VIDEO SCRIPT: Phase Noise Measurement Techniques for RF Engineers

---

[OPENING — 0:00–0:45]

Every RF engineer has seen it — a pristine sine wave on paper, perfectly periodic, perfectly predictable. But in the real world, no oscillator is ideal. Every signal source carries a ghost — a subtle, random fluctuation in the timing of its zero crossings. That ghost has a name. It's called phase noise. [BEAT]

Phase noise is one of the most critical specifications in modern RF and microwave system design. It limits the sensitivity of receivers, degrades the performance of radar systems, corrupts the constellation diagrams of digital communications, and introduces reciprocal mixing that can mask weak signals entirely. [GRAPHIC: System block diagram showing a local oscillator feeding a mixer — phase noise spreading onto the IF output.]

If you're an RF engineer designing oscillators, synthesizers, or frequency converters, you don't just want to measure phase noise — you need to measure phase noise. And you need to measure it accurately, repeatably, and across a wide range of offset frequencies.

Today, we're going to walk through the three principal techniques used to measure phase noise: the spectrum analyzer method, the phase detector method, and the cross-correlation method. We'll discuss how each works, what their strengths and limitations are, and where that famous specification — minus one hundred dBc per Hertz at ten kilohertz offset — fits into the picture.

Let's get started.

---

[SECTION 1: WHAT IS PHASE NOISE? — 0:45–2:30]

Before we discuss measurement techniques, let's make sure we're grounded in the fundamentals.

An ideal oscillator produces a signal of the form: V of t equals A times cosine of two pi f-zero t.

That's a pure tone — a single spectral line at frequency f-zero. In the real world, however, random noise processes — thermal noise, flicker noise, shot noise — perturb both the amplitude and the phase of the signal. The result is: V of t equals A of t times cosine of two pi f-zero t plus phi of t.

The term phi of t represents the random phase fluctuation. When we analyze the spectral consequences of this fluctuation, we get what's called single-sideband phase noise, denoted script L of f, and measured in decibels relative to the carrier per Hertz of bandwidth — written as dBc/Hz. [GRAPHIC: Annotated phase noise plot. Carrier at center. Noise skirts rise from right. Offset frequency axis labeled. Y-axis in dBc/Hz. A marker at 10 kHz offset reading −100 dBc/Hz.]

On a phase noise plot, the x-axis is the offset frequency — the distance in Hertz from the carrier — and the y-axis is the power spectral density of the phase noise in dBc/Hz. A good oscillator will have a plot that drops as you move away from the carrier, ideally reaching a noise floor set by the measurement system itself.

When someone specifies a phase noise performance of negative one hundred dBc per Hertz at ten kilohertz offset, they're saying: at a frequency separation of ten kilohertz from the carrier, the noise power in a one-Hertz bandwidth is one hundred decibels below the carrier power. For many communication and radar oscillators, this is a meaningful — and sometimes demanding — benchmark.

Now let's talk about how we actually measure it.

---

[SECTION 2: THE SPECTRUM ANALYZER METHOD — 2:30–4:45]

The most intuitive approach to measuring phase noise is to simply look at the signal on a spectrum analyzer. You connect your device under test — your DUT — to the analyzer, set the resolution bandwidth and video bandwidth appropriately, and observe the noise sidebands around the carrier. [GRAPHIC: Spectrum analyzer display showing a carrier with visible noise skirts. Cursor measuring noise at 10 kHz offset.]

To convert the displayed noise level to script L of f in dBc/Hz, you apply a normalization correction: Script L of f equals P-noise minus P-carrier minus ten log base ten of the resolution bandwidth, plus a correction factor for the detector and bandwidth shape.

This correction accounts for the fact that the analyzer measures noise in a bandwidth determined by the resolution bandwidth filter, and we need to normalize that to a one-Hertz equivalent noise bandwidth. [BEAT]

Now, the spectrum analyzer method has clear advantages. It's simple. It requires no special hardware beyond the analyzer itself. It provides a quick visual picture of the noise profile, and it works across a wide range of offset frequencies.

