The chip photograph of this up-conversion mixer is shown in Fig. 3.35 with a chip size of 0.79×0.61mm2 including all pads. All of the dc pads are wire bonded to an external PCB board. The LO source is generated by a signal generator (Keysight E8267D). To provide a differential IF signal, the IF source from a Keysight E8267D signal generator is divided by a Marki BAL-0026 balun. Moreover, the output signal at the RF port is observed through an Agilent E4448 spectrum analyzer. The measurement setup is shown in Fig. 3.36. The chip is measured though on wafer probing with GSG probes at LO and RF ports, and GSSG probes at IF ports. To ensure the function of up-conversion mixer, the measurement of CG versus different LO power is conducted at first and the result is shown in Fig. 3.37. Compared to the simulation, the measured
results are lower in the large LO power region. The inconsistency between measurement and simulation is caused by the not-well consideration of the post-simulation. There are three main reasons which are discussed as follows. Firstly, in the LO port design, a bypass capacitor is located at the middle of secondary wind which is used for the virtual ground for differential operation, as shown in Fig. 3.13. To produce larger capacitance in the small area, the bypass capacitor is realized by the lateral flux capacitor. However, in the simulation, the interdigital capacitor with the same capacitance is used to reduce the time spent on the EM simulation. The capacitance of lateral flux capacitor and interdigital capacitor used in the LO transformer circuit is shown in Fig. 3.38. Since the parasitic inductance of lateral flux capacitor is much lower than that of the interdigital capacitor due to capacitance density, the self-resonant frequency of lateral flux capacitor is much higher. As a result, the simulation results of LO transformer with lateral flux capacitor and interdigital capacitor are different. So, considering this discrepancy, the measurement and the re-simulation of CG versus different LO power is exhibited in Fig.
3.39. Secondly, the measured and simulated results can be more consistent with each other when there are -5% process variation in the poly resistors (R1, R2) located in the drain of transistors M9, M10, as shown in Fig. 3.21. Considering these factors, the measurement and the re-simulation of CG versus different LO power is exhibited in Fig.
3.40. Finally, the DC blocking capacitors used in the transconductance stage are 1.5 pF which are composed of the lateral flux capacitor from metal 8 to metal 2 due to the lack of the CTM and CBM layers in the 28 nm CMOS. The capacitance of this proposed capacitor is larger than that of the interdigital capacitor in the small area. The implementation of this capacitor is shown in Fig. 3.41. The bottom layer from Metal 1 to Metal 3 is defined as the ground. As a result, there are additional parasitic capacitance (Cp) induced between the capacitor and the ground which are not considered in the
original post simulation. The capacitor of Cp is 0.1 pF which is roughly 7 % variation of the DC blocking capacitors. Since the capacitance of DC blocking capacitor is not large enough for IF input signals which will be considered as a matching element, CG of the proposed mixer considering the additional coupling effect would be lower than the original CG of the proposed mixer. Taking all of the factors into consideration, the measured and simulated results can be more consistent with each other. Fig. 3.42 shows the measurement and the simulation of CG versus different LO power after considering all of the unexpected factors. The measured and simulated result agree well to each other. The LO power is set to 2 dBm in the following measurement which is the same as the LO power in the simulation. Moreover, the measurement of conversion gain versus RF frequency with a fixed LO frequency at 27 GHz is then conducted. In Fig. 3.43, the measured and simulated frequency responses of CG also agree reasonably. The proposed mixer exhibits -9.3 to -6.4 dB of conversion gain from 17 to 29 GHz of RF frequency at upper sideband. Moreover, the measured and simulated CG versus IF frequency are shown in Fig. 3.44. The measured results demonstrate maximum CG of -10.9 dB at 1 GHz. The measured and simulated results have a little difference since the Marki balun [49] is unable to provide accurate differential signal. Considering this effect, the measurement and re-simulation of CG versus IF frequency with LO power of 2 dBm and LO frequency of 27 GHz is shown in Fig. 3.45. Besides, the measured LO leakage to RF port is plotted in Fig. 3.46 which can achieve -35 dB over the bandwidth.
Fig. 3.47 shows the simulated and measured CG, output power versus input power with 27 GHz LO frequency and 1 GHz IF frequency. The proposed mixer achieves -2.2 dBm OP1dB. The measured IM3 results are shown in Fig. 3.41 and the OIP3 is -10.2 dBm. The IMD3 versus input power at different IF spacing is shown in Fig. 3.49 and Fig. 3.50.
Similarly, the other different bias conditions with roughly the same CG is compared in
ttttttttt
Fig. 3.36 The measurement setup of the up-conversion mixer.
Fig. 3.37 The measured and simulated CG versus LO power with LO frequency of 27 GHz, and IF frequency of 1.0 GHz.
Fig. 3.38 The capacitance of lateral flux capacitor and interdigital capacitor used in the LO transformer circuit.
Fig. 3.39 The measured and re-simulated CG versus LO power with LO frequency of 27 GHz, and IF frequency of 1.0 GHz after considering the discrepancy in LO port matching.
Fig. 3.40 The measured and re-simulated CG versus LO power with LO frequency of 27 GHz, and IF frequency of 1.0 GHz after considering the process variation in poly resistors.
Fig. 3.41 Schematic diagram of the parasitic effects in implementation.
Fig. 3.42 The measured and re-simulated CG versus LO power with LO frequency of 27 GHz, and IF frequency of 1.0 GHz after considering all of the unexpected factors.
