System-on-Package MHMIC Milimeter-Wave Frequency Synthesizer for 60 GHz WPANs

We present a low-cost millimeter-wave frequency synthesizer with ultralow phase noise, implemented using system-on-package (SoP) techniques for high-data-rate wireless personal area network (WPAN) systems operating in the unlicensed 60GHz ISM band (57–64 GHz). The phase noise speciﬁcation of the proposed frequency synthesizer is derived for a worst case scenario of an 802.11.3c system, which uses a 64-QAM 512-carrier-OFDM modulation, and a data rate of 5.775 Gbps. Our design approach adopts commercial-of-the-shelf (COTS) components integrated in a low-cost alumina-based miniature hybrid microwave integrated circuit (MHMIC) package. The proposed design approach reduces not only the system cost and time-to-market, but also enhances the system performance in comparison with system-on-chip (SoC) designs. The synthesizer has measured phase noise of − 111 . 5 dBc/Hz at 1MHz o ﬀ set and integrated phase noise of 2.8 ◦ (simulated: 2.5 ◦ ) measured at 57.6 GHz with output power of


Introduction
The current demand for high-definition video streaming as well as the need for high-data-rate transmission in the range of multigigabit/s, attracts the use of the 60-GHz unlicensed ISM band (57-64 GHz). The main reason for the interest in the 60 GHz ISM band is attributed to the availability of 7 GHz of unlicensed bandwidth. Furthermore, the high oxygen absorption, and line-of-sight use, of the 60 GHz band makes this band well suited for frequency reuse which increases the system capacity; in addition, it minimizes harmful cochannel interferences, and increases the security of communication.
Although the 60-GHz band has many advantages, the design of low-cost high-performance frequency synthesizers that meet the system requirements of low-phase noise presents a design challenge, particularly, when CMOS system-on-chip (SoC) is the technology to be used (see Table 2 for comparison). Such a challenge is due to the lossy silicon-substrate of CMOS technology. This is the main reason for using GaAs-based COTS components in our proposed design. The advantage of SoP integration is that, it enables the realization of the passive components on the same substrate of the packaging. In addition, active devices can be selected from different technologies to optimize the system performance. As an example, CMOS components can be used for high-density logic and analog circuits, SiGe and GaAs for high-speed microwave circuits, and GaN for high power.
This paper will derive synthesizer phase noise specification and present design and measurements of the synthesizer. System analysis is performed in Section 2, with calculation of best-case SNR required. In Section 3 an analysis is performed on the influence of the phase noise on SNR and synthesizer requirements are derived. Section 4 discusses the proposed  frequency synthesizer. Finally the measured data is presented in Section 5.

System Analysis
Before presenting the analysis and design of the synthesizer we need to understand how the synthesizer performance affects the overall system performance. The following sections will describe how the SNR of the system is derived from top level specifications and how the SNR is degraded by phase noise of the synthesizer, leading to synthesizer phase noise specification. The model presented here begins with an ideal case and builds on the nonidealities of various system components and the effect each has on the degradation of SNR. This work begins by presenting the system's link budget, followed by peak power requirement, and finally by phase noise degradation of SNR.

System
Overview. The proposed synthesizer is designed to be used in an experimental setup (shown in Figure 1) to test transmission in the 60-GHz range. For simplicity, the setup would take several channels from 802.11n (5.5 GHz) signal and upconvert them to 60 GHz.

Link Budget.
Following the standard specifications for 802.15.3c [5] the required system specifications is derived. The required path loss of the radio link is considered next. First we define the system specifications in Table 1

Peak Power
Requirement. An OFDM signal has a higher peak-to-average power ratio (PAPR) than a single carrier (SC) signal. The worst case PAPR as a function of number subcarriers [6] is: PAPR dB = 10 · log 10 (N). (1) For 802.15.3c HSI mode N is 352, and PAPR is 25.5 dB. This means that average power would have to be 25.5 dB below the 27 dBm peak power specified by FCC [3]. Therefore, the average transmit power would be 1.5 dBm (27 dBm−25.5 dB) due to peak power constraint. Maximum SNR for peak power requirement would degrade to 18.5 dB.
International Journal of Microwave Science and Technology  Figure 5: ω n normalized to intersection frequency (a) and damping constant ξ (b) of PLL loop filter for smallest integrated phase noise. r is the capacitance ratio of the 2nd order PLL loop filter.

