Chapter 2 High Linearity Gilbert Upconversion Mixer
2.1 High-Linearity Bias-Offset TCA
2.1.2 Gilbert Upconversion Mixers With NMOS/PMOS Bias-Offset TCA
(2.3)
because 2
V
IFV
IF＝V
CMis a constant dc,
1 2 ( ) 2 1 3)( 2 3
o o IF IF CM x Tn B
i i V V ( －k k V V V )+ k V . (2.4) Thus,
g
m 2kV if k k
B, 1k
3 . The gm is a constant and no third-order inter-modulation occurs ideally. However, the short-channel effect degrades the linearity, especially when a large gate-to-source voltage is applied on the device with a shorter gate length in an advanced CMOS technology.Gilbert upconversion mixers with NMOS/PMOS bias-offset TCAs are implemented, respectively, and fully compared in next section.
2.1.2 Gilbert Upconversion Mixers With NMOS/PMOS Bias-Offset TCA
VB
CC
Fig. 2-4 Schematic of the Gilbert upconverters with NMOS/PMOS-type bias-offset TCAs and a single-band LC current combiner.
The schematics of the Gilbert upconversion mixers with NMOS-type and folded PMOS-type input stages are shown in Fig. 2-4. The bias-offset differential pair employed in the input stage has the transconductance of 2kVB if the MOS characteristic is still in a square-law region with a long channel transfer function, as introduced in Section 2.1.1.
The NMOS and PMOS I-V characteristics are shown in Fig. 2-5(a) and (b) with the gate lengths of 0.5 μm, 1 μm, and 2 μm and the correspondent widths of 25 μm, 50 μm, and 100 μm, respectively.
0.0 0.5 1.0 1.5 2.0 2.5
Fig. 2-5 (a)NMOS (b) PMOS I-V characteristics with different gate lengths.
The drain saturation voltage, Vdsat, is less than the gate-overdrive voltage,
V
gs−V
T=V
OV, if the short channel effect takes place. However, if VOV is still small, the transistors are still in the long channel region because the electric field is not large enough to saturate the velocity of the electrons as depicted in Fig. 2-5(a). Contrarily, the PMOS is almost in the long channel region even if the gate-overdrive voltage is large as shown in Fig. 2-5(b). Moreover, if the device is still in the long channel region, the drain saturation currents are almost the same if the W/L ratio is kept at a constant as shown in Fig. 2-5(b). On the other hand, the saturation current of a short channel device is lower than that of the long channel device when the W/L ratio is kept the same as shown in Fig. 2-5(a). In addition, a shorter gate length results in a smaller output resistance of the MOS transistor.The gate length of the NMOS transistors, M1-M4, in the input stage of the first chip is 0.5 μm while the gate length of the PMOS transistors, M1-M4, of the other chip is 1μm. For the lower mobility of PMOS transistors, the widths of the PMOS transistors are designed much wider to achieve similar transconductance gain when compared to the upconverter with the NMOS TCA. As a result, the IF bandwidth of the upconverter with the PMOS TCA is much narrower than that with the NMOS TCA.
Instead of using the MOS transistors in the Gilbert switching quad with a large LO switching voltage requirement of 2VOV(gate-overdrive voltage), the bipolar transistors Q1-Q4 are utilized in this work with only about 0.1-V LO voltage swing requirement to make the current fully switch. The current commutation mechanism is highly linear and only performs the frequency translation. A single band current combiner with a π-shape is shown in Fig. 2-6 with the differential input current I+ and
I
−. The combination of the LC current combiner and the current source I+ can berepresented by its Norton equivalence Iout and Zout as shown in Fig. 2-6. The equivalent current source (Iout) and the output equivalent impedance (Zout) can be obtained by of the ABCD matrix
1 2
Fig. 2-6 Block diagram of the LC current combiner and its equivalent circuit at the resonant frequency.
The output shunt-shunt feedback TIA is employed to translate the combined current output Itot to the voltage signal. Moreover, the output impedance is reduced by the factor of (1+A
for the shunt-type feedback and thus the output matching is easily achieved. The 20-GHz bandwidth of the feedback TIA is designed with strong feedback to reduce the nonlinearity effect.The die photo of the high linearity Gilbert upconverters are demonstrated by
utilizing NMOS and PMOS TCAs are shown in Fig. 2-7(a) and (b), respectively.
On-wafer measurement facilitates the RF performance.
(a) (b)
Fig. 2-7 Die photo of the Gilbert upconverter (a) using an NMOS TCA (b) using a PMOS TCA.
3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5
-20 -18 -16 -14 -12 -10 -8 -6 -4 -2 0
Conversion Gain (dB)
RF frequency(GHz)
NMOS PMOS
Fixed IF =20 MHz
Fig. 2-8 Conversion gain with respect to RF frequency.
The peak conversion gain of the upconverters with NMOS/PMOS input TCAs occurs at both RF=4.4 GHz, IF=20 MHz and the LO power is −5 dBm as shown in Fig. 2-8. The power performance of the upconverters with NMOS/PMOS TCAs is shown in Fig. 2-9. The OP1dB and OIP3 are −11/−11 dBm and 5.5/9.5 dBm for the upconverters with NMOS and folded PMOS TCAs, respectively.
-30 -25 -20 -15 -10 -5 0 5 10 15 20
Fig. 2-9 Power performance.
1 10 100 1000
Fig. 2-10 IF bandwidth.
4.0 4.2 4.4 4.6 4.8 5.0 5.2
Fig. 2-11 LO-to-RF isolation.
The IF bandwidth of the upconverters with NMOS and PMOS TCAs are 500 MHz and 100 MHz, respectively, as shown in Fig. 2-10. The upconverter with a PMOS TCA has a 20.5-dB difference between the OIP3 and OP1dB which is larger than a 16.5-dB difference for the upconverter with an NMOS TCA at the cost of a narrower IF bandwidth for a similar gain design purpose. However, both designs are much more linear than a conventional Gilbert upconverter. The output RF return loss is better than 16 dB for both circuits over 20 GHz. The LO-to-RF isolation of each upconverter is better than 28/26 dB when LO frequency ranging from 4-5.2 GHz as shown in Fig. 2-11. The power consumption of each circuit is 40/46 mW, respectively.
TABLE. 2.1 compares the proposed high-linearity Gilbert upconversion mixers using NMOS/PMOS TCA to the state-of-the-art circuits in literatures [8], [10]-[12].
TABLE.2.1PERFORMANCE COMPARISON OF THE GILBERT UPCONVERSION MIXERS
w/ NMOS