A Current Bleeding Cmos Mixer Featuring Lo Amplification Based on Current-reused Topology

A double balanced Gilbert-cell class-A amplifier bleeding mixer (DBGC CAAB mixer) is proposed and implemented. The injection current is utilized to amplify the local oscillator (LO) signal to improve the performance of the transcon-ductor stage. The DBGC CAAB mixer achieves a conversion gain of 17.5 dB at −14 dBm LO power, and the noise figure is suppressed from 45 dB to 10.7 dB. It is important to stress that the new configuration will not drain additional power in contrast to the former current bleeding mixers. This topology dramatically relieves the requirement of the LO power. The DBGC CAAB mixer is implemented by using 0.18-μm RFCMOS technology and operates at the 2.4 GHz ISM application with 10 MHz intermediate frequency. The power consumption is 12 mA at 1.5 V supply voltage. The DBGC CAAB mixer features the highest FOM figure within a wide range of LO power.


Introduction
With the rapid proliferation of modern wireless market, RF transceivers play a more and more important role in our daily lives.Mixer is one of the key blocks in the signal chain performing frequency translation before signals are further processed in the intermediate frequency circuits.The noise and the gain performance of the mixer, in general, determine the performance of the whole system at large [1].In past decades, many research scholars made deep insights into the noise mechanism [2][3][4][5] and gain enhancement [6][7][8] in mixers.In order to achieve good performance including gain, noise, isolation and linearity (higher-order components), Gilbert-cell [9] is commonly used as the mixer core topology.
Amongst most factors, the conversion gain (CG) and the third-order intercept point (IP3) of the mixer are the key elements determining the mixer performance.Both CG and IP3 are proportional to the square root of the bias current (I b1 ) of the driver stage [10,11], as shown in Figure 1(a).In order to improve CG and IP3, a simple method is to increase the bias current of the driver stage M1, namely I b1 .However, this improvement is at the expense of other degradations.For instance, the increase of I b1 will also increase the currents, I b2 and I b3 , in the switch pair M2 and M3 (referred as M2-M3 thereafter).
As a result, the noise contributions from M2-M3 are also increased [12][13][14].In addition, in order to keep all the transistors working in saturation region, the load resistance Z L should be decreased to avoid too much voltage drop across them, especially in modern sub-micro CMOS technology.The load resistance reduction will in turn cause the gain compression.Hence it is concluded that increasing the bias current solely of the driver stage is not an efficient way to improve the overall performance, including gain, linearity and noise of the mixer.
In order to retain the benefits of increasing the bias current but without degrading other performances, a current bleeding topology, by applying a current source in parallel with M2-M3, is used [15].Analysis shows that such topology [15], [16] facilitates the quiescent current in the driver stage to be independent from M2-M3.shows the illustrative topology.The bleeding current source drives a higher driver stage current through the higher load resistance of M1 by virtue of the fact that part of the driver bias current is steered from M2-M3.An additional advantage of using the bleeding source is that M2-M3 can operate at a lower gate-source voltage and thus rendering a compact size.Lower gatesource voltage helps to improve the conversion efficiency as fewer charges are necessary to turn M2-M3 on and/or off.
In this paper, a double balanced Gilbert-cell class-A amplifier bleeding mixer (DBGC CAAB mixer) is developed.By inserting a class-A amplifier in parallel with M2-M3, the local oscillator (LO) swing is increased to drive M2-M3 into a hard switching fashion.An additional advantage is that the bias current of the class-A amplifier will enhance the bias of the driver stage of the mixer, hence improving the overall performance.The present investigation is an extension of our previous work [17] with a detailed analysis of noise reduction and gain boosting.Detailed measurement results are also provided.The paper is organized as follows: Section 2 describes the motivation of this work; Section 3 describes the topology, the design and the analysis results, Section 4 summarizes the performance of the developed mixer by including a FOM comparison and, Section 5 gives the conclusion.

