus $P_{IN}$: high gain setting (black) and low gain setting (blue). RF freq. = 10 GHz, IF freq. = 100 MHz, LO power = 0 dBm [Reference Sakalas, Joram and Ellinger7].

Frequency doubler

Frequency doublers are usually designed by using differential pair, bootstrapped mixer, or even distributed topologies [Reference Nguyen, Nguyen, Fujii and Pham16–Reference Henderson, Avery, Sacks, Clements and Pham18]. However, all these topologies suffer either from very high DC power consumption, low bandwidth, or poor harmonic rejection ratio. To overcome the wideband matching problem and to inherently implement a good fundamental and harmonic frequency rejection ratio, a balanced, ultra-wideband, and very simple frequency doubler technique was designed (Fig. 10).

Fig. 10. Chosen ultra-wideband, balanced frequency multiplier topology [Reference Sakalas, Joram and Ellinger9].

The quasi-passive topology utilizes the diode connected HBT devices and resistors to set the operational point. With the symmetrical design, the frequency doubler strongly suppresses the fundamental and the odd harmonic frequencies, leaving just the signals at $2 \times f_{0}$ and the $4 \times f_{0}$. The fourth harmonic frequency is spectrally far apart and in some applications, can be suppressed by filtering. Alternatively, adding a fully balanced amplifier on the output the fourth harmonic could be strongly suppressed. More information about the circuit can be found in [Reference Sakalas, Joram and Ellinger9].

RF switch and active balun

Two versions of the RF switch and active balun circuit were designed. The first version stood out for its good isolation and flat $G_{T}$ response, however the relatively low amplitude and especially phase imbalance was considered critical for directly using it as an LO driver of a Gilbert cell type mixer [Reference Sakalas, Joram and Ellinger19]. The second version was designed to address these issues. Next to some further layout optimization, additional fully balanced large signal amplifiers at the LB and HB outputs were added. These amplifiers were optimized for the corresponding 4–12 GHz and 5–20 GHz frequencies and a good common mode suppression to minimize the phase and amplitude imbalances. In addition for minimizing the DC power consumption and to improve the isolation, these amplifiers feature tail current deactivation (Fig. 11). In case an LB channel is active, the tail current of the HB amplifier is switched off, preserving the circuit from drawing DC power and turning the deactivated HB large signal amplifier into an attenuator. This further improves the isolation of the deactivated LB, or HB channel. Furthermore, two extra inverters are employed in order to fully push, or pull the control signal to GND, or $V_{DD}$. By adding the two dedicated fully differential amplifiers, the G _T could be increased by approximately 11 dB. Due to biasing the differential amplifiers with a supply voltage of 3 V, rather than 2 V as is done for the RF switch itself, the saturated output power of the combined circuit was increased from $-$1 to 2.3 dBm. This enabled a better driving of the LO input on the LB mixer side and the frequency multiplier on the HB side. Furthermore, due to the increased $G_{T}$, the output power saturation was reached at the input drive level of approximately $-7\, {\rm dBm}$. This gave an advantage that the mixer and the frequency multiplier circuits and hence the RX front-end performance became less sensitive to the LO power level applied. This effect will be shown in the final front-end measurement results. The simulated phase and amplitude imbalance is presented in Fig. 12.

Fig. 11. Schematic diagram of the second version RF switch and active balun IC.

Fig. 12. Simulated ($\phi$) imbalance (left Y -axis, solid) and $\vert A\vert$ imbalance (right Y -axis, triangles), LB (blue) and HB (black) respectively.

Low drop-out regulator

The LDO circuit is employed as a protection helping to keep the RF switch transistors from reaching the collector–emitter break-down voltages $V_{CE{\rm 0}}$ when in the off state. The schematic of the LDO circuit is presented in Fig. 13.

Fig. 13. Simplified schematic of the LDO circuit.

