Differential clock and timing devices are commonly characterized for phase noise using baluns. While deceptively simple to use, baluns perform a fairly complicated process that can unknowingly introduce artifacts into measured results. This article describes such artifacts, discusses why they appear and how to eliminate them. Recommendations are provided for selecting a balun and using it to accurately characterize devices for phase noise.

Differential clock signals are widely used for a broad range of applications spanning data communications, wireless, instrumentation and medical. Differential signaling uses a pair of conductors; ideally, each carries a signal of the same magnitude but opposite phase. Examples include LVPECL, LVDS and CML. Compared with single-ended signals, differential signals have lower voltage swings on each conductor, which enables them to operate at higher frequencies. A composite (differential) swing, however, can be larger than a single-ended swing with the same power supply, which increases its signal-to-noise ratio.

Figure 2

Figure 2 Measurement setups for oscilloscope (a) combiner balun (b) splitter balun (c) and single-ended phase noise (d) with the DUT termination used for LVPECL (e) and LVDS (f) outputs.

Differential signals also perform better in noisy environments by rejecting common-mode noise. In addition, they provide more precise timing, since the crossover location for a differential signal is easier to control than for a single-ended signal (which depends on a voltage crossing an absolute reference level).1

Phase noise quantifies the short-term phase fluctuations in a signal2 and is arguably the most important parameter for evaluating clock and timing devices used in critical timing applications. Phase noise (along with amplitude noise) can be measured using a spectrum analyzer or a dedicated phase noise analyzer. These instruments, however, analyze only single-ended signals. To convert a device’s differential signal into a single-ended signal, an active probe, differential-to-single-ended amplifier or passive balun is required. For low noise measurements, baluns are preferred, since they do not add amplifier noise to the measurement. Broadband baluns are particularly attractive because a single device can support measurements over a wide range of frequencies.

This article explores how to use baluns for characterizing phase noise in differential clock signals. It begins by discussing artifacts that a balun can introduce into the measurement data. Example test results are shown to illustrate and are not intended to represent typical or worst-case scenarios. In practice, whether or not a balun impairs phase noise data, and by how much, is difficult to predict. This article analyzes the factors involved, such as the choice of balun, the device being measured and the cabling and components connecting the device to the balun. Experimental techniques are presented to determine if the balun is impacting phase noise measurements. Finally, recommendations are presented for how to select a balun and use it to accurately characterize devices for phase noise. To the authors’ knowledge, this is the first published article studying the impact of baluns on phase noise measurements.


Figure 3

Figure 3 156.25 MHz LVPECL (left) and 312.5 MHz LVDS (right) crystal oscillator waveforms using baluns with (bottom) and without (top) isolation.

Figure 1 illustrates the role of a balun in converting a balanced impedance (or differential signal) to an unbalanced impedance (or single-ended signal). The balun itself is deceptively easy to use, requiring only three connections (two inputs and one output) and no power. Being a reciprocal device, a balun can be driven in either direction. A balun that converts a signal from single-ended to differential is called a splitter. When operated in reverse, it is called a combiner. In normal operation, the differential ports J2 and J3 ideally provide equal and opposite signals. The unbalanced port J1 is matched to the transmission line impedance, typically 50 Ω.

A balun’s performance can be summarized with a few key metrics.3 Amplitude balance (in dB) is the differential insertion loss from the unbalanced port to one differential port versus the other. Phase balance (in degrees) is the differential phase shift on the differential ports. Insertion loss (in dB) is the additional loss in signal power — beyond the nominal loss caused by splitting the signal — introduced by inserting the balun in the signal path. Isolation (dB) is the ratio of signal power entering one differential port (e.g., J2) that appears at the other differential port (e.g., J3). Return loss (in dB) or voltage standing wave ratio (VSWR) represents how well the device is matched to a specific load and source impedance, typically 50 Ω. Finally, common mode rejection ratio (CMRR) (in dB) is the ratio of common-mode to differential-mode gains, and reflects how well the balun attenuates common-mode signals passing from balanced to unbalanced ports. CMRR can be calculated from the amplitude and phase balance, based on vector cancellation.