Fig. 3.43 The simulated and measured CG versus RF frequency with LO power of 2 dBm and IF frequency of 1 GHz.
Fig. 3.44 Measured and simulated CG versus IF frequency with LO power of 2 dBm and LO frequency of 27 GHz.
Fig. 3.45 Measured and re-simulated CG versus IF frequency with LO power of 2 dBm and LO frequency of 27 GHz.
Fig. 3.46 The measured LO leakage versus RF frequency of the proposed mixer.
Fig. 3.47 The simulated and measured CG, output power versus IF input power at RF frequency of 28 GHz.
Fig. 3.48 The two-tone measurement result of RF power versus IF power at RF frequency of 28 GHz 5 MHz.
Fig. 3.49 The measured IMD3 result at RF frequency of 28 5 MHz with and without linearizer.
Fig. 3.50 The measured IMD3 result at RF frequency of 28 12.5 MHz with and without linearizer.
the two-tone measurement result. At the linearizer-off state, the transistor of M6 and M7
is biased at 0.5 V and the transistor of M5 and M8 is also biased at 0.7 V where transistors M5-M8 are biased closed to saturated region. As a result, the gm2 of M6 and M7 is closed to near-zero region and the magnitude of positive IM3 current is reduced.
The proposed mixer exhibits around 10 dBc of IM3 cancellation with the IM2 signal injection technique in the wide power region. Moreover, in the large power region, the output power of the proposed mixer is improved around 3dB at the same IMD3 power level with the linearizer. From these measurement result, the IM3 difference shows clearly that the linearization technique is indeed effective.
3.3.2 Digital Modulation
The measurement setup of the modulated signal is shown in Fig. 3.51. Because the
IF input of the proposed mixer is differential, the differential I/Q IF signal given by the Agilent M8190A AWG is connected to the Agilent E8267D. Then, the output signal from the Agilent E8267D is divided by a Marki BAL-0026 balun. The LO power is conducted through Agilent E8267D and the RF output signal is down-converted and demodulated by Keysight N9030B spectrum analyzer (PXA). For the high-linearity test, both 64-QAM and 256-QAM signals are tested for the proposed mixer. For the single-carrier 64-QAM signal with a symbol rate of 50 MHz Baud rate signals, the constellation diagram and the error vector magnitude (EVM) measurement result with the linearizer turning on and off is shown in Fig. 3.52. The proposed mixer demonstrates the output power of -3.3 dBm and -2.8 dBm where EVM equals to -25.27 dB and -25.16 dB with the linearizer turning on and off, respectively. Moreover, the measured EVM versus output power of a 64-QAM signal with a symbol rate of 50 MHz Baud signal is shown in Fig. 3.53. As can be seen, the output power where EVM is around -25 dB is roughly at same level when the linearizer turning on and off. Here, the EVMs are calculated and normalized by the peak power (EVMmax). According to the global standard (Release 16) established by 3GPP, the minimum channel bandwidth in the ttttttt
Fig. 3.51 Measurement setup of the modulated signal.
Fig. 3.52 (a) Measured constellation diagram and (b) Waveforms of the single carrier 64-QAM signals with 50 MHz baud rate upon linearizer turning on. (c) Measured constellation diagram and (d) Waveforms of the single carrier 64-QAM signals with 50 MHz baud rate upon linearizer turning off.
Fig. 3.53 The measured EVM (dB) versus Pout (dBm) with and without linearizer of the 64 QAM 50 MHz Baud signal.
Fig. 3.54 (a) Measured constellation diagram and (b) Waveforms of the single carrier 256-QAM signals with 20 MHz baud rate upon linearizer turning on. (c) Measured constellation diagram and (d) Waveforms of the single carrier 256-QAM signals with 20 MHz baud rate upon linearizer turning off.
Fig. 3.55 The measured EVM (dB) versus Pout (dBm) with and without linearizer of the 256 QAM 20 MHz Baud signal.
millimeter wave is 50 MHz. As seen in Fig. 3.43, the conversion gain at 38 and 39 GHz is -9.3 and -10.4 dB. Since the conversion gain of proposed mixer over RF frequency drops when frequency is higher 28 GHz, the proposed mixer requires higher PAPR when testing the 256-QAM signal with a symbol rate of 50 MHz baud rate signals.
Therefore, to confirm the proposed IM3 improvement technique is beneficial to the improvement of the output power level of high-order modulation measurement, the single carrier 256-QAM signal with a symbol rate of 20 MHz baud rate signal is also tested. The constellation diagram and the error vector magnitude (EVM) measurement results with the linearizer turning on and off are shown in Fig. 3.54. The proposed mixer shows that the output power of -4.4 and -3.7 dBm where EVM equals to -29.75 and -23.57 dB with the linearizer turning on and off, respectively. The measured EVM versus output power of a 256-QAM signal with a symbol rate of 20 MHz Baud signal is shown in Fig. 3.55. According to the IEEE 802.11a standard, the required rms EVM for 256-QAM modulated signal is formulated below -30 dB. With the proposed technique, the mixer shows an improvement of the modulated output power of 256-QAM signal.
Besides, the improvement of the modulated output power of 256-QAM signal under -30 dB is roughly similar to the improvement of output power under the same IMD3 power level. As a result, this proves that the IMD3 distortion dominates the SNR of mixer in the high IF input power region.