Microwave Journal
www.microwavejournal.com/articles/33095-the-phase-noise-challenge-pacing-the-race-to-5g

The Phase Noise Challenge Pacing the Race to 5G

November 14, 2019

The wireless communication industry’s move toward mmWave frequencies, driven by 5G cellular, is posing a challenge to existing oscillator technologies, particularly phase noise. New techniques and approaches may be required.

5G is pushing the state-of-the-art in virtually every area of cellular radio technology, including higher channel frequencies. To meet the most ambitious 5G goals, including peak data rates of 10 Gbps, cell edge data rates of 100 Mbps and 1 ms end-to-end latency, near-mmWave frequency bands above 20 GHz are needed in the U.S. One of the many challenges posed by near-mmWave frequencies is managing radio link noise and interference. In 1948, Claude Shannon showed that radio system capacity depends not only on the signal strength and bandwidth, but also on the radio link noise level.1

Radio link noise has two broad sources: internal and external. External noise, also termed interference, is related to the environment, while internal noise is related to the radio system’s electronic circuitry. The main interest of this article is the internal noise generated in the local oscillator (LO), i.e., the phase noise. From Shannon’s Law, it is the key limiter of radio channel capacity. Quoting James Buckwalter, a professor at the University of California Santa Barbara, “Oscillator phase noise is a silent killer in interference limited systems.”2

Oscillator phase noise is defined as the oscillator’s short-term instability resulting in random fluctuation in the frequency or phase of its output (see Figure 1). Phase noise is measured as the power spectral density for each 1 Hz frequency of a single sideband relative to the power spectral density of the oscillator’s central frequency, in dBc/Hz. A well-known empirical model for phase noise, the Leeson equation was developed to describe and predict LC tank circuit phase noise performance.3

Figure 1

Figure 1 Oscillator phase noise.

Figure 2

Figure 2 Typical QAM (I/Q) modulator.

In Leeson’s equation, F is an empirically determined constant for curve fitting, k is Boltzmann’s’ constant, T is the absolute temperature in Kelvin, Psig is the tank power dissipation, ω0 is the oscillation frequency, Q is the loaded oscillator quality factor, Δω is the offset from the oscillation frequency and ω1 is the corner frequency between the 30 dB/decade and 20 dB/decade slope regions.

EFFECT OF LO PHASE NOISE

LO phase noise performance is critical to modern radio system performance, especially to high data rate orthogonal frequency division multiplexing (OFDM) systems.4 OFDM is the data modulation technique used by most data transmission systems today, including LTE (4G cellular), Wi-Fi, cable and DSL networks. OFDM enables transmission systems to operate close to the Shannon theoretical capacity, overcoming frequency specific interference but susceptible to oscillator phase noise.

In an ideal OFDM modulator, the data stream is mixed with the ideal oscillator frequency (labeled carrier oscillator in Figure 2) to produce ideal modulated symbols (see Figure 3a). In real life, however, the LO generates the carrier frequency and additional close-in frequencies called additive phase noise. These frequencies, i.e., the carrier plus the additive phase noise, are mixed with the data to produce the modulated signal. The addition of the phase noise around the central carrier frequency produces an error in the phase angle of the resulting symbol, called its error vector magnitude (EVM)5, leading to a shift in the placement of the symbol in the constellation (see Figure 4). Too great an error obscures the symbol, making demodulation questionable or impossible, as shown in Figure 3b.

The EVM, an expression of the symbol position in the decoded constellation relative to ideal, is an important specification that cellular system equipment must meet to qualify for commercial use6 (see Table 1). Symbol position errors, measured as EVM, have multiple causes. The most important source of vector error for high data rate OFDM radio communications is LO phase noise.7 Symbol vector errors lead to inter-symbol interference, which is measured as symbol error rate. Symbol errors, which corrupt the data stream, slow the data rate by forcing some data to be resent, degrading link performance. In this way, LO phase noise degrades radio link performance.

Figure 3

Figure 3 I/Q constellation (a). If the inter-symbol interference is too high, accurate demodulation is impossible (b).

Figure 4

Figure 4 Error vector magnitude.



Phase Noise Power

Table 1

Total phase noise power is key to understanding the impact of oscillator phase noise on radio link performance. Shannon’s Law says with the noise power equal to the signal power, the channel capacity is equal to the bandwidth of the broadcast signal, e.g., 1 Mbps for a 1 MHz broadcast bandwidth. It also says that with every reduction of noise power by a factor of 10 relative to the signal power, the channel capacity goes up by a factor of 3.3. If the noise power is 10 percent of the carrier’s power, the channel capacity is 3.3x the bandwidth of the broadcast signal, e.g., 3.3 Mbps for a 1 MHz broadcast bandwidth. If the noise power is 1 percent of the carrier’s power, the channel capacity is 6.6x the bandwidth and so on.