Test data is measured using one of four basic setups as shown in Figures 2a through d. Clock signals are analyzed for signal integrity using a high speed real-time oscilloscope, while phase noise is measured with a signal source analyzer.4 While clock devices from many manufacturers were analyzed for this study, two devices under test (DUT) are presented here to illustrate key findings. Both DUTs are commercially available 5 mm × 7 mm surface-mount crystal oscillators (XO). The first DUT is a 156.25 MHz LVPECL XO based on analog multiplication. The second DUT is an LVDS XO whose output frequency can be changed by modifying an internal phase-locked loop (PPL) feedback divider to provide 78.125 MHz or 312.5 MHz. The termination for each DUT is shown in Figures 2e and f, which is appropriate to drive 50 Ω test equipment. Unless otherwise noted, Figure 2 setups use 0.1 µF AC coupling capacitors, and 0 Ω series termination resistors.

Figure 4

Figure 4 Operating a balun as a splitter results in a cleaner differential signal than operating it as a combiner.

Some setups use connectorized baluns with or without fixed coaxial attenuators (i.e., padding). While baluns from several manufacturers were analyzed for this study, two broadband baluns are presented here to illustrate key findings. Both baluns are from the Marki Microwave test and measurement product line,5 namely BAL0006 (200 kHz to 6 GHz) and BAL0036 (300 kHz to 36 GHz). They were selected for their different levels of isolation. For the DUT frequencies analyzed, BAL0006 has a standard isolation of 6 dB; BAL0036 has an improved isolation of 10 dB, much more at higher frequencies. For clarity, this article refers to BAL0036 and BAL0006 as the baluns “with” and “without” (added) isolation, respectively.

Since a phase noise analyzer’s RF input port can only accept AC signals, a DC block must be inserted somewhere in the signal path between the DUT and the instrument. In general, DC can be blocked at either side of the balun. However, if the balun has its ports DC shorted to ground (refer to its datasheet), then DC blocks must be placed at the balun’s input ports when it is used as a combiner. For this reason, it’s probably best to develop a habit of placing DC blocks at the balun’s input (differential) ports, as shown in Figure 2.

Figure 5

Figure 5 A balun with poor isolation (a) allows signal leakage inside the balun, which can be reduced by increasing the isolation (b) or adding external attenuation (c). Baluns operating as splitters (d) don’t have a leakage signal, resulting in a cleaner output waveform.


Oscilloscopes typically have more than one input, so a balun is not required to perform measurements. Nevertheless, viewing a balun’s output signal in the time domain can provide insight into its operation. Figure 3 shows waveforms for two different XOs and two different baluns. The left and right sides of Figure 3 correspond to LVPECL 156.25 MHz and LVDS 312.5 MHz waveforms, respectively. The bottom and top correspond to baluns with and without (added) isolation, respectively. The balun without isolation provides a noisier waveform, which external attenuation tends to clean up. For reference, each plot contains a “no balun” curve, which was measured using two oscilloscope channels (shown in Figure 2a), where one channel’s data was subtracted from the other’s to compute the differential signal.

The balun’s insertion loss is apparent as all signals obtained with baluns swing less than the “no balun” reference waveform. The balun without isolation is observed to degrade both the LVPECL and LVDS waveforms’ signal integrity. The bumps in the logic levels of the waveforms are indicative of signal distortion caused by feedback from the balun differential ports back to the DUT. By comparison, the balun with isolation provides significantly cleaner waveforms. Inserting external attenuators at the balun’s differential ports (shown in Figure 2b) improves the signal integrity of the waveform in proportion to the level of attenuation. In this example, 9 dB of external attenuation is required at the balun without isolation’s inputs to recover the “no balun” waveform shape (that is, normalizing each curve by their peak amplitude produces overlapping curves).

Figure 6

Figure 6 LVPECL 156.25 MHz XO phase noise measured using baluns without (a) and with (b) added isolation, showing the effect of external attenuation.

Interestingly, Figure 4 shows that operating a balun as a splitter provides much cleaner waveforms compared to operating it as a combiner. The signals shown in Figure 4 were measured as shown in Figures 2a and c using the balun without isolation, the LVPECL XO and no external attenuation.