Motivation
In the mixer analysis, the drain current of M2-M3 are presented as: (1) The voltage of Local Oscillator (LO), v LO , is defined as: (3) From KCL: and, for the differential operation of M2-M3, the output current is: Based on the physics of the MOS device, Equations ( 1) and ( 2) are supported by the following relationships: In order to characterize the relationship of the output current and the performance of M2-M3, the output current is derived as a function of LO v : This formula is derived based on the condition that both M2 and M3 are at ON state.When 2 LO o , M2-M3 acts as a hard switch, and the output current, I out , can be modeled as a sgn function: where, Figure 2 shows the I-V characteristics of M2-M3 graphically.The transconductance of M2-M3, G m , is given by: It will be explained in (13) that increasing v LO will improve the mixer performance; hence the dependence of G m on v LO needs to be investigated.G m is derived as a function of LO v   : Noise is an important parameter in mixer design and is now discussed.In general, there are three main noise decreases and eventually approaches zero.sources in conventional mixer circuits, namely the transconductor stage, the switch pair M2-M3 and the load.In Terrovitis [5] and Darabi [2] models, noise contributed to the output exactly when both switching transistors are at ON state-indicated as the switch interval Δ (will be discussed in Section 3).The noise PSD (power spectrum density), introduced within Δ is presented as: where K-Boltzmann's constant T-Absolute temperature γ-noise coefficient (= 2/3 for long channel device) g to vanish and thus suppressing the noise.Noise PSD can further be converted [5] to the relationship of bias current and noise contribution: where I B is the bias current of M2-M3.It is seen from ( 13) that, by increasing v LO , Noise PSD is suppressed efficiently.Furthermore, by reducing I B , PSD will also be suppressed, thus further verifying the current bleeding principle.
The other key parameter of a mixer is gain.In order to enhance the gain, the conversion gain (CG) of the mixer is also analyzed [5].

CG c
  (14) where the multiplier c is given by: It is seen from ( 15) that increasing c, and/or g m1 , is an efficient way to increase CG.The implementation will be discussed in Section 3.4.
Based on the above discussion, the novelties of the proposed design are: A1) Novel bleeding source: In order to improve the gain, linearity and noise, the bias current of the driver stage should be increased.However, such a direct increase will cause other related problems, including the high noise contribution from M2-M3 as well as the excessive voltage drop on the load resistor.It is important to note that the current bleeding structure improves only the bias of the driver stage but without increasing the bias of M2-M3.
A2) LO amplification: Higher amplitude of V LO helps to minimize the noise PSD from the M2-M3 by reducing Δ, and at the same time, maximizing In this investigation, a Gilbert-cell mixer based on the current bleeding technique is explored.Inspired from the benefit of high gain and low noise (referring to (A1) and (A2)), and taking advantage of the linearity, a class-A amplifier (referred as Amp thereafter), is implemented as the bleeding sources to amplify the LO signals.Hence the class-A amplifier bleeding source is employed to replace the traditional simple current source.After performing the LO amplification, the DC bias current of the class-A amplifier is steered to the driver stage of the mixer to improve the gain, noise and linearity characteristics.It will be explained that, comparing to the conventional current bleeding mixer, the CAAB structure can reuse the bleeding current more efficiently.
 to reach the upper limit.

The Amp Design
The Amp in disturbance of I a may cause gain ripples and non-linear degradations in the driver stage M1.In this case, since 0.18-μm fabrication process is used, the channel length of the transistor is small, rendering I a more and more susceptible to the channel length modulation [18]: The large output voltage swing normally appearing on the drain of the transistor will change the V DS dramatically and periodically.In principle, in order to avoid the channel length modulation, a cascode topology can be employed.However, this structure will limit the output voltage swing.An alternative method is to use a longer channel length.In this investigation, 0.35-μm gate length is preferred as the operating frequency is not very high (2.5 GHz).
It is important to stress that, in order to lower the inherent noise in the CAAB mixer, the Amp is realized as a low noise amplifier.In Figure 4, the components L g , C gs , C sr , L sr and L s are used for noise matching [19].In order to further analyze the amplifier, the equivalent circuit is shown in  The input impedance Z in is then degenerated to be: At resonance, the imaginary part of Z in vanishes, and the real part is left for 50 Ω matching [19].The Amp operates at a DC bias current of 4.5 mA.

The DBGC CAAB Mixer and Implementation
The Amp is then designed and analyzed by SpectreRF in Cadence.Compromising the power consumption, LO amplification and stability, the gain of the Amp to be designed is chosen to be about 14 dB.The component values in Figure 5 are optimized as: 6 shows the final characteristics of Amp. Figure 7 shows the voltage waveforms at the input and the output of the Amp at various LO input power levels.An investigation of the voltage waveforms reveals that the amplitude of the LO signal is amplified by a factor of five (5).The achieved input 1 dB compression point (P 1dB,in ) is 3 dBm, thus surpassing the traditional performance that P 1dB,in < 0 dBm.The DBGC CAAB mixer is realized by a 0.18-μm 1-poly 6-metal RFCMOS technology.A microphotograph of the device is shown in Figure 8.The chip occupies an area of 1.2 × 1.3 mm 2 .Under nominal operation, the mixer extracts 12 mA from the 1.5 V supply.The chip was bonded to a FR4-PCB board with gold-wire for measurement.The operating frequency is 2.4 GHz with 10 MHz IF as the output frequency for testing purpose.