Phase Noise Effect on SNR
According to [7] the SNR degradation due to phase noise in an OFDM signal can be calculated as follows: where σ ϕ corresponds to the integrated phase noise in radians (also referred to as RMS jitter).
Since ECC can significantly change the required SNR, we need to look at the effect of SNR degradation due to phase noise at SNR levels required with ECC. low-density-paritycheck (LDPC) coding, used in 802.15.3c, can get very close [8] to the Shannon limit. The SNR at Shannon limit is [9]: SNR min = 10 · log 10 2 r·M − 1 , where r is code rate, M is constellation size (M × M QAM).

4
International Journal of Microwave Science and Technology  LDPC coding can achieve SNR values as close as 1 dB from the Shannon limit [8]. Allowing us to see how phase noise would influence ECC-coded signal using (2) and (3). This is shown for various modulation schemes and code rates ( Figure 2).
The SNR at Shannon limit for code rate 5/8 64-QAM signal is 15 dB (SNR min from (3)). By adding LDPC encoding we expect it to increase to16 dB. The SNR margin would then be 18.5 dB−16 dB = 2.5 dB. From Figure 2 this means that the system can tolerate 4 • of RMS jitter while still conforming to specifications in Table 1.

PLL Design.
For simplicity a standard PLL topology [14] is used as shown in Figure 3.
For better phase noise a maximum channel spacing of 3.2 GHz is chosen, based on 100 MHz PFD frequency.

PLL Loop Filter Design.
Loop filter is chosen to be a commonly used 2nd order integrator-lead RC filter [14]. Higher order filters have been simulated producing no noticeable improvement on overall phase noise. The loop filter topology in is employed as shown in Figure 4.
The loop bandwidth and loop damping constant are optimized based on the VCO slope, C 1 /C 2 capacitance ratio (r) for which a rule of thumb value of 10 is used, and the fact that the reference + PLL phase noise is nearly thermal around the loop 3 dB frequency. Using MATLAB, a sweep of loop natural frequency (ω n ) and damping constant (ξ) is performed for various VCO slopes and r's. The ω n and ξ values producing smallest integrated phase noise are shown in Figure 5.
Based on the plots of Figure 5, and using the fact that VCO has a slope of 28 dB/dec and with capacitance ratio of 10, the optimal ω n /ω x and ξ are 0.7 and 0.88, respectively, where ω x is intersection of open loop phase noise of VCO and the reference, which is at 2π · 65 KHz ( Figure 6). From this, loop component values that produce minimal integrated phase noise are calculated using [14]. The result is shown in Table 2. The plot in Figure 6 shows component phase noise based on open loop measurements illustrating loop bandwidth selection.

Implementation.
The reference, DC supplies, and digital control signals are provided from an PCB board, which is connected to the MHMIC SoP with bond wires. The 60 GHz output signal is bond-wired from the amplifier to a coaxial V-connector. The synthesizer with populated components is shown on Figure 7.

Measurements
The synthesizer phase noise is measured at 57.6 GHz with a Rohde & Schwarz FSUP signal source analyzer. The plot in Figure 8 shows measured phase noise versus simulated. Table 3 compares measured phase noise with published measurements (all normalized to 57.6 GHz).
The power consumption of VCO and ×4 multiplier is 2.1 W, which makes the applications of the proposed synthesizer suitable to large size installations such as set-top boxes, kiosks, and point-to-point radios.

Conclusions
A cost-effective high-performance SoP synthesizer is designed and manufactured based on the derived phase noise requirements from system analysis. Worst case 802.15.3c MCS-index-7 signal is used for derivation of SNR. The worst case SNR is derived to be 18.5 dB for 10 m link with BER of 10 −6 . The Integrated phase noise specification is derived to be 4 • . The synthesizer has been designed and manufactured in MHMIC process which exceeds this  specification, achieving phase noise of −111.5 dBc/Hz (1 MHz offset) and the integrated phase noise of 2.8 • at 57.6 GHz, degrading SNR by only 1.8 dB. To authors' knowledge this is the best phase noise reported, at the time of writing, at frequency close to 60 GHz.