The above signal integrity impairments can largely be attributed to limited isolation in the balun. Figure 5a illustrates that a balun without isolation has appreciable signal leakage between its differential ports. The leakage signal from one differential port interferes with the forward signal at the other differential port. The leakage signal also appears at the DUT output drivers, which, depending on the driver architecture, can affect its operation.

The balun-with-isolation waveforms shown in Figure 3 have better signal integrity because the additional isolation inside the balun attenuates this leakage current (shown in Figure 5b). Adding external attenuation to the balun without isolation, as shown in Figure 5c, doesn’t prevent leakage between differential ports, but the signal that does leak is attenuated compared to when no external attenuation is used. Additionally, external attenuation reduces the leakage signal appearing at the DUT output driver. This leakage signal is actually attenuated twice (once for each attenuator) as it travels from one output driver through the balun to the other output driver. Judging by the balun-without- isolation curves for “balun” and “balun + 9 dB attenuation” in Figure 3, whose shapes match well (after normalizing them by their peak amplitude), the effect of the leakage signal appearing at the DUT output driver circuitry is a major source of noise in the balun’s output signal.

Finally, a balun operated as a splitter as shown in Figures 4 and 5d provides cleaner waveforms than when used as a combiner, because the DUT output driver does not observe a leakage signal from the balun.


Phase noise is measured using the combiner balun setup in Figure 2b. Phase noise is a measure of phase variation in the frequency domain. The phase noise data can be integrated to obtain a phase jitter value in seconds RMS. The region of the phase noise curve that dominates this integral is located by lowering a -10 dB/decade line to where it first intersects the curve.6

Figure 7

Figure 7 LVPECL 156.25 MHz XO waveform and phase noise, measured using the balun without isolation and 6" (a) and 18" (b) coaxial cables.

Figure 6a shows how external attenuation can dramatically change the measured phase noise for the LVPECL XO. With no attenuation, the balun without isolation phase noise measurement provides overly optimistic and pessimistic results below and above, respectively, about 600 kHz offset frequency. Adding 3 dB of attenuation significantly reduces balun artifacts in the measured phase noise. As more attenuation is added, the level of improvement eventually diminishes. Phase noise measured with 6 dB of attenuation (not shown) overlaps the 9 dB attenuation curve.

The balun with isolation phase noise shown in Figure 6b is independent of the level of external attenuation, indicating the added isolation inside the balun eliminates a significant source of balun-induced artifacts in the measurement. Therefore, adding external attenuation to the balun without isolation provides a similar benefit as isolation in the balun with isolation.

One disadvantage of adding external attenuation to remove balun artifacts from phase noise data is that it reduces the signal power entering the phase noise analyzer, which can lead to less accurate data. The Keysight phase noise analyzer used for these measurements has diode-based phase detectors inside its PLLs, which must be biased with current. An input signal power of 0 to 5 dBm is recommended for this reason. Adding external attenuation essentially pushes the signal further into the noise floor of the instrument. Enabling cross correlation in the measurement can help recover the signal; however, cross correlation increases test time and, depending on how deep the signal lies below the instrument’s noise floor, may not always help. The influence of the instrument’s noise floor on the measured phase noise data can be seen in Figure 6, in which the 9 dB external attenuation curves have more noise at the lowest phase noise levels (i.e., above 2 MHz offset frequencies) compared to curves with less external attenuation.

It is therefore important to use baluns with high isolation between differential ports. If external attenuation is required, use the least amount required to stabilize the data. The optimum attenuation can be determined by increasing attenuation by small increments until the phase noise data no longer changes. Then, select the smallest amount of attenuation that produces this data. In Figure 6a, the optimum attenuation is 6 dB (not shown). In Figure 6b, no external attenuation is needed.

Figure 8

Figure 8 Phase noise for an LVDS 312.5 MHz XO (a), from which spurs are detected at 39 and 78 MHz (b) and their respective magnitudes measured (c) and (d).