The supply voltage ($V_{supply}$) is reduced by the output resistance of the transistors ($M_{9}$ and $M_{10}$). These are controlled by the two-stage amplifier with a feedback to ensure a constant output voltage ($V_{output}$). Two voltage dividers, consisting of $R_{1}$, $R_{2}$ and $R_{3}$, $R_{4}$, placed at the input of the amplifier are used to set the required output voltage ($V_{output}$) level. A capacitor $C_{Stab}$ of 16 pF is employed to assure the stability of the feedback circuit. Simulation results are presented in Fig. 14.

Fig. 14. Simulated time domain performance of the LDO circuit. Output voltage (solid) versus input voltage (dashed, blue).

It can be seen that after switch on, it takes some time for the output voltage to set to the required 2 V reading. This is because the large stabilizing capacitor ($C_{Stab}$) needs to be charged. The value of $C_{Stab}$ can be decreased, however this would reduce the gain/phase margin and is not a necessary condition for a constant ON operation.

IF switch and IF amplifier

For addressing the 10–100 MHz frequency range, HV-FET devices could be chosen for the design. This gave the advantage of being able to choose a higher supply voltage, ranging from 3 to 5 V. Choosing 5 V supply voltage enables larger output voltage amplitudes without driving the IF amplifier into compression. In such a way larger direct path signals can be handled by choosing an ADC with corresponding maximum peak-to-peak input voltage. In this way the dynamic range of the Rx itself is improved. Due to the single-ended input of the ADC, the output of the IF switch and IF amplifier circuit was chosen to be single-ended. The chosen topology of the IF switch is presented in Fig. 15. The IF switch consists of two differential pair circuits, one for the LB signal (marked on the left side of the schematic) and one for the HB (marked on the right side respectively) and the inverters for selecting between the LB, or the HB channels. In order to enhance the gain compression, a common gate input configuration is employed. The switching between the LB and HB is implemented by controlling the bias of the input transistors. To maximize the high- and low-switching voltage amplitudes, two inverter pairs are employed. In case the LB signal is needed, a logic 1 (3–5 V) is applied to the LB/HB input, in this way activating the LB differential pair and fully shutting off the HB differential pair. The chosen circuit topology is presented in Fig. 16. A two-stage topology with a resistive feedback was employed to enable a flat gain response. A differential configuration was chosen, where one side of the differential pair is connected to the input and the other is used for the feedback. The resistive feedback is employed to make the gain of the circuit more stable against temperature variations.

Fig. 15. Simplified schematic view of the designed IF switch circuit. The LB part is circled left and the HB circled right.

Fig. 16. Simplified schematic view of the designed IF amplifier circuit.

Rx front-end IC

The designed components were combined into a single RX IC unit. The fabricated IC photograph with denotations is shown in Fig. 17. The HB input is seen at the bottom left, whereas the LB input is placed in the middle right. Both, LB and HB input GSGSG pads are placed on the bottom side with some distance apart, this is for the reason to have sufficient spacing between the on-board differential TLs. On the HB side, the first component is the variable gain LNA followed by the mixer component. The down-converted signal is routed to the output of the IC to the off-chip capacitors for high-pass filtering. The same approach is seen on the LB side, with one difference that the first component here is the active power limiter. The LO input is located on the top side of the IC in order to separate the on-board LO TL from the LB and HB inputs. The 4–20 GHz LO signal is passed to RF switch, and commuted either to the LB or the HB channel. In case the HB channel is active, the LO signal is passed to a frequency doubler to up-convert the frequency from 5–20 GHz to 10–40 GHz. The following fully differential amplification stage is responsible for boosting the signal amplitude above 0 dBm and to reduce the amplitude and phase imbalances. The same approach is seen on the LB channel, with the difference that no frequency multiplication is required. On the right side of the IC, the IF switch and IF amplifier circuits are seen. With one input being activated at a time, it chooses whether the LB, or the HB signal is passed through, or attenuated and amplifies it to the required power level to drive the ADC input. On the top right side, the IF output pads are seen and on the right bottom side the control and bias voltage inputs are located.

Fig. 17. Microscopic view of the designed RX IC with denotation of separate components.