From Equation 2, the LC tank circuit oscillator phase noise power increases with the square of the frequency. For example, comparing the phase noise of 2 and 20 GHz VCOs (see Figure 5), the total phase noise power for the 2 GHz VCO is −46 dBc, while the total phase noise power for the 20 GHz VCO is −27 dBc, 19 dBc higher. The results in the figure were calculated using a piecewise linear model assuming the phase-locked loop (PLL) is flat at −70 dBc in both cases. Considering only the phase noise power and not other interference sources, the calculated channel capacity of the system with the 2 GHz VCO and a signal bandwidth of 1 MHz is 15.3 Mbps, while the capacity of the system with the 20 GHz VCO and the same signal bandwidth is only 9 Mbps, a 41 percent reduction in channel capacity (see Figure 6). The reduced capacity is due to the significantly higher phase noise power of the 20 GHz VCO compared to the 2 GHz VCO, i.e., 19 dB or almost 80x.

Challenges at mmWave Frequencies

Figure 5

Figure 5 Phase noise and total noise power of 2 and 20 GHz VCOs.

A fundamental question for 5G and other high modulation density radio systems is what oscillator technology to use. Figure 7 illustrates today’s available technologies: SAW, VCO, dielectric resonance oscillator (DRO) and YIG. Tunability is an important feature, as is phase noise. Today’s commercially available VCOs, although tunable, suffer increased phase noise between 2 and 3 dBc per GHz, free running, based on publicly available data sheets for commercial VCOs. Phase locking with a PLL will improve noise performance, which is how VCOs are typically used in these applications. SAW oscillators have good phase noise performance, but they are not available above 6 GHz, nor are they tunable. Neither the VCOs nor SAW oscillators appear to be ideal for frequencies above 20 GHz. DROs perform better than VCOs above 20 GHz; however, they also are not tunable. The question of which oscillator technology is best for high bandwidth connections in the near-mmWave spectrum has not been answered. Significant research continues to focus on finding a solution, including exotic VCO/PLL architectures, advanced SAW oscillators, micro/chip-scale atomic clocks and YIG-tuned oscillators. As published in Microwave Journal, “The technical approach for supporting mmWave frequencies in handsets is being developed, but the technology is not as mature as for the sub-6 GHz bands.”8

Figure 6

Figure 6 Oscillator channel capacity comparison at 1 MHz signal bandwidth.

Figure 7

Figure 7 Oscillator technology phase noise vs. frequency at 100 kHz offset.

YIG Oscillators

Of the technological options to provide high quality frequency sources in the near-mmWave and mmWave frequencies, YIG may be a promising option. As background, YIG-tuned oscillators exploit the property that YIG RF permeability is variable, depending on the strength of the encompassing DC-biased magnetic field. The frequency at peak YIG RF permeability tracks the strength of the encompassing magnetic field linearly, at a rate of 2.8 MHz/gauss and is modeled as a parallel resonant inductor and capacitor (i.e., LC tank circuit), with the inductor value varying with the magnetic field strength. The peak YIG RF permeability frequency has very high Q (> 1000), resulting in low phase noise when the YIG is used as an oscillator tank circuit. While an electrically-coupled tank circuit depends on electron movement to realize the resonant frequency, a magnetically-coupled tank circuit depends on an oscillating magnetic field to couple energy in and out of the tank circuit at the resonant frequency of the tank. Instead of direct electron flow of the electrically connected tank, the magnetically connected tank operates as a transformer, inductively conducting energy between two electrically unconnected inductors.

Figure 8

Figure 8 Traditional YIG assembly.

Figure 9

Figure 9 Differential resonant ring oscillator schematic (a) and early prototype (b).

Traditional YIG-tuned oscillators use a “negative resistance” topology to create a high performance oscillator. The negative resistance topology has been proven for decades and is optimized for power, size and performance. However, traditional YIG oscillator designs have not progressed significantly in more than 20 years, since Verticom introduced permanent magnet biasing in 1997.9 In addition, with micro-formed wire loops hand-placed and hand-tuned, traditional YIG devices are inherently high cost, with manual assembly and tuning a barrier for high volume applications (see Figure 8). YIG-tuned oscillator design is ripe for innovation.

Next-Generation YIG Oscillators

Next-generation YIG-tuned oscillators use an oscillator design integrated onto a custom MMIC to couple to the YIG resonator. This design approach enables high performance, increased functionality, low power, SMT packaging and low-cost. Although it has been tried repeatedly over the years, the negative resistance topology used for traditional YIG oscillators has never been successfully integrated onto an IC. This failure is due, in part, to the relatively low RF power in the circuit, requiring a long sense wire to couple the electromagnetic energy back and forth between the YIG and the negative resistance driver circuit. Next-generation YIG-tuned oscillators using a differential resonant ring topology10,11 overcome the limitations of the negative resistance topology (see Figure 9). By controlling the RF power of the YIG-tuned circuit and controlling parasitic reactances and magnetic interference modes, high performance, low power, small size and low-cost oscillators can be produced.