But it also has significant limitations. [GRAPHIC: Comparison chart — Spectrum Analyzer vs. other methods. Highlight limitations.] First, the analyzer's own local oscillator contributes phase noise to the measurement. You're not measuring just the DUT — you're measuring the sum of the DUT noise and the analyzer's internal oscillator noise. If the analyzer's LO is noisier than the DUT, your measurement is meaningless. Second, spectrum analyzers have limited dynamic range near the carrier. The shape factor of the resolution bandwidth filter — typically Gaussian — means that the skirts of the carrier itself can mask the true phase noise at small offset frequencies. Third, the method is inherently insensitive. A typical spectrum analyzer might achieve a measurement floor of around negative 100 to negative 110 dBc/Hz at offsets of ten kilohertz and beyond. That's fine for many applications, but if you're trying to characterize a low-noise synthesizer that needs to meet a spec of negative one hundred dBc per Hertz at ten kilohertz, you need to be confident that you're actually measuring the DUT — not the analyzer's noise floor. [BEAT]

For quick, first-order characterization, the spectrum analyzer method is invaluable. But for precision measurement — especially of high-performance sources — we need something better. We need to remove the analyzer's noise from the equation. That brings us to the phase detector method.

---

[SECTION 3: THE PHASE DETECTOR METHOD — 4:45–7:30]

The phase detector method — sometimes called the PLL method or the homodyne method — is the foundation of precision phase noise measurement. It's the principle behind dedicated phase noise analyzers from manufacturers like Keysight, Rohde & Schwarz, and Microsemi.

Here's the basic idea.

You take the DUT signal and a reference signal — from a second, independent oscillator — and you feed both into a double-balanced mixer operating as a phase detector. You adjust the two signals to be at the same frequency and set in quadrature — that is, ninety degrees apart in phase. [GRAPHIC: Phasor diagram showing two signals in quadrature. Phase fluctuation converts to amplitude fluctuation at the mixer IF port.]

When the two signals are in quadrature, any small phase deviation of either signal produces a proportional voltage change at the mixer's output. The mixer acts as an analog multiplier, and at quadrature, its output is directly proportional to the phase difference between the two inputs.

That output is a slowly varying baseband signal — it contains the beat note of the phase fluctuations. You then feed this signal into a low-frequency spectrum analyzer or an FFT-based baseband analyzer, and the result is the phase noise spectral density of the two sources combined. [GRAPHIC: Equation — S-phi of f equals V-noise squared divided by K-d squared, where K-d is the mixer phase sensitivity in volts per radian.]

To maintain quadrature over time — because oscillators drift — a slow feedback loop (a PLL with very low bandwidth, typically below one hertz) is used to lock the phase relationship. This loop must be slow enough that it doesn't corrupt the phase noise measurement at the offset frequencies of interest. [BEAT]

The big advantage of the phase detector method is sensitivity. Because the carrier is suppressed — the mixer outputs zero DC when in perfect quadrature — the baseband analyzer only sees the noise. There's no large carrier signal to saturate the input or limit dynamic range. This allows you to measure phase noise far below what a spectrum analyzer could see — routinely down to negative 160 to negative 170 dBc/Hz at offset frequencies from one hertz to one megahertz.

But there's a critical limitation. [GRAPHIC: Highlighted box — "Reference source must be quieter than the DUT."]

The measurement gives you the combined phase noise of the DUT and the reference. Mathematically, it's the power sum of the two. If the reference oscillator is significantly quieter than the DUT, you can approximate the result as the DUT's noise alone. But what if the reference is not quiet enough? What if it's comparable to — or worse than — the DUT? Then you can't separate the two contributions with a single measurement.

This is where the cross-correlation technique comes in — and it's a genuinely elegant solution.

---

[SECTION 4: THE CROSS-CORRELATION METHOD — 7:30–10:15]

The cross-correlation method is the gold standard for precision phase noise measurement. It was pioneered in the 1980s and 1990s and is now implemented in modern instruments from Keysight, Rohde & Schwarz, and others.

Here's the concept.

