Chapter 3 Concurrent Dual-Band Receiver Front-End
3.3 Design of Concurrent Dual-Band Receiver Front-End
In this section a new concurrent dual-band receiver using only one frequency synthesizer with tuning range of around 2.4 GHz for WLAN applications is introduced first. Figure 3.3.1 shows the concurrent dual-band receiver block diagram which has been proposed in [4]. It provides a RF concurrent dual-band receiver solution for IEEE 802.11a/b/g. The receiver consists of a differential concurrent dual-band LNA, a sub-harmonic mixer for 2.45GHz, a Gilbert-cell mixer modified from sub-harmonic topology for 5.25GHz, a quadrature voltage-controlled oscillator (VCO) and a multi-modulus frequency synthesizer. Appling such mixer operating at 2.45 GHz or 5.25 GHz with the same architecture can reduce the design complexity significantly. On-chip IF Gm-C filters are used for noise bandwidth limiting and anti-aliasing reasons. The concurrent dual-band receiver front-end is designed for this receiver block diagram as the marked area in Figure 3.3.1, which is designed and implemented cooperatively by the author and the other one [21].
Based on the comparisons of differential receiver architectures in last section, we choose low-IF receiver architecture in this work because of high degree of integration.
The IF frequency is chosen at 10MHz because of the noise and receiver architecture considerations. The receiver frequency plan is shown in Figure 3.3.2. It can be seen that the tow LO frequencies are very close because of the usage of sub-harmonic mixer. Hence one frequency synthesizer is enough to provide the tuning range of LO signals around 2.4GHz. Compared with traditional topology with two Gilbert-cell mixers two frequency synthesizers may be needed owning to large frequency difference of two LO signals for two bands.
Dual-Band
Figure 3.3.1 New concurrent dual-band receiver and concurrent dual-band front-end
Figure 3.3.2 Receiver frequency plan (a) 2.45GHz (b) 5.25GHz
The architecture of concurrent dual-band LNA for receiver front-end is shown in Figure 3.3.3. It has a similar architecture as the one discussed in last chapter except the spiral inductor Ls is replaced by the bondwire inductor Lbondwire1. Two other bondwires are needed in the RF input pad and power supply pad, so the input matching network and output matching network must be redesigned by considering the effect of the parasitic inductor from the bondwire. The inductance of bond wire is predicted as 1nH per 1mm length. Figure 3.3.4 illustrates the comparison of operation principles of conversional mixer and sub-harmonic mixer. The role of switching transistor (Qs) is evenly distributed to two parallel-connected transistors (Qs1, Qs2) in sub-harmonic mixer, thus it needs only half LO frequency compared to conversional mixer. Figure 3.3.5 and 3.3.6 show the topologies of the two mixers for receiver front-end. The design details about sub-harmonic mixer and Gilbert-cell mixer modified from sub-harmonic mixer can be found in [21].
The challenge of integrating LNA and mixers comes from the inter-stage design. In the design procedure we try to match the output matching of differential dual-band LNA and RF input matching of two mixers to the same impedance, for instance, 500 ohms parallel with 100pF, rather 50ohms. Large coupling capacitors are added between LNA and mixers for RF signal coupling and dc isolation. Some other circuits, like quadrature balun, balun, LO port matching network, and IF low-pass-filter, are implemented on PCB with lumped elements. The chip layout occupies area of 1.45mm x 1.45mm, and is shown in Figure 3.3.7. Figure 3.3.8 is the chip photograph of this work.
Vdd=1.8V
Figure 3.3.3 Concurrent dual-band LNA for receiver front-end
LNA
Figure 3.3.4 Basic concept of (a) conversional mixer (b) sub-harmonic mixer
M1 M2
Figure 3.3.5 Gilbert-cell mixer for 2.45GHz front-end
M1 M2
Lop_5g Lon_5g
Lop_2g Lon_2g Figure 3.3.7 Chip layout of concurrent dual-band front-end
Figure 3.3.8 Chip photo of concurrent dual-band front-end RFp
RFn
IFn_2g IFp_2g IFp_5g IFn_5g LOpj_5g
LOnj_5g
3.4 Experimental Results and Discussions
The concurrent dual-band receiver front-end is measured by two PCB boards, 2.45GHz and 5.25GHz, rather one PCB board, because of large size off-chip passive baluns, too many on-board decoupling capacitors, and complicated dc bias routing for circuits. As shown in section 3.3, a balun for 2.44GHz and a quadrature balun for 2.62GHz are needed to provide differential and quadrature LO signals, which are shown in Figure 3.4.1 and 3.4.2, respectively.
