For the PDF version of this article, click here.

Click here to read part I of this article

The mixers presented here have been designed to downconvert an 8 MHz bandwidth signal from RF (1800 MHz) to an intermediate frequency (IF) of 40 MHz. The local oscillator (LO) frequency is 1840 MHz. The fabrication process used to make the mixers is a standard 0.8µm SiGe process.

Among active mixers the most common schema is the Gilbert cell. The first mixer is a Gilbert cell that has been optimized to obtain low noise. It is called mixer NF in this article. The second and the third mixers are Gilbert cells designed to improve linearity, using two different methods: emitter degeneration and class AB RF input stage, respectively. The mixer called mixer Ze is a Gilbert cell with emitter degeneration. In the mixer called mixer AB, the class AB has been used as the amplification stage.

Mixer NF is a Gilbert cell designed to obtain low noise and high gain. Its microphotography is presented in Figure 1. This mixer has been designed to obtain the minimum noise figure in this technology for a Gilbert cell.

• Noise figure

  To reduce thermal noise from differential input pair, transistors were chosen with large emitter area and double base contact to reduce r_b contribution. Each transistor includes an emitter area of 240 × 0.8 µm². It has been divided into eight smaller transistors of 30 × 0.8 µm² with double base contact connected in parallel.

  The value of β depends on the technology employed and, in this case, it is 90, high enough to allow us to neglect shot noise contribution from I_b. So to reduce shot noise contribution from I_C, collector current I_C has been selected relatively high at 10.4 mA.

• Gain

  To avoid the influence of stages following the differential pair in the noise figure of the mixer, mixer gain should be high. To increase gain, as I_C has been fixed relatively high at 10.4 mA, collector resistance R_C has been chosen as high as possible: 240 Ω. These values cause a voltage drop of 2.5 V in the collector terminals of transistors in the mixing stage. This drop cannot be higher without switching off the LO switches.

• Linearity

  To minimize the noise figure, emitter degeneration was not used to linearize the standard schema. In this mixer the only way to not reduce linearity was to ensure equality between differential branches through symmetry in layout design and to drive enough LO power to properly switch the transistors of mixer core when making measurements.

This mixer is a Gilbert cell with high linearity obtained through emitter degeneration. It is presented in the microphotography included in Figure 2. This mixer has been designed to increase the linearity obtained with the standard Gilbert cell. The linearization technique used has been emitter degeneration.

• Linearity

  To obtain a better IP3 value, a 2.3 nH inductor has been used as emitter degeneration in the emitter terminals of transistors of the differential pair in the input stage. Moreover, the size of the switches has been chosen as small as possible, taking into account the current density that they have to drive, which is established by the foundry. Each switch is a 32 × 0.8 µm², obtained through the parallel of four 8 × 0.8 µm² smaller transistors with double base contact.

• Noise figure

  Due to emitter degeneration elements, noise figure has been increased because of the inherent gain compression due to emitter degeneration. To reduce the thermal noise of the differential pair, the size of the input transistors is similar to that of the transistors in mixer NF (300 × 0.8 µm² transistors, divided in smaller parts and with double base contacts) collector current value I_C has been chosen high in order to reduce shot noise from I_C (9.5 mA). Recalling that g_i depends directly on I_c, a high collector current reduces noise because of β in 1800 MHz is still high enough to make dominant the 1/g_i noise.

• Gain

  As collector current has been fixed to reduce shot noise, the only value to increase mixer gain is collector resistor. In this case, because the voltage drop must be lower, 2.1 V because of emitter degeneration has been included. The collector resister has been chosen as a 225 Ω resistor.

When the mixer Ze values have been set, the expression is used to calculate the theoretical improvement in linearity:

where σ = I_C Z_E/V_T, and Z_E is the emitter degeneration impedance evaluated at input frequency. By substituting mixer Ze design values, a theoretical improvement of 19.46 dB from mixer NF to mixer Ze iIP3 value and the degradation in noise figure we arrive at:

where σ = I_C R_E/V_T, and R_E is the parasitic resistance associated to the emitter degeneration elements. The parasitic resistance associated to the 2.3 nH inductor is approximately 5 Ω. By substituting these values, the mixer Ze noise figure should be 4.62 dB higher than mixer NF noise figure.

The microphotography of Gilbert cell with class AB input stage is included in Figure 3.

In mixer AB, the two linearization techniques were combined to achieve the highest improvement in linearity: class AB input stage and emitter degeneration.

• Linearity

  Class AB linearization input stage is composed of three transistors in class AB configuration. In emitter degeneration, 100 Ω resistors have been used to increase linearity and to match input impedance to 50 Ω. However, because of this high value, the collector current must be low because of the voltage drop in emitter degeneration. It has been set to 2 mA. Transistors in switching stage are small transistors (20 µm × 0.8 µm2) to make switching faster.

• Gain

  Because of the small current value, conversion gain of the mixer is not high. To increase it, the resistance of the collector resistor has been set to 650 Ω, which results in a 1.3 V voltage drop.

• Noise figure

  Because of the small current value, the shot noise of I_C has greater importance, where a shot noise of I_C includes the 1/g_i elements. In addition, as the gain is low, noise in the next stages becomes more influential.