Noise Analysis
In essence, the LO signal is a sinusoid rather than a square wave, rendering that the current switching to approximate a soft-switch with switch interval Δ (see Fig- ure 9).When v LO < V X , (V X is the threshold voltage of M2-M3) both transistors of M2-M3 are "ON", and the current division ratio is biased dependently such that the total bias current of M2-M3 is constant: when M2 is biased in high level, the current flow through M2 increases, and the current in M3 decreases.When v LO > V X , the current will flow through one of the transistors and the other counterpart is turned off.At this point, the "ON" transisacts as the cascode stage of the driver stage.The output noise component from M2-M3 is dictated by the relationship in (10) and (12), in which g m2 and g m3 are time-varying transconductances of M2-M3 under large LO drive.When v LO > V X , either M2 or M3 is cutoff and g m vanishes.Consequently, G m , as well as PSD, also vanish.When v LO < V X , the non-zero G m proliferates M2-M3 to contribute noise to the output.
Let PSD = S sw,within when v LO < V X and PSD = S sw,outside when v LO > V X .For analysis purpose, S sw,within is normalized to a train of pulses [4] operating at a rate 2f LO , as shown in Figure 9.The width of Δ is V X /λ, where λ is the slope of LO waveform at the zero-crossing point, and is given by: It was illustrated in Section 3.4 that by devising the Amp, v LO has been amplified by five (5) times, thus the switch interval Δ 2 (after amplification) has been compressed to one fifth of Δ 1 (before amplification) (see Figure 9).
To further analyze the relationship between the noise PSD and the slope of LO waveform, λ, at the zerocrossing point, ( 12) is modified and derived as follows: From ( 19), it is analyzed and concluded that the noise from M2-M3 is reduced by a factor of five (5).Hence it is analyzed that: Additional noise improvement comes from the increasing bias of the driver stage due to the current bleeding source (here, it is DC bias of the Amp).Based on the original 1.5 mA DC bias current, the current bleeding source feeds additional 4.5 mA DC current to the driver stage.From [18], , consequently, g m1 is improved by a factor of 4 .For double-balanced mixer, the single-side band noise figure, NF [5], is rearranged as: where γ 1 and γ 3 represent the noise coefficient of M1 and M3 respectively.The poly resistance of the gate is indicated as r g1 and r g3 .It is seen from ( 20) that NF is inversely proportional to g m1 .Thus it is seen that the current bleeding improves g m1 dramatically and conesquently suppresses the noise figure efficiently.
To examine the performance improvement of the DGBC CAAB mixer, an identical conventional Gilbertquad mixer, without current bleeding-Amp, but having the same transistor size in both the driver stage and the switch stage, is also designed with same process.Both mixers operate at 2.4 GHz input frequency with 10 MHz IF output.LO drive.Figure 11(b) intuitively shows the trend of L sw,dB (in logarithm scale) due to soft switching.Table 1 lists the reduction of L sw,lin due to the LO amplitude amplification at four different LO cases.In the extreme case that Δ = 0.45 (when LO voltage is very weak), the gain improvement can be as high as 3 dB.It is also seen that when Δ is one fourth of the LO period (T LO ), which is the most probably case, L sw,lin can be improved from T 

Conversion Gain Analysis
In the CAAB mixer, the conversion gain is affected by the switch stage M2-M3 and the driver stage.Recapitulated from ( 14), a Δdecreases, approaches to unit.As a result, the CG increases until reaching the upper limit 11 illustrates the switch loss, L sw , versus Δ. Figure 11(a) shows L sw,lin , versus Δ due to the non-ideal square wave . As a result, 0.88 dB CG is gained.
It is shown in Section 2 that g m1 is improved by a factor of 4, hence an additional 12 dB gain is obtained.Thus, it is concluded that the current bleeding structure will boost g m1 , provoking a higher CG.
To examine the gain performance of the DBGC CAAB mixer, a comparison with the conventional mixer and other published works is shown in Figure 12.For the conventional mixer, the maximum gain achieved is 11 dB when LO port is fed by 10 dBm signal.In contrast, a gain of 17.5 dB is achieved when −14 dBm is fed to the LO port of the DBGC CAAB mixer.Thus, it is con-   cluded that by incorporating the current bleeding LO Amp, not only the conversion gain is enhanced, but also the LO power requirement is relieved by more than 20 dB.The comparison with other published works [20][21][22] reveals that the DBGC CAAB mixer has the highest gain but requiring the lowest LO power requirement.