Figure 3.4.1 Balun for 2.44GHz
The measured transmission coefficients of the 2.44GHz rat-race is
[ ]
0.029 117.4 0.678 135.56 0.689 135.4 0.016 94.2 0.678 135.14 0.045 0.2 0.024 161.7 0.675 42.5 0.688 134.95 0.023 162.5 0.029 34.6 0.677 137.62 0.016 93.8 0.675 42.1 0.678 138.15 0.057
rat race
when all other ports are terminated with matched loads. The measured transmission coefficients of the quadrature balun composed of two rat-races and quadrature hybrid from port1 to port 2-port 5 are
21 0.444 125.75
S = ∠ − ° ; S31 =0.442 54.82∠ ° ;
41 0.452 142.04
S = ∠ ° ; S51 =0.452∠ −35.27° Selecting port 3 as phase reference we have phase relationship as
Port 2：179.43∘ ; Port 3：0∘ ; Port 4：87.22∘ ; Port5：269.91∘
The characteristic of quadrature balun satisfies the requirement for the LO port of sub-harmonic mixer though there are small phase and magnitude errors.
PCB layouts and practical FR4 PCB circuits with SMA connectors are shown in Figure 3.4.3 and 3.4.4. There are some comments on PCB boards. Firstly the width of RF and LO signal paths on PCB are drawn as 50 ohms-line for impedance matching.
Lumped Coupling capacitors (1uF) are placed in the RF paths for dc isolation. To filter out the ineluctable noise and spur from the power supplies we add four lumped decoupling capacitors (100pF, 10nF, 100nF, and 1uF) between each dc voltage and ground. IF low-pass-filters composed of lumped capacitors and resistors are placed at the IF outputs to depress the high frequency noise. The signal lines for differential or quadrature signals should be symmetric to avoid the phase error caused by the PCB transmission lines.
The block diagram of PCB on board testing for dual-band receiver front-end is
shown in Figure 3.4.5. LO port has two paths for 2.45GHz front-end because of differential balun and four paths for 5.25GHz front-end because of quadrature balun.
Two RF baluns are needed in the measurement, one for 2.45GHz and the other for 5.25GHz, to convert the RF signal from single to differential. The oscilloscope will be connected to the IF port to measure the output waveform because of 1M high input impedance.
LO IN
RF IN IF OUT
(a)
LOi IN RF IN
IF OUT
LOj IN
(b)
Coupling Capacitors Decoupling Capacitors IF LPF Figure 3.4.3 PCB layout for (a)2.45GHz (b) 5.25GHz front-end
(a) (b)
Figure 3.4.4 Photograph of PCB board for (a)2.45GHz (b)5.25GHz front-end
Figure 3.4.5 Block diagram of PCB on-board testing for dual-band front-end
Table 3.4.1 summaries the performance of this work, including simulation and measurement results. The concurrent dual-band receiver front-end was fabricated using 0.18um CMOS 1P6M process. The RF input return loss of LNA are -15.9 dB and -15.8 dB at 2.45GHz and 5.25GHz, as shown in Figure 3.4.6. The LO port input return loss of two mixers are -13.4 dB and -13.1 dB, as shown in Figure 3.4.7 and 3.4.8. Figure 3.4.9 ~ Figure 3.4.12 show the measured linearity of the front-end characterized by the overall RF-to-IF -21.0 dBm and -15.3 dBm P1dB and the overall RF-to-IF -4.2 dBm and 4.9 dBm IIP3 for RF signals in two frequency bands. It demonstrates 17.2 dB and 11.8 dB voltage gain, 7.22 dB and 10.78 dB noise figure concurrently at two frequency bands with 28.8mw power dissipation. Finally the 10MHz output waveforms measured by oscilloscope are shown in Figure 3.4.13.
