Chapter 1 INTRODUCTION
1.2 Thesis Organization
Because this ultra-wideband mixer is designed for low power, the topology of mixer is very important. The double balanced mixer is the most common topology. It can reject the feedthrough both from RF and LO to IF port. However the single and differential ended LO inputs of double balance mixer have different characteristics. Each topology has advantages and drawbacks itself. Which topology is better for the low power design described in section 2.3.1.
After choosing the topology, the first thing is to design every part of the topology.
ultra-wideband mixer comprises seven parts, including current mirror, current injection, transconductance stage, switching stage, load, and RF and LO input matching network. The basic principle of mixer and the methods to design for ultra-wideband mixer are shown from section 2.3.2 to 2.3.7.
From pre-simulation to post-simulation, there is a very important problem. Because the operation frequency band is from 3.1 to 15GHz, the parasitic effect is obvious. Therefore the performances of the pre-simulation won’t be matched for ones in post-simulation. The topology will determine the routes of RF signal, therefore the parasitic effect can be predicted. The skill of layout is very important to reduce the effect of parasitic. It is shown the layout to explain how the parasitic effect affects performance in section 2.3.8.
In section 2.4, the measurement results and comparing the performances among pre-simulation, post-simulation and other papers are discussed.
In section 2.5, the IF frequency response and the phase and amplitude of signal at some nodes in the circuit are discussed.
Finally, we make the conclusion and present the future prospects in Chapter 3.
Chapter 2
L OW P OWER U LTRA- W IDEBAND M IXER F OR 3.1GHz ~ 15GHz
2.1
Introduction
A low power RF device becomes a tendency as applied in the portable wireless communication systems. However, the performance including linearity and conversion gain will be degraded when we reduce the power or the supply voltage of RF mixer circuit.
Hence, the implementation of the mixer with low power consumption, high linearity and high conversion gain would be a challenge in the RF front-end circuit. Figure 2-1 shows the architecture of MB-OFDM receiver [7]. The mixer of this receiver will be introduced in this thesis. Because the UWB is emphasized low power for short range wireless communication, how to reduce the power consumption, cost of circuit and have better performances are very important. Therefore the low power UWB mixer is designed shown in Figure 2-2.
Figure 2-1 Architecture of MBOA Receiver
Figure 2-2 Schematic of the Ultra-Wideband Mixer
Although there are many topologies of mixer designed for UWB. When the mixer is applied for low power and low voltage, the circuit will be influenced by DC Offset. The DC Offset comes from LO self-mixing. The problem can be solved by the double balanced mixer with differential-ended LO inputs. Because the isolation of LO-to-RF can be increased by differential-ended LO ports. And the details will be introduced in section 2.3.
2.2
Architecture
The architecture of the ultra-wideband mixer is shown in Figure 2-2. In order to have better port-to-port isolation, the architecture of double-balanced mixer is best choice. M3, M4 are transconductance stage in Figure 2-2, their function is to transform voltage to current. And passive components, L1, L2, L3, R1, R2, R3, C1, C2 and C3, compose three
matching networks of L form. Making the RF and LO ports be meted to 50Ω for measurement. The inductor adopts the model of TSMC 0.18um process. The gate of M3 is connected to ground for ac by connecting a capacitor to ground. Using transconductance stage translates RF voltage signals into two current signals in different direction. Then the architecture just needs single ended RF input, the numbers of RF matching network can be reduced to one. Switching stage is composed of M5 ~ M8. When LO is positive period, M5 and M8 turn on. When LO is negative period, M6 and M7 turn on. M9 and M10 are current injection circuit, it can reduce the current of Switch MOS (M5 ~ M8). Therefore Rloadn and Rloadp can be larger to increase conversion voltage gain. Buffer circuits are added, M11 and M12, to make mixer can provide larger current for driving the next stage. M1 and M2 consist of current mirror to provide a large output resistance and bias current. The small signal can be divided between the sources of M3 and M4 to generate two signal current with phase difference in 180 degrees.
2.3
Analysis of Ultra-Wideband Mixer
This section introduces the principles of design and the performances of ultra-wideband mixer including RF and LO matching network, P1dB, IIP3, Conversion Gain and Noise Figure. There are two circuits of mixer introduced in section 2.3.1, they are mixer with single-ended LO input and differential-ended LO inputs respectively. The advantages and drawbacks of each topology will be described in detail. We can determine which the topology is better for Ultra-Wideband Mixer from these advantages and drawbacks.