Critical elements of the design of the MMIC-based YIG oscillator include: assuring a uniform magnetic bias field, efficiently coupling the RF energy into and out of the YIG sphere, achieving 360 degree loop phase shift and managing stray magnetic fields from circuit conductors. The DC magnetic bias field is based on the proven combination of permanent and tunable electromagnets to minimize power and size. RF magnetic coupling is managed within the MMIC, where power levels and impedances are tightly managed with a combination of design elements and active control.

A key difference between the negative resistance and differential resonant ring architectures is the operational mode of the YIG sphere. In the negative resistance circuit, the YIG sphere is modeled as a magnetically-coupled parallel tank circuit. In the differential resonant ring architecture, the YIG sphere is used as a filter element, where only the energy at the tuned frequency is transmitted between the loops in the ring. Naturally, the design of the oscillator must conform to the basic tenants: the loop phase shift must be an integer multiplier of 360 degrees and the loop gain must be greater than 1 at the frequencies of interest.

The differential resonant ring MMIC architecture simplifies the oscillator magnetic design, greatly reducing stray electromagnetic flux. Reducing the stray flux has eliminated the need for the mitigations necessary with traditional YIG oscillators, eliminating any need for hand tuning. A direct consequence of eliminating hand tuning is the ability to use advanced manufacturing assembly processes used for ICs and multi-chip modules. Next-generation YIG oscillators are designed to use this manufacturing flow to keep assembly costs low and yields high.

Beyond the basic improvements achieved with IC integration, next-generation YIG oscillators will offer additional features and functionality, capitalizing on Moore’s Law. For example, with the differential resonant ring architecture, doubling the frequency is much simpler compared to traditional single-ended YIG oscillators, as well as dividing the frequency. Combining a new topology with IC manufacturing processes, the next-generation YIG oscillator can evolve into microwave and mmWave synthesizers on a chip.

References

  1. C. Shannon, “A Mathematical Theory of Communication,” The Bell System Technical Journal, Vol. 27, No. 3 and 4, pp. 379-423, 623-656, July, October, 1948.
  2. J. Buckwalter, “Reciprocal Mixing: The Trouble with Oscillators,” [Online]. Available: www.ece.ucsb.edu/Faculty/rodwell/Classes/ece218c/notes/Lecture4_ReciprocalMixing.pdf. [Accessed 1 April 2019].
  3. D. B. Leeson, A Simple Model of Feedback Oscillator Noise Spectrum, Vol. 54. No. 2, Proceedings IEEE, February 1966, pp. 329-330.
  4. M. R. Khanzadi, “ Phase Noise in Communication Systems Modeling, Compensation, and Performance Analysis,” Chalmers University Technology, Göteborg, Sweden, 2015.
  5. A. Georgiadis, “Gain, Phase Imbalance, and Phase Noise Effects on Error Vector Magnitude,” IEEE Transactions on Vehicular Technology, Vol. 53, No. 2, pp. 443-449, March 2004.
  6. European Telecommunications Standards Institute (ETSI), “5G,” European Telecommunications Standards Institute (ETSI), [Online]. Available: https://www.etsi.org/technologies/5g. [Accessed 3 April 2019].
  7. A. G. Armada, “Understanding the Effects of Phase Noise in Orthogonal Frequency Division Mulitplexing (OFDM),” IEEE Transactions on Broadcasting, Vol. 47, No. 2, pp. 153-159, 2001.
  8. B. Thomas, “5G Brings New RF Challenges for Handsets,” Microwave Journal, October 9, 2018. [Online]. Available: www.microwavejournal.com/articles/print/31177-g-brings-new-rf-challenges-for-handsets. [Accessed 21 3 2019].
  9. Verticom, Inc. (1997). Microwave Ferrite Resonator with Parallel Permanent Magnet Bias. 5,677,652.
  10. R. Parrott and A. A. Sweet, “A Wide band, Low Phase Noise Differential YIG Tuned Oscillator,” WAMICON, pp. 1-3, June 2014.
  11. M. van Delden, N. Pohl, K. Aufinger, C. Vaer and T. Musch, “A Low-Noise Transmission-Type Yttrium Iron Garnet Tuned Oscillator based on a SiGe MMIC and Bond-Coupling Opertating up to 48 GHz,” IEEE Transactions on Microwace Theory and Techniques, Accepted but not yet published, 2019.