In addition to signal impairments caused by poor port-to-port isolation in the balun, reflections can also occur at interfaces that are not terminated at the characteristic impedance of the transmission line (typically 50 Ω). These reflections are synchronous with the forward wave, which can combine to form a standing wave. Here, the level of voltage (and current) at both DUT and balun ends of the cable is a function of cable length, which can affect the operation of the DUT and/or balun. The VSWR metric measures the ratio of maximum standing wave amplitude to minimum standing wave amplitude. A perfectly terminated component would have a VSWR of 1, indicating the voltage (and current) at any location in the cable remains constant. In practice, components have a VSWR greater than one. The impedance a DUT driver observes looking towards the balun is therefore a function of the cable length connecting the DUT to the balun. Figure 7 shows how changing this cable length can affect signal integrity and phase noise characteristics.

In theory, the transmission line effects discussed become more noticeable at longer cable lengths. In short cables, where the delay from DUT to balun is less than the signal’s transition time, the reflections settle before they are noticed. In the frequency domain, longer cables result in more of a phase variation as the frequency changes. In the time domain, longer cables result in a longer delay between when reflections occur, and more reflections contribute to standing waves and interference effects. These effects can be minimized by using baluns with excellent return loss (preventing the initial reflection) and when measuring a device with good return loss (preventing subsequent reflections).


Although a phase noise analyzer measures raw phase noise in dBc/Hz, it can post process this data to detect spurious phase noise in dBc. Phase noise data in dBc/Hz may be plotted along with the spurious data in dBc, where the spurious data is plotted using a different color to distinguish its change in units (since both share the y-axis scale). Figures 8a and b illustrate this process for two spurs in the 312.5 MHz LVDS XO.

Figures 8c and d quantify the spur magnitudes using horizontal lines for the single-ended signals, as measured for the configuration of Figure 2d. Single-ended spur magnitudes are shown using horizontal lines and differential spur magnitudes are shown using bars as a function of attenuation. The lines correspond to OUT+ and OUT−. For this DUT, the spurious levels between outputs are different. The bars in Figures 8c and d indicate the spurious magnitudes measured using the balun without isolation as shown in Figure 2b. Here, increasing the level of external attenuation reduces the spurious magnitude to roughly the mean dBc value of the single-ended (i.e., OUT+ and OUT−) spur magnitudes.

Figure 9

Figure 9 DUT phase noise vs. output driver resistance and attenuation for a 156.25 MHz LVPECL XO (a) and 78.125 MHz LVDS XO (b) using a balun without isolation.


To further analyze the effect that component reflections and balun isolation have on measured phase noise data, the DUT outputs are matched to their transmission lines using a series termination and re-characterized with the balun without isolation. Specifically, the LVPECL XO output impedance measured 35 Ω at 156 MHz, so a value of 15 Ω is used for Rs in Figure 2e. Likewise, the LVDS XO output impedance measured 3 and 13 Ω at 78 and 312 MHz, respectively, so 47 and 37 Ω are used, respectively, for Rs in Figure 2f. In each case, the resulting phase noise data is more accurate using a series termination.

Figure 9 summarizes the test results for two cases. The red curve is the original phase noise data, measured using the balun without isolation with no series termination and no external attenuation. The green curve uses a similar setup as the red curve, but includes a series termination. The blue curve uses a similar setup as the green curve, but adds more external attenuation than needed to stabilize the phase noise curve (e.g., 3 dB less attenuation produces the same blue curve data shown in Figure 9). The blue curve is also the same data as measured using the balun with isolation, and therefore represents the most accurate phase noise data for the device. A series termination (green curve) improves the phase noise measurement.

Since neither the balun, the cables, nor the DUT output circuit are perfectly matched to 50 Ω or to each other, reflections occur that can lead to standing waves and resonances. These reflections, together with any leakage signal resulting from poor balun isolation, form a reverse signal that travels from balun to DUT. If the DUT has poor isolation between its output buffer and its internal VCO, oscillator circuitry or other components, the phase noise output by the DUT can be affected. Matching the DUT to the transmission line impedance using a series termination absorbs the reverse signal, preventing it from traveling back and forth between DUT and balun. A series termination is observed to have a similar effect on phase noise as adding external attenuation (compare Figure 9a with Figure 6a). Introducing attenuation between the DUT and balun effectively improves the load return loss by twice the attenuation value.