2.3.1
The suitable topology for Ultra-Wideband Mixer
Figure 2-3 Schematic of Mixer with Single-ended RF and LO Port
There are two topologies tried to implement low power ultra-wideband mixer in this thesis. One of them is shown in Figure 2-2, the LO signal is differential ended. The other is shown in Figure 2-3, the LO signal is single ended. The mixer with single-ended LO port in Figure 2-3 is discussed first. There are some advantages and drawbacks for this topology. In advantages, there is only one LO and RF input. Therefore the RF and LO ports just need two matching networks. It can reduce the chip size and the complexity of layout. However there is a serious problem. Because the LO signal is single ended, the LO signal will feedthrough to RF terminal. Then it will be conversed down to DC by LO switching stage so that DC offset appears at two branches. It not only affects the DC level of output, but also the bias condition of M4. There are some pictures shown in Figure 2-4 to Figure 2-6, these phenomenons can be observed apparently. These figures show that the higher frequency the higher DC offset. The RF return loss is also influenced shown in Figure 2-7.
Figure 2-4 IF Output Signal RF@ 3.1GHz
Figure 2-5 IF Output Signal RF@ 7GHz
Figure 2-6 IF Output Signal RF@ 10.6GHz
Figure 2-7 RF Return Loss
The mixer with differential-ended LO ports shown in Figure 2-2 is suitable for low power ultra-wideband mixer. It can avoid the two drawbacks of the mixer with single-ended LO port shown in Figure 2-4 to Figure 2-7. The differential-ended LO signals will reduce the feedthrough effect because LO signals will be canceled each other at RF port. It means that the LO-to-RF isolation will be improved. The DC offset and the bias voltage of M4 can be improved. And Figure 2-8 to Figure 2-11 shows the improvement after changing the topology from single-ended LO input to differential-ended LO inputs. We can see that not only the DC Offset but also the RF return loss are improved.
Figure 2-8 IF Output Signal RF@ 3.1GHz
Figure 2-9 IF Output Signal RF@ 7GHz
Figure 2-10 IF Output Signal RF@ 10.6GHz
Figure 2-11 RF Return Loss
2.3.2 The RF and LO matching networks
(a) (b)
Figure 2-12 Matching Network (a) L-Shape (b) L-Shape with Resistance
Matching network has many kinds like L-shape, Chebyshev polynomial, etc. L shape is better for this design. Because the mixer needs three matching networks for RF and LO terminals. In order to decrease the complexity and chip size, hence it must be simple structure. The L-shape matching network is the best choice like Figure 2-12 (a). However it is suitable for narrowband. The network can be changed into the topology like Figure 2-12 (b). The difference between (a) and (b) in Figure 2-12 is a resistance. This resistance can decrease the Q factor of LC matching network to achieve wideband matching. When the resistance increases, the band of matching will be wider. However the return loss will get worse. Then it must be traded off between these. From the Eq. (2.3), Zin can be designed that the RF Return Loss is better than 10 dB. However it is very hard to match the demands all the bandwidth from 3.1 to 15 GHz. The matlab software can be used to calculate the complex equations. It will facilitate the work.
SC SL
2.3.3
Conversion Gain
According the relationship of transconductance in traditional mixer architecture, which implies that the voltage gain will be portion to gm and RL. Therefore the way to increase the voltage gain is to raise bias current or load. However the gm and load are impeded by each other. Adding the current injection can resolve this problem [8]. Because the current provided by current injection does not go through the load to generate additional voltage drop, the gm and RL can be increased at the same time.
2.3.4
Effects of Nonlinearity
In older to simplify the analysis of nonlinearity, consider a memoryless and time-variant systems. Then the transfer function of transconductance stage can be expressed as the equation of (2.5).
In Eq. (2.6), the second term with input frequency is called the ‘‘fundamental’’ and the higher-order terms are called ‘‘harmonics’’. The parameter,α1, α2 and α3, can be determined by the bias current. From the relation of the Vgs and bias current at transconductance stage, the values of these parameters can be got [9]. It shows that α1 and
α3 have different polarity.