Here are some discussions about the experimental results. The good RF input return loss may be owing to the accurate prediction of bondwire inductance and on-chip
circular spiral inductors which were designed, measured, and modeled by our group, rather foundry. The good LO input return loss comes from the accurate LO matching network composed of lumped inductors and capacitors. It may take great efforts to tune the matching network from the finite lumped element libraries. Although this work has good port input return loss, the performance of gain and noise figure does not meet our anticipation. There are three major factors. First, the inter-stage design may be interfered by the parasitic capacitors and resistors, causing the impedance mismatch between the output of differential dual-band LNA and RF input of mixers.
Second, the quality factor Q values of the inductors are not good enough due to parasitic resistances. The Q-values of these inductors involved in this work is from 7.08 to 8.27. The gain and output matching of the concurrent dual-band LNA will be seriously affected by the poor Q-value of inductors. Finally the absence of output buffers at IF output impacts the driving capability of the front-end. These factors may depress the gain and increase noise figure of the concurrent dual-band receiver front-end.
Table 3.4.1 Performance summary of dual-band receiver front-end 2.45GHz Front-End 5.25GHz Front-End
Sim. Mea. Sim. Mea.
LO Power (dBm) -3 8 -3 7
RF Return Loss (dB) -18.4 -15.9 -13.4 -15.8 LO Return Loss (dB) -13.2 -13.4 -18.3 -13.1 Conversion Gain (dB) 14.7 6.0 2.57 -12.0 Voltage Gain (dB) 26.5 17.2 19.9 11.8 Noise Figure (dB) 3.77 7.22 7.28 10.78
P1dB (dBm) -20.6 -21.0 -22.1 -15.3
IIP3 (dBm) -7.8 -4.2 -4.5 4.9
Power (mw) 17.9 28.8 17.9 28.8
0 1G 2G 3G 4G 5G 6G 7G -25
-20 -15 -10 -5 0
RF P ort Re turn Los s (dB)
Frequency (Hz)
Simulation Measurement
Figure 3.4.6 Comparison between simulation and measurement RF input return loss
2.0G 2.2G 2.4G 2.6G 2.8G 3.0G
-20 -15 -10 -5 0 5
LO Port Ret urn Loss (dB)
Frequency (Hz)
Simulation Measurement
Figure 3.4.7 Comparison between simulation and measurement LO input return loss of 2.45GHz Gilbert-cell mixer
2.0G 2.2G 2.4G 2.6G 2.8G 3.0G
L O P ort I nput Retur n Los s ( dB)
Frequency (Hz)
Simulation Measurement
Figure 3.4.8 Comparison between simulation and measurement LO input return loss of 5.25GHz sub-harmonic mixer
-40 -35 -30 -25 -20 -15 -10
Conversion Power Gain (dB)
Input RF Power (dBm)
Simulation Measurement
Figure 3.4.9 Comparison between simulation and measurement of P
-40 -35 -30 -25 -20 -15 -10
Conversion Power Gain (dB)
Input RF Power (dBm)
Simulation Measurement
Figure 3.4.10 Comparison between simulation and measurement of P1dB
of 5.25GHz front-end
-30 -25 -20 -15 -10 -5 0
IF Ou tp ut Po wer (d Bm)
RF Input Power (dBm)
Sim_Fundamental Sim_IM3
Mea_Fundamental Mea_IM3
Figure 3.4.11 Comparison between simulation and measurement of IIP3 for 2.45GHz front-end
-30 -25 -20 -15 -10 -5 0 5
IF Output Pow er (dBm)
RF Input Power (dBm)
Sim_Fundamental Sim_IM3
Mea_Fundamental Mea_IM3
Figure 3.4.12 Comparison between simulation and measurement of IIP3 for 5.25GHz front-end
VIFp = Figure 3.4.13 Output waveform of (a) 2.45GHz (b) 5.25GHz front-end
3.5 Comparisons
Table 3.5.1 shows the comparisons of this work and other recently dual-band receiver front-end papers. Compared with other dual-band front-end this work achieves comparable performances with nearly equal chip area and lower power dissipation under concurrent operation for two frequency bands.
Table 3.5.1 Comparisons of dual-band receiver front-end Ref [2] 2004 [22] 2004 [23] 2005 This Work
Condition Mea. Mea. Mea. Sim. Mea.