Instead of using one phase detector with one reference, you use two identical measurement channels, each with its own independent reference oscillator and its own phase detector. Both channels are fed by the same DUT signal, split via a power divider. [GRAPHIC: Signal flow — DUT → splitter → Channel 1 (Ref 1 + Mixer 1 + FFT 1) and Channel 2 (Ref 2 + Mixer 2 + FFT 2) → Cross-correlation processor.]

Each channel independently measures the phase noise of the DUT combined with its own reference. The key insight is this:

The DUT noise is common to both channels — it's correlated. The reference noise in each channel is independent — it's uncorrelated. The measurement system noise — from amplifiers, digitizers, and so on — is also uncorrelated between channels.

When you compute the cross-spectral density — essentially the cross-correlation in the frequency domain — between the two channels, the uncorrelated noise components average toward zero. The correlated component — which is the DUT's true phase noise — survives. [GRAPHIC: Mathematical illustration. Show noise floor dropping as number of correlations increases. Plot showing improvement from N=1 to N=100 to N=10,000. Noise floor drops as ten times the log of N.]

The more correlations you compute, the deeper the measurement. Specifically, the noise floor improves by five times the log base ten of N decibels, where N is the number of correlations. So after one hundred correlations, you gain ten dB. After ten thousand, you gain twenty dB. After a million, thirty dB.

This is extraordinary. It means you can use noisy reference oscillators — even references that are worse than the DUT — and still extract the DUT's true phase noise, given enough averaging time. [BEAT]

Let's tie this back to our specification of interest. [GRAPHIC: Phase noise plot. Marker at 10 kHz offset showing −100 dBc/Hz. Dashed line showing analyzer noise floor at −155 dBc/Hz after cross-correlation.]

Suppose you need to verify that an oscillator meets a spec of negative one hundred dBc per Hertz at ten kilohertz offset. Using the spectrum analyzer method, you might hit a measurement floor of around negative one hundred and ten to negative one hundred and fifteen dBc/Hz — leaving you only ten to fifteen dB of measurement margin. That's tight. You can't be confident the number is accurate.

With a single-channel phase detector setup, if your reference is clean enough, you might reach a floor of negative one hundred and forty dBc/Hz — giving you forty dB of margin. That's much better.

With the cross-correlation method, after sufficient averaging, you can push the effective measurement floor to negative one hundred and sixty dBc/Hz or beyond — giving you sixty decibels of margin below the specification. At that point, you can trust your measurement with high confidence, and you can also characterize the close-in noise — at offsets of one hertz, ten hertz, and one hundred hertz — where the phase noise is typically much higher and where system performance is most vulnerable. [BEAT]

The trade-off, of course, is time. Cross-correlation requires many averages to achieve deep noise floors, and the measurement can take minutes to hours depending on the offset frequency range and the desired sensitivity. But for characterization, qualification, and production testing of high-performance oscillators, it's the method of choice.

---

[SECTION 5: COMPARISON AND PRACTICAL GUIDANCE — 10:15–11:30]

Let's summarize the three methods side by side. [TABLE ANIMATES IN:]

In practice, many engineers use all three methods at different stages. A spectrum analyzer for a quick sanity check during prototyping. A phase detector setup for production verification. And a cross-correlation analyzer for full characterization, especially when the target specification is demanding — like that negative one hundred dBc per Hertz at ten kilohertz spec.

---

[CLOSING — 11:30–12:00]

Phase noise is invisible, but its effects are everywhere — in the sensitivity of your receiver, the clarity of your signal, the accuracy of your radar. Understanding how to measure it correctly is not optional for the RF engineer. It's fundamental.

Whether you're grabbing a quick number with a spectrum analyzer, setting up a PLL-based phase detector, or running a long cross-correlation measurement overnight, the goal is the same: to see the truth about your signal — jitter, sidebands, and all.

Thanks for watching. If you found this useful, subscribe for more deep dives into RF test and measurement fundamentals. Until next time. [END CARD — 12:00]

--- [FADE TO BLACK] Total Word Count: ~1,520