Some design considerations are applied to all the presented mixers. The mixer and output stage current sources were designed with complementary metal oxide semiconductor (CMOS) transistors because their lower gm provides more stability to supplied current. Furthermore, MOS current sources exhibit three times less noise than their bipolar counterpart. Common centroid techniques have been applied to minimize the influence of dispersion in transistors, resistors and capacitors from the two branches that formed differential schemas. In addition, dummy structures have been included in resistors and capacitors.

Die area has been filled with capacitors connected between voltage supply and ground, to filter any undesired signal from voltage supply that could modify the circuit performance. The elements of mixer (transistors, capacitors, resistors and inductors) have been surrounded with many substrate contacts to ensure proper performance.

The three mixers have been fabricated using a standard 0.8µm-SiGe process. To measure gain, isolation and linearity, the E4407B spectrum analyzer was used. The noise figure measurements were obtained with the HP8970B noise figure meter. All those measurements were carried out “on wafer.” Figure 4 shows the setup used to measure the gain of mixer Ze.

Measurement results are presented in Table 1. All measurements were taken in the same conditions for all three mixers.

The performance of different mixers is summarized in Table 1. The results in terms of absolute values are not important in our case as we focus on what might be achieved using various design approaches, disregarding what might be gained using another fabrication process capable of manufacturing smaller transistors. The important point is the relative difference among all three designs that is not influenced by the technology used to fabricate them.

The first mixer, mixer NF, is a standard Gilbert cell designed in SiGe 0.8µ-process to obtain the minimum possible noise figure in this particular process. This mixer has been chosen as a reference for the other designed mixers. It is a 13.5 dB gain and 9.7 dB of single sideband (SSB) NF. However, its linearity cannot go beyond -5 dBm because of the theoretical limit of differential pairs.

Comparing mixer NF and mixer Ze reveals that if emitter degeneration is used in Gilbert cells, a great improvement of 17.4 dB in linearity can be obtained. However, mixer Ze gain is 10.7 dB lower than the values obtained with the classic Gilbert cell, mixer NF, and its noise figure is 4.5 dB higher. In this point, it is necessary to compare these differences in linearity and noise figure between mixer NF and mixer Ze with the predicted improvement in linearity and degradation in noise figure. The improvement in linearity, 17.4 dB, is slightly lower than the predicted value (19.46 dB) and the noise figure degradation, (4.5 dB) has been correctly calculated.

Comparing mixer NF and mixer AB shows that if a class AB stage is used as amplification stage then the linearity can be improved by 12.2 dB. In this case the gain of the mixer AB is only 3.8 dB lower than the gain of the mixer NF and its noise figure is 5.3 dB higher.

Both linearizing techniques are compared through the parameters of mixer Ze and mixer AB included in Table 1. Mixer with emitter degeneration (mixer Ze) achieves an iIP3 5.2 dB higher than the iIP3 of mixer AB. However, mixer Ze gain is 6.9 dB lower than mixer AB. The difference in NF is only 0.7 dB. Both linearizing techniques imply trade-offs between linearity and gain but NF is equal in both mixers. When a linear mixer is needed, the best choice is emitter degeneration, mainly if RF allows us to use inductive degeneration. When linearity and gain are important, the mixer with class AB as input stage is the best option.

Three Gilbert cells have been designed in a SiGe 0.8 µm process. The first of them has been optimized to obtain high gain and low noise figure. The others have been designed to achieve high linearity with different linearizing techniques.

The results obtained in measurements confirm the usefulness of these design guidelines. With the measurements included in Table 1, a comparison between mixer topologies can be obtained.

The standard Gilbert cell, which serves as a reference for the other designs, had 13.5 dB gain and 9.7 dB of NF. The classic Gilbert cell is the best option to obtain active mixers with high gain and the lower noise allowed by fabrication process.

If emitter degeneration is used in Gilbert cells, an improvement of 17.4 dB of linearity can be obtained. However, because of the trade-off between linearity and gain, the later is substantially reduced by 10.7 dB. Gilbert cell with emitter degeneration is the best choice if very linear mixers with low gain are required and when the noise figure is not important. To make the noise figure as small as possible, inductors are the best choice for the degenerative elements, if the input frequency is high enough.

If a class AB stage is used as an amplification stage, then the linearity can be improved by 12.2 dB. Although this increase is 5.2 dB lower than the improvement with emitter degeneration, the gain in mixers including class AB stage is 6.9 dB higher. The class AB stage replacing the input RF stage in a standard Gilbert cell is the best alternative to linearize an active mixer without reducing the gain too much.

In previous works, only one mixer designed was presented and its performance was compared with that of other authors. However, its particular performance depends not only on the mixer topology but also on the fabrication process employed. So it is impossible to compare them reliably when the fabrication processes are significantly different, for example, the doping density of the base in the transistors or the maximum number of base-emitter contacts. This undesired influence has been avoided by comparing mixers integrated with the same fabrication process to find out the improvement range for mixer parameters that can be achieved purely by the mixer circuit design process.

N. Rodríguez, Ph. D. is a telecommunications engineer for the Centro de Estudios e Investigaciones Técnicas de Gipuzkoa (CEIT), San Sebastián, Spain. She can be reached via e-mail at nrodriguez@ceit.es.

E. Hernández, G. Bistué, J. Presa and R. Berenguer are also with CEIT. Bistué is an IEEE member.

I. Gutiérrez is with Tecnun, Escuela Superior de Ingenieros de San Sebastián (Universidad de Navarra), San Sebastián, Spain.

RSS    Save to Del.icio.us  Digg This