α1=
Figure 2-13 Mixer Linearity
When the power of signal isn’t large enough, the circuit can amplifier the signal linearly. If the signal is large enough, the fundamental term will be affected seriously by the third harmonic terms. The conversion gain of fundamental term will be getting smaller as the input signal is getting larger. When the conversion gain decreases 1 dB, the point of P1dB will be determined by input power shown in Figure 2-13. When the linear term and the third-order intermodulation term cross, the point of IIP3 will be determined by input power shown in Figure 2-13.
In order to increase the linearity, source degeneration is usually used [10]. The nonlinearity can be improved. However the conversion gain will be decreased. It must be increased the power to maintain the conversion gain. Therefore it must trade off between the power, conversion gain and linearity. However, the noise figure will also be affected. The current injection circuit can solve this trade off relation between linearity and noise figure [11].
This is due to the fact that a large biasing current is needed for the RF input stage to achieve high gain, while a fairly low current is required for the LO switching quad to realize better noise performance.
2.3.5
Noise
The noise contribution of the loads, transconductance and switches is presented [12].
More accurate analytic methods have been represented in [13]. The principle of noise figure is discussed in A, B, C and D four parts.
A. Load Noise
Flicker noise in loads of downconversion mixer interfere the signal in zero-IF or low-IF receiver. PMOSFET has lower flicker noise than NMOSFET [14][15]. Using resistors as load in this design, which are free of flicker noise, need expense of voltage headroom.
B. Transconductance Noise
The noise in transconductance stage includes white noise and flicker noise. The white noise and flicker noise must be translated in frequency by switching stage. The flicker noise will shift the frequency ωlo and its odd harmonics, hence it doesn’t appear at IF. The white noise at ω and its odd harmonicsrf is downconverted to IF.
C. Direct Switch Noise
The direct switch noise is that the noise at the gate of switching stage interferes in the switch of switching stage. The switching stage won’t switch at the frequency ωlo. Because
noise can interfere with LO signal at zero-crossing, the switch will advance or retard.
Therefore the noise goes through to IF by this effect. The output superposed with a pulse train of random width Δt and amplitude of 2I at frequency of 2ωlo. Over one period the average value of the output current is
T noise can be reduce to minimum by increasing the amplitude of LO. The reason can be explained from why the noise generates. It is shown in Figure 2-14. From the Figure 2-14, we can know that the higher slope of LO at zero-crossing, the fewer direct switch noise at IF.
Figure 2-14 Noise Source of Direct Switch
D. Indirect Switch Noise
The noise source as shown in Figure 2-15, , applied at the gate of the MOS will generates noise at IF output signal in two ways. First one is described in direct switch noise.
The second way is introduced in indirect switch noise. The noise will charge in the Cp capacitor, hence it appears at IF by the bias current. Considering the square-wave LO, the magnitude of the current is Eq.(2.8) for zero IF.
Vn
n p n
o C V
i T2
, = (2.8)
Figure 2-15 Single-balanced mixer with switch noise modeled at gate.
In general, there are two methods to reduce the Noise Figure: (1) Large LO power can suppress the noise [16] (2) Using current injection circuit to facilitate the switch of switching stage. When designing the Ultra-Wideband Mixer, the noise figure is always a troublesome problem. Considering the ultra-wideband mixer, bandwidth of the matching networks is from 3.1GHz to 15 GHz. The interferences in this bandwidth have more impacts on circuit’s performances than traditional narrowband matching. These interferences will pass through the matching networks and appear at the gates of switching stages and input transconductance stage. The interferences and white noise of transconductance stage will be translated to IF by switching stage. Also, the interferences of switching stage becomes serious by the same mechanism. Therefore the direct and indirect switch noise will get worse than those of narrow band mixer.
2.3.6
Port Isolation
Figure 2-16 LO to RF Isolation
The isolation is important element for the choice of architecture. The reason is explained in section 2.3.1. The isolation of LO-to-RF port can be increased using mixer with differential-ended LO ports. Figure 2-16 shows the isolation of RF-to-LO and LO-to-RF port. Finally, we will talk about the isolation of IF port. Although many high frequency signals will feedthrough to IF port, these effects aren’t important. Because low pass filter is usually designed at the IF port to filter these signals. From the description in section 2.3.1, mixer with differential-ended LO ports indeed have better LO-to-RF isolation than mixer with single-ended LO port.
2.3.7
Design flow
In this section the design flow of ultra-wideband mixer will be discussed in detail.
When designing a mixer, the DC bias point is very important. If the DC bias point is wrong,
the conversion gain of the circuit will be very small. Therefore the first step is to determine the DC bias point. Then the circuit performances can be designed following the analysis from the section 2.3.1 to 2.3.6. The parameters of elements in circuit can be determined.
However there is one thing, it must be noticed. It is the condition of the transistor. The condition of the transistor will determine the stability. The transistor must be designed in the condition of saturation. Some simulations about stability are shown in Table 2 and Table 3.
The design of LO switching stage must be careful. The performance of noise figure and conversion gain will be decreased, if LO switching stage can’t turn on or off completely.
The transistors of this stage are also designed in saturation region. Finally, adding a low-pass filter to filter the noise and buffers at IF port. These are all the design flows. If one of the performances is not satisfied, repeat the design flows until the performances are satisfied.
Table 3 Stability Simulation – (2)
There are many layout guidelines for high frequency RF circuits. The high frequency parasitic effects will influence the performances of design. There are several common rules:
The first one is that the routes of high frequency signals must be as straight as possible.
Because the more parasitic effects are generated at corners, the performances of ultra-wideband mixer will be changed. The second one is that the width of power lines must be wide enough to prevent from being burned. And the third one is that the routes of high frequency signal must be as short as possible. The parasitical capacitor from the route will affect our performance. The topology affects the parasitical capacitor, for example, the single-ended and differential-ended input mixers have one and two RF input respectively.
The differential-ended input mixer needs two matching networks, it means that it has two
inductors. Because inductor has bigger area, the routes of RF signal will be extended. The topology of single RF input mixer is the better choice in this thesis. Finally, the parasitic effect of bond-wires will greatly influence the high frequency impedance matching. On wafer circuit measurement with PCB bias network is the best method. RF and LO signals are provided by 3 pins and 5 pins probe respectively. DC bias and ground are provided by the bond-wires connected to power supply. And the total chip size is 1×1 mm2. The layout is shown in Figure 2-17.
Figure 2-17 The layout of proposed low-power UWB mixer
2.4
Measurement of Ultra-Wideband Mixer
2.4.1
Measurement Consideration
The operation frequency is from 3.1 to 15GHz. The bond wire will destroy the function of RF and LO input matching networks. The method of measurement must be on wafer circuit measurement with PCB bias network. The Die Photograph of UWB Mixer is shown in Figure 2-18. In order to facilitate the measurement, the length of the PCB is increased shown in Figure 2-19 and Figure 2-20. The SMA connector must be outside the plane of probe station, otherwise it can’t make the PCB connected with probe plane tightly. In addition, we need a Balun to generate differential LO signals from 3.1 to 15GHz. CIC provides two kinds of Balun. One covers 4 ~ 8GHz, the other covers 6 ~ 20GHz. The measurement can be completed by using these Baluns. The other consideration is the DC Blocking, because the DC Blocking isn’t added on chip. The measurement methods are shown in Figure 2-21. Figure 2-21 shows the methods to measure conversion gain, P1dB, RF and LO input return loss and two-tone linearity of IIP3.
Figure 2-18 Die Photograph of UWB Mixer
Figure 2-19 PCB
Figure 2-20 Practical PCB test board of UWB Mixer
(a)
(b)
(c)
Figure 2-21 Measurement setup for (a) conversion gain (b) input return loss (c) two-tone IIP3 testing
2.4.2 Measurement Results
The power of the circuit is 23.4mW at post-simulation, when VDD is 1.8 volt. The current of the circuit is 13mA. However the current is measured about 9.695mA. Figure 2-22 shows the environment of measurement including probe station, signal generator, spectrum analyzer and DC Power Supply. The performances of measurement will be shown
The power of the circuit is 23.4mW at post-simulation, when VDD is 1.8 volt. The current of the circuit is 13mA. However the current is measured about 9.695mA. Figure 2-22 shows the environment of measurement including probe station, signal generator, spectrum analyzer and DC Power Supply. The performances of measurement will be shown