Linearity, LO Power Leakage and Operation Bandwidth
The output voltage swing is examined by terminating an oscilloscope at the IF (10 MHz) port.The maximum voltage swing (peak-to-peak voltage) was measured to be 1.1 V when the RF port was fed by −11 dBm input power (see Figure 13).Figure 14 shows the measured third order intermodulation (IP3) of the DBGC CAAB mixer at different LO power levels.When the switch pair is fed with a −5 dBm LO power, the IIP3 obtained is 3 dBm.When the LO power is −17 dBm, the mixer still has an IIP3 point of −9 dBm.In the DBGC CAAB mixer, the LO amplitude is amplified by five (5) times.Hence, attention is drawn to the LO power leakage.By examining the LO power at RF and IF port, it is found that the LO to IF port leakage is smaller than −90 dBm.The LO to RF power leakage is measured to be less than −60 dBm (see Figure 15).The good LO-IF isolation achieved is attributed to use of the double-balanced topology.The DBGC CAAB mixer is designed for 2.4 GHz ISM band application.Figure 16 shows the gain performance at 2.4 -2.4835 GHz when the IF frequency is 10 MHz.In the whole band, the conversion gain of the proposed mixer varies from 19 dB to 21 dB, featuring good gain flatness.

Performance Summary and Comparison
The FOM takes incorporates important parameters including Gain, NF, IIP3, LO power, DC power and operating frequency into considerations.The FOM values are listed in Table 2 for comparison.From the comparison, it is found that the DBGC CAAB mixer has the lowest LO power requirement (−17 dBm) and the highest gain (15.7 dB), while maintaining a relatively lowest noise figure of 9.7 dB.

Conclusion
A current bleeding with LO amplification mixer based on current reuse topology is designed and implemented.The developed double balanced Gilbert-cell class-A amplifier bleeding mixer (DBGC CAAB mixer) has the highest conversion gain at the lowest LO power when compared to mixers formerly investigated.The DBGC CAAB mixer is implemented by using 0.18-μm CMOS technology and operates at the 2.4 GHz ISM application with 10 MHz intermediate frequency.The power consumption is 12 mA at 1.5 V supply voltage.With the novel LO amplification and current reuse technique, the mixer features an excellent high gain of 17.5 dB at a very low LO power feeding of −14 dBm.The noise performance is also good.The DBGC CAAB mixer features a noise figure of 10.7 dB, thus rendering the resulting noise to be suppressed to [8.7 dB, 12.4
As shown in Figure 2, m LO reaches its maximum value as LO approaches 0. When
Trans-conductance of M2-M3 At the zero-crossing point of v LO , both M2 and M3 are at ON state, thus resulting in non-zero value of 2 m g and 3 m g .From (11), large LO voltage swing will diminish G m , rendering either 2 m g or 3 m

Figure 3
Figure 3 shows the proposed double-balanced Gilbertcell mixer incorporating CAAB, namely DBGC CAAB mixer.As shown in Figure 1(b), taking advantage of the symmetry, one half of the circuit is used for analysis.In conventional mixers, the bias current, I b1 , of the driver stage M1 is the sum of current of I b2 (of M2) and I b3 (of M3), i.e.I b1 = I b2 + I b3 .However, in a DBGC CAAB mixer, the bias current of M1 is the sum of I b2 , I b3 and I a , i.e.I b1 = I b2 + I b3 + I a .By devising component values for Amp, V LO of M2-M3 is increased, thus reducing Δ, and rendering a noise reduction in M2-M3.

Figure 5 .
Figure 5. Equivalent circuits of the class-A amplifier (Amp) in Figure 4.

Figure 6 .Figure 7 .
Figure 6.The simulated voltage gain of the Amp.

Figure 10
shows the noise comparison at LO power from −20 dBm to 10 dBm.It is observed that the conventional mixer presents a noise figure in the range of 13 dB to 46 dB whereas the CAAB mixer features a much lower noise figure, namely from 9 dB to 11.5 dB.The inset of Figure 10(b) shows the measured noise figures versus LO power at 10 MHz intermediate frequency.The measured NF varies from 12.4 dB to 8.7 dB when the LO power is varied from −20 dBm to 10 dBm.

Figure 12 .
Figure 12.The conversion gain comparison between the DBGC CAAB mixer and other works.

Figure 13 .Figure 14 .Figure 15 .
Figure 13.IF peak-to-peak voltage versus RF input power.(The result is measured with the termination of oscilloscope at IF port).

Figure 16 .
Figure 16.Conversion gain of the DBGC CAAB mixer versus frequency when IF frequency = 10 MHz.

Table 2
summarizes the performances of the DBGC CAAB mixer.To evaluate the mixer comprehensively, a benchmarking figure of merit, FOM, is presented: dB].In contrast, in the conventional mixer, the noise figure varies from 13 dB to 46 dB at the same LO feed.It is important to point out that, compared to the other mixer investigations, the DBGC CAAB mixer features the highest FOM figure within a wide range of LO power.