Although a differential clock output buffer is designed to drive a 50 Ω load, it doesn’t typically have a 50 Ω output impedance. This isn’t a problem when connecting devices to test equipment terminated with 50 Ω, but becomes an issue when components are used with non-ideal loads. While it may not be practical to series terminate devices for the purpose of routinely measuring phase noise, reflections can be minimized by selecting components with higher return loss (lower VSWR) values. In addition, the impact of standing waves from reflections or poor isolation can be reduced by using the shortest possible cables (between the DUT and balun to prevent resonances).


A reverse leakage signal can travel from balun to device caused by poor balun isolation and reflections from components having non-ideal loads. Since the device’s output impedance itself is not matched to the transmission line, this reverse signal again reflects at the device output buffer and travels back and forth between device and balun. This results in a standing wave that can resonate and potentially affect device operation. If the device has poor isolation between its output buffer and internal VCO, oscillator, and/or other components, then the phase noise generated by the device may also change.

The effect that a given balun will have on measuring phase noise for a particular device is difficult to predict. Balun impairments may be observed rarely or often, depending on many complex factors. These impairments can cause the phase noise data to appear better or worse than observed by an actual system. The following is a prioritized list of recommendations to minimize measurement artifacts:

  1. Select a balun primarily for high isolation (balanced port to balanced port) and high return loss. The balun should also have high common-mode rejection and good amplitude and phase balance. Everything else being equal, select a balun with low insertion loss.
  2. Use short, phase-matched cables between the device and balun.
  3. Use the smallest amount of external attenuation between device and balun necessary for the measured phase noise data to remain unchanged compared to measurements with larger attenuation.
  4. Place DC blocks at the balun’s differential ports if the balun’s ports are DC shorted to ground.

From a phase noise measurement perspective, the balun market may be divided into generic and high performance products. Generic baluns generally have ±1 dB amplitude balance, ±10º phase balance, 6 dB isolation, 10 dB return loss and 20 dB CMRR — or worse. High performance baluns generally have ±0.5 dB amplitude balance, ±5º phase balance, 15 dB isolation, 15 dB return loss and 25 dB CMRR — or better. While a few high performance baluns include isolation, the term “180º hybrid combiner/divider” is often used to describe a balun with high isolation. Regardless of the terminology, baluns used to measure phase noise should target the high performance specifications above, especially isolation and return loss. Phase noise measurements will also benefit from low insertion loss baluns, which can vary from 3.5 to 6.5 dB, depending on architecture.


The authors wish to acknowledge useful discussions with Dr. Bob Temple, formerly with Agilent Technologies; Tony Wade, with Keysight Technologies; Dan Nehring, with CTS Corp.; Stuart Rumley, with Valon Technology; and Pierre Guebels and Boris Drakhlis, with Microchip Technology.


  1. D. Brooks, “Differential Signals: Rules to Live By,” Printed Circuit Design, CMP Media Inc., October 2001, https://www.ieee.li/pdf/essay/differential_signals.pdf.
  2. T. H. Lee and A. Hajimiri, “Oscillator Phase Noise: A Tutorial,” IEEE Journal of Solid-State Circuits, Vol. 35, No. 3, March 2000, pp. 326–336.
  3. D. Jorgesen and C. Marki, “A Tutorial on Baluns, Balun Transformers, Magic-Ts, and 180° Hybrids,” Marki Microwave, Application Note, 2014,  https://www.markimicrowave.com/Assets/appnotes/balun_basics_primer.pdf.
  4. “E5052B Signal Source Analyzer,” Keysight Technologies, Product Brochure 5989-6389EN, 2009.
  5. “Product Catalog for Broadband Test and Measurement Baluns,” Marki Microwave, www.markimicrowave.com/3595/Baluns_/_Inverters.aspx?ShowTab=32.
  6. G. Giust, “Determine the Dominant Source of Phase Noise By Inspection,” JitterLabs, NOTE-4, Technical Note, 2016, www.jitterlabs.com/support/publications.