Architecture
*：IF mixer is included
Chapter 4
Low-Voltage Micromixer
4.1 Review of Basic Micromixer
The down-conversion mixer is a key building block in a receiver system. Its main function is to translate the incoming RF signal to an intermediate frequency for further processing. It dominates the system linearity and determines the performance requirements of its adjacent blocks. Among many proposed active mixers the Gilbert-cell mixer has been widely used because of it’s LO suppression at the IF output. However the circuit linearity is limited by MOSFET transistor linearity, which is the common source MOSFET transconductance [24]. The small-signal linearity of the input stage, and thus the third-order intercept point, can be greatly improved using several techniques, notably, source degeneration, the multi-tanh doublet and triplet.
However the 1-dB gain compression point still falls short of what may be required in handling large input signals without significant intermodulation. Further these RF stages do not provide an accurate match to the source [25]. Therefore the micromixer was proposed in [25] to overcome these problems. The topology of the basic micromixer is shown in Figure 4.1.1.
The micromixer follows the general form of Gilbert-cell mixer except for the use of a bisymmetric class-AB RF stage based on the translinear principles while the mixer core is identical to the Gilbert-cell mixer. The class-AB RF stage provides
Figure 4.1.1 Basic micromixer
Although the micromixer does not have inherent gain compression in RF stage, the 1-dB compression point of the micromixer will often be determined by limitations on the output IF signal amplitude, rather than by the RF stage. The noise figure of the micromixer depends on design details and is acceptable for many receiver applications although it is generally not as low as in mixers specially optimized for noise performance.
In Figure 4.1.1, Q1 can be viewed as a grounded-base stage. It delivers its output I1
to the mixer pair QM1-QM2 in phase. It can, in principle, handle unlimited amounts of current during large negative excursion of VGEN. On the other hand, the current mirror sub-cell Q2-Q3 can handle essentially unlimited amounts of current during positive excursion of VGEN both at its input node and at its inverted-phase current output I , which drives QM3-QM4. Acting together, these two sub-cells provide an
overall transfer characteristic which is symmetric to both positive and negative inputs, and which is in principle not limited by the choice of bias level. The differential current output I1-I3 is linear with IRF, although the individual currents are quite nonlinear. [25]
Because of the advantage of easily matching and wide dynamic range the micromixer is also applied to the CMOS process in recently years [26]. Replacing the BJT with MOSFET, we can derive two simple expressions for low-frequency small-signal input resistance and voltage gain under the assumption of ideal transistors and neglecting parasitic effects for simplifying [27]. The low frequency small-signal input resistance of RF input stage is approximately
( )
which implies the micromixer RF input stage can be matched to 50Ω as long as we choose proper bias current. Assume perfect impedance matching to 50Ω, the low frequency small-signal voltage gain is approximately
L
These two equations will be very helpful when designing the micromixer.
4.2 Low-Voltage Micromixer
In recent years low-voltage circuit design has become an important issue because of the consideration of battery design and power reduction. However the traditional micromixer is inapplicable for the low voltage design due to the stack of the four stage cascode architecture. Here we propose a modified micromixer applicable for
The main improvement of the low-voltage micromixer is the RF stage, while the switch-stage of the low-voltage micromixer is identical to the basic micromixer. The RF signal is feed in between R1 and M2, and coupled to the RF stage by CcRF1 and CcRF2. We bias the transistors M1 and M2 separately using Vg1 and Vg2. The improved RF stage overcomes the bias-relative problem and retains the characteristic of class
Figure 4.2.1 Low-voltage micromixer CcLOn
CcLOp
LLOn
LLOp Rbn
Rbp
CLOn2
CLOp2 CLOn1
CLOp1 LOn
LOp
VbLO VbLO
LOng
LOpg
Figure 4.2.2 LO matching network
AB stage in the basic micromixer. The pi-matching network is added at the LO port for the narrow band input matching to 50 ohms for measurement consideration.
Figure 4.2.2 shows the topology of the LO port on-chip pi-matching network composed of two MIM capacitors and one spiral inductor. The LO stage bias voltage is feed with bias resistors in the matching network. To keep the output IF waveform symmetric the two resistors R1 and R2 in the RF stage adjust the transconductance and current balance of M1 and M2.
In the low-voltage micromixer we adopt the charge injection method to improve the gain [28]. According the relationship of transconductance and IP3 with current in the traditional mixer architecture
2
which imply the mixer gain and IP3 are proportional to the bias current flowing in the input MOSFETs, ISS . Because the micomixer has identical operational model as the Gilbert-cell mixer, the two equations are also applicable to the micromixer. The charge injection method can improve the micromixer gain and linearity, compensating the disadvantage of low supply voltage and low transconductance in CMOS process.
In Figure 4.2.1, M7 and M8 work as current sources, and provide extra charge current feeding into the RF stage. R7 and R8 provide high impedance to prevent the small signal from going to the current sources so that the charge injection stage will not interfere with the function of low-voltage micromixer.
4.3 Layout and Measurement Considerations
The circuits elements of low-voltage micromixer are all on-chip except for the IF port low pass filters, so we choose the PCB (printed circuit board) on-board testing for the micromixer. The layout of low-voltage micromixer is shown in Figure 4.3.1 and chip photo is shown in Figure 4.3.2. The circuit occupies chip area of 1mm x 0.85mm. Figure 4.3.3 shows the on-board testing PCB layout. The photograph of the realized PCB with chip is shown in Figure 4.3.4.
The circuit ground and substrate are separated in the layout and the bondwire works as RF choke to prevent the circuit from the noisy substrate. In the design process the parasitic effects of bondwires and bond-pads have been taken into consideration.
Typically, the inductance of bond wire is about 1nH per 1mm length and the parasitic capacitance of a 100umx100um bond-pad is approximate 150fF to the ground. We also consider the process variation by the TT, FF and SS corner simulations with libraries provided by the foundry.
Two extra circuits are needed in the measurement of low-voltage micromixer.
First the LO ports use a differential 2.44GHz signal so we need a balun suitable for 2.44GHz to convert the signal generator output to differential form. Secondly to filter out the high frequency noise in the 10MHz output waveform, the IF low pass filters composed with lumped resistors and capacitors are made on board at the IF output pads. The simplified block diagram of PCB on-board testing is shown in Figure 4.3.5. We can follow the simplified block diagram to measure the RF and LO port input return loss, conversion gain, third-order intercept point, and noise figure of the low-voltage micromixer. It should be noted that the losses of cable, balun, SMA connectors, and PCB board itself must be taken account for calibration and calculation in measurement results.
Vss IFp Vdd IFn bulk
Vg1 RF Vg3 Figure 4.3.1 Chip layout of low-voltage micromixer
LOp LOn
Vblo Vblo
Coupling Capacitors Decoupling Capacitors IF LPF Figure 4.3.3 PCB layout for low-voltage micromixer
Figure 4.3.4 Photograph of PCB for low-voltage micromixer
Spectrum
Figure 4.3.5 Simplified block diagram of PCB on-board testing for micromixer
4.4 Experimental Results and Discussions
The low-voltage micromixer was simulated and fabricated using CMOS 0.18um process. The measurement results shows that it has 14.9 dB RF port return loss, 8.28 dB conversion voltage gain, -5.63 dBm P1dB, and 4.21 dBm IIP3. The total power dissipation of the low-voltage micromixer is 1.72mw from 1V voltage supply. Figure 4.4.1 shows the RF port input return loss is better than 10 dB between 2.1GHz and 4.2GHz, which proves the well-defined input impedance of micromixer topology.
Figure 4.4.2 shows the measured optimum LO power is 0 dBm while the simulated one is -5 dBm for the maximum conversion voltage gain. The measured conversion voltage gain is a little bit less than simulation, which may be caused by less power
compression point and third-order intercept point. Figure 4.4.3 shows the 1dB compression point and Figure 4.4.4 shows the third-order intercept point. In summary the measurement results are very close to simulations. The low-voltage micromixer has good RF port matching, high conversion gain, high linearity, and very low power consumption under 1V low power supply. The differential 10MHz IF output waveforms are shown in Figure 4.4.5. Table 4.4.1 summaries the simulation and measurement performance of the low-voltage micromixer.
compression point and third-order intercept point. Figure 4.4.3 shows the 1dB compression point and Figure 4.4.4 shows the third-order intercept point. In summary the measurement results are very close to simulations. The low-voltage micromixer has good RF port matching, high conversion gain, high linearity, and very low power consumption under 1V low power supply. The differential 10MHz IF output waveforms are shown in Figure 4.4.5. Table 4.4.1 summaries the simulation and measurement performance of the low-